N-glycosylation blocks and simultaneously fosters different receptor-ligand binding sites: the chameleonic CD44–hyaluronan interaction

While protein function is largely encoded in DNA, it is also modulated by a complementary and often invisible layer of information encoded by glycans attached to the protein surface. The CD44–hyaluronan complex involved in inflammatory responses and cell migration is a prime example of the significance of glycans, as the ability of the cell surface receptor CD44 to bind its ligand, hyaluronan, is modulated by N-glycosylation. Intriguingly, how glycans can regulate this binding process, and the activation of CD44, has remained unclear. In this work, based on atomistic simulations and NMR, we provide evidence that CD44 has multiple distinct binding sites for hyaluronan, and that N-glycosylation modulates their respective roles. While non-glycosylated CD44 is found to favor the canonical submicromolar binding site, glycosylated CD44 binds hyaluronan with an entirely different micromolar binding site. Our findings show for the first time how glycosylation can alter receptor affinity by shielding specific regions of the host protein, thereby promoting weaker binding modes. The mechanism revealed in this work emphasizes the importance of glycosylation in protein function and constitutes a challenge for protein structure determination where glycosylation is typically not observed.


Introduction
Glycosylation is a fundamental process where proteins are linked to complex oligosaccharides, glycans (1). Most of the proteins at the extracellular side of eukaryotic cells contain covalently linked glycans (2). Their structural roles include the mediation of interactions with the surrounding environment (3), facilitation of correct folding (4)(5)(6), and involvement in the assembly of membrane proteins (7), also by direct interaction with lipids (8). Glycans are also known to modulate the binding of ligands with several proteins, e.g., by masking the binding site (9)(10)(11). Such regulation is relevant, especially in most immune processes, such as activation and homing, guided by regulated remodeling of the glycans (12). However, the details of these modulation mechanisms are often poorly understood due to the glycans' structural flexibility and dynamic nature (13,14). The transmembrane protein called CD44 is a key exam-ple of glycoproteins, whose functions are modulated by Nglycosylation (9,(15)(16)(17)(18)(19). Its primary task is to serve as a receptor for a carbohydrate polymer, hyaluronic acid (hyaluronan (HA)) (20,21). This ligand-protein interaction mediates a variety of physiological processes such as white blood cell homing, healing of injuries, embryonic development, and controlled cell death (22). Recently, the CD44-HA interaction has also been utilized in the design of functional biomaterials (23). CD44 binds HA exclusively via its lectin-like hyaluronate binding domain (HABD). In the canonical form, CD44 is a 722 residue-long type I transmembrane protein from which HABD comprises the first 150 amino acids  after the signal peptide (24, 25). Notably, human CD44-HABD contains five possible N-glycosylation sites (N25, N57, N100, N110, and N120) (24) (see Fig. 1) that are known to be occupied by highly branched N-glycans, especially in various cancer cell lines (18,19,26). The N-glycans elicit a dual effect on HA binding: while some glycan content favors the recognition of HA, the presence of negatively-charged sialic acids generally interferes or even blocks it (16,27,28). However, the molecular mechanisms underlying such a dual effect remain unclear. In fact, most of the currently available structural data of HA-CD44 complexes are derived from non-glycosylated constructs (24, 25, 29, 30), leaving the structural details of fully N-glycosylated HABD elusive.
A shallow groove on the surface of the HABD forms the canonical binding site for HA. There the residue R41 stabilizes the binding in a pincer-like fashion (25, 31,32). In addition to this so-called crystallographic binding mode (blue chain in Fig. 1), in our previous work, we postulated the existence of two potential lower-affinity binding modes called parallel (green chain in Fig. 1) and upright (red chain in Fig. 1) modes (14). These modes occupy the same general face of CD44-HABD, sharing to a large extent the R41-containing binding epitope. Additionally, each of these modes occupies a second arginine residue that is distinct from that of the other binding poses (14). As a result, each mode covers a unique region of the CD44-HABD surface. Such separation of the binding sites allows their selective silencing via antibodies that target different regions of CD44 (33). It also implies that the presence of N-glycosylations will affect each of the binding modes differently. This idea is the central hypothesis of this work.
In this study, we employed atomistic molecular dynamics (MD) simulations to unravel how complex N-glycans at N25, N100, and N110 cooperatively cover the canonical binding groove of CD44-HABD. This sugar shield hinders the accessibility and ligand availability of the canonical binding groove significantly, thereby promoting the secondary upright HA-CD44 binding mode over the crystallographic binding site. We then used NMR complemented by atomistic MD simulations to show that a few HA oligomers can bind CD44-HABD simultaneously at distinct binding sites. The observed binding sites correspond to the previously characterized crystallographic (25), parallel, and upright binding modes (14). We further reveal that anti-CD44 antibody MEM-85 does not cross-block the canonical HA binding site in non-glycosylated CD44. Instead, it blocks HA binding to glycosylated CD44 (34,35). These findings provide compelling evidence for the existence of a lower-affinity upright binding mode for HA. This binding mode overlaps with the binding site of MEM-85 and is promoted by N-glycosylation. The results demonstrate the existence of a new mechanism to control the ligand binding affinity of receptor proteins by promoting alternate binding sites by N-glycosylation.

Materials and Methods
A. Simulation system construction and models. We generated computational simulation models of glycosylated CD44-HABD. As the primary oligosaccharides, we employed fucosylated complex-type triantennary N-glycans, containing zero (asialo) or one (monosialo) terminal sialic acids per antenna, i.e., non-reducing termini. These oligosaccharide structures represent the predominant types in the so-called inducible (monosialo) hyaluronate binding phenotypes, together with a non-sialylated reference (asialo) (9,18,26). To mimic the predominant CD44 glycovariants found recently in mouse myeloma cells (26), we glycosylated N25, N57, N100, and N110 with the above-described complex type N-glycans and N120 with a triantennary high-mannose type structure without fucosylation (Fig. 2C). We call these glycoforms myeloma asialo and myeloma monosialo, depending on the degree of sialylation. Additionally, to emulate the mutant proteins lacking the N25 and N120 glycans that also lead to the inducible phenotype (19), we constructed a monosialo glycoform, which lacks N-glycans at N25 and N120 (partial monosialo). Finally, we designed a fourth glycoform, where each of the five N-glycans is a chargeneutral core pentasaccharide (full pentasaccharide), to represent mildly glycosylated, less-cancerous cell types.
We constructed the simulation systems using the crystal structure of human CD44-HABD (PDB:1UUH (36)). We then followed the steps described in our previous work to curate the 1UUH structure (14). This was followed by the in silico N-glycosylation of the HABD structure with the doGlycans (37) tool. Before simulations, we inspected the ready-made glycan structures visually (38) to confirm their correct configuration and stereochemistry. In the main text, we employed the mutually-compatible AMBER99SB-ILDN (proteins) (39) and GLYCAM06 (carbohydrates) (40) force fields (systems G1-4 in Table S1). Additionally, all the critical simulations were repeated in SI with the CHARMM36 force field (systems C1-2 in Table S1), and their conclusions were consistent with the combination of AMBER99SB-ILDN and GLYCAM06. In every system, sodium and chloride ions were added to reach the physiological salt concentration of 150 mM, and to neutralize the charge of the system (Dang ions (41) for AMBER99SB-ILDN and default ions for CHARMM36 systems). The systems were solvated with the recommended TIP3P water model (42). For each GLYCAM06-modeled glycoform (systems G1-4 in Table S1), we generated three different N-glycan starting configurations. Each configuration was used to start five replica simulations of 1000 ns, totalling to 15 replicas per glycoform. The CHARMM36 systems were simulated with three replicas. We also added a hyaluronate oligomer (18 monosaccharide units) to the CHARMM36 systems (systems C1-2 in Table S1), initially to study its binding to HABD, yet the oligomer never bound during the trajectories. Hence, we do not expect the hyaluronate to interfere with the folding of the N-glycans in those systems. We also constructed an additional GLYCAM06-modeled system (20 replicas of 1000 ns) having non-glycosylated CD44-HABD together with three unbound (i.e. 2-4 nm from the protein surface) hyaluronate hexamers to study their spontaneous and simultaneous binding (system G5 in Table S1). Similarly, we generated systems (10 replicas of 1000 ns each) with CD44-HABD and two hyaluronate hexamers from which one was initially complexed to the crystallographic binding site, while the other was unbound (system G6 in Table S1). All simulation data are publicly available in zenodo.org (To be provided upon acceptance).

B. Parameters for molecular dynamics simulations.
Simulations were conducted using the GROMACS simulation software package (43). For every simulation, we employed the following protocol. First, to relax clashes produced in the building process, we performed a short energy minimization run with the steepest descent algorithm (1000 steps). Subsequently, we performed 1 and 2 ns equilibration runs in the NVT and NpT ensembles, respectively, with coordinates of the protein and glycans restrained. Finally, we conducted production runs of different lengths (see Table S1). The production runs, along with equilibration, employed the leap-frog integrator with a time step of 2 fs. During the runs, periodic boundary conditions were used in all three directions, and the LINCS algorithm was used to keep all bonds constrained (44). Electrostatic interactions were treated with particle-mesh Ewald (PME) (45) with a cut-off of 1.0 nm for the real part. Lennard-Jones interactions were cut off at 1 nm. Neighbour searching for long-range interactions was carried out every ten steps. The V-rescale (46) thermostat was used to couple the systems to a heat bath of 310 K, while the Parrinello-Rahman (47) barostat was employed to keep the pressure at 1 bar. At the beginning of each production simulation, we assigned random initial velocities using the Boltzmann distribution at the target temperature. The CHARMM36 simulations used the default parameters provided by CHARMMGUI v1.7 (48). The simulation trajectories were saved every 100 ps. For other non-specified parameters, we refer to the GROMACS 4.6.7 (49, 50) defaults for the AMBER99SB-ILDN/GLYCAM06 systems or to the GROMACS 5.1.4 (43) defaults for the CHARMM36 systems.

D. Complex N-glycans on CD44-HABD can cooperatively block its canonical binding site for hyaluronate.
To characterize how N-glycans behave and fold on CD44-HABD, we in silico glycosylated a HABD structure (PDB:1UUH) with myeloma asialo, myeloma monosialo, partial monosialo, and full pentasaccharide N-glycan profiles ( Fig. 2C and systems G1-4, respectively, in Table S1). We then simulated each glycoform through 15 replicas. An average minimum distance between the complex N-glycans and the protein, as mapped onto the surface of HABD (Fig. 2B), reveals that in the myeloma glycoforms, the Nglycans cover a significant fraction of the protein surface. That is, with the complex oligosaccharides in myeloma monosialo and myeloma asialo glycoforms, the N25 glycan can interact intimately with the nearby N100 and N110 glycans, forming a sugar shield that covers the canonical binding site of hyaluronate ( Fig. 2A). Furthermore, the contact map for the five N-glycans in the myeloma monosialo glycoform (Fig. 2C) Table S1). CD44-HABD is colored pale, and different colors separate the glycans. Glycans are depicted at every 50 ns in a trajectory of 1000 ns. b: CD44-HABD with the surface colored according to the minimum distance to the N-glycans (not shown). Pale color corresponds to CD44-N-glycan distances over 15 Å, whereas bright red corresponds to distances less than 3 Å. The distance data have been averaged over 15 replicas (myeloma monosialo glycoform). c: Glycoforms used in this study. The symbols follow the Symbol Nomenclature for Graphical Representations of Glycans (52). d: Number of contacts (defined as distance < 0.6 nm) between the five N-glycans on CD44-HABD (myeloma monosialo glycoform). The results have been averaged over time and 15 replicas. Another simulation force field (CHARMM36) in Fig. S1 and Table S2 shows consistent data.
to establish, on average, several hundred intermolecular contacts, which are possible only if the three N-glycans become interconnected in the region that resides over the crystallographic hyaluronate binding groove. These results clearly show how complex N-glycans, facilitated by inter-N-glycan interactions, shield a significant portion of the hyaluronate binding face of HABD.
E. N-glycans foster the occupancy of a secondary hyaluronate-CD44 binding mode. Table 1 lists the occupancy (see Methods in Supplementary Information (SI)) of each of the three binding sites by the N-glycans. In the tested glycoforms (systems G1-4 in Table S1), amino acid residues distinct to the CD44-HABD binding modes exhibit a coverage of about 20 to 50 %. In all cases, the crystallographic binding site is most significantly obstructed by the N-glycans, while the upright site is obstructed the least. Furthermore, occupancy values calculated for the key hyaluronate binding residues of CD44-HABD (Fig. S2) reveal how the key residues that are specific to the upright mode, such as K38 and R162, are generally less covered by the N-glycans. These observations imply that the lower-affinity upright mode is the most accessible binding configuration in a glycosylated CD44-HABD. Table 1. Total N-glycan occupancy of the residues involved in each binding mode. Data are calculated from systems G1-4 in Table S1. The contributions of each residue to each binding mode are extracted from our previous work (14). The results indicate how much of the CD44-HABD surface (that is critical to hyaluronate binding) is covered by N-glycans. For details of the analysis, see SI.
Binding Strikingly, we observe minimal differences between the myeloma monosialo and myeloma asialo glycoforms, where the oligosaccharides are of the same length. However, the occupancy values decrease notably with reduced glycan content in the partial monosialo or the shorter full pentasaccharide glycoforms. Like the myeloma-derived CD44-HABDs, the partial monosialo glycoform also displays a large number of contacts between the glycans N100 and N110 (Fig. S1). Their interaction is, however, less prone to disturb the crystallographic binding site as the N25-linked glycan is missing (Fig. S1). The full pentasaccharide glycoform is fully glycosylated but entails shorter oligosaccharides, which therefore limit the degree of protein coverage. These observations suggest that it is predominantly the degree of glycosylation and the size of the attached oligosaccharides that determine the coverage of the binding site. The inclusion of sialic acids has little effect on the coverage when compared to similar-sized non-sialylated N-glycans. Residues from the hyaluronate-perturbed region such as K38, N39, and G40 exhibit similar spectral behaviour for the mixture of hyaluronate and antibody as for hyaluronate alone, i.e., their signals disappear. On the other hand, residues from the antibody-perturbed region (33) such as A138, I145, and G159 -necessary for upright mode-exhibit similar perturbations for the complex with both hyaluronate and the antibody, as in that of the antibody alone. Moreover, we calculated the histograms of the minimal combined chemical shift perturbation with respect to the free CD44-HABD spectra along its sequence (Fig. 3C). The obtained chemical shifts indicate that the spectra of the complex of CD44-HABD with both the antibody and hyaluronate still possesses the antibody-induced changes (residues within mainly the C-terminal segment of CD44-HABD) in addition to the hyaluronate-induced changes (residues within mainly the N-terminal segment of CD44-HABD). This clearly demonstrates the simultaneous binding of both hyaluronate hexamer and scFv MEM-85 to the non-glycosylated recombinant CD44-HABD.
In addition, the signals in the spectrum obtained for 15 N-CD44-HABD in the presence of both antibody and hyaluronate hexamer are significantly broadened relatively to the signals in the spectra obtained for binary mixtures, as expected for a higher molecular weight of the ternary complex.

G. Short hyaluronate oligomers bind to CD44-HABD simultaneously at distinct binding sites.
We analyzed the individual signals in the HSQC spectra for 15 N-CD44-HABD titrated with hyaluronate hexamer; signals located in crowded areas of the spectra, including R41, were not taken into account to avoid ambiguity. This analysis revealed two trends (Fig. 4). Certain backbone amide group signals exhibited an instant shift or disappearance already at the hyaluronate to CD44-HABD ratio of 1:1, which indicates a strong interaction in the sub-μM range of the respective residues with hyaluronate, while other signals shifted gradually during the individual titration steps, suggesting a relatively weaker interaction (>10 μM)(see Fig. 4A). In addition, several signals exhibited doubling, connected either with an instant shift or a gradual shift (see Fig. 4B). This points out residues which interact only in a fraction of CD44-HABD molecules with hyaluronate, and/or interact in two different modes. Specifically, the signals of the following residues exhibited instant disappearance: K38, G40, G80, Y114; instants shift: S43, I44, Y79; instant shift with doubling: D140, R150, R154, V156, T174; gradual shift: D23, N25, E37, E75, I96, Y105, Q113, E127, V148, G159, R162, E166; and gradual shift with doubling: N39, R78, L107, K158, N164, D175. Next, we mapped the critical residues involved in either strong or weak interaction with hyaluronate (Fig. 4C) and the residues interacting with hyaluronate in a single/double mode (Fig. 4D, E) onto the surface of a computational model of CD44-HABD (residues 20-169) (14). This illustrates that the surface patches associated with all the three modes are affected in our hyaluronate titration experiments. Notably, the linear patch including residues K38, S43, I44, Y79, G80, Y105, Q113, Y114, R162, and E166 outlines the binding site for the upright binding mode. Moreover, the doubling of the signals in the C-terminal portion of CD44-HABD (residues D140, R150, R154, V156, K158, and N164) indicates the coexistence of the parallel and upright modes with the crystallographic mode. The NMR data, therefore, demonstrate that the short hyaluronate hexamer can, especially in higher molar excess, bind to non-glycosylated recombinant CD44-HABD simultaneously in several modes at distinct binding sites.
To further explore the simultaneous binding of hyaluronate on CD44-HABD, we performed a set of MD simulations with two hyaluronate hexamers binding to CD44-HABD. Fig. 4C shows the probability of the HABD surface to be in contact with HA (simulation G6 in Table S1), which correlates with the CSPs observed in NMR. Additionally, Figure S3 shows a contact profile similar to the chemical shift profile recorded in NMR (Fig. 3), indicating that our experimental and computational results are in agreement.

Discussion
We employed atomistic MD simulations and NMR to shed light on ligand-receptor interactions of CD44 and hyaluronate to unravel how N-glycosylation modulates the interactions. MD simulations showed that in the crystallographic mode (sub-μM), N-glycans on CD44-HABD collectively shield the primary binding residues for hyaluronate. The shielding effect in this canonical binding mode is the strongest when complex type N-glycans occupy the Nglycosylation sites N25, N100, and N110. They are the most typical oligosaccharides found in these N-glycosylation sites (26) and are sufficiently long to interlock over the canonical hyaluronate binding groove, thereby severely hindering its availability for the ligand.
We also found that the N-glycosylation of CD44-HABD promotes a secondary, less shielded but weaker (>10 μM) hyaluronate binding site, which corresponds to the upright binding mode characterized previously by us (14) and also suggested by others (24). The results also revealed the degree of glycosylation and the size of the attached oligosac-Cryst.
Parallel Upright  ), and residues with a gradual shift, i.e., weak interaction (magenta color). b: The signals are sorted in two groups -residues with a single signal, i.e., single binding mode (orange color), and residues with a doubled signal, i.e., double binding mode (cyan color). Hyaluronate hexadecamers are shown in three possible binding modes distinguished by shades of gray -the crystallographic (cryst.), parallel (parallel), and upright (upright) mode. c: Hyaluronate-perturbed residues in simulations. The colored surface displays the probability of a given residue to be in contact with HA6 in our simulations (G6 in Table S1). Filled circles highlight the positions of selected residues from both the crystallographic and upright modes, which were perturbed by hyaluronate binding in our NMR experiments (cf. panel d). These marks are also colored based on the predominant binding mode of the highlighted residues in our earlier simulations (14). Lines between the residues are drawn to guide the reader. The position of residues belonging to the R41-containing binding epitope is also shown (cf. panel e). d e: Selected 15 N/ 1 H HSQC signals are shown for free 15 N-CD44-HABD (grey), 15 N-CD44-HABD with equimolar hyaluronate hexamer (green) and two-fold (blue) and a three-fold (red) molar excess of hyaluronate hexamer. Colored circles (top left) indicate to which category the residue falls in panels A and B. Probability of a given residue to interact with HA in each binding mode in our earlier simulations (14) is indicated in the top right corner of each graph.
charides to be the key factors in determining the coverage of the binding site, while the inclusion of single sialic acids to the glycan termini was found to have only a minor additional effect when glycans of equal length were compared to one another. Thus, it can be speculated, in the case of CD44-HA binding, that the binding-inhibiting role of monosialic acids stems from the more extended nature of the oligosaccharides and the resulting increase in the degree of coverage. The negative charge may play a more significant role in the case of polysialylated sugars. Furthermore, if the glycosylation site N25 lacks sufficiently long glycans, the propensity to interlock with the N100 and N110 glycans decreases, thereby substantially decreasing the coverage of the crystallographic site, resulting in a more exposed site to the ligand. This is in line with findings that have suggested some glycosylation patterns do not decrease the hyaluronate binding (16).
Our NMR experiments revealed distinct hyaluronate binding sites on non-glycosylated CD44-HABD, which provides substantial evidence for the existence of separate hyaluronate binding modes. Strikingly, the residues perturbed in NMR match those involved in the crystallographic, parallel, and upright binding modes. In our previous computational work, we illustrated the dynamic nature of the HABD-HA interactions outside the R41 epitope, especially in the case of the crystallographic binding mode (see Fig. S4). Similarly, the strong versus weak CSPs in Fig. 4A show both the R41 epitope and upright groove to give predominantly strong interaction signals, while other regions flanking the R41 epitope tend to give out weak interaction signals, corresponding with the increased mobility of the bound HA in those regions. Despite the dynamic interactions, the importance of such weak binding sites to the overall strength of the binding is found to be high in a related protein-carbohydrate interaction (53). The dynamics of the bound HA can be visualized in Fig. S4. The experimental results also agree well with the findings of our simulations of multiple hyaluronate hexamers with CD44-HABD, showing a similar hyaluronate-HABD binding profile (Fig. S3). The NMR readouts also show that the anti-CD44 antibody MEM-85 co-binds with hyaluronate on a non-glycosylated CD44, thus having a minimal effect on hyaluronate binding in this case. Conversely, the literature clearly states that MEM-85 blocks the hyaluronate binding of a glycosylated CD44 (34,35), implying the existence of a lower-affinity binding mode, whose binding site overlaps with the binding site of MEM-85. The MEM-85 epitope is known to be located around the residues Glu160, Tyr161, and Thr163 (33). As these residues are also a part of the upright mode, our results confirm the existence of such binding. Providing further evidence for the existence of the upright mode, when CD44-Ig (immunoglobulin) fusion proteins were expressed in COS cells and hence were presumably glycosylated, both MEM-85 and hyaluronate binding were significantly reduced by the mutation of K38 to arginine (34). According to our previous work, K38 is exclusive to the upright mode (14), which further implies that glycosylated CD44 favors to bind hyaluronate with the upright mode over the canonical crystallographic binding. We also note that distinct N-glycosylation profiles, e.g., ones that include an increasing amount of sialic acids, might cause different alterations to the binding. CD44-HA interaction is known to display glycosylationdependent levels of activation (9) and binding affinities (16). The activation levels have been attributed to varying degrees of sialylation (19,54), yet the glycosylation dependent binding affinities could stem from the simultaneous masking of high-affinity binding sites and promotion of secondary sites. Such activation-dependent regulation of glycan remodeling is undoubtedly known to be a major mechanism driving cell motility, e.g., in the immune response (12). CD44, in particular, is a hyaluronate-dependent leukocyte homing receptor that mediates both rolling interactions (55) and cellular transmigration (56). In such processes, tightly regulated affinity is required to enable dynamic velcro-like interactions between leukocytes and endothelial cells at inflamed tissue. It is known that glycans stabilize or promote specific protein conformations (4-6), dimer interfaces (57), or orientations (8,58), which ultimately affect ligand binding. There is also evidence of oligosaccharides that mask and shield specific parts of the protein surface (11,59). N-glycosylations are also generally quite well known to protect large regions of the protein surface from, e.g., non-specific interactions or proteolytic cleavage (60). The novelty of the present work lies in the fact that, in addition to all these features, N-glycosylation has an extremely valuable and hitherto unknown mechanism of action: N-glycosylation can control the affinity of ligand-receptor interaction by selectively blocking binding sites and promoting others.   Figure S3 shows that three initially unbound HA 6 molecules bind to non-glycosylated CD44 HABD during 20 × 1000 ns trajectories. The occupied epitopes in HABD contacting the HA molecules are at residues 20-26, 38-42, 75-79, 87-100, 105-115, 141-161. These residues match the known epitopes for all three binding modes (14). Qualitatively, we observe only one recognizable binding mode across 20 simulation replicas, with one HA fragment binding to the crystallographic binding mode. Yet, less distinct binding of the HA fragments occurred frequently at distinct binding sites. These observations along with the binding profile agree with the results of the NMR that show simultaneous binding of short HA fragments. Figure S3 also shows similar binding behavior for initially unbound HA 6 molecule that binds to HABD already occupied with another HA 6 in the crystallographic binding groove. The major binding epitopes locate roughly at residues 38-42, 105-115, 144-169. The C-terminal amino acids, 144-169, correspond to both the upright mode-specific HA binding residues and the residues that are influenced by MEM-85. This observation therefore agrees with the experimental observations and fortifies the notion of a second, lesser affinity binding site at the C-terminal portion of HABD.

R41 R150
R154 R162 R78 R41 R150 R154 R78 Fig. S4. Dynamics of the HA ligand in crystallographic (left) and upright (right) binding modes. The tan surface depicts the protein surface in the first frame of a simulation. The red sticks represent the ligand drawn at every 50 ns into the simulation trajectory. R41 is colored light green. Other key arginines are colored yellow and labeled accordingly. Data are extracted from our previous simulations (14).