Small Conformational Changes Underlie Evolution of Resistance to NNRTI in HIV Reverse Transcriptase

Abstract

Despite achieving considerable success in reducing the number of fatalities due to acquired immunodeficiency syndrome, emergence of resistance against the reverse transcriptase (RT) inhibitor drugs remains one of the biggest challenges of the human immunodeficiency virus antiretroviral therapy (ART). Non-nucleoside reverse transcriptase inhibitors (NNRTIs) form a large class of drugs and a crucial component of ART. In NNRTIs, even a single resistance mutation is known to make the drugs completely ineffective. Additionally, several inhibitor-bound RTs with single resistance mutations do not exhibit any significant variations in their three-dimensional structures compared with the inhibitor-bound RT but completely nullify their inhibitory functions. This makes understanding the structural mechanism of these resistance mutations crucial for drug development. Here, we study several single resistance mutations in the allosteric inhibitor (nevirapine)-bound RT to analyze the mechanism of small structural changes leading to these large functional effects. In this study, we have shown that in absence of significant conformational variations in the inhibitor-bound wild-type RT and RT with single resistance mutations, the protein contact network analysis of their static structures, along with molecular dynamics simulations, can be a useful approach to understand the functional effect of small local conformational variations. The simple network analysis exposes the localized contact changes that lead to global rearrangement in the communication pattern within RT. Furthermore, these conformational changes have implications on the overall dynamics of RT. Using various measures, we show that a single resistance mutation can change the network structure and dynamics of RT to behave more like unbound RT, even in the presence of the inhibitor. This combined coarse-grained contact network and molecular dynamics approach promises to be a useful tool to analyze structure-function studies of proteins that show large functional changes with negligible variations in their overall conformation.

Significance

Emergence of resistance against antiretroviral drugs targeting human immunodeficiency virus-1 remains one of the biggest challenges in the treatment of acquired immunodeficiency syndrome. This is complicated by the fact that conformational changes in reverse transcriptase, resulting from binding of drugs and the emergence of resistance mutations, are subtle. Here, we elaborate and characterize these small conformational changes using an approach based on network analysis and molecular dynamics. We illustrate that these small conformational changes drastically alter the communication paths between different subdomains and perturb the underlying community structure. Importantly, we clearly show reversal of perturbations in network parameters, community structure, and dynamics due to presence of resistance mutations, as compared with the wild-type reverse transcriptase bound to drug, thereby distinctly delineating the structural basis of drug resistance.

Introduction

Human immunodeficiency virus-1 (HIV-1) has been the underlying cause for the devastating pandemic of acquired immunodeficiency syndrome (AIDS). HIV, being a retrovirus, has RNA as its genome and needs to undergo the process of reverse transcription for replication and integration into the host genome and to produce the viral proteins to make copies of itself inside the host cells (1). Reverse transcription is carried out by the enzyme reverse transcriptase (RT), which is a polymerase encoded by the viral gene pol, and catalyzes the polymerization of DNA using the RNA template. HIV-1 RT is a heterodimeric enzyme composed of two chains—p66, having 560 amino acids, and p51, having 440 amino acids, derived by the cleavage of Gag-Pol polyprotein (2). It is a structurally remarkable metamorphic protein with two chains having exactly the same amino acid sequence but different three-dimensional structures (3). RT has two enzymatic active sites: a “polymerase active site” for DNA polymerization using both DNA and RNA as template, and an “RNase H active site” for hydrolysis of RNA in the DNA-RNA hybrid. Both the active sites are present in the p66 subunit of heterodimer. The p66 subunit of HIV-1 RT has a canonical right-handed shaped structure (Fig. 1, A and C, Protein Data Bank, PDB: 1VRT). It is composed of “thumb” (salmon color), “finger” (purple), “palm” (olive), “connection” (cyan), and “RNase H” (yellow) domains. The polymerase active site, where the viral DNA binds, is present in the palm domain.

Figure 1.

Structural alignment and residuewise RMSD for 98 crystal structures of HIV-1 RT. (A) Alignment of all 98 structures is shown. Different subdomains are shown in different colors: finger domain, purple; palm domain, olive; thumb domain, salmon; connection domain, cyan; RNase H domain, yellow; and p51 chain, gray. (B) Residuewise RMSD of Cα atoms of common residues in all the structures is shown. The straight line represents the average RMSD over all the residues in all structures. The shaded regions at the top of the plot correspond to residues belonging to different subdomains. (C) The structure of HIV-1 RT (PDB: 1VRT) is shown. Domains are depicted in the same color code as in (A). The NNRTI (nevirapine) binding site is highlighted with a black box. (D) Nevirapine (dark red ball and stick model) bound to RT (black box in C) is shown. Inhibitor binding residues are shown in a pink stick model. Resistance mutation sites are shown in an orange ball and stick model. (E) Projection of all the crystal structures on the PC1 versus PC2 space. The four clusters are as follows: cluster 1, navy; cluster 2, maroon; cluster 3, orange; and cluster 4, green. (F) Superposition of representative structures from four clusters, colored as per the clusters to which they belong. To see this figure in color, go online.

Being one of the most important drug targets in the antiretroviral therapy (ART) against HIV-1, multiple drugs have been developed against RT (4,5), and possibilities of developing better inhibitors of RT are still being explored (6,7). Nearly half of the anti-HIV drugs target the polymerase activity of RT (2). Nucleoside reverse transcriptase inhibitors (NRTIs) are structurally diverse analogs of naturally occurring nucleosides. These participate in the polymerase reaction and block the reaction from proceeding, thus directly inhibiting the replication.

A different class of drugs, the non-nucleoside reverse transcriptase inhibitors (NNRTIs), on the other hand, bind to an allosteric site and inhibit the polymerase activity of RT noncompetitively. Although these drugs have proven to be highly effective, continued use of the drugs leads to development of resistance in the virus through mutations in RT, thereby making the drugs ineffective. Several drug resistance mutations have been clinically identified, and by understanding these drug resistance mutations better, drugs and/or cocktails of drugs with higher genetic barrier to resistance have been developed (8). Being allosteric inhibitors, NNRTIs bind to nonconserved sites at some distance from the polymerase active site (Fig. 1 D). Consequently, the high mutation rate of RT accompanied by the selection pressure exerted by the drug leads to development of a resistance mutation in the RT against the NNRTIs. Understanding the structural basis of the mechanism of drug resistance emanating for these mutations is crucial for both developing better drugs and to improve the clinical outcomes of the ART. Based on multiple computational and experimental studies, different mechanisms have been proposed to effect resistance mutations in RT (2,4,5).

Some of the commonly observed resistance mutations against NNRTIs are K103N, L100I, K101E, Y181C, Y188C, G190A, and E138K (Fig. 1 D). Out of these, K103N and Y181C are the most common resistance mutations (5). K103, K101, and E138 are the boundary residues that line the rim of the NNRTI binding pocket and usually do not interact directly with the inhibitor (5). The mechanism of resistance mutations at these sites is still poorly understood in terms of the protein structure and conformational changes associated with mutations. The mechanism for other mutations involving residues that interact directly with NNRTI seem mostly to be derived based on either the loss of crucial interactions with the inhibitor (Y181C, Y188C) (9,10) or steric clashes introduced by the mutations (L100I).

Based on hundreds of crystal structures determined in different complex states and several biochemical as well as biophysical experiments, it is now widely accepted that the HIV-1 RT exhibits remarkable conformational heterogeneity and that this conformational dynamics plays a crucial role in its function (11, 12, 13, 14, 15, 16). Presence of multiple crystal structures provides an opportunity to understand the conformational heterogeneity of the protein as exhibited in different states (11,17). A previous study analyzed 52 crystal structures of HIV-1 RT in various complex states and found that the RT structures clustered into three groups based on the structural and dynamic information encoded in those structures (12). Apart from this, several studies have been performed to understand the role of dynamics in inhibition and resistance mutation (14,18, 19, 20, 21). However, one of the most interesting and unusual confounding issues has been that the binding of NNRTI, as well as emergence of resistance mutation, is accompanied by small conformational changes across the RT enzyme (as evidenced by their insignificant differences in RMSD), which leads to a remarkable change in the functional behavior of the enzyme. Thus, there is still a lack of clarity and consensus on the mechanism of inhibition and resistance mutation in RT. In this study, we propose a simple yet useful method using a combination of static graphs from the crystal structures (protein contact networks) and molecular dynamics simulations to address this problem. We use this combined framework to investigate the small conformational changes, accompanied by binding of NNRTI and occurrence of resistance mutation in several RT structures, to elucidate their mode of action.

Protein contact network (PCN) is a relatively new coarse-grained representation of protein structures that has been used to understand several biological phenomena (22,23). This method is especially useful in identifying small contact changes in protein structures that underlie large functional changes, which are otherwise not easily detectable (24, 25, 26). However, the crystal structures and the PCNs derived from them provide a static representation. To comprehend the structure-function relationship in proteins, molecular dynamics simulations are crucial for understanding their solution dynamics (27). It forms a very important computational approach to decipher the structural mechanisms underlying the protein’s function.

In this study, keeping in mind that RT structures (unliganded or apo, DNA bound, inhibitor bound, and with resistance mutations) have conformational variations, we have first analyzed 98 crystal structures of HIV-1 reverse transcriptase to gain an understanding of the extent of their conformational heterogeneity. Then, we chose a smaller set of structures, composed of ApoRT, drug-bound wild-type RT, and drug-bound RT with single resistance mutations, with a maximum overall conformational variation of 2.26 Å. Using a combined network and dynamics approach, we have clearly elucidated the small structural changes that could accompany such large and important alterations in functions in the multiple RT structures with single resistance mutations.

Methods

Data set

We selected 98 structures of HIV-1 reverse transcriptase with resolution <3 Å (listed in Table S3) from the PDB (28). Most of the structures had nonterminal missing residues. For residuewise comparison between different structures, common resolved residues in all the structures were determined. For analysis of single resistance mutations, a smaller data set was selected. It is composed of crystal structures of HIV-1 reverse transcriptase in unbound state (PDB: 1RTJ) and nevirapine-bound state (PDB: 1VRT) and with single resistance mutations with the nevirapine-bound state (PDB: 1FKP, 1JLB, 1JLF, 1S1U, 2HND, and 2HNY). The details of the structures used in this analysis are given in Table S1. The 1FKP (with mutation K103N) has been chosen as an example to show many results because it is reported to be one of the most common resistance mutations of HIV-1 RT (5).

Structural alignment and residuewise root mean-square deviation

The Cα atoms of common residues in all the crystal structures were aligned using Biopython (29,30) and residuewise root mean-square deviation (RMSD) calculated using the following equation:

$$RMS{D}_i=\sqrt{\underset{n}{\frac{1}{n}\sum {\delta }_i^2}},$$

where RMSD_i is the RMSD for the Cα of the i-th residue, and δ_i is the distance between the Cα atoms of each of the n structures and the reference structure.

Protein contact network generation

PCN is a graph theoretic representation of the three-dimensional structure of a protein in terms of a set of interconnected nodes and edges. The protein structure as determined from X-ray crystallography or NMR is converted into a network in which nodes are the Cα atoms of the amino acid residues, and the edges or links between nodes are defined based on the cutoff distance between two Cα atoms (25,26,31).

For PCN construction, distances between pairs of all Cα atoms of the protein with N residues are calculated, giving an N X N distance matrix D. From this distance matrix, an adjacency matrix (A) is generated based on following criteria:

$${\mathrm{A}}_{\mathrm{i}\mathrm{j}}=\mathrm{1}\mathrm{i}\mathrm{f}{\mathrm{D}}_{\mathrm{i}\mathrm{j}}<\mathrm{7}\mathrm{\AA }$$

$${\mathrm{A}}_{\mathrm{i}\mathrm{j}}=\mathrm{0}\mathrm{i}\mathrm{f}{\mathrm{D}}_{\mathrm{i}\mathrm{j}}\ge \mathrm{7}\mathrm{\AA }$$

$${\mathrm{A}}_{\mathrm{i}\mathrm{j}}=\mathrm{0}\mathrm{i}\mathrm{f}\mathrm{i}=\mathrm{j},$$

where D_ij is the distance between the i-th and j-th residue of the protein. The cutoff of 7 Å to define contact between residues in coarse-grained representation has been found to be appropriate in previous studies (25,26,32). The major results in this study have been checked to hold for both 6.5- and 7.5-Å cutoffs. This adjacency matrix (A) represents an unweighted, undirected network that we refer to as the “Protein Contact Network.” These are coarse-grained networks, as all other atoms and side chains in each amino acid are collapsed onto the spatial position of its Cα atom and taken as a node. Many questions in protein science require fine-scale networks in which all atoms and side chains are considered (33,34). The PCN construction was done using in-house python scripts, which can be made available freely upon request for academic purposes.

Network analysis

Network parameters provide metrics that summarize the properties of the topology of the network at both local (node) and global (whole-network) levels. Network parameters like degree, betweenness centrality, eigenvector centrality (EVC), and shortest path of the PCNs were calculated using igraph package (35) of R (36), as described previously (25).

Communities were detected in each of the PCNs using the fast-greedy algorithm (37) as implemented in NetworkX (38). Cytoscape 3.7 (39) was used for visualization of networks and communities.

The residues used for calculating the shortest paths between different domains were found based on the interaction of nucleotide and DNA with RT as determined in previous crystal structure (40). These residues are W24, P25, F61, L74, and Q151 from finger domain; I94, D110, V111, D113, Y115, F116, P157, Y183, M184, D185, D186, M230, and G231 from Palm domain; and Q255, C258, K259, G262, K263, N265, W266, R284, G285, T286, and L289 from thumb domain.

Molecular dynamics simulations

Crystal structures obtained from the PDB were used as starting models for the MD simulations (Table S2). Missing regions in the structures were modeled using Modeler (41). Crystallographic waters were removed, and only the protein and ligands (in inhibitor-bound and resistance mutation simulations) were retained in the final model. These were solvated in a dodecahedron box filled with TIP3P water molecules. The systems were neutralized with counterions. The final system was minimized until the maximum force on any atom reached 1000 kJ/mol/nm. The minimized system was first subjected to NVT equilibration for 500 ps, followed by NPT equilibration for 500 ps. The systems were maintained at 300 K using modified Berendson thermostat (42), and pressure was maintained at 1 bar using Parinello-Rahman barostat (43). After equilibration, the unbound RT and wild-type inhibitor-bound RT were simulated for 100 ns. The resistance mutations carrying RT with bound inhibitors were simulated for 50 ns each. The MD simulations were performed using GROMACS 4.5 (44).

Principal component analysis

Principal component analysis (PCA) was performed on the coordinates of the trajectories by diagonalizing the covariance matrix using the g_covar function in GROMACS. The trajectories were projected on the eigenvectors using the g_anaeig function. PCA was performed on the coordinates of the backbone atoms. The coordinates of the same residues from individual MD simulations were then projected on the first two principal components.

Results and Discussion

Conformational heterogeneity in crystal structures of RT and its complexes

Owing to its importance as a target in the antiretroviral drug therapy of HIV-1, hundreds of crystal structures of HIV-1 RT have been determined in different complex states and with different resistance mutation. We analyzed this large data set to gain insight into the conformational heterogeneity of HIV-1 RT as represented in the ensemble of crystal structures. Ninety-eight crystal structures were selected for cross-structure comparisons after correcting for missing residues (see Methods). The structural alignment of these proteins shows regions of high flexibility in the HIV-1 RT (Fig. 1 A). To quantify the flexibility observed in different regions of the protein across different structures, residuewise RMSD was calculated (Fig. 1 B). The entire thumb domain shows high RMSD in this data set, suggesting remarkable conformational heterogeneity of this domain in multiple crystal structures. The motions of the thumb and finger subdomains have been associated with the functional motions of the reverse transcriptase (20), explaining the conformational heterogeneity in this region in WT, inhibitor-bound, and resistance mutants of RT. Some residues in the palm domain (residue numbers 111–116, 185, 223, 230–231) also show high RMSD. These residues correspond to the region near the inhibitor binding site, which might be because of different complex and activity states represented in all the crystal structures (Fig. 1, C and D).

To understand the overall conformational variability across different crystal structures, we performed PCA on the ensemble of 98 crystal structures. The first and second PCs explain ∼80% (41.2 and 37.3% each) of the variance of the RT conformational heterogeneity. In the PC1-PC2 space, the crystal structures show four distinct clusters (Fig. 1 E: cluster 1 in navy, cluster 2 in maroon, cluster 3 in orange, and cluster 4 in green). Table S3 gives the PDB identifiers of the structures belonging to each of the different clusters. Clusters 1, 2, and 3 are separated along PC2 and show negative values for PC1 (Fig. 1 E; Fig. S1, A–C). Cluster 4 (shown in green in Fig. 1, E and F; Document S1. Figs. S1–S9 and Tables S1–S9, Document S2. Article plus Supporting Material) is separated from these three clusters along PC1 and shows positive values for PC1. This cluster comprises crystal structures of RT with DNA bound to them. This suggests that the conformation of RT in the DNA-bound form is considerably different from the other conformations observed in the crystal structures. The thumb and finger subdomain of RT are much closer in all the structures in this cluster and take up a closed conformation (Document S1. Figs. S1–S9 and Tables S1–S9, Document S2. Article plus Supporting Material). This closed conformation differs markedly from the DNA-unbound form of RT and is evident in the PC1-PC2 plot (Fig. 1 E). The rest of the structures that cluster in three different groups separated along PC2 have unbound RT and inhibitor-bound RT, as well as those with resistance mutations present in them. The motion along PC2 shows slight rotation and bending of the thumb and finger subdomains. These structures have little overall conformational heterogeneity and show variation primarily in the slightly different conformations of the thumb and finger subdomains.

The above analysis clearly indicates that, though there are significant differences in the DNA-unbound (clusters 1–3) and DNA-bound (cluster 4) states of the RT, it is difficult to recognize the subtle structural differences between the unbound RT, inhibitor-bound RT, and the inhibitor-bound RT structures bearing resistance mutations, despite their having very significant functional variations. To gain insight into these subtle structural variations and, consequently, to understand the mechanism of inhibition and resistance mutation, we focused on a smaller set of structures for further analysis.

Nevirapine (NVP) is one of the major NNRTIs used for the first line of treatment against HIV-1. However, the use of NVP is often followed by the development of resistance mutations in the HIV-1 genome, making this drug ineffective. In the case of NVP, which is an allosteric inhibitor, a single point mutation near the inhibitor binding site is often enough to cause resistance, without any effect on NVP binding (45,46). For our study, we have chosen a subset of NVP-bound RT structures with only single mutations that specifically exhibit NVP resistance. To study the effect of NVP binding and single point mutations causing resistance to the NVP, the eight different structures chosen are 1) unbound (henceforth referred as ApoRT), 2) inhibitor (NVP) bound (henceforth referred to as NvpWTRT), and 3) six structures with single resistance mutations (RMs) (henceforth referred to according to their resistance mutation, i.e., NvpK103NRT, etc.). Among these, ApoRT (PDB: 1RTJ) and three structures with single resistance mutations (PDB: 1FKP, 1JLB, 1JLF) belong to cluster 1, whereas the nevirapine-bound wild-type RT (PDB: 1VRT) and the other three structures with resistance mutations (PDB: 1S1U, 2HND, 2HNY) belong to cluster 2 in the PC space (Fig. 1 E). To clearly identify the effect of the inhibitor binding and resistance mutations on the overall structures, we analyzed these eight crystal structures of HIV-1 RT in three different states (unbound/WT, inhibitor bound, and inhibitor bound with resistance mutations) using network theory and molecular dynamics simulations. The details of this smaller set of structures used in the analysis have been provided in Table S1. Because inhibitor is present in all the resistance mutation structures analyzed here, it indicates that the mutation is not disabling drug binding but is inducing conformational changes to enable DNA binding and gain of function by RT. We elaborate this using network analysis below.

Network analysis identifies subtle contact pattern changes upon inhibitor binding and resistance mutations in RT

The structural comparison of the chosen eight crystal structures reveals that the overall conformation between these different states is very similar, with RMSD ranging between 0.26 and 2.26 Å with an average of 1.4 Å (Fig. S2 A; Table S4). This suggests that the changes in inhibition of polymerase function of RT due to drug binding at an allosteric site, or single mutations releasing this inhibition even when the drug is bound, are not due to large conformational changes but are caused by small local changes. It is also important to note that all six RM structures, chosen from the different clusters, have the single mutations quite close to each other on the RT (Fig. S2 B). Thus, we modeled the eight crystal structures as protein contact networks (Methods) and compared the local and global network properties and their contact patterns.

Contact analysis reveals their distributed reorganization on binding of inhibitor and their reversal in resistance mutations

Binding of the inhibitor causes several small conformational changes in the RT structure. In terms of contact network, these small conformational changes are represented as gain and loss of contacts between the nodes (residues) due to nevirapine binding and resistance mutations. Compared with the PCN of ApoRT, in the PCN of NvpWTRT, 148 contacts were lost, and 123 new contacts were made. The contacts were lost throughout the structure including chain B of the dimer. Some of these contacts might be playing an important role in the function of the enzyme, and consequently, their loss due to allosteric nevirapine binding could be responsible for the inhibition of RT. To identify these contacts, we compared the PCN of NvpWTRT (i.e., PCN of PDB: 1VRT) with the PCNs of RT with different resistance mutations. The total number of contacts gained and lost in PCNs of RM (compared with the PCN of NvpWTRT) ranged between 58 and 129 (Table S5), indicating differential regulation of the loss of inhibition due to different allosteric site mutations (even though all mutations were close to the nevirapine binding site). This indicates that the resistance mutations may be using different paths to yield the same effect on its function—the DNA binding to RT. We further analyzed the domainwise location of these contacts in the structure and classified the changes happening in contacts at intradomain and interdomain levels (Fig. 2).

Figure 2.

Contacts lost and gained in different RM structures as compared with NvpWTRT (PDB: 1VRT). The fraction of intradomain (A) contacts lost and (B) contacts gained is shown. The fraction of interdomain (C) contacts lost and (D) contacts gained is shown. To see this figure in color, go online.

In all the RM structures, most intradomain contacts are lost within the palm (where the viral DNA binds) and, for some, in the connection domains, whereas the contacts are gained almost uniformly within all the domains except in the thumb domain (Fig. 2, A and B). The maximum number of interdomain contacts is lost between palm and thumb, followed by finger and palm. Conversely, most contacts are gained between finger-palm and RNase-H-chain-B (Fig. 2, C and D). This analysis also shows that in all of the six different RMs, most of the contact rearrangements are intradomain, suggesting local changes, with only a few interdomain long-range changes in the contact patterns. It is known that the palm, thumb, and finger domains in RT are primarily involved in DNA binding. Thus, our contact change analysis uncovers the contacts in specific regions that play a significant role in the loss of inhibition due to these single mutations. For a virus such as HIV, which has a very high mutation rate, this strategy of reorganizing local contacts seems most useful in the scenario of selection under widespread drug usage.

Change in network properties due to inhibitor binding shows reversal in RMs

Network properties are descriptors of the topology and connectivity patterns of the whole graph. Changes in network connectivity are reflected in the node and edge specific network parameters. Thus, these parameters can be useful indicators of both local and global (average) properties of a network. Because PCNs are constructed based on the protein structure coordinates, the network parameters describe specific structural aspects of the protein. To study the small variations in the NvpWTRT and RM PCNs on drug binding and due to resistance mutations, we studied the network parameters both at the global and local node level. As expected, the average (global) network parameters show no difference among different PCNs (Table S6) because the overall conformation of the proteins is highly similar and, in concordance with the previous analysis, the changes in the PCNs are mostly small local contact changes.

To understand the role of local changes in the contact patterns, we compared the node-level network parameters for all the PCNs. The difference in the degree between all nodes of ApoRT and NvpWTRT shows that most of the changes are small (Fig. S3 A). However, there were a few residues that showed large changes in degree upon binding of the inhibitor. However, most importantly, in the RM networks, the degree of some of these nodes or the nearby nodes changed back to the ApoRT network level. Similar results are also seen for the network parameter betweenness centrality (Fig. S3 B), which is an indicator of the role of specific nodes in transmitting information in different parts of the network. An important parameter of a network is the EVC. The EVC of a node describes the importance of the node in the network with respect to its neighbors based on their connectivity pattern. EVC of the residues shows variations in different states of the RT (Fig. S4)—the ApoRT, NvpWTRT, and the six RMs. In the case of ApoRT, the residues in the RNase H domain of chain A show high EVC, whereas in the inhibitor-bound form, the residues from connection and finger domains, with some residues from the palm domain, show high EVC. Interestingly, in the case of RM networks, the EVC profiles show variability, but they can be grouped based on some patterns: 1) NvpY181CRT and NvpY188CRT, 2) NvpK101ERT and NvpE138KRT, and 3) NvpK103NRT and NvpL100IRT show very similar EVC profiles. This indicates that EVC is sensitive to small changes in the connectivity patterns arising due to different mutations.

Despite these variations in EVC in individual RMs, all of them show reversal when the differences are compared (Fig. 3, shown for three representative RMs from the three groups). The EVC of the residues in the finger subdomain (shaded purple) remains unchanged; however, it increases for several residues in the palm and thumb subdomains (shaded olive and salmon, respectively). Thus, the connectivity patterns in all the resistance mutation PCNs alter the node centralities (in magenta) on the opposite directions, restoring it closer to the unbound RT network.

Figure 3.

Changes in eigenvector centrality in PCNs of inhibitor-bound RT and RM: (A) NvpY181C RT; (B) K103N RT; (C) NVPK101E RT. Differences between NvpWTRT and ApoRT are shown in cyan. Differences between RM and NvpWTRT are shown in magenta. To see this figure in color, go online.

For allosterically bound inhibitors, such as the NNRTIs, change in connectivity patterns due to inhibitor binding can strongly affect communication between different regions of the protein and, consequently, its function. Polymerization reaction carried out by the RT involves several conformational changes and persistent communication between different domains of the p66 subunit of the protein (13,14). Binding of the double-stranded nucleic acid template in the thumb subdomain changes the conformation from a closed to open state. When dNTP approaches, the finger subdomain closes upon it, bringing the polymerase active site, 3′ OH of the primer, and the phosphate group of dNTP together for the reaction (2). Such a mechanism of polymerization entails the effective communication between different regions of the enzyme, and perturbation of this communication may lead to inhibition of the enzyme. Perturbation in communication between different regions of the protein should be reflected in the changes in the network parameter “shortest path.” In the simple coarse-grained, unweighted, undirected network description of these proteins, shortest path would simply measure the smallest number of nodes to be traversed to reach another node. Also, nevirapine being an allosteric inhibitor, the functionally relevant changes in the shortest paths due to inhibitor binding and the resistance mutations could be distributed in the entire protein. We wished to see whether, even in this coarse-grained description, the RMs showed reversal in this property when compared with the inhibitor-bound PCN.

We investigated the pathways of communication between different regions of the RT by calculating the shortest paths between all pairs of nodes (residues) in the corresponding PCNs of ApoRT, NvpWTRT, and one of the RMs carrying the resistance mutation K103N (NvpK103NRT). Fig. 4, A and B shows the heat maps of “differences in shortest paths” between all residue pairs in RT in different states—between ApoRT and NvpWTRT (Fig. 4 A) and that between the NvpK103NRT and NvpWTRT (Fig. 4 B).

Figure 4.

Changes in the shortest path between different regions of RT in different states. Shown are shortest path differences between (A) unbound and inhibitor (Nvp)-bound RT and (B) RM K103N and inhibitor-bound RT. Different domains of the protein have been depicted as shaded regions at the bottom of the plot. Shown are the finger domain, purple; palm domain, green; thumb domain, red; connection domain, cyan; RNase H domain, yellow; and chain B, gray. To see this figure in color, go online.

The heat maps in Fig. 4 indicate that, even though most residue pairs show small changes in their shortest paths on inhibitor binding, as compared with ApoRT, there are certain regions in which the changes show a clear reversal in pattern on comparison of Fig. 4, A and B. The residues 250–290 (belonging to the thumb domain in chain A) and 1–250 (belonging to the finger and palm domains in chain A), show a decrease in shortest paths upon inhibitor binding (Fig. 4 A). This suggests that the shortest paths between the thumb and the finger and palm domains get perturbed upon binding of the allosteric inhibitor nevirapine. However, in the case of the resistance mutation K103N (NvpK103NRT), the residue pairs in the same regions of the PCN show an increase in shortest paths (Fig. 4 B). This feature is seen in all of the six RM PCNs studied. This is also supported by the fact that a higher fraction of interdomain contacts is lost than gained between palm and thumb domains in the RM structures (Fig. 2, C and D).

We further explored the changes in shortest paths by mapping these paths between functionally relevant residues (see Methods) from finger, palm, and thumb domains (Fig. S5; Table S7). The average path length between residues of finger and thumb domains, as well as palm and thumb domain, decreases in inhibitor-bound RT and shows an increase in the resistance mutation PCNs. The actual path between residue F61 in finger domain and Q255 in thumb domain shows major rearrangement in inhibitor-bound state and resistance mutation PCN (Fig. S5). Most importantly, binding of inhibitor causes the shortest paths to bypass chain B residues, whereas in the case of resistance mutation structure, chain B residues are again utilized in the shortest paths. We also found that binding of the inhibitor leads to loss of two interchain contacts (181A-139B and 182A-138B) involving functionally relevant residues, and one (181A-139B) of these contacts is gained back in the resistance mutation PCN of NvpK103NRT. The communication between p51 and p66 has been shown to be important in previous computational and experimental studies (14,47). The loss and gain of contacts at the p66 and p51 interface is reminiscent of this communication between the two chains and their role in functional changes.

Thus, these results, obtained by simple network analysis, clearly expose the involvement of perturbation in the communication between the thumb and the finger and palm domains due to inhibitor binding and its reversal due to mutations at an allosteric site leading to resistance to the drug action.

Contact changes lead to rearrangement of community structures of RT in different states

Community structure is an important property of networks that indicates the clustering of nodes in tightly connected groups (48). Proteins are organized in multiple scales, and their modular architecture has been implicated to have functional relevance (49,50). Perturbations in local contacts have been previously known to cause changes in the overall community structure of the PCNs in the absence of any significant conformational changes (25,51). We calculated the community structure of each of the eight PCNs and compared them to understand the role of small variations in contacts due to inhibitor binding and resistance mutations on the community memberships of residues in unbound RT. The first column in Fig. 5, A, C, and E shows all the PCNs in the three states of RT, which get divided into five to six large and a few small communities. The second column (Fig. 5, B, D, and F) shows the groups of residues in the RT structures colored according to the major communities (details in Table S8). In the case of the ApoRT or unbound RT network, the residues from finger, thumb, and palm domains primarily form two large communities (C2 and C3 in Fig. 5 A) with almost all PAS residues (shown as black nodes in Fig. 5) forming part of a single community (C3 in Fig. 5 A and the green residues in Fig. 5 B). The IB residues are divided into three communities (C1, C2, and C3 in Fig. 5 A). The community structure of inhibitor-bound RT network changed radically (Fig. 5 C). In this, the residues of finger, thumb, and palm domains now get distributed into three communities (C1, C4, and C5 in Fig. 5 C). The PAS residues get distributed into two communities: C1 contains residues belonging to the palm domain (94, 110–111, 113, 183–186, 230–231), and C5 contains residues belonging primarily to the finger domain (24–25, 61, 74, 151) in addition to three from the palm domain (115–116, 157). Interestingly, now all the IB residues in the inhibitor-bound RT network come together in the single community (C1 in Fig. 5 C). This clearly shows that change in contact pattern upon binding of inhibitor perturbs the communication between functional residues in different domains of the protein (Fig. 5, C and D).

Figure 5.

Community structures of HIV-1 RT PCNs. Community structure and mapping of residues in different communities on the RT structure of (A and B) unbound, (C and D) inhibitor (NVP)-bound, and (E and F) with resistance mutation K103N. Colors of the nodes in all the networks are according to community membership in unbound network (A). The PAS residues are shown as black nodes in all community networks. To see this figure in color, go online.

In the case of the K103N resistance mutation, the whole network is divided into five large and a few smaller communities (Fig. 5 E). Although the nodes of the finger, palm, and thumb domains are still distributed into three communities, the community comprising the thumb domain nodes (C2 in Fig. 5 E) extends to include nodes of the palm domain, essentially making this community very similar to the C2 in the unbound RT network (C2 in Fig. 5, A, B, and F). Rearrangement of communities also leads to most of the PAS residues joining a large community, comprising primarily the finger domain residues (C5 in Fig. 5 E), and a smaller community, comprising palm domain nodes (C6 in Fig. 5 E). The IB residues, in this RM network, also get distributed into three different communities (C1, C2, and C6 in Fig. 5 E), making it similar to the unbound RT network. The community structures of the other resistance mutation networks also show similar rearrangements, making them similar to the unbound RT network (Fig. S6; Table S8).

These remarkable changes in the distribution of PAS and IB residues between different communities in ApoRT, NvpWTRT, and the RM networks clearly indicate that allosteric inhibitor binding and resistance mutations, which did not yield significant conformational change in the proteins, caused small contact changes leading to perturbation in the communication within the networks, which finally resulted in complete change in function (inhibition and its release). Both the network properties and community structure analysis of the PCNs, described above, clearly depict that the resistance mutation networks show remarkable similarities with the unbound RT network despite the RM structures carrying the bound inhibitor. Based on the network analysis of these very similar static crystal structures of proteins, we have shown that the small conformational changes in the RT structures lead to inhibition of RT upon binding of inhibitor, and further small changes accompanied by the mutations lead to a reversal effect in making the inhibitor ineffective, which leads to resistance to NNRTI.

NNRTI binding and resistance mutations significantly change the RT dynamics: MD simulations

Protein dynamics plays a very important role in protein function. It assumes special significance in the case of RT and its different states, in which conformational changes are very small. HIV-1 RT undergoes functional motions during the polymerization reaction, which primarily involves movement of the thumb and finger subdomains that grasp the DNA during HIV genome replication. To understand the effect of local conformational changes caused by binding of inhibitor and the resistance mutations on the overall dynamics of RT, we performed MD simulations for ApoRT, nevirapine-bound RT (NvpWTRT), and four nevirapine-bound RTs with different single resistance mutations in each (RMs: NvpK103NRT, NvpY181CRT, NvpY188CRT, NvpE138KRT).

Root mean-square fluctuations (RMSFs) of the Cα atoms of each residue were calculated for all the proteins and are shown in Fig. 6. In chain A, higher flexibility is observed for residues belonging to the thumb (237–318) and finger (1–85, 118–155) subdomains. Most of the residues of the palm, connection, and RNase H domains show low flexibility (Fig. 6 A). To understand the effect of inhibitor binding and resistance mutation on the dynamics of RT, we compared the RMSF of Cα atoms among all states of RT. The binding in NVP causes suppression in the overall flexibility of RT (Fig. S7, cyan bars). This suppression is more significant in certain regions of the protein. Similar quenching in the flexibility upon binding of NVP was also reported in an earlier experimental study using hydrogen-deuterium exchange mass spectrometry (13). Many residues showing a decrease in flexibility were found to be in the regions identified in the previous study, corroborating our observations using MD simulations. In accordance with the experimental observations made earlier, we also observed the suppression in flexibility not only in the orthosteric but also at allosteric sites. These regions mostly involve residues from the thumb and finger subdomains. Looking for the effect of resistance mutation on the dynamics, we observed that the suppression in flexibility seen in the NvpWTRT is consistently reversed in all simulation systems with resistance mutation (Fig. S7, magenta bars). This increase in flexibility is indicative of the fact that even a single resistance mutation can cause the RT to become functional despite the presence of the allosteric inhibitor NVP.

Figure 6.

Change in RT dynamics on inhibitor binding and resistance mutation. (A) RMSFs of Cα atoms in ApoRT (cyan), NvpWTRT (magenta), and four resistance-mutation-carrying RTs (NvpK103N-RT, dark green; NvpY181C-RT, olive; NvpY188C-RT, yellow-green; and NvpE138K-RT, lawn green). Shown is the distribution of distance between Cα atoms of (B) A181 and B139 and (C) A182 and B138. To see this figure in color, go online.

We further analyzed the distance between Cα atoms of the residues that showed changes in contact pattern in the network analysis of static structures (Fig. 6, B and C). The distribution of distances between the Cα atoms of A181 and B139 (Fig. 6 B), and A182 and B138 (Fig. 6 C), showed considerable shift toward larger values in the case of NvpWTRT as compared with ApoRT. This suggests that the contact loss observed in the static network analysis is reminiscent of small but crucial structural rearrangement. This is further corroborated by the fact that, in the case of resistance mutation simulations, the distribution moves toward lower distances.

Analysis of MD trajectories exposes opposing changes in the conformational landscape upon inhibitor binding and resistance mutations

Molecular dynamics simulations help to illustrate the dynamics of the protein in solution. However, owing to the high dimensionality of the motions exhibited in the overall dynamics of the systems, it often becomes difficult to study the large-scale or dominant motions evinced by the protein molecule. PCA of the MD trajectory provides a way to understand these dominant motions (52). To understand the difference in dynamics in different states of RT, PCA was performed on the MD trajectories of the unbound, inhibitor-bound, and resistance-mutation-carrying RT. The first five principal components explain most of the variance in the motions in all the systems (Table S9). The primary dominant motion in all the proteins involves the closing motion of the thumb domain. This is also evident in the fluctuation along the PC1 (Fig. S8). To understand the conformational landscape of RT in different states, we calculated free energy landscape (FEL) using the top two principal components, PC1 and PC2, as reaction coordinates (Fig. 7).

Figure 7.

Conformational landscape of RT varies in different states. Shown is the free energy landscape along the first two principal components obtained from PCA of trajectories for (A) ApoRT, (B) NvpWTRT, (C) NvpK103NRT, and (D) NvpY188CRT MD simulations. The distance between the centroid of the thumb and finger domain of each representative structure is mentioned. To see this figure in color, go online.

The differences in the FEL between different systems indicate changes in the conformational dynamics of RT in different states. The unbound RT shows primarily two minima separated along PC1 (Fig. 7 A). The distance between the center of mass of the finger and thumb domains (hereafter, COM distance) in representative structures from each minimum in the FEL shows that there are primarily two distributions that exist in the unbound RT form. In the case of inhibitor-bound RT, changes in the conformational landscape exhibit a deeper minimum in the FEL with much larger COM distance, for the representative structure of this minimum (Fig. 7 B). This suggests that binding of inhibitor causes one conformation corresponding to a larger COM distance to become a dominant one with only a few frames exhibiting a different conformation corresponding to the second minimum.

The FEL corresponding to the mutated RTs further shows difference from the inhibitor-bound state (Fig. 7, C and D). The NvpK103NRT and NvpY181CRT MD trajectories show three distinct minima, suggesting an increase in the conformational heterogeneity despite the presence of inhibitor in the binding pocket (Fig. 7 C; Fig. S9 A). Similar to unbound RT, NvpY188CRT shows two distinct minima with a shallower third minima (Fig. 7 D). The FEL of the NvpE138K is considerably different from all other systems with a broad single minimum (Fig. S9 B).

The two minima in the FEL of unbound trajectory show two distinct relative conformations of finger and thumb subdomains, with one showing the closed form and the other showing the open form (Fig. 7 A). The binding of NVP alters the conformational landscape, with a majority of frames showing an open kind of conformation with larger distance between the finger and thumb domain (Fig. 7 B). The representative from one out of the three minima observed in the FEL of NvpK103NRT shows a lower distance as compared with the representative conformation from minima in NvpWTRT (Fig. 7 C, bottom). For NvpY188CRT, the representative from two minima shows smaller COM distance, with one being much closer to the unbound RT (40.3 Å) (Fig. 7 D). Thus, the studies on the dynamics of RT in these three states—unbound, inhibitor-bound, and resistance mutation—also show that the small structural changes that inhibitor binding confers leads to the inhibition of DNA binding activity through changes in the flexibility and conformational dynamics of RT. Also, single resistance mutations cause changes in the conformational dynamics leading to the release of such inhibition.

Conclusion

This study addresses the relatively unexplored area of “structure-function” relationship in proteins for cases when complete change in function occurs for insignificant overall conformational changes due to mutation or allosteric ligand binding. In this work, we have analyzed 98 crystal structures of HIV-1 RT in unbound, allosteric drug-bound, and RT carrying mutations that led to resistance to the drug. We find that the conformational heterogeneity exhibited by these static structures of RT does not delineate the small conformational differences in various states of the enzyme (Fig. 1), except perhaps the DNA/RNA-bound complexes. For detailed analysis, we chose the ApoRT structure, the RT with allosterically bound drug nevirapine inhibiting RT function, and six nevirapine-bound RT structures carrying single mutations that conferred resistance to the drug, i.e., release from inhibition. All eight structures showed very low average cross-structure RMSD, indicating overall similarity in their conformations. Obviously, understanding the structural basis of these large alterations in functions of RT arising from small local changes in positions of the residues in space, and their dynamics, has important implications in understanding the appearance of resistance mutations against ART drugs for HIV-1 treatment.

To understand these small local changes, we model the proteins as coarse-grained networks of amino acids and find that the changes in the contact patterns in the unbound to drug-bound, and its reversal in resistance-mutation-carrying RT networks, clearly depict the structural basis of the observed functional changes. This approach also shows that nodewise changes in the network properties in the inhibitor-bound complex, as compared with the unbound complex, gets largely reversed in the RT complexes with resistance mutations (Fig. 3; Fig. S3), even though the average network properties do not show any difference between these different states of RT. Reversal of the shortest paths in the functionally important regions in the RT carrying resistance mutations, when compared with the inhibitor-bound RT, clearly shows that the changed communication paths are reversed and that RT function is restored, despite the presence of the bound inhibitor (Fig. 4; Fig. S5). As observed for the local network properties, the community structure analysis also exhibits remarkable reorganization of the functionally relevant polymerase-binding and inhibitor binding residues in the RT at different states and shows the same trend of resistance mutation network communities becoming similar to the unbound RT network (Figs. 5 and S6; Table S8). These clearly suggest that the network analysis of the RT structures in different states can successfully expose the small conformational changes and elucidate how communication between different regions of the protein are altered. This simple framework has also been shown previously to analyze the subtle structural changes associated with thermostability and other functional changes in proteins (25,26).

To elucidate how small local contact changes can lead to global functional alterations through local residue dynamics, we perform MD simulations on unbound, inhibitor-bound, and resistance mutation RT structures. Here, the presence of resistance mutations also tends to reverse the suppression in flexibility observed in the inhibitor-bound complex (Fig. 6; Fig. S7). We also show that this in turn has an effect on the overall conformational landscape of the protein, with resistance mutation RT complex showing palm, thumb, and finger domain motions, which are primarily involved in DNA binding, similar to unbound RT (Fig. 7; Fig. S9).

Since the insightful study on the big role of small changes in protein structures (53), the use of alternative approaches, such as protein contact network representation and their analysis has slowly grown in importance (54,55). In this work on HIV-1 reverse transcriptase protein, we have used the combined approach of static network analysis and molecular dynamics simulations to demonstrate the effect of small local contact changes arising due to drug binding and single mutations on global changes in the communication pattern and dynamics of the protein. This can further lead to the functional modifications, such as emergence of resistance to drugs, thus making them ineffective. Thus, our local contact change analysis in protein networks can uncover the specific contacts that play a significant role in the loss of inhibition due to single mutations. For a virus, such as HIV-1, having very high mutation rate, this strategy of reorganizing local contacts while the inhibitor is bound seems most useful in the scenario of selection under widespread drug usage. Our study also suggests that, to develop more potent drugs with the lowest susceptibility to resistance mutations, it is imperative to take into consideration the effect of these mutations on the overall structure and dynamics of the protein.

Author Contributions

Designed research, A.S. and S.S.; Performed research, A.S. and V.B.; Analyzed data, A.S. and S.S.; Wrote the manuscript, A.S. and S.S.

Editor: Jane Dyson.