InSty: a ProDy module for evaluating protein interactions and stability

Abstract

ProDy is a widely used application programming interface for analyzing the collective dynamics of proteins and their complexes, offering enhanced capabilities to address the growing needs of the computational biology community to bridge structure and function. Here, we introduce InSty, a new module integrated into ProDy to identify and quantify intra- and intermolecular interactions critical to protein stability and structural dynamics. InSty analyzes the non-covalent interactions using conformational ensemble data from both experiments and computational predictions, assesses their time evolution and persistence during molecular dynamics simulations as well as their conservation across homologs. It provides insights into the significance of these interactions in achieving function and/or supporting stability. InSty outputs lend themselves to statistical evaluation, visualization, and automated ensemble analysis for interpreting the significance of the interactions in the context of protein dynamics, sequence evolution, and allostery. Consolidation of InSty with various ProDy modules enables its efficient usage as a versatile tool that supports mutagenesis studies and identifies critical spots for functional interactions. The InSty module is available as part of the ProDy package at https://github.com/prody/ProDy.

Graphical abstract

Introduction

Non-covalent interactions are vital for maintaining the functional structures of biomolecules, ensuring their stability, biological actions, and adaptability [1, 2]. Additionally, they play essential roles in assembly or binding events underlying various biological processes [3]. A comprehensive understanding of the type, strength, and distribution of non-bonded interactions, both within individual proteins and across multiple components of a complex, is crucial for enabling precise modulation of protein-protein and -ligand interactions. Such insights are useful for design purposes, e.g., enhancing enzyme stability and optimizing their applications in biotechnology [4], as well as assessing the impact of point substitutions. The ability to predict, understand, leverage, or replace dominant interactions holds great significance for advancing medicine, biotechnology, and related fields.

Various methods and tools have been developed for identifying non-covalent interactions in proteins and their complexes: LigPlot+ is widely used for generating 2D plots of protein-ligand and -protein interactions, visualizing spatial arrangements of hydrogen bonds and hydrophobic interactions within protein binding sites [5]; HBPLUS specializes in hydrogen bond analysis and identification of hydrogen bond networks [6]; PLIP (Protein-Ligand Interaction Profiler) assists in the analysis of non-covalent interactions between proteins and ligands, as well as interactions within multiprotein complexes [7]; Arpeggio computes both intra- and interatomic interactions within or between proteins, DNA, or small organic molecules, encompassing various types of contacts, including van der Waals, ionic, metal, and hydrophobic contacts, and halogen bonds [8]. PPI3D [9] characterizes and models protein-protein interactions using 3D structural properties. As recently reviewed [10], several recent tools focus on the effect of mutations on interactions toward engineering biological function; examples are Mutabind [11], mCSM-PPI2 [12], DynaMut [13], and SKEMPI [14].

While these methods are effective in analyzing interactions and the impact of mutations, they do not usually take into consideration the time dependency of the interactions, whether they are transient and thus insignificant over a time span, or critical to collective structural dynamics that underlie function, or to family conservation. A thorough characterization must account for adaptabilities conferred by conformational flexibility and sequence diversity. The significance of protein dynamics in defining the impact of mutations is now well established [15]. Furthermore, there is a need to analyze interactions for ensembles of conformers rather than single snapshots, hence efforts for generating representative ensembles of conformations [16] or analyzing the signature properties of families of proteins rather than single members [17].

To address these needs, we developed InSty - a Python interface integrated into the widely-used ProDy application programming interface (API) [18]. InSty incorporates the effects of conformational dynamics/flexibilities and permits the interpretation of the outputs in relation to the collective dynamics of the examined system. InSty enables automated evaluation of specific types of interactions using time-evolved and/or ensemble data from experiments or simulations, as described below. InSty is designed to provide insights into conserved and robust interactions, along with the determinants of biomolecular structural and functional intricacies. The different functionalities of InSty are described in http://www.bahargroup.org/prody/tutorials/insty_tutorial/ by way of applications to case studies. We present four of them here, illustrating the use of InSty, with more information available in the Supplementary Material.

1. Methods

Description of InSty Inputs and Outputs

The InSty module accepts a range of inputs (Figure 1), broadly categorized as experimental or computational. Experimental inputs include: (a) a single file from the Protein Data Bank (PDB) [19], including those with multiple NMR models; (b) ensembles of PDB structures resolved for the same protein under different conditions or functional states, such as the open/closed or bound/unbound forms, or alternative conformers along allosteric cycles; and (c) an ensemble of PDB structures resolved for structural homologs retrieved via Dali [20] and other methods using ProDy’s signature dynamics module SignDy [17]. Computational inputs include: (d) conformations (or snapshots) collected during molecular dynamics (MD) simulations; and (e) ProDy-predicted conformational ensembles along the most collective motions predicted by the anisotropic network model (ANM [23, 24]). Hybrid methods such as ClustENMD [21] and ANMD [22] (see Supplementary Methods) are used for generating such conformers intrinsically accessible to the protein. These inputs capture conformational variabilities associated with the protein’s intrinsic dynamics, interactions, and sequence variations.

Figure 1. Overview of InSty features.

Results are illustrated for a single-chain structure (PDB: 5KQM). InSty can use three types of input using either experimental or computational data, as described (top); it evaluates seven types of non-covalent interactions (central box), and returns different types of outputs (bottom), which will be elaborated in the context of specific examples below.

ProDy reads MD trajectories in DCD format, such as those generated by NAMD [25] and CHARMM [26]. However, it is possible to convert various trajectory formats into DCD using existing and freely available packages such as MDTraj [27] or MDAnalysis [28]. ProDy permits users to construct ensembles of X-ray or cryo-electron microscopy (cryo-EM) structures (input b) and save them as multi-model PDB or DCD files, both supported by InSty.

The outputs are various objects related to 3D coordinates, as well as data for various plots (e.g., probabilistic distribution of interactions) facilitated by integrated Python libraries and visualized by VMD using TCL scripts. Supplementary Fig. S1 illustrates some of these functionalities.

Evaluating different types of non-covalent interactions

We have implemented classes for predicting seven types of non-covalent interactions: hydrogen bonds, salt bridges, hydrophobic interactions, electrostatic repulsions, disulfide bonds, π–stacking and π–cation interactions (Figure 1). Their definitions/criteria for detecting and analyzing these interactions are given in ProDy selection tools. For hydrophobic interactions, we developed a geometric method (in C++ and as a Python extension) to precisely identify the hydrophobic cores (see Supplementary Material and Supplementary Figs. 2-3). The InSty module is flexible, enabling users to modify interaction criteria (e.g., distances, angles, residue types) and focus on specific regions (e.g., residues, domains, or subunits) or protein-protein interactions. Analyses can use structures/conformations fetched from the PDB, local files, or ProDy pipelines. Missing atoms, such as hydrogens, are added by ProDy using PDBfixer/OpenMM [29] or OpenBabel [30]. Users have the option to count interactions and evaluate approximate energies (using options energy=True and energy_list_type in several functions). These include those deduced from solvent-mediated (𝑒𝑛𝑒𝑟𝑔𝑦_𝑙𝑖𝑠𝑡_𝑡𝑦𝑝𝑒 = “𝐼𝐵_𝑠𝑜𝑙𝑣”) or residue-mediated (“𝐼𝐵_𝑛𝑜𝑠𝑜𝑙𝑣”) distance-independent knowledge-based potentials [31] as well as energies from comprehensive umbrella sampling MD simulations with state-of-the-art ff19SB Amber force field and OPC explicit water model (“𝐶𝑆”; see Supplementary Methods).

Output interaction types and their quantitative features (e.g., angles, distances) are visualized using color-coded ribbon diagrams of PDB files compatible with various visualization tools (Figure 1, red indicates the highest number of interactions). TCL scripts can be generated for VMD to highlight residues involved in specific interactions, with dashed lines color-coded by type: blue for hydrogen bonds (HBs), yellow for salt bridges (SBs), grey for hydrophobic (HPh), red for electrostatic repulsions or repulsive ionic bonding (RIB), green for π-stacking (PiStack), orange for π-cation (PiCat), and black for disulfide bonds (DiBs). The same colors display interaction distributions (Figure 1, bottom right). Dense interaction regions are shown as heat maps (Figure 1 in dark red). Users can also characterize the interactions at the interfaces between chains.

Ensemble analysis and implementation of SignInt for Signature Interactions

InSty is a powerful tool for analyzing multiple structures (e.g., those resolved for the same protein under different conditions), structural models (e.g., NMR structures), structural homologs, and MD trajectories. It provides a statistical analysis of the results collected from different conformations provided. The user has access to the raw data, which can be used to generate the time evolution of each interaction type as well as a detailed analysis of pair-wise interactions (Supplementary Fig. S1). The graph presents pairs of interactions and their duration. The user can also obtain information about the mean distances between pairs and their standard deviations, and the distribution of geometric properties (e.g., distance or angle). Additionally, InSty provides a function for comparing selected conformations. This type of analysis can be used to see differences between alternative structures (including wild type and mutant) or different conformations (e.g., NMR data or different stages of a trajectory).

Within the InSty module, we have developed a pipeline called Signature Interactions (SignInt) to identify and download structural homologs for any query protein (by inputting its PDB file). SignInt leverages well-established tools, such as BLAST for sequence alignment or Dali [20] or Foldseek [32] for structure alignment. Corresponding newly created functions are runBLAST, runDali, and runFoldseek. Access to Dali and BLAST was implemented in ProDy previously [18, 33]; that to Foldseek is newly implemented. The list of structures extracted using those tools can be filtered, refined to complete the missing atoms, structurally aligned, and analyzed using InSty, with the function calcSignatureInteractions, to obtain a fingerprint of interactions for the family of structural homologs.

InSty uses residue mappings to characterize and track the persistence of specific types of interactions across homologs. The frequent interactions identified by InSty represent signature interactions (e.g., salt bridges, hydrogen bond networks) between evolutionarily conserved amino acid pairs, or interactions shared by multiple members of a family (e.g., interdomain contacts, coordination of active site residues), which may be maintained by compensating mutations or substitutions of similar amino acids. The former group is sequentially conserved, the second originates from the conservation of a structural motif. Either type of signature interactions can be important to function or stability, as will be illustrated in the case study presented for the lipoxygenase family below.

2. Results

To demonstrate the usefulness and versatility of InSty as a module embedded in the ProDy API, we present below four types of analyses: (i) intermolecular interactions between a viral spike protein and the host antibody, using a single PDB file; (ii) intramolecular interactions observed during MD simulations between the two subunits of a calcium-independent phospholipase A₂β (iPLA₂β) dimer, (iii) signature interactions across structural homologs retrieved by Dali [20] for the lipoxygenase (LOX) family, and finally (iv) the most significant interactions mediating the global motions of the N-terminal domain of a metabotropic glutamate receptor (mGluR).

(i). InSty evaluation of intermolecular interactions between a viral spike and a human antibody, using single structure data

We applied InSty to antibody-antigen complexes formed by the broadly neutralizing human antibody VIR-7229 and the receptor-binding domain (RBD) if the SARS-CoV-2 spike, focusing on recent variant BA.2.86 (PDB: 9ASD) [34], resolved by cryo-EM. The broad neutralization capacity of the antibody has been attributed largely to non-specific interactions, such as backbone hydrogen bonds and hydrophobic contacts, which were maintained in the spike mutants L455F and F456L, nicknamed FLip when present together in dominant variants.

Our analysis with InSty confirmed the dominance of non-specific interactions, especially at the interface between the RBD (designated as R chain) and the heavy chain (H) of the antibody (Supplementary Figure S4a-b). In addition several hydrogen bonds were detected at the interface of R with both heavy and light (L) chains of the antibody (Supplementary Fig. S4b-c). There was also one π-cation interaction between L Y97 and R K460. The color-coded bar plots in panel d display the distributions of different types of interactions for the R-H (left) and R-L (right) pairs, and the corresponding energies based on solvent-mediated knowledge-based potentials [31] are presented in panels e and f. Panel f displays the residue-based potentials, obtained by summing up all the interaction energies experienced by a given residue, thus permitting us to identify the residues (here L103 and L455 on the antibody H chain) that play an important role in stabilizing the interfacial interactions

We also examined the effects of various inputs and parameter choices. A few hydrogen bonds (0 or 1 per antibody chain) when using standard parameters for HB donor-acceptor distance ( ≤ 3.5Å) and donor-hydrogen-acceptor angle (≤ 40° deviation from 180°) with PDBFixer to add missing atoms (Supplementary Fig. S5a). This was in stark contrast to the total of 16 hydrogen bonds reported earlier [34]. However, the use of OpenBabel to further add hydrogens increased the number of detected hydrogen bonds to 3–4 per interface (Supplementary Fig. S5b), suggesting that the hydrogen atoms placed by OpenBabel are more reasonable, in line with our WatFinder results [35]. Likewise, increasing the HB distance and angle tolerance to ≤ 4Å and ≤ 90° allowed us to capture eight hydrogen bonds at each interface in accordance with previous work [34] (Supplementary Fig. S4b-d). Many hydrogen bonds were formed by RBD backbone atoms, such as those between the light chain Y34 hydroxyl and the backbone N of RBD N417, which has seen mutations in several variants.

InSty also estimates the energetics of pairwise interactions using knowledge-based potentials in the presence of solvent mediation (Supplementary Fig. S4e-f) or residue mediation (Supplementary Fig. S5c) [31], or using potentials of mean force derived from MD simulations of amino acid pairs carried out in the present study (Supplementary Fig. S5d). Summation over all the interactions of a given residue yields the cumulative-per-residue energies (Supplementary Figs. S4f and S5c-d), which helps identify residues playing a role in maintaining stability. The strongest interactions are robustly distinguished regardless of the approach adopted for evaluating the energetic contributions. In the current application, we observed such robust interactions between the heavy chain H residues L101-G109 and RBD residues L455-Y473, and between light chain L residues N33-Y34 and RBD N405 and Q409.

(ii). InSty evaluation of iPLA₂β intermonomer interactions using MD data

Here, we illustrate the use of InSty for evaluating the time-resolved interactions between two monomers of the phospholipase iPLA₂β dimer, inferred from MD simulations. Phospholipase iPLA₂β is a calcium-independent phospholipase A₂ [36, 37], which, in a dimeric form, hydrolyzes peroxidized phospholipids, eliminating ferroptotic cell death signals [38]. Here, we performed a 300 ns MD simulation of its catalytic domain in a dimeric form and monitored the changes in intermonomer interactions using InSty. The approach helps determine the time evolution of the different types of interactions at the intermonomer interface (Supplementary Fig. S6a). Detailed bar plots for specific types of interactions provide information on the identity of residue pairs, their average distances and corresponding standard deviations, as illustrated for salt bridges in the inset of Supplementary Fig. S6a. We note that interfacial hydrogen bonds were stabilized after 50 ns of MD simulation, and hydrophobic interactions and salt bridges contributed significantly to dimer stability. InSty also provides a graphical representation of the most frequent interactions observed during the course of simulations (Supplementary Fig. S6b). Residues are color-coded by their physicochemical properties (charged: red/blue; polar: green; non-polar: grey). Interacting residues are connected by lines, with thinner and darker lines showing closer and stronger interactions. Intermonomer salt bridges, particularly those between D609 and R693/R691 (black dotted ovals, Supplementary Fig. S6b-c) are found to be key stabilizers, along with hydrogen bonds Q538-V647 and H615-E697 and hydrophobic interactions R687-L602 and L602-W695.

(iii). Detecting structurally conserved interactions in LOX family members using the Signature Interaction function of InSty

We developed SignInt implemented in InSty to identify the conserved interactions shared between structural homologs. SignInt processes the structural homologs retrieved from the Dali server [20] using ProDy, and the sequence alignments retrieved from the Pfam database [39]. SignInt also integrates BLAST and Foldseek data, offering flexibility for the users, and helping consolidate the results from alternative methods. By focusing on an ensemble of structural homologs, SignInt uncovers functional regions and key residue positions that may not be apparent through sequence analysis, and discloses structure-encoded evolutionary adaptations.

The utility of SignInt is illustrated by way of its application to LOX family members. The members of this family are structurally conserved despite their low (< 30%) sequence identity [40]. They catalyze the oxygenation of polyunsaturated fatty acids into fatty acid hydroperoxides, thus affecting inflammation, immune response, and cell death mechanisms [41–43]. LOXs are known to contain conserved dynamics and conserved interaction motifs [40]. Using 15LOX-2 (PDB: 4NRE [42]) as a reference, we retrieved 1,581 structural homologs with Dali and selected 140 (including the reference structure) that met the requirements of sequence coverage ≥0.5 and RMSD <1.0 Å. After the addition of hydrogens and structural alignment using runDali, we used InSty to characterize the conserved interactions. Apart from the traditional visualization of interactions as dotted lines in ribbon diagrams (Figure 1) or lines of different thicknesses in graphical representations (Supplementary Fig. S4b-c), we designed a function that displays interactions by color-coded spheres localized between the interacting residues. This approach helps visualize the regions subject to the most extensive interactions and their types (by color codes), as illustrated in Figure 2.

Figure. 2. SignInt results for lipoxygenase (LOX) family members.

The LOX fold represented by 15LOX-2 (PDB: 4NRE) colored by sequence conservation from blue (variable) to red (conserved). The sequence conservation data were obtained using the Evol tool within ProDy based on 871 LOX sequences (with sequence identity >90%, and coverage >70%) parsed from the Pfam database (ID: PF00305). Colored spheres mark clusters of non-covalent interactions detected by InSty. Interaction clusters were derived from 140 structural homologs (139 homologs and reference PDB structure) with ≥50% sequence coverage and ≥5% sequence identity in the PDB. Conserved interactions present in at least 25 structures are displayed, showing hydrogen bonds (blue), salt bridges (yellow), electrostatic repulsions (red), π-cation interactions (orange), and π-stacking (green) (a), and hydrophobic interactions (dark gray) (b). Insets (1)-(4) present enlarged views of selected interaction motifs. Red arrows show sequentially variable regions detected by SignInt to engage in structurally conserved interactions.

Notably, some of the conserved interactions are not sequentially conserved but structurally conserved (denoted by red arrows in the insets 1–4), such as the salt bridge and hydrogen bonds near R361 (for 15LOX-2, Figure 2a inset-(1)), and the interactions near residues R654 and F640 (Figure 2a-(2)). R654 and F640 themselves are involved in conserved interactions, mainly a salt bridge (R654-D476) and π-stacking interaction (F487-F640), playing a role in coupling two parallel helices. Interestingly, we identified an interaction motif that serves as an entrance for O₂ molecules for LOX catalytic activity [44]. This interaction motif is distinguished by several conserved hydrogen bonds (Figure 2a-(3)), which shape the entrance but are not sequentially conserved. We also identified another interaction motif rich in conserved interactions between the β-barrel and catalytic domain (Figure 2a-(4)), which has been recently reported [45, 46] to be a potential allosteric site for LOX inhibitors. InSty detected conserved fingerprints for specific (Figure 2a) and hydrophobic (Figure 2b) interactions.

(iv). Redistribution of interactions during global rearrangements: combining InSty and ANMD to study the clamshell closure of metabotropic glutamate receptors

To illustrate InSty’s potential to analyze changes in interactions accompanying conformational transitions, we used it in conjunction with ANMD for examining the mGluR1 N-terminal Venus fly trap (VFT) domain. This domain forms a clamshell structure that opens and closes in order to bind the neurotransmitter L-glutamate, leading to additional rearrangements of this dimeric class C G-protein-coupled receptor [47].

We started with the glutamate-bound active structure of the VFT dimer (PDB: 1EWK [48]), which has chain A in the closed state and chain B in the open state. Structural alignment of the two chains permits one to visualize the conformational change occurring between the open and closed forms (Figure 3a), shown in green-orange and blue diagrams, respectively. We focused on this clamshell closure mechanism, which has been shown to be captured by a single ANM mode [49, 50]. The first bar in Figure 3b shows that the overlap (correlation cosine absolute value) between ANM mode 1 accessible to open conformer (computationally predicted) and the conformational change undertaken during the open-to-closed transition (experimentally observed) is approximately 0.90, i.e., reconfiguration of the structure along this mode achieves a significant part of the transition. Figure 3c (blue curve) displays the RMSD between the open and closed forms as successive conformers along mode 1 are generated by ANMD, starting from the open form. The RMSD decreases from its original value of 4.53 Å to 1.79 Å at the fourth step. In parallel, the RMSD from the original open conformation increases from zero to ~5Å (green curve). Therefore, as expected, the ANMD ensemble along mode 1 enables a clamshell closure (Figure 3c and Movie 1), helping traverse a large portion of the transition path between the open and closed forms. Furthermore, the ANMD ensemble along mode 2 (not shown) exhibits a complementary twisting motion (Movie 2) in line with previous observations [49, 50].

Figure 3. Change in metabotropic glutamate receptor mGluR1 N-terminal domain inter-lobe interactions, as the undergoes clamshell-like motions between its open and closed forms.

(a) Overlaid structures of the N-terminal domain open (green and orange) and closed (blue) conformations from two views. The second view is used in panels d-f. (b) Correlation cosine overlaps showing contribution of open chain ANM normal modes to the conformational change from open to closed form. Mode 1 strongly dominates (overlap 0.93). (c) RMSDs in the coordinates of Cα atoms with respect to the starting (open) conformation as ANMD ensembles of conformers are generated in multiple steps along mode 1. The stepwise deviation away from the starting structure is shown by the green curve, and the approach towards the target (closed) structure, by the blue curve. (d-f) Interaction maps (top) and zoomed structures (bottom) showing the changes in interfacial interactions for the open state (d), the closed state (e) and a computationally generated structural model along mode 1 (f; conformation 5 in c). The points in the maps correspond to interactions between residues on the upper lobe along the X-axis and lower lobe along the Y-axis, and indicated by the dark red and bright green labels in the maps and ribbon diagrams. The square colors correspond to the number of atom pairs that make contacts as indicated by the color bars with purple corresponding to one interacting atom pair and brighter colors corresponding to more interacting pairs.

Having generated a series of intermediate conformations along the transition pathway, we analyzed the changes in interactions between the upper lobe (orange in Figure 3a, d-f; residues R36-P206 and E346-K472) and the lower lobe (green; residues A214-K341 and R478-K511). The open state showed few inter-lobe contacts, which were localized at the hinge regions at the back and lip of the clamshell (Figure 3d). The closed state showed many more inter-lobe interactions (Figure 3e). Panel f displays the interfacial contacts observed in the conformation 5 (Figure 3b) generated by ANMD computations. As described above, this conformation, generated by moving along mode 1 on the conformational space, starting from the open form, represents an intermediate conformation close to the closed state, which may still retain some of the open state contacts. In fact, panel f shows 3 of the 4 contacts characterizing the closed state (panel e) are formed in this intermediate, but one of the contacts observed in the open state is still retained (Figure 3f). This type of analysis of intermediate conformers provides insights into not only key interactions broken/formed during a transition, but the sequence of such events, which may be explored with the help of InSty.

Conclusion

The first version of the ProDy API was developed as a Python package for structure-based analyses, aimed at studying the structural dynamics of proteins under equilibrium conditions using elastic network models and normal mode analyses. Over the years, numerous functions have been systematically added to ProDy, making it more powerful and widely used worldwide. The new features provided by InSty presented here greatly extend ProDy’s capability to analyze intra- and intermolecular interactions at full atomic resolution, providing information at a completely new level.

Major advantages of the new tool are: its ability to analyze time-resolved interactions as well as ensemble-averaged ones; its versatile applicability to multiple conformations of a given protein or complex (observed experimentally or computationally) as well as families of structural homologs; and its use to efficiently disclose conserved interaction patterns, or signature interactions shared by structural homologs. It is well-established that structures exhibit conserved motifs while sequences diverge. The present analysis further shows that well-defined interaction fingerprints support the stability or function of biomolecular systems, and those interaction patterns are conserved among family members even when the involved amino acids are not sequentially conserved. As demonstrated in various applications, InSty can serve as a powerful tool for detecting such signature interactions that stabilize complexes, and for assessing their redistribution during biological function, benefiting from the elaborate ProDy framework for exploring the conformational space and analyzing sequence evolutionary patterns.

Highlights.

• InSty identifies conserved intramolecular and intermolecular interactions within proteins and complexes

• Integrated in the ProDy Python package, InSty has customizable pipelines for characterizing interaction patterns of time-evolved conformations

• InSty provides quantitative results and plots, plus structure visualisation in VMD

• New SignInt pipelines enable signature interaction analysis for structural homologs