Molecular origins of binding affinity: seeking the Archimedean point

Growing ecosystem of deep learning methods for modeling protein-protein interactions

Numerous cellular functions rely on protein-protein interactions. Efforts to comprehensively characterize them remain challenged however by the diversity of molecular recognition mechanisms employed within the proteome. Deep learning has emerged as a promising approach for tackling this problem by exploiting both experimental data and basic biophysical knowledge about protein interactions. Here, we review the growing ecosystem of deep learning methods for modeling protein interactions, highlighting the diversity of these biophysically informed models and their respective trade-offs. We discuss recent successes in using representation learning to capture complex features pertinent to predicting protein interactions and interaction sites, geometric deep learning to reason over protein structures and predict complex structures, and generative modeling to design de novo protein assemblies. We also outline some of the outstanding challenges and promising new directions. Opportunities abound to discover novel interactions, elucidate their physical mechanisms, and engineer binders to modulate their functions using deep learning and, ultimately, unravel how protein interactions orchestrate complex cellular behaviors.

A dynamical view of protein-protein complexes: Studies by molecular dynamics simulations

Protein-protein interactions are at the basis of many protein functions, and the knowledge of 3D structures of protein-protein complexes provides structural, mechanical and dynamical pieces of information essential to understand these functions. Protein-protein interfaces can be seen as stable, organized regions where residues from different partners form non-covalent interactions that are responsible for interaction specificity and strength. They are commonly described as a peripheral region, whose role is to protect the core region that concentrates the most contributing interactions, from the solvent. To get insights into the dynamics of protein-protein complexes, we carried out all-atom molecular dynamics simulations in explicit solvent on eight different protein-protein complexes of different functional class and interface size by taking into account the bound and unbound forms. On the one hand, we characterized structural changes upon binding of the proteins, and on the other hand we extensively analyzed the interfaces and the structural waters involved in the binding. Based on our analysis, in 6 cases out of 8, the interfaces rearranged during the simulation time, in stable and long-lived substates with alternative residue-residue contacts. These rearrangements are not restricted to side-chain fluctuations in the periphery but also affect the core interface. Finally, the analysis of the waters at the interface and involved in the binding pointed out the importance to take into account their role in the estimation of the interaction strength.

Rational design of affinity ligands for bioseparation

Affinity adsorbents have been the cornerstone in protein purification. The selective nature of the molecular recognition interactions established between an affinity ligands and its target provide the basis for efficient capture and isolation of proteins. The plethora of affinity adsorbents available in the market reflects the importance of affinity chromatography in the bioseparation industry. Ligand discovery relies on the implementation of rational design techniques, which provides the foundation for the engineering of novel affinity ligands. The main goal for the design of affinity ligands is to discover or improve functionality, such as increased stability or selectivity. However, the methodologies must adapt to the current needs, namely to the number and diversity of biologicals being developed, and the availability of new tools for big data analysis and artificial intelligence. In this review, we offer an overview on the development of affinity ligands for bioseparation, including the evolution of rational design techniques, dating back to the years of early discovery up to the current and future trends in the field.

Conformational Dynamics and Cooperativity Drive the Specificity of a Protein-Ligand Interaction

Molecular recognition is critical for the fidelity of signal transduction in biology. Conversely, the disruption of protein-protein interactions can lead to disease. Thus, comprehension of the molecular determinants of specificity is essential for understanding normal biological signaling processes and for the development of precise therapeutics. Although high-resolution structures have provided atomic details of molecular interactions, much less is known about the influence of cooperativity and conformational dynamics. Here, we used the Tiam2 PSD-95/Dlg/ZO-1 (PDZ) domain and a quadruple mutant (QM), engineered by swapping the identity of four residues important for specificity in the Tiam1 PDZ into the Tiam2 PDZ domain, as a model system to investigate the role of cooperativity and dynamics in PDZ ligand specificity. Surprisingly, equilibrium binding experiments found that the ligand specificity of the Tiam2 QM was switched to that of the Tiam1 PDZ. NMR-based studies indicated that Tiam2 QM PDZ, but not other mutants, had extensive microsecond to millisecond motions distributed throughout the entire domain suggesting structural cooperativity between the mutated residues. Thermodynamic analyses revealed energetic cooperativity between residues in distinct specificity subpockets that was dependent upon the identity of the ligand, indicating a context-dependent binding mechanism. Finally, isothermal titration calorimetry experiments showed distinct entropic signatures along the mutational trajectory from the Tiam2 wild-type to the QM PDZ domain. Collectively, our studies provide unique insights into how structure, conformational dynamics, and thermodynamics combine to modulate ligand-binding specificity and have implications for the evolution, regulation, and design of protein-ligand interactions.

Determination of oligomerization state of Drp1 protein in living cells at nanomolar concentrations

Biochemistry in living cells is an emerging field of science. Current quantitative bioassays are performed ex vivo, thus equilibrium constants and reaction rates of reactions occurring in human cells are still unknown. To address this issue, we present a non-invasive method to quantitatively characterize interactions (equilibrium constants, K_D) directly within the cytosol of living cells. We reveal that cytosolic hydrodynamic drag depends exponentially on a probe's size, and provide a model for its determination for different protein sizes (1-70 nm). We analysed oligomerization of dynamin-related protein 1 (Drp1, wild type and mutants: K668E, G363D, C505A) in HeLa cells. We detected the coexistence of wt-Drp1 dimers and tetramers in cytosol, and determined that K_D for tetramers was 0.7 ± 0.5 μM. Drp1 kinetics was modelled by independent simulations, giving computational results which matched experimental data. This robust method can be applied to in vivo determination of K_D for other protein-protein complexes, or drug-target interactions.

Refining evERdock: Improved selection of good protein-protein complex models achieved by MD optimization and use of multiple conformations

A method for evaluating binding free energy differences of protein-protein complex structures generated by protein docking was recently developed by some of us. The method, termed evERdock, combined short (2 ns) molecular dynamics (MD) simulations in explicit water and solution theory in the energy representation (ER) and succeeded in selecting the near-native complex structures from a set of decoys. In the current work, we performed longer (up to 100 ns) MD simulations before employing ER analysis in order to further refine the structures of the decoy set with improved binding free energies. Moreover, we estimated the binding free energies for each complex structure based on an average value from five individual MD snapshots. After MD simulations, all decoys exhibit a decrease in binding free energy, suggesting that proper equilibration in explicit solvent resulted in more favourably bound complexes. During the MD simulations, non-native structures tend to become unstable and in some cases dissociate, while near-native structures maintain a stable interface. The energies after the MD simulations show an improved correlation between similarity criteria (such as interface root-mean-square distance) to the native (crystal) structure and the binding free energy. In addition, calculated binding free energies show sensitivity to the number of contacts, which was demonstrated to reflect the relative stability of structures at earlier stages of the MD simulation. We therefore conclude that the additional equilibration step along with the use of multiple conformations can make the evERdock scheme more versatile under low computational cost.

Heterodimer Binding Scaffolds Recognition via the Analysis of Kinetically Hot Residues

Physical interactions between proteins are often difficult to decipher. The aim of this paper is to present an algorithm that is designed to recognize binding patches and supporting structural scaffolds of interacting heterodimer proteins using the Gaussian Network Model (GNM). The recognition is based on the (self) adjustable identification of kinetically hot residues and their connection to possible binding scaffolds. The kinetically hot residues are residues with the lowest entropy, i.e., the highest contribution to the weighted sum of the fastest modes per chain extracted via GNM. The algorithm adjusts the number of fast modes in the GNM's weighted sum calculation using the ratio of predicted and expected numbers of target residues (contact and the neighboring first-layer residues). This approach produces very good results when applied to dimers with high protein sequence length ratios. The protocol's ability to recognize near native decoys was compared to the ability of the residue-level statistical potential of Lu and Skolnick using the Sternberg and Vakser decoy dimers sets. The statistical potential produced better overall results, but in a number of cases its predicting ability was comparable, or even inferior, to the prediction ability of the adjustable GNM approach. The results presented in this paper suggest that in heterodimers at least one protein has interacting scaffold determined by the immovable, kinetically hot residues. In many cases, interacting proteins (especially if being of noticeably different sizes) either behave as a rigid lock and key or, presumably, exhibit the opposite dynamic behavior. While the binding surface of one protein is rigid and stable, its partner's interacting scaffold is more flexible and adaptable.

Surface energetics and protein-protein interactions: analysis and mechanistic implications

Understanding protein-protein interactions (PPI) at the molecular level is a fundamental task in the design of new drugs, the prediction of protein function and the clarification of the mechanisms of (dis)regulation of biochemical pathways. In this study, we use a novel computational approach to investigate the energetics of aminoacid networks located on the surface of proteins, isolated and in complex with their respective partners. Interestingly, the analysis of individual proteins identifies patches of surface residues that, when mapped on the structure of their respective complexes, reveal regions of residue-pair couplings that extend across the binding interfaces, forming continuous motifs. An enhanced effect is visible across the proteins of the dataset forming larger quaternary assemblies. The method indicates the presence of energetic signatures in the isolated proteins that are retained in the bound form, which we hypothesize to determine binding orientation upon complex formation. We propose our method, BLUEPRINT, as a complement to different approaches ranging from the ab-initio characterization of PPIs, to protein-protein docking algorithms, for the physico-chemical and functional investigation of protein-protein interactions.

Affinity and competition for TBP are molecular determinants of gene expression noise

Cell-to-cell variation in gene expression levels (noise) generates phenotypic diversity and is an important phenomenon in evolution, development and disease. TATA-box binding protein (TBP) is an essential factor that is required at virtually every eukaryotic promoter to initiate transcription. While the presence of a TATA-box motif in the promoter has been strongly linked with noise, the molecular mechanism driving this relationship is less well understood. Through an integrated analysis of multiple large-scale data sets, computer simulation and experimental validation in yeast, we provide molecular insights into how noise arises as an emergent property of variable binding affinity of TBP for different promoter sequences, competition between interaction partners to bind the same surface on TBP (to either promote or disrupt transcription initiation) and variable residence times of TBP complexes at a promoter. These determinants may be fine-tuned under different conditions and during evolution to modulate eukaryotic gene expression noise.

Antibody Binding Selectivity: Alternative Sets of Antigen Residues Entail High-Affinity Recognition

Understanding the relationship between protein sequence and molecular recognition selectivity remains a major challenge. The antibody fragment scFv1F4 recognizes with sub nM affinity a decapeptide (sequence 6TAMFQDPQER15) derived from the N-terminal end of human papilloma virus E6 oncoprotein. Using this decapeptide as antigen, we had previously shown that only the wild type amino-acid or conservative replacements were allowed at positions 9 to 12 and 15 of the peptide, indicating a strong binding selectivity. Nevertheless phenylalanine (F) was equally well tolerated as the wild type glutamine (Q) at position 13, while all other amino acids led to weaker scFv binding. The interfaces of complexes involving either Q or F are expected to diverge, due to the different physico-chemistry of these residues. This would imply that high-affinity binding can be achieved through distinct interfacial geometries. In order to investigate this point, we disrupted the scFv-peptide interface by modifying one or several peptide positions. We then analyzed the effect on binding of amino acid changes at the remaining positions, an altered susceptibility being indicative of an altered role in complex formation. The 23 starting variants analyzed contained replacements whose effects on scFv1F4 binding ranged from minor to drastic. A permutation analysis (effect of replacing each peptide position by all other amino acids except cysteine) was carried out on the 23 variants using the PEPperCHIP® Platform technology. A comparison of their permutation patterns with that of the wild type peptide indicated that starting replacements at position 11, 12 or 13 modified the tolerance to amino-acid changes at the other two positions. The interdependence between the three positions was confirmed by SPR (Biacore® technology). Our data demonstrate that binding selectivity does not preclude the existence of alternative high-affinity recognition modes.

Conformational Destabilization of Immunoglobulin G Increases the Low pH Binding Affinity with the Neonatal Fc Receptor

Crystallographic evidence suggests that the pH-dependent affinity of IgG molecules for the neonatal Fc receptor (FcRn) receptor primarily arises from salt bridges involving IgG histidine residues, resulting in moderate affinity at mildly acidic conditions. However, this view does not explain the diversity in affinity found in IgG variants, such as the YTE mutant (M252Y,S254T,T256E), which increases affinity to FcRn by up to 10×. Here we compare hydrogen exchange measurements at pH 7.0 and pH 5.5 with and without FcRn bound with surface plasmon resonance estimates of dissociation constants and FcRn affinity chromatography. The combination of experimental results demonstrates that differences between an IgG and its cognate YTE mutant vary with their pH-sensitive dynamics prior to binding FcRn. The conformational dynamics of these two molecules are nearly indistinguishable upon binding FcRn. We present evidence that pH-induced destabilization in the CH2/3 domain interface of IgG increases binding affinity by breaking intramolecular H-bonds and increases side-chain adaptability in sites that form intermolecular contacts with FcRn. Our results provide new insights into the mechanism of pH-dependent affinity in IgG-FcRn interactions and exemplify the important and often ignored role of intrinsic conformational dynamics in a protein ligand, to dictate affinity for biologically important receptors.

Updates to the Integrated Protein-Protein Interaction Benchmarks: Docking Benchmark Version 5 and Affinity Benchmark Version 2

We present an updated and integrated version of our widely used protein-protein docking and binding affinity benchmarks. The benchmarks consist of non-redundant, high-quality structures of protein-protein complexes along with the unbound structures of their components. Fifty-five new complexes were added to the docking benchmark, 35 of which have experimentally measured binding affinities. These updated docking and affinity benchmarks now contain 230 and 179 entries, respectively. In particular, the number of antibody-antigen complexes has increased significantly, by 67% and 74% in the docking and affinity benchmarks, respectively. We tested previously developed docking and affinity prediction algorithms on the new cases. Considering only the top 10 docking predictions per benchmark case, a prediction accuracy of 38% is achieved on all 55 cases and up to 50% for the 32 rigid-body cases only. Predicted affinity scores are found to correlate with experimental binding energies up to r=0.52 overall and r=0.72 for the rigid complexes.

Allosteric Dynamic Control of Binding

Proteins have a highly dynamic nature and there is a complex interrelation between their structural dynamics and binding behavior. By assuming various conformational ensembles, they perform both local and global fluctuations to interact with other proteins in a dynamic infrastructure adapted to functional motion. Here, we show that there is a significant association between allosteric mutations, which lead to high-binding-affinity changes, and the hinge positions of global modes, as revealed by a large-scale statistical analysis of data in the Structural Kinetic and Energetic Database of Mutant Protein Interactions (SKEMPI). We further examined the mechanism of allosteric dynamics by conducting studies on human growth hormone (hGH) and pyrin domain (PYD), and the results show how mutations at the hinge regions could allosterically affect the binding-site dynamics or induce alternative binding modes by modifying the ensemble of accessible conformations. The long-range dissemination of perturbations in local chemistry or physical interactions through an impact on global dynamics can restore the allosteric dynamics. Our findings suggest a mechanism for the coupling of structural dynamics to the modulation of protein interactions, which remains a critical phenomenon in understanding the effect of mutations that lead to functional changes in proteins.

Protein-protein docking: from interaction to interactome

The protein-protein docking problem is one of the focal points of activity in computational biophysics and structural biology. The three-dimensional structure of a protein-protein complex, generally, is more difficult to determine experimentally than the structure of an individual protein. Adequate computational techniques to model protein interactions are important because of the growing number of known protein structures, particularly in the context of structural genomics. Docking offers tools for fundamental studies of protein interactions and provides a structural basis for drug design. Protein-protein docking is the prediction of the structure of the complex, given the structures of the individual proteins. In the heart of the docking methodology is the notion of steric and physicochemical complementarity at the protein-protein interface. Originally, mostly high-resolution, experimentally determined (primarily by x-ray crystallography) protein structures were considered for docking. However, more recently, the focus has been shifting toward lower-resolution modeled structures. Docking approaches have to deal with the conformational changes between unbound and bound structures, as well as the inaccuracies of the interacting modeled structures, often in a high-throughput mode needed for modeling of large networks of protein interactions. The growing number of docking developers is engaged in the community-wide assessments of predictive methodologies. The development of more powerful and adequate docking approaches is facilitated by rapidly expanding information and data resources, growing computational capabilities, and a deeper understanding of the fundamental principles of protein interactions.

Predicting Binding Free Energy Change Caused by Point Mutations with Knowledge-Modified MM/PBSA Method

A new methodology termed Single Amino Acid Mutation based change in Binding free Energy (SAAMBE) was developed to predict the changes of the binding free energy caused by mutations. The method utilizes 3D structures of the corresponding protein-protein complexes and takes advantage of both approaches: sequence- and structure-based methods. The method has two components: a MM/PBSA-based component, and an additional set of statistical terms delivered from statistical investigation of physico-chemical properties of protein complexes. While the approach is rigid body approach and does not explicitly consider plausible conformational changes caused by the binding, the effect of conformational changes, including changes away from binding interface, on electrostatics are mimicked with amino acid specific dielectric constants. This provides significant improvement of SAAMBE predictions as indicated by better match against experimentally determined binding free energy changes over 1300 mutations in 43 proteins. The final benchmarking resulted in a very good agreement with experimental data (correlation coefficient 0.624) while the algorithm being fast enough to allow for large-scale calculations (the average time is less than a minute per mutation).

Information flow and protein dynamics: the interplay between nuclear magnetic resonance spectroscopy and molecular dynamics simulations

Proteins participate in information pathways in cells, both as links in the chain of signals, and as the ultimate effectors. Upon ligand binding, proteins undergo conformation and motion changes, which can be sensed by the following link in the chain of information. Nuclear magnetic resonance (NMR) spectroscopy and molecular dynamics (MD) simulations represent powerful tools for examining the time-dependent function of biological molecules. The recent advances in NMR and the availability of faster computers have opened the door to more detailed analyses of structure, dynamics, and interactions. Here we briefly describe the recent applications that allow NMR spectroscopy and MD simulations to offer unique insight into the basic motions that underlie information transfer within and between cells.

Redesigning protein cavities as a strategy for increasing affinity in protein-protein interaction: interferon- γ receptor 1 as a model

Combining computational and experimental tools, we present a new strategy for designing high affinity variants of a binding protein. The affinity is increased by mutating residues not at the interface, but at positions lining internal cavities of one of the interacting molecules. Filling the cavities lowers flexibility of the binding protein, possibly reducing entropic penalty of binding. The approach was tested using the interferon-γ receptor 1 (IFNγR1) complex with IFNγ as a model. Mutations were selected from 52 amino acid positions lining the IFNγR1 internal cavities by using a protocol based on FoldX prediction of free energy changes. The final four mutations filling the IFNγR1 cavities and potentially improving the affinity to IFNγ were expressed, purified, and refolded, and their affinity towards IFNγ was measured by SPR. While individual cavity mutations yielded receptor constructs exhibiting only slight increase of affinity compared to WT, combinations of these mutations with previously characterized variant N96W led to a significant sevenfold increase. The affinity increase in the high affinity receptor variant N96W+V35L is linked to the restriction of its molecular fluctuations in the unbound state. The results demonstrate that mutating cavity residues is a viable strategy for designing protein variants with increased affinity.

Structural basis for assembly and function of the Nup82 complex in the nuclear pore scaffold

The yeast Nup82 complex forms an unusual asymmetric structure with a dimeric array of subunits that mediate its anchorage to the NPC scaffold and its concomitant interaction with the soluble nucleocytoplasmic transport machinery.

Nuclear pore complexes (NPCs) are huge assemblies formed from ∼30 different nucleoporins, typically organized in subcomplexes. One module, the conserved Nup82 complex at the cytoplasmic face of NPCs, is crucial to terminate mRNA export. To gain insight into the structure, assembly, and function of the cytoplasmic pore filaments, we reconstituted in yeast the Nup82–Nup159–Nsp1–Dyn2 complex, which was suitable for biochemical, biophysical, and electron microscopy analyses. Our integrative approach revealed that the yeast Nup82 complex forms an unusual asymmetric structure with a dimeric array of subunits. Based on all these data, we developed a three-dimensional structural model of the Nup82 complex that depicts how this module might be anchored to the NPC scaffold and concomitantly can interact with the soluble nucleocytoplasmic transport machinery.

Information-driven structural modelling of protein-protein interactions

Protein-protein docking aims at predicting the three-dimensional structure of a protein complex starting from the free forms of the individual partners. As assessed in the CAPRI community-wide experiment, the most successful docking algorithms combine pure laws of physics with information derived from various experimental or bioinformatics sources. Of these so-called "information-driven" approaches, HADDOCK stands out as one of the most successful representatives. In this chapter, we briefly summarize which experimental information can be used to drive the docking prediction in HADDOCK, and then focus on the docking protocol itself. We discuss and illustrate with a tutorial example a "classical" protein-protein docking prediction, as well as more recent developments for modelling multi-body systems and large conformational changes.

A minimal model of protein-protein binding affinities

A minimal model of protein-protein binding affinity that takes into account only two structural features of the complex, the size of its interface, and the amplitude of the conformation change between the free and bound subunits, is tested on the 144 complexes of a structure-affinity benchmark. It yields Kd values that are within two orders of magnitude of the experiment for 67% of the complexes, within three orders for 88%, and fails on 12%, which display either large conformation changes, or a very high or a low affinity. The minimal model lacks the specificity and accuracy needed to make useful affinity predictions, but it should help in assessing the added value of parameters used by more elaborate models, and set a baseline for evaluating their performances.

Prediction and redesign of protein-protein interactions

Understanding the molecular basis of protein function remains a central goal of biology, with the hope to elucidate the role of human genes in health and in disease, and to rationally design therapies through targeted molecular perturbations. We review here some of the computational techniques and resources available for characterizing a critical aspect of protein function - those mediated by protein-protein interactions (PPI). We describe several applications and recent successes of the Evolutionary Trace (ET) in identifying molecular events and shapes that underlie protein function and specificity in both eukaryotes and prokaryotes. ET is a part of analytical approaches based on the successes and failures of evolution that enable the rational control of PPI.

HADDOCK(2P2I): a biophysical model for predicting the binding affinity of protein-protein interaction inhibitors

The HADDOCK score, a scoring function for both protein-protein and protein-nucleic acid modeling, has been successful in selecting near-native docking poses in a variety of cases, including those of the CAPRI blind prediction experiment. However, it has yet to be optimized for small molecules, and in particular inhibitors of protein-protein interactions, that constitute an "unmined gold reserve" for drug design ventures. We describe here HADDOCK(2P2I), a biophysical model capable of predicting the binding affinity of protein-protein complex inhibitors close to experimental error (~2-fold larger). The algorithm was trained and 4-fold cross-validated against experimental data for 27 inhibitors targeting 7 protein-protein complexes of various functions and tested on an independent set of 24 different inhibitors for which K(d)/IC50 data are available. In addition, two popular ligand topology generation and parametrization methods (ACPYPE and PRODRG) were assessed. The resulting HADDOCK(2P2I) model, derived from the original HADDOCK score, provides insights into inhibition determinants: while the role of electrostatics and desolvation energies is case-dependent, the interface area plays a more critical role compared to protein-protein interactions.

Predicting the Impact of Missense Mutations on Protein-Protein Binding Affinity

The crucial prerequisite for proper biological function is the protein's ability to establish highly selective interactions with macromolecular partners. A missense mutation that alters the protein binding affinity may cause significant perturbations or complete abolishment of the function, potentially leading to diseases. The availability of computational methods to evaluate the impact of mutations on protein-protein binding is critical for a wide range of biomedical applications. Here, we report an efficient computational approach for predicting the effect of single and multiple missense mutations on protein-protein binding affinity. It is based on a well-tested simulation protocol for structure minimization, modified MM-PBSA and statistical scoring energy functions with parameters optimized on experimental sets of several thousands of mutations. Our simulation protocol yields very good agreement between predicted and experimental values with Pearson correlation coefficients of 0.69 and 0.63 and root-mean-square errors of 1.20 and 1.90 kcal mol^-1 for single and multiple mutations, respectively. Compared with other available methods, our approach achieves high speed and prediction accuracy and can be applied to large datasets generated by modern genomics initiatives. In addition, we report a crucial role of water model and the polar solvation energy in estimating the changes in binding affinity. Our analysis also reveals that prediction accuracy and effect of mutations on binding strongly depends on the type of mutation and its location in a protein complex.