Glycobiology, 2016, vol. 26, no. 3, 251–260 doi: 10.1093/glycob/cwv101 Advance Access Publication Date: 3 November 2015 Original Article

Computational Biology

Roles of glycans in interactions between gp120 and HIV broadly neutralizing antibodies Yifei Qi2, Sunhwan Jo3, and Wonpil Im1,2 2

Department of Molecular Biosciences and Center for Computational Biology, The University of Kansas, 2030 Becker Drive, Lawrence, KS 66047, USA, and 3Leadership Computing Facility, Argonne National Laboratory, 9700 Cass Ave Bldg. 240, Argonne, IL 60439, USA 1

To whom correspondence should be addressed: Tel: +1-785-864-1993; Fax: +1-785-864-5558; e-mail: [email protected]

Received 1 October 2015; Revised 28 October 2015; Accepted 30 October 2015

Abstract Many novel broadly neutralizing antibodies against human immunodeficiency virus (HIV) have been identified during the past decade, providing promising templates for the development of an effective HIV-1 vaccine. Structural studies reveal that the epitopes of some of these antibodies involve one or more crucial glycans, without which the binding is completely abolished. In this study, we have investigated the critical roles of glycans in interactions between HIV-1 gp120 and two broadly neutralizing antibodies PG9 (targeting V1/V2) and PGT128 (targeting V3) that are able to neutralize more than 70% of HIV-1 isolates. We have performed molecular dynamics simulations of a number of systems including antibody–gp120 complex with and without glycans, antibody, gp120 with and without glycans, and glycan-only systems. The simulation results show that the complex structures are stabilized by the glycans, and the multivalent interactions between the antibody and gp120 promote cooperativities to further enhance the binding. In the free gp120, the glycans increase the flexibility of the V1/V2 and V3 loops, which likely increases the entropy cost of the antibody recognition. However, the antibodies are able to bind the flexible interface by recognizing the preexisting glycan conformation, and penetrating the glycan shield with flexible complementarity determining region loops that sample the bound conformations occasionally. Key words: PG9, PGT128, protein–glycan interaction

Introduction Human immunodeficiency virus (HIV) is one of the viruses mostly threatening to public health. So far, the most successful treatment of HIV infection is the use of antiretroviral drugs that control the virus concentration at a low level and thus turns the infection into a chronic disease. However, the burden of daily medicine, side effects and the emergence of drug resistance make it highly desirable to develop more effective therapies. As an alternative strategy, considerable efforts have been devoted to developing HIV vaccines that are expected to induce antibodies targeting envelope protein gp120 on the viral surface. It is believed that a successful vaccine should not only stimulate profound T-cell immune responses, but also produce broadly neutralizing antibodies (Stamatatos 2012). On the envelope trimer, each gp120 subunit has on average 25 N-linked glycosylation sites, and ∼50% of its mass is made up of

carbohydrates (Leonard et al. 1990). Analysis of these glycans released from recombinantly expressed gp120 has revealed an unusual high and conserved population of oligomannose-type glycans (Pritchard et al. 2015). These glycans contribute to the structure and dynamics of gp120. For example, structural studies show that a glycan at N262, which is 99% conserved, stabilizes the gp120 structure (Kong et al. 2015). Molecular dynamics (MD) simulation of gp120 with and without glycans at the V3 loop suggests that the dynamics of the V3 loop is modulated by the glycans (Wood et al. 2013). The prevalence, specificity and conservation of the HIV gp120 glycans make these glycans promising targets for vaccine development (Horiya et al. 2014; Crispin and Doores 2015), as glycans have already been used as cell antigenic determinants for antibacterial, anticancer and antiviral vaccine developments (Astronomo and Burton 2010; Morelli et al. 2011).

© The Author 2015. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: [email protected]

251

252

Y Qi et al.

Recently, a collection of broadly neutralizing monoclonal antibodies (bnmAbs), including 2G12 (Trkola et al. 1996), PG9/PG16 (Walker et al. 2009), PGTs (Pejchal et al. 2011; Walker et al. 2011), CH01-04 (Bonsignori et al. 2011), VRC01 (Wu et al. 2010), 35O22 (Huang et al. 2014), etc., have been identified, and they are able to neutralize HIV-1 isolates with high potency and breadth (Walker et al. 2009, 2011; Wu et al. 2010). Structural studies show that some of these bnmAbs not only bind to gp120, but also recognize several glycans attached to gp120 (McLellan et al. 2011; Pejchal et al. 2011; Kong et al. 2013; Pancera et al. 2013; Garces et al. 2014). This binding mode is in contrast to a traditional view in which gp120 glycans act as a shield to protect HIV from antibodies’ recognition (Reitter et al. 1998; Wei et al. 2003). With the continuing discoveries of glycan-recognizing bnmAbs (Eroshkin et al. 2014), the roles of these glycans in the antibody recognition have been an active topic in recent studies. In fact, these glycans turn out to be an indispensable part of the bnmAbs’ binding process. Mutation at the glycosylation sites significantly reduces the binding and the neutralizing effects of some bnmAbs (McLellan et al. 2011; Pejchal et al. 2011). Although individual protein–carbohydrate interactions could be weak, the bnmAbs are able to achieve tight recognition through multivalent binding to multiple glycans (supersite or glycan patch) (Kong et al. 2013) as well as gp120 protein residues. To develop more effective HIV vaccines, it is crucial to understand the roles of these glycans in interactions between gp120 and bnmAbs. In this study, we focus on two bnmAbs, PG9 (McLellan et al. 2011) and PGT128 (Pejchal et al. 2011). The complex structures of these bnmAbs, which target the V1/V2 (PG9) and V3 (PGT128) regions of gp120, respectively, have been recently solved with the gp120 binding regions grafted to small scaffold proteins ZM109 (with V1/V2 region, referred to as V1V2 scaffold hereinafter) and eODmV3 (with V3 region, referred to as V3 scaffold hereinafter). These structures show that binding of the

bnmAbs to gp120 involves recognition of one or two critical glycans on the V1/V2 or V3 region (McLellan et al. 2011; Pejchal et al. 2011). In this study, we aim to understand the roles of these mannose-rich glycans (GlcNAc2Man5 in V1/V2 and GlcNAc2Man9 in V3) in PG9-V1V2 and PGT128-V3 interactions from a structural and dynamics point of view using long MD simulations, which could provide insight into glycan– antibody interactions in a broader context.

Results Two bnmAbs systems were simulated in this study (Table I and Figure 1). In the PG9-V1V2 system, PG9 is in complex with 1FD6ZM109, in which the V1/V2 region from the HIV strain ZM109 is grafted to a scaffold protein (referred to as V1V2 scaffold, Figure 2A) (McLellan et al. 2011). In the PGT128-V3 system, PGT128 is in complex with eODmV3, a glycosylated gp120 outer domain with a truncated V3 loop (referred to as V3 scaffold, Figure 2B) (Pejchal et al. 2011). Six systems of each complex were built and simulated: glycan-only, antibody, scaffold-gp120 protein with and without glycans, and the complex with and without glycans. The two glycans in each complex were assumed to have the same sequence (GlcNAc2Man5 in V1V2 and GlcNAc2Man9 in V3: Supplementary data, Figure S1), and several missing sugar residues in the crystal structures were modeled. In the crystal structure, PG9 forms intensive interactions with N160 glycan, and backbone–backbone interactions between CDR H3 and strand C of the V1/V2 loop stub (Figure 2A). PGT128 is in contact with N301 glycan and N332 glycan, and CDR H3 interacts with the V3 loop (Figure 2B).

Glycans stabilize the complex To examine the overall effect of the glycans on the complex stability, we calculated the root mean square deviation (RMSD) of the complex

Table I. Simulation systems and their notationsa PG9-V1V2 system Name #Atom

glycan 23410

V1V2 scaffoldb 51738

V1V2-glycan 51655

PG9 55588

PG9-V1V2 124904

PG9-V1V2-glycan 124851

V3 scaffoldb 59438

V3-glycan 97113

PGT128 48103

PGT128-V3 147050

PGT128-V3-glycan 146918

PGT128-V3 system Name #Atom

glycan 22289

a

Each system was simulated for 200-ns equilibration using NAMD and 1.5-μs production using Anton. The scaffold protein without the glycans.

b

Fig. 1. The PG9-V1V2-glycan and PGT128-V3-glycan simulation systems. The antibody heavy chain is shown in red, the light chain in cyan, the V1V2 and V3 scaffold in green and the glycans in sticks. The water box is shown in crystal surface. This figure is available in black and white in print and in color at Glycobiology online.

Glycan’s roles in gp120–antibody interactions

253

Fig. 2. Binding interface of (A) PG9-V1V2 (PDB:3U2S) and (B) PGT128-V3 (PDB:3TYG). This figure is available in black and white in print and in color at Glycobiology online.

Fig. 3. Stability of PG9-V1V2. (A) Comparison of the last MD snapshot of the glycan-free system (green for V1V2 scaffold and yellow for PG9) to the crystal structure (cyan for V1V2 scaffold and blue for PG9). (B and C) RMSD and number of native hydrogen bonds in PG9-V1V2 system with and without glycans as a function of time (including 200-ns NAMD equilibration). This figure is available in black and white in print and in color at Glycobiology online.

structure using the Cα atoms. In PG9-V1V2-glycan, the antibody– protein complex system with glycans is stable during the simulation time with a RMSD value of 7 Å (Figure 3). The individual components (V1V2 scaffold and PG9) are stable with RMSD values around 1–4 Å (Supplementary data, Figure S2A and S3A). The deviations from the crystal structure mostly arise from the flexibility of the V1V2 scaffold, i.e., a structural change of the part distant from the binding interface. When the glycans are removed in the complex, the PG9-V1V2 system shows larger deviations with the RMSD values of 10 Å. Comparison of the crystal structure to the last simulation snapshot shows that the V1V2 scaffold rotates almost 90° and the native hydrogen bonds are completely broken (Figure 3). In PGT128-V3-glycan, the glycans

show similar stabilization effect on the complex structure (Figure 4). After removing the glycans, the V3 scaffold rotates significantly, with some native contacts remaining between CDR H3 and V3 loop within the simulation time (Figure 4C), but the individual V3 scaffold and PGT128 antibody remain stable in the absence of the glycans (Supplementary data, Figure S2B and S3D). These simulations suggest that the glycans are essential components in forming the correct protein–protein interface, in agreement with experimental results that removing the glycans abolishes the binding completely (McLellan et al. 2011; Pejchal et al. 2011). The glycans appear to stabilize the complex by fixing the relative orientation of the antibody and the V1V2/V3 scaffold.

254

Y Qi et al.

Fig. 4. Stability of PGT128-V3. (A) Comparison of the last MD snapshot of the glycan-free system (green for V3 scaffold and yellow for PGT128) to the crystal structure (cyan for V3 scaffold and blue for PGT128). (B and C) RMSD and number of native hydrogen bonds with and without glycans as a function of time (including 200-ns NAMD equilibration). This figure is available in black and white in print and in color at Glycobiology online.

Cooperativity exists between glycan–antibody interactions and gp120–antibody protein–protein interactions It is known that PG9 and PGT128 do not bind to deglycosylated gp120 (McLellan et al. 2011; Pejchal et al. 2011), and glycans usually bind to proteins with weak affinity with a dissociation constant greater than micromolar (Liang et al. 2007). For example, PG9 binds to GlcNAc2Man5-Asn with only 1.6 ± 0.9 mM (McLellan et al. 2011). However, the glycans and gp120 together bind to the antibodies with a much higher affinity. PG9 binds to glycosylated V1V2 scaffold with 5.67 ± 0.19 μM (McLellan et al. 2011), and PGT128 binds to glycosylated V3 scaffold with an apparent affinity of 46 µM (Pejchal et al. 2011). The strong binding is believed to arise from multivalent binding to multiple glycans and protein residues within one Fab (Calarese et al. 2003, 2005), and it has been speculated that cooperativity may exist between glycan–antibody interactions and gp120– antibody protein–protein interactions. However, whether such cooperativity exists or which component contributes to the cooperativity is unknown. To explore these questions, we used information theory transfer entropy to characterize the cooperativity quantitatively (Schreiber 2000). In transfer entropy analysis, the net information flow between two time series x and y is quantified as a normalized directional index, Dxy, between −1 and 1, in which a negative value means that time series x responses to y and a positive value means that time series x drives y. In the current study, we used the number of hydrogen bonds between glycan–antibody and gp120–antibody residue pairs as the time series. Two drive–response relationships were identified with |Dxy| > 0.1 and P-value < 0.05 (Table II). In PG9-V1V2-glycan, the number of hydrogen bonds between GlcNAc1 in N160 glycan and Arg100B in PG9 heavy chain drives the hydrogen bond between GLN170 in V1V2 scaffold and Asp100I in PG9 heavy chain, meaning that the glycan–antibody interaction drives

Table II. Transfer entropy analysis of glycan–antibody interactions and gp120–antibody protein–protein interactions System

Dxy

Residue pair x

PG90.13 N160 glycan:GlcNAc1V1V2-glycan PG9:Arg100B PGT128−0.12 N332 glycan:AMAN5V3-glycan PGT128:Trp52F

Residue pair y V1V2:Gln170PG9:Asp100I V3:Arg116PGT128:Asp100D

the protein–protein interaction between V1V2 scaffold and the antibody (Figure 5A). Interestingly, in PGT128-V3-glycan, Dxy is negative, indicating that the hydrogen bond between MAN5 in N332 glycan and Phe52 in PGT128 heavy chain responses to the hydrogen bond between Arg116 in V3 scaffold and Asp100D in PGT128 heavy chain, i.e., the glycan–antibody interaction responses to the protein– protein interaction between V3 scaffold and the antibody (Figure 5B). This result suggests that the cooperativity indeed exists between the glycan–antibody interactions and gp120–antibody protein–protein interactions, and the direction may be different in different systems.

Glycans affect the structure and dynamics of V1/V2 and V3 loops Glycans make up about half of the molecular mass of HIV gp120 protein (Wyatt and Sodroski 1998; Pantophlet and Burton 2006). These glycans not only play important roles in shielding gp120 from antibody recognition, but also affect the structure and dynamics of gp120 (Wood et al. 2013). To investigate the effects of the glycans on the proteins, the structure and dynamics of the V1/V2 and V3 loops in V1V2 and V3 scaffold systems are compared with and without glycans (Figure 6A, B and D). In both glycosylated and deglycosylated V1V2 scaffold, the V1 loop does not show obvious change, while

255

Glycan’s roles in gp120–antibody interactions

Fig. 5. The hydrogen bonds that show drive–response interactions in transfer entropy analysis: (A) PG9-V1V2-glycan and (B) PGT128-V3-glycan. The hydrogen bonds are shown in black dashed lines. The black arrows indicate the directions of the driving interactions. This figure is available in black and white in print and in color at Glycobiology online.

Fig. 6. Structure and dynamics of the V1/V2 and V3 loops. (A, B and D) Comparison of the last MD snapshot (green) to the crystal structure (cyan) in (A) deglycosylated V1V2 scaffold, (B) V1V2-glycan and (D) V3-glycan systems. The glycans are shown in sticks and the V2 loop is shown in red. (C) Time series of the distance between the Cα atoms of Ser178G and Tyr191 (black line in A) to characterize the motion of the V2 loop. (E) The dihedral angle between the Cα atoms of residues 108, 100, 119 and 123 to characterize the motion of V3 loop. (F) Histogram of the dihedral angles with and without glycans. This figure is available in black and white in print and in color at Glycobiology online.

the V2 loop folds back and forms interactions with the scaffold protein (Figure 6A and B). In the glycosylated system, N173 glycan interacts with V2 and slows down the folding back process, making the V2 loop more flexible as seen from the distance between the Cα atoms of Ser178G and Tyr191 (Figure 6C). This interaction is over all N173 glycan residues and is transient (Supplementary data, Figure S4A) and likely nonspecific. In V3 scaffold systems, the V3 loop also has large flexibility in the glycosylated system, which is caused by the interactions between N301 glycan and N332 glycan (Figure 6D). Most of the interactions between the glycans are transient, but several pairs between the base residues of the glycans show 30–40% interacting probability (Supplementary data, Figure S4B). On the envelope protein, the interactions between neighboring glycans are possible and observed in a crystal structure of a fully glycosylated HIV-1 gp120 core (Kong et al. 2015). To further show the flexibility change, we calculated the dihedral angle between the Cα atoms of residues 108, 100, 119 and 123 (Figure 6E). The dihedral has larger fluctuations in V3-glycan than that in the deglycosylated V3 scaffold (Figure 6F).

In total, the glycans make the V2 and V3 loops more flexible in both V1V2-glycan and V3-glycan systems and likely increase the entropic cost for the antibody binding. These results suggest that glycans not only act as a shield to the antibody but also are able to change the flexibility and dynamics of the V1/V2 and V3 loops.

The antibody recognizes the preexisting glycan conformations When attached to proteins, glycans are able to affect the local protein structures (Lee, Qi, et al. 2015). On the other hand, the glycosylated protein also impacts the conformation of the glycans (Jo et al. 2016). In this section, we sought to investigate the conformations of the glycans in the free form, attached to V1V2/V3 scaffold and in the scaffold–antibody complex by comparing the dihedral angles of the glycosidic linkages (Figure 7). Compared to the free form, the glycans did not show large deviations when attached to proteins. The only noticeable change is that the minor population of φ around −45° is gone

256

Y Qi et al.

Fig. 7. φ and ψ dihedral angle distributions of the glycans. (A) Glycans in PG9-V1V2 systems; the rows from top to bottom are from free glycan, N160 glycan in V1V2-glycan system, N173 glycan in V1V2-glycan system, N160 glycan in PG9-V1V2-glycan system, and N173 glycan in PG9-V1V2-glycan system. (B) Glycans in the PGT128-V3 systems; the rows from top to bottom are from free glycan, N332 glycan in V3-glycan system, N301 glycan in V3-glycan system, N332 glycan in PGT128-V3-glycan system, and N301 glycan in PGT128-V3-glycan system. The arrows indicate values from the crystal structure. This figure is available in black and white in print and in color at Glycobiology online.

in linkage 1–2 and 2–3 in the glycosylated system. This suggests that the protein exerts its impact mostly on the base of the glycans, which is near the protein surface. Although the glycans interact with each other in the glycosylated V1V2/V3 system (e.g., Supplementary data, Figure S4), these nonspecific interactions do not change the conformational space significantly. Upon binding of the antibody, several glycosidic linkages show narrower distributions (e.g., linkages 3-4, 4-C and C-D1 in V3 system). However, the narrower distribution of the glycosylated form is still similar to that of the free glycan systems. This suggests that the antibody recognizes the existing glycan conformations and stabilizes part of the accessible conformations.

CDR H3 loop in free PG9 and PGT128 is able to sample the bound conformations The structure of antibodies PG9 and PGT128 remain stable in all the systems, with the RMSD values around 1–3 Å (Supplementary data, Figure S3). PG9 shows larger RMSDs due to the long CDR H3 loop (Supplementary data, Figure S5). The antibodies are more stable in terms of root mean square fluctuation (RMSF) in the complex structure with glycans, and less stable in the free form and the complex structure without glycans. Not surprisingly, the CDR regions show the largest flexibility change in different systems. A striking feature of PG9 and PGT128 is to have a long CDR H3 loop (30 residues in

257

Glycan’s roles in gp120–antibody interactions PG9 and 21 residues in PGT128). To examine the dynamics of CDR H3 in the different systems, we aligned the antibody structure without the CDR regions and calculated the RMSD of the CDR H3 tip residues (from H95 to H100H in PG9 and from H96 to H100I in PGT128, following Kabat numbering: Figure 8). The CDR H3 tip is most stable in the PG9-V1V2-glycan system, although the RMSD reaches as high as 12 Å due to rotation of V1V2 scaffold. In PG9-V1V2, the RMSD value has large fluctuations because of the large deviation of V1V2 scaffold in the absence of glycans. In the PG9-only system, the RMSD value is around 10 Å in most of the simulation time due to the high flexibility of CDR H3 tip. However, the CDR H3 tip is able to occasionally switch to the conformation toward the crystal structure with a RMSD as low as 3 Å. In PGT128 systems, the CDR H3 tip is more stable than that in the free PG9, with an average RMSD of 3.5 Å and a minimal RMSD value of 1.3 Å after the first 200 ns simulation (Figure 8D). This is likely because the CDR H3 loop in PGT128 is shorter than that in PG9. In PGT128-V3 and PGT128-V3-glycan systems, the tip is stable with a minimal RMSD of less than 1 Å. To further compare the tip conformation in the free antibody and in the complex, we performed PCA on the tip conformations in the gp120 scaffold–glycan–antibody systems and projected the trajectories to the eigenvectors that correspond to the first two largest eigenvalues (Figure 9). The tip conformation in the crystal structures is within the accessible space of the tip conformations in the free PG9 and PGT128 systems. Interestingly, the crystal conformation is outside the MD trajectory in PG9-V1V2-glycan (Figure 9B), which is in agreement with the large RMSD values due to deviation from the crystal structure (Figure 8C). Nonetheless, there are some overlaps between the conformations in the free antibody (Figure 9A and C) and the conformations in the V1V2/V3 scaffold–glycan–antibody complex systems (Figure 9B and D). These results suggest that the bound conformation of CDR H3 tip is accessible in the free PG9 and PGT128.

Discussion and conclusions In this study, we have used MD simulations to study two bnmAbs systems PG9 and PGT128, with the focus on the recognition of the glycans by the antibodies. The simulations show that the V1V2/V3 scaffold–antibody complex becomes distorted within 1 μs of simulations without the glycans, in agreement with experimental results that the glycans are indispensable to the gp120–antibody binding. Protein–glycan interactions are weak due to glycan’s soluble nature that prefers interactions with water rather than with protein. Our information theory transfer entropy analysis quantitatively shows that there is cooperativity between the protein–glycan and protein–protein interactions. Interestingly, the direction of the cooperativity is system dependent. In PG9, the formation of glycan–antibody hydrogen bonds drives the protein–protein hydrogen bonds between the gp120 scaffold and antibody, but vice versa in PGT128, highlighting the importance of both glycan–protein and protein–protein interactions in the antibody recognition process. The multivalent interactions between the antibody and gp120 not only stabilize the complex, but may also promote cooperativity to further enhance the binding. The five variable loop regions (V1–V5) are the most flexible parts on gp120. Our simulations show that the V1/V2 and V3 loops are more flexible in the presence of the glycans. Detailed analysis reveals that the increased flexibility is due to glycan–glycan and glycan– protein interactions. Contradictory conclusions have been made in other computational studies. For example, Wood and coworkers studied the influence of N-linked glycans on the V3 loop dynamics using MD simulation and found that the glycans narrow down the conformational space of the V3 loop (Wood et al. 2013), whereas Lee and coworkers performed a systemic investigation of the effects of N-glycosylation on the conformation and dynamics of proteins and found that N-glycosylation does not induce significant changes in protein structure, but decreases protein dynamics (Lee, Qi, et al. 2015). However, we have observed that the glycans on the V3 loop enhances

Fig. 8. RMSD time series of the CDR H3 tip residues in PG9 and PGT128: (A) free PG9, (B) PG9-V1V2, (C) PG9-V1V2-glycan, (D) free PGT128, (E) PGT128-V3, and (F) PGT128-V3-glycan. This figure is available in black and white in print and in color at Glycobiology online.

258

Y Qi et al.

Fig. 9. Projection of the tip conformations to the first two eigenvectors from PCA: (A) PG9, (B) PG9-V1V2-glycan, (C) PGT128, and (D) PGT128-V3-glycan. The red stars show the projections of the crystal structures. This figure is available in black and white in print and in color at Glycobiology online.

the dynamics of the loop. We speculate this is because the systems used in this study have more frequent glycan–glycan and glycan–protein interactions that can modulate the loop dynamics, whereas those two studies either have glycans not directly present on the loop (Wood et al. 2013) or have less frequent glycan–glycan interactions (Lee, Qi, et al. 2015). The different effects observed in these studies highlight the complexity of how the glycans modulate the structure and dynamics of the protein. The glycans on gp120 make major contributions to the flexibility of the envelope. To our surprise, the conformations of the glycans in different systems do not significantly deviate from each other in the simulations. Antibody recognition only narrows down the accessible space of the φ and ψ dihedral angles. Although the glycans interact with each other frequently in the simulations, the individual glycan structures do not change a lot. High-resolution crystal structures are most powerful in elucidating the conformations of the glycans. To compare the glycan structures in different states, we used GS-align (Lee, Jo, et al. 2015) to superimpose the N332 glycan structures in three bnmAbs–gp120 complex structures (Supplementary data, Figure S6). The alignment shows that the glycans are in very similar conformations. The simulation results and the structural comparison imply that the structural conservation of these glycans is likely the reason that PG9 and PGT128 have a broad neutralizing effect across different clades. Both PG9 and PGT128 are featured by a long CDR H3 loop that can penetrate the glycan shield and form interactions with gp120. Our RMSD comparison and PCA show that the tip of the CDR H3 loop is flexible in the free antibodies and is able to sample the bound conformation, which likely increases the possibility to penetrate the glycan shield. In conclusion, our simulation study and detailed analyses highlights the key roles of glycans in the recognition of gp120 by antibodies PG9 and PGT128. The dynamics from the microsecondlong simulations provide a deeper insight into the systems that are not available from the static crystal structures. On the native

envelope trimer, the situation is more complicated as there are more glycan–glycan and protein–glycan interactions. With the recent success of structural characterization of the envelope trimer (Julien et al. 2013; Lyumkis et al. 2013; Do Kwon et al. 2015), it would be interesting to simulate the effects of glycans in their native environment.

Materials and methods For each of the PG9-V1V2 and PGT128-V3 complexes, the following six systems were simulated: the complex with and without glycans, antibody, scaffold-gp120 protein with and without glycans, and glycan-only (Table I). The initial structures were obtained from the crystal structures of PDB:3U2S (PG9-V1V2) (McLellan et al. 2011) and PDB:3TYG (PGT128-V3) (Pejchal et al. 2011). The missing glycans in the crystal structures were modeled using the glycan fragment database (www.glycanstructure.org) (Jo and Im 2013). To reduce the system size, the constant domain of the heavy and light chains from the antibodies was removed. The missing loops in the crystal structures were built with the ModLoop server (Fiser and Sali 2003). The simulation systems were built using the Glycan Reader module in CHARMM-GUI (Jo et al. 2008, 2011; Brooks et al. 2009) with TIP3P water and 0.15 M KCl (Figure 1). The systems were simulated for 200 ns using NAMD (Phillips et al. 2005) with the constant particle number, pressure and temperature (NPT) ensemble and the CHARMM36 force field (Huang and MacKerell 2013; Patel et al. 2014). The van der Waals interactions were smoothly switched off at 10–12 Å by a force-switching function (Steinbach and Brooks 1994). The particle mesh Ewald algorithm was applied to calculate electrostatic forces (Essmann et al. 1995). A time step of 2 fs was used in all simulations. Temperature was maintained at 300 K using Langevin dynamics with a coupling coefficient of 1 ps−1. Nosé–Hoover Langevin–piston method (Martyna et al. 1994; Feller et al. 1995) was used to maintain constant pressure (1 bar) with a piston period of 50 fs and a piston decay of 25 fs.

259

Glycan’s roles in gp120–antibody interactions After the NAMD simulation, each system was further simulated for 1.5 μs on Anton (Shaw et al. 2008) using the CHARMM36 force field (Huang and MacKerell 2013; Patel et al. 2014). The constant particle number, volume and temperature (NVT) ensemble was used with temperature maintained at 300 K using the Nosé–Hoover method (Nose 1984; Hoover 1985). The time step was 2 fs and trajectories were saved every 240 ps. Short-range nonbonded and long-range electrostatic interactions were evaluated every 2 and 6 fs, respectively. The nonbonded cutoff parameters were automatically determined using the Anton optimization protocol. Long-range electrostatics was calculated using the k-Gaussian split Ewald method (Shan et al. 2005) with a 64 × 64 × 64 grid. SHAKE was used to constrain all bonds involving hydrogen atoms (Ryckaert et al. 1977). Trajectory analysis was performed using CHARMM (Brooks et al. 2009), ST-analyzer (Jeong et al. 2014) and VMD (Humphrey et al. 1996). The structural figures were prepared with VMD and Pymol (Schrodinger 2010). The dihedral angles of the glycans were calculated using the following definitions: φ = O5-C1-Oi′-Ci′ and ψ = C1-Oi′-Ci′-Ci-1′. Information theory transfer entropy analysis was performed to reveal the drive–response relationship between the glycan–antibody interactions and gp120–antibody protein–protein interactions. The time series in the analysis was the number of hydrogen bond between residue pairs in the gp120–antibody binding interface. The embedding dimension, which is the number of steps in the past to be included in the probability calculation, was set to 1 as described previously (Qi and Im 2013). In principal component analysis (PCA), the antibody structures were aligned using Cα atoms without the complementarity determining regions (CDR). The average position of the CDR H3 tip residues was calculated based on the aligned trajectory, and the covariance matrix of the fluctuations from the average position was constructed. After the eigenvector and eigenvalue of the covariance matrix were obtained, the trajectories were projected to the first two eigenvectors that have the largest eigenvalues. In the antibody-only PG9 and PGT128 systems, the trajectories were projected to the eigenvectors from the complex PG9-V1V2-glycan and PGT128-V3-glycan systems, respectively.

Supplementary data Supplementary data for this article are available online at http:// glycob.oxfordjournals.org/.

Acknowledgements The Anton machine at PSC was generously made available by D.E. Shaw Research.

Funding This work was supported by the University of Kansas General Research Fund allocation #2301552, NSF IIA-1359530, NIH U54GM087519 and XSEDE MCB070009. Anton computer time was provided by the National Center for Multiscale Modeling of Biological Systems (MMBioS) through Grant P41GM103712-S1 from the National Institutes of Health and the Pittsburgh Supercomputing Center (PSC).

Conflict of interest statement None declared.

Abbreviations bnmAbs, broadly neutralizing monoclonal antibodies; CDR, complementarity determining region; HIV, human immunodeficiency virus; MD, molecular dynamics; NPT, constant particle number, pressure and temperature; NVT, constant particle number, volume and temperature; PCA, principal component analysis; RMSD, root mean square deviation; RMSF, root mean square fluctuation.

References Astronomo RD, Burton DR. 2010. Carbohydrate vaccines: Developing sweet solutions to sticky situations? Nat Rev Drug Discov. 9:308–324. Bonsignori M, Hwang KK, Chen X, Tsao CY, Morris L, Gray E, Marshall DJ, Crump JA, Kapiga SH, Sam NE, et al. 2011. Analysis of a clonal lineage of HIV-1 envelope V2/V3 conformational epitope-specific broadly neutralizing antibodies and their inferred unmutated common ancestors. J Virol. 85:9998–10009. Brooks BR, Brooks CL 3rd, Mackerell AD Jr, Nilsson L, Petrella RJ, Roux B, Won Y, Archontis G, Bartels C, Boresch S, et al. 2009. CHARMM: the biomolecular simulation program. J Comput Chem. 30:1545–1614. Calarese DA, Lee HK, Huang CY, Best MD, Astronomo RD, Stanfield RL, Katinger H, Burton DR, Wong CH, Wilson IA. 2005. Dissection of the carbohydrate specificity of the broadly neutralizing anti-HIV-1 antibody 2G12. Proc Natl Acad Sci USA. 102:13372–13377. Calarese DA, Scanlan CN, Zwick MB, Deechongkit S, Mimura Y, Kunert R, Zhu P, Wormald MR, Stanfield RL, Roux KH, et al. 2003. Antibody domain exchange is an immunological solution to carbohydrate cluster recognition. Science. 300:2065–2071. Crispin M, Doores KJ. 2015. Targeting host-derived glycans on enveloped viruses for antibody-based vaccine design. Curr Opin Virol. 11:63–69. Do Kwon Y, Pancera M, Acharya P, Georgiev IS, Crooks ET, Gorman J, Joyce MG, Guttman M, Ma X, Narpala S, et al. 2015. Crystal structure, conformational fixation and entry-related interactions of mature ligand-free HIV-1 Env. Nat Struct Mol Biol. 22:522–531. Eroshkin AM, LeBlanc A, Weekes D, Post K, Li Z, Rajput A, Butera ST, Burton DR, Godzik A. 2014. bNAber: Database of broadly neutralizing HIV antibodies. Nucleic Acids Res. 42:D1133–1139. Essmann U, Perera L, Berkowitz ML, Darden T, Lee H, Pedersen LG. 1995. A smooth particle Mesh Ewald method. J Chem Phys. 103:8577–8593. Feller SE, Zhang YH, Pastor RW, Brooks BR. 1995. Constant-pressure molecular-dynamics simulation—the Langevin Piston method. J Chem Phys. 103:4613–4621. Fiser A, Sali A. 2003. ModLoop: Automated modeling of loops in protein structures. Bioinformatics. 19:2500–2501. Garces F, Sok D, Kong L, McBride R, Kim HJ, Saye-Francisco KF, Julien JP, Hua Y, Cupo A, Moore JP, et al. 2014. Structural evolution of glycan recognition by a family of potent HIV antibodies. Cell. 159:69–79. Hoover WG. 1985. Canonical dynamics—Equilibrium phase-space distributions. Phys Rev A. 31:1695–1697. Horiya S, MacPherson IS, Krauss IJ. 2014. Recent strategies targeting HIV glycans in vaccine design. Nat Chem Biol. 10:990–999. Huang J, Kang BH, Pancera M, Lee JH, Tong T, Feng Y, Imamichi H, Georgiev IS, Chuang GY, Druz A, et al. 2014. Broad and potent HIV-1 neutralization by a human antibody that binds the gp41–gp120 interface. Nature. 515:138–142. Huang J, MacKerell AD Jr. 2013. CHARMM36 all-atom additive protein force field: Validation based on comparison to NMR data. J Comput Chem. 34:2135–2145. Humphrey W, Dalke A, Schulten K. 1996. VMD: Visual molecular dynamics. J Mol Graph. 14:33–38, 27–38. Jeong JC, Jo S, Wu EL, Qi Y, Monje-Galvan V, Yeom MS, Gorenstein L, Chen F, Klauda JB, Im W. 2014. ST-analyzer: A web-based user interface for simulation trajectory analysis. J Comput Chem. 35:957–963. Jo S, Im W. 2013. Glycan fragment database: A database of PDB-based glycan 3D structures. Nucleic Acids Res. 41:D470–D474. Jo S, Kim T, Iyer VG, Im W. 2008. CHARMM-GUI: A web-based graphical user interface for CHARMM. J Comput Chem. 29:1859–1865.

260 Jo S, Qi Y, Im W. 2016. Preferred conformations of N-glycan core pentasaccharide in solution and in glycoproteins. Glycobiology. 26:19–29. Jo S, Song KC, Desaire H, MacKerell AD Jr, Im W. 2011. Glycan Reader: Automated sugar identification and simulation preparation for carbohydrates and glycoproteins. J Comput Chem. 32:3135–3141. Julien JP, Cupo A, Sok D, Stanfield RL, Lyumkis D, Deller MC, Klasse PJ, Burton DR, Sanders RW, Moore JP, et al. 2013. Crystal structure of a soluble cleaved HIV-1 envelope trimer. Science. 342:1477–1483. Kong L, Lee JH, Doores KJ, Murin CD, Julien JP, McBride R, Liu Y, Marozsan A, Cupo A, Klasse PJ, et al. 2013. Supersite of immune vulnerability on the glycosylated face of HIV-1 envelope glycoprotein gp120. Nat Struct Mol Biol. 20:796–803. Kong L, Wilson IA, Kwong PD. 2015. Crystal structure of a fully glycosylated HIV-1 gp120 core reveals a stabilizing role for the glycan at Asn262. Proteins. 83:590–596. Lee HS, Jo S, Mukherjee S, Park SJ, Skolnick J, Lee J, Im W. 2015. GS-align for glycan structure alignment and similarity measurement. Bioinformatics. 31:2653–2659. Lee HS, Qi Y, Im W. 2015. Effects of N-glycosylation on protein conformation and dynamics: Protein Data Bank analysis and molecular dynamics simulation study. Sci Rep. 5:8926. Leonard CK, Spellman MW, Riddle L, Harris RJ, Thomas JN, Gregory TJ. 1990. Assignment of intrachain disulfide bonds and characterization of potential glycosylation sites of the type 1 recombinant human immunodeficiency virus envelope glycoprotein (gp120) expressed in Chinese hamster ovary cells. J Biol Chem. 265:10373–10382. Liang PH, Wang SK, Wong CH. 2007. Quantitative analysis of carbohydrate– protein interactions using glycan microarrays: Determination of surface and solution dissociation constants. J Am Chem Soc. 129:11177–11184. Lyumkis D, Julien JP, de Val N, Cupo A, Potter CS, Klasse PJ, Burton DR, Sanders RW, Moore JP, Carragher B, et al. 2013. Cryo-EM structure of a fully glycosylated soluble cleaved HIV-1 envelope trimer. Science. 342:1484–1490. Martyna GJ, Tobias DJ, Klein ML. 1994. Constant-pressure moleculardynamics algorithms. J Chem Phys. 101:4177–4189. McLellan JS, Pancera M, Carrico C, Gorman J, Julien JP, Khayat R, Louder R, Pejchal R, Sastry M, Dai K, et al. 2011. Structure of HIV-1 gp120 V1/V2 domain with broadly neutralizing antibody PG9. Nature. 480:336–343. Morelli L, Poletti L, Lay L. 2011. Carbohydrates and immunology: Synthetic oligosaccharide antigens for vaccine formulation. Eur J Org Chem. 2011:5723–5777. Nose S. 1984. A unified formulation of the constant temperature moleculardynamics methods. J Chem Phys. 81:511–519. Pancera M, Shahzad-Ul-Hussan S, Doria-Rose NA, McLellan JS, Bailer RT, Dai K, Loesgen S, Louder MK, Staupe RP, Yang Y, et al. 2013. Structural basis for diverse N-glycan recognition by HIV-1-neutralizing V1-V2-directed antibody PG16. Nat Struct Mol Biol. 20:804–813. Pantophlet R, Burton DR. 2006. GP120: Target for neutralizing HIV-1 antibodies. Annu Rev Immunol. 24:739–769. Patel DS, Pendrill R, Mallajosyula SS, Widmalm G, MacKerell AD Jr. 2014. Conformational properties of alpha- or beta-(1–>6)-linked oligosaccharides: Hamiltonian replica exchange MD simulations and NMR experiments. J Phys Chem B. 118:2851–2871. Pejchal R, Doores KJ, Walker LM, Khayat R, Huang PS, Wang SK, Stanfield RL, Julien JP, Ramos A, Crispin M, et al. 2011. A potent and

Y Qi et al. broad neutralizing antibody recognizes and penetrates the HIV glycan shield. Science. 334:1097–1103. Phillips JC, Braun R, Wang W, Gumbart J, Tajkhorshid E, Villa E, Chipot C, Skeel RD, Kale L, Schulten K. 2005. Scalable molecular dynamics with NAMD. J Comput Chem. 26:1781–1802. Pritchard LK, Vasiljevic S, Ozorowski G, Seabright GE, Cupo A, Ringe R, Kim HJ, Sanders RW, Doores KJ, Burton DR, et al. 2015. Structural constraints determine the glycosylation of HIV-1 envelope trimers. Cell Rep. 11:1604–1613. Qi Y, Im W. 2013. Quantification of drive-response relationships between residues during protein folding. J Chem Theory Comput. 9:3799–3805. Reitter JN, Means RE, Desrosiers RC. 1998. A role for carbohydrates in immune evasion in AIDS. Nat Med. 4:679–684. Ryckaert J-P, Ciccotti G, Berendsen HJC. 1977. Numerical integration of the Cartesian equations of motion of a system with constraints: Molecular dynamics of n-alkanes. J Comput Phys. 23:327–341. Schreiber T. 2000. Measuring information transfer. Phys Rev Lett. 85:461–464. Schrodinger, LLC. 2010. The PyMOL Molecular Graphics System, Version 1.3r1. Shan Y, Klepeis JL, Eastwood MP, Dror RO, Shaw DE. 2005. Gaussian split Ewald: A fast Ewald mesh method for molecular simulation. J Chem Phys. 122:54101. Shaw DE, Deneroff MM, Dror RO, Kuskin JS, Larson RH, Salmon JK, Young C, Batson B, Bowers KJ, Chao JC, et al. 2008. Anton, a specialpurpose machine for molecular dynamics simulation. Communications of the ACM. 51:91–97. Stamatatos L. 2012. HIV vaccine design: The neutralizing antibody conundrum. Curr Opin Immunol. 24:316–323. Steinbach PJ, Brooks BR. 1994. New spherical-cutoff methods for long-range forces in macromolecular simulation. J Comput Chem. 15:667–683. Trkola A, Purtscher M, Muster T, Ballaun C, Buchacher A, Sullivan N, Srinivasan K, Sodroski J, Moore JP, Katinger H. 1996. Human monoclonal antibody 2G12 defines a distinctive neutralization epitope on the gp120 glycoprotein of human immunodeficiency virus type 1. J Virol. 70:1100–1108. Walker LM, Huber M, Doores KJ, Falkowska E, Pejchal R, Julien JP, Wang SK, Ramos A, Chan-Hui PY, Moyle M, et al. 2011. Broad neutralization coverage of HIV by multiple highly potent antibodies. Nature. 477:466–470. Walker LM, Phogat SK, Chan-Hui PY, Wagner D, Phung P, Goss JL, Wrin T, Simek MD, Fling S, Mitcham JL, et al. 2009. Broad and potent neutralizing antibodies from an African donor reveal a new HIV-1 vaccine target. Science. 326:285–289. Wei X, Decker JM, Wang S, Hui H, Kappes JC, Wu X, Salazar-Gonzalez JF, Salazar MG, Kilby JM, Saag MS, et al. 2003. Antibody neutralization and escape by HIV-1. Nature. 422:307–312. Wood NT, Fadda E, Davis R, Grant OC, Martin JC, Woods RJ, Travers SA. 2013. The influence of N-linked glycans on the molecular dynamics of the HIV-1 gp120 V3 loop. PLoS One. 8:e80301. Wu X, Yang ZY, Li Y, Hogerkorp CM, Schief WR, Seaman MS, Zhou T, Schmidt SD, Wu L, Xu L, et al. 2010. Rational design of envelope identifies broadly neutralizing human monoclonal antibodies to HIV-1. Science. 329:856–861. Wyatt R, Sodroski J. 1998. The HIV-1 envelope glycoproteins: Fusogens, antigens, and immunogens. Science. 280:1884–1888.