Molecular dynamics (MD) simulations are a popular method of studying protein structure and function, but are unable to reliably sample all relevant conformational space in reasonable computational timescales. A range of enhanced sampling methods are available that can improve conformational sampling, but these do not offer a complete solution. We present here a proof-of-principle method of combining MD simulation with machine learning to explore protein conformational space. An autoencoder is used to map snapshots from MD simulations onto the conformational landscape defined by a 2D-RMSD matrix, and we show that we can predict, with useful accuracy, conformations that are not present in the training data. This method offers a new approach to the prediction of new low energy/physically realistic structures of conformationally dynamic proteins and allows an alternative approach to enhanced sampling of MD simulations.

Multi-domain proteins are not only formed through natural evolution but can also be generated by recombinant DNA technology. Because many fusion proteins can enhance the selectivity of cell targeting, these artificially produced molecules, called multi-specific biologics, are promising drug candidates, especially for immunotherapy. Moreover, the rational design of domain linkers in fusion proteins is becoming an essential step toward a quantitative understanding of the dynamics in these biopharmaceutics. We developed a computational framework to characterize the impacts of peptide linkers on the dynamics of multi-specific biologics. We constructed a benchmark containing six types of linkers that represent various lengths and degrees of flexibility and used them to connect two natural proteins as a test system. The microsecond dynamics of these proteins generated from Anton were projected onto a coarse-grained conformational space. The similarity of dynamics among different proteins in this low-dimensional space was further analyzed by a neural network model. Finally, hierarchical clustering was applied to place linkers into different subgroups based on the neural network classification results. The clustering results suggest that the length of linkers used to spatially separate different functional modules plays the most important role in regulating the dynamics of this fusion protein. Given the same number of amino acids, linker flexibility functions as a regulator of protein dynamics. In summary, we illustrated that a new computational strategy can be used to study the dynamics of multi-domain fusion proteins by a combination of long timescale molecular dynamics simulation, coarse-grained modeling, and artificial intelligence.

Inorganic pyrophosphatases (PPases) catalyze the hydrolysis of pyrophosphate to phosphates. PPases play essential roles in growth and development, and are found in all kingdoms of life. Human possess two PPases, PPA1 and PPA2. PPA1 is present in all tissues, acting largely as a housekeeping enzyme. Besides pyrophosphate hydrolysis, PPA1 can also directly dephosphorylate phosphorylated JNK1. Upregulated expression of PPA1 has been linked to many human malignant tumors. PPA1 knockdown induces apoptosis and decreases proliferation. PPA1 is emerging as a potential prognostic biomarker and target for anti-cancer drug development. In spite of the biological and physiopathological importance of PPA1, there is no detailed study on the structure and catalytic mechanisms of mammalian origin PPases. Here we report the crystal structure of human PPA1 at a resolution of 2.4 Å. We also carried out modeling studies of PPA1 in complex with JNK1 derived phosphor-peptides. The monomeric protein fold of PPA1 is similar to those found in other family I PPases. PPA1 forms a dimeric structure that should be conserved in animal and fungal PPases. Analysis of the PPA1 structure and comparison with available structures of PPases from lower organisms suggest that PPA1 has a largely pre-organized and relatively rigid active site for pyrophosphate hydrolysis. Results from the modeling study indicate the active site of PPA1 has the potential to accommodate double-phosphorylated peptides derived from JNK1. In short, results from the study provides new insights into the mechanisms of human PPA1 and basis for structure-based anti-cancer drug developments using PPA1 as the target.

Cysteine (Cys) is the most reactive amino acid participating in a wide range of biological functions. In-silico predictions complement the experiments to meet the need of functional characterization. Multiple Cys function prediction algorithm is scarce, in contrast to specific function prediction algorithms. Here we present a deep neural network-based multiple Cys function prediction, available on web-server (DeepCys) (https://deepcys.herokuapp.com/). DeepCys model was trained and tested on two independent datasets curated from protein crystal structures. This prediction method requires three inputs, namely, PDB identifier (ID), chain ID and residue ID for a given Cys and outputs the probabilities of four cysteine functions, namely, disulphide, metal-binding, thioether and sulphenylation and predicts the most probable Cys function. The algorithm exploits the local and global protein properties, like, sequence and secondary structure motifs, buried fractions, microenvironments and protein/enzyme class. DeepCys outperformed most of the multiple and specific Cys function algorithms. This method can predict maximum number of cysteine functions. Moreover, for the first time, explicitly predicts thioether function. This tool was used to elucidate the cysteine functions on domains of unknown functions (DUFs) belonging to cytochrome C oxidase subunit-II (COX2) like transmembrane domains. Apart from the web-server, a standalone program is also available on GitHub (https://github.com/vam-sin/deepcys)

We have investigated the pressure- and temperature-induced conformational changes associated with the low complexity domain of hnRNP A1, an RNA-binding protein able to phase separate in response to cellular stress. Solution NMR spectra of the hnRNP A1 low-complexity domain fused with protein-G B1 domain were collected from 1 to 2,500 bar and from 268 K to 290 K. While the GB1 domain shows the typical pressure-induced and cold temperature-induced unfolding expected for small globular domains, the low-complexity domain of hnRNP A1 exhibits unusual pressure and temperature dependences. We observed that the low-complexity domain is pressure sensitive, undergoing a major conformational transition within the prescribed pressure range. Remarkably, this transition has the inverse temperature dependence of a typical folding-unfolding transition. Our results suggest the presence of a low-lying extended, and fully solvated state(s) of the low-complexity domain that may play a role in phase separation. This study highlights the exquisite sensitivity of solution NMR spectroscopy to observe subtle conformational changes and illustrates how pressure perturbation can be used to determine the properties of metastable conformational ensembles.

One way in which trichocyte keratin intermediate filament proteins (keratins) and keratin associated proteins (KAPs) differ from their epithelial equivalents is in their higher levels of cysteine residues. Interactions between these cysteine residues within a mammalian fiber, and the putative regular organization of interactions (i.e., types of disulfide bond) are likely important for defining fiber mechanical properties, and thus biological functionality of hairs. Here we extend a previous study of cysteine accessibility under different levels of exposure to reducing compounds to explore a finer set of levels associated with interactions between keratins and KAPs. We found that most of the cysteines in the KAPs were close to either the N- or C- terminal domains of these proteins. The most accessible cysteines in keratins were present in the head or tail domains indicating their function in readily forming intermolecular bonds with KAPs. Some of the more buried cysteines in keratins were discovered either close to or within the rod region in positions previously identified in human epithelial keratins as being involved in crosslinking between the heterodimers of the tetramer. Our present study therefore provides a deeper understanding of the accessibility of disulfides especially in keratins and thus proves that there is some specificity to the disulfide bond interactions leading to these intermolecular bonds stabilizing the fiber structure.

Human histone H1 subtypes interaction networks was constructed to show a spectrum of their activities realized through the protein-protein interactions. Histone H1 subtypes participate in over half a thousand interactions with nuclear and cytosolic proteins engaged in the enzymatic activity and binding of nucleic acids and proteins. Small scale networks created by H1 subtypes are similar in their topological parameters (p > 0.05) but hub proteins of the networks formed with subtype H1.1 and H1.4 differ from those of subtype H1.3 and H1.5 in the closeness centrality, clustering coefficient and neighborhood connectivity (p < 0.05). Molecular function and biological process of the networks hubs is related to RNA binding and ribosome biogenesis (subtype H1.1 and H1.4), cell cycle and cell division (subtype H1.3 and H1.5) and protein ubiquitination and degradation (subtype H1.2). Such a disparity between H1 subtypes is also manifested by enriched GO terms of their interacting proteins. The residue propensity and secondary structures of interacting surfaces as well as a value of equilibrium dissociation constant indicate that a type of H1 subtypes interactions is transient in term of the stability and medium-strong in relation to the strength of binding. Histone H1 subtypes bind interacting partners in the intrinsic disorder–dependent mode, according to the coupled folding and binding and mutual synergistic folding mechanism. These results evidence that multifunctional H1 subtypes operate via protein interactions in the networks of crucial cellular processes and, therefore, confirm a new histone H1 paradigm relating to its functioning in the protein-protein interaction networks.

The structure of heterotetrameric sarcosine oxidase (HSO) contains a highly complex system composed of a large cavity and tunnels, which are essential for the reaction and migration of the reactants, products, and intermediates. Previous molecular dynamics (MD) simulation of HSO has identified the regions containing the water channels from the density distribution of water. The simulation is consistent with the selective transport hypothesis of the migration of the iminium intermediate, 5-oxazolidinone (5-OXA), of the enzyme reaction whereby tunnel T3 is the exit pathway of 5-OXA. In the present study, the potential of mean force (PMF) for the transport of 5-OXA through tunnels T1, T2, and T3 was calculated using umbrella sampling (US) MD simulations and the weighted histogram analysis method. The maximum errors of the calculated PMF were estimated by repeating the US simulations using different sets of initial positions. The PMF profiles for the three tunnels support the notion that tunnel T3 is the exit pathway of 5-OXA and that 5-OXA tends to stay at the middle of the tunnel. The PMF profile for the transport of glycine through tunnel T3 was also calculated to investigate where 5-OXA is converted into glycine, and how glycine is released to the outside of HSO was explained.

Polyene polyketides amphotericin B (AMB) and nystatin (NYS) are important antifungal drugs. Thioesterases (TEs), located at the last module of PKS, control the release of polyketides by cyclization or hydrolysis. Intrigued by the tiny structural difference between AMB and NYS, as well as the high sequence identity between AMB TE and NYS TE, we constructed four systems to study the structural characteristics, catalytic mechanism, and product release of AMB TE and NYS TE with combined MD simulations and QM/MM calculations. The results indicated that compared with AMB TE, NYS TE shows higher specificity on its natural substrate and R26 as well as D186 were proposed to a key role in substrate recognition. The energy barrier of macrocyclization in AMB-TE-Amb and AMB-TE-Nys systems were calculated to be 14.0 and 22.7 kcal/mol, while in NYS-TE-Nys and NYS-TE-Amb systems, their energy barriers were 17.5 and 25.7 kcal/mol, suggesting the cyclization with their natural substrates were more favorable than that with exchanged substrates. At last, the binding free energy obtained with the MM-PBSA.py program suggested that it was easier for natural products to leave TE enzymes after cyclization. And key residues to the departure of polyketide product from the active site were highlighted. We provided a catalytic overview of AMB TE and NYS TE including substrate recognition, catalytic mechanism and product release. These will improve the comprehension of polyene polyketide TEs and benefit for broadening the substrate flexibility of polyketide TEs.

To greatly expand the druggable genome, fast and accurate predictions of cryptic sites for small molecules binding in target proteins are in high demand. In this study, we have developed a fast and simple conformational sampling scheme guided by normal modes solved from the coarse-grained elastic models followed by atomistic backbone refinement and sidechain repacking. Despite the observations of complex and diverse conformational changes associated with ligand binding, we found that simply sampling along each of the lowest 30 modes is near optimal for adequately restructuring cryptic sites so they can be detected by existing pocket finding programs like fpocket and concavity. We further trained machine-learning protocols to optimize the combination of the sampling-enhanced pocket scores with other dynamic and conservation scores, which only slightly improved the performance. As assessed based on a training set of 84 known cryptic sites and a test set of 14 proteins, our method achieved high accuracy of prediction (with area under the receiver operating characteristic curve > 0.8) comparable to the CryptoSite server. Compared with CryptoSite and other methods based on extensive molecular dynamics simulation, our method is much faster (1-2 hours for an average-size protein) and simpler (using only pocket scores), so it is suitable for high-throughput processing of large datasets of protein structures at the genome scale.

The Protein Data Bank (PDB) file format remains a popular format used and supported by many software to represent coordinates of macromolecular structures. It however suffers from drawbacks such as error-prone manual editing. Because of that, various software toolkits have been developed to facilitate its editing and manipulation, but, to date, there is no online tool available for this purpose. Here we present PDB-Tools Web, a flexible online service for manipulating PDB files. It offers a rich and user-friendly graphical user interface that allows users to mix-and-match more than 40 individual tools from the pdb-tools suite. Those can be combined in a few clicks to perform complex pipelines, which can be saved and uploaded. The resulting processed PDB files can be visualized online and downloaded. The web server is freely available at https://wenmr.science.uu.nl/pdbtools.

Accurate prediction of protein secondary structure (alpha-helix, beta-strand and coil) is a crucial step for protein inter-residue contact prediction and ab initio tertiary structure prediction. In a previous study, we developed a deep belief network-based protein secondary structure method (DNSS1) and successfully advanced the prediction accuracy beyond 80%. In this work, we developed multiple advanced deep learning architectures (DNSS2) to further improve secondary structure prediction. The major improvements over the DNSS1 method include (i) designing and integrating six advanced one-dimensional deep convolutional/recurrent/residual/memory/fractal/inception networks to predict secondary structure, and (ii) using more sensitive profile features inferred from Hidden Markov model (HMM) and multiple sequence alignment (MSA). Most of the deep learning architectures are novel for protein secondary structure prediction. DNSS2 was systematically benchmarked on two independent test datasets with eight state-of-art tools and consistently ranked as one of the best methods. Particularly, DNSS2 was tested on the 82 protein targets of 2018 CASP13 experiment and achieved the best Q3 score of 83.74% and SOV score of 72.46%. DNSS2 is freely available at: https://github.com/multicom-toolbox/DNSS2.

The focal adhesion kinase (FAK) and the proline-rich tyrosine kinase 2-beta (PYK2) are implicated in cancer progression and metastasis and represent promising biomarkers and targets for cancer therapy. FAK and PYK2 are recruited to Focal Adhesions (Fas) via interactions between their Focal Adhesion Targeting (FAT) domains and conserved segments (LD motifs) on the proteins Paxillin, Leupaxin and Hic-5. A promising new approach for the inhibition of FAK and PYK2 targets interactions of the FAK domains with proteins that promote localization at Focal Adhesions. Advances toward this goal include the development of surface plasmon resonance, HSQC-NMR and fluorescence polarization assays for the identification of fragments or compounds interfering with the FAK-Paxillin interaction. We have recently validated this strategy, showing that Paxillin mimicking polypeptides with 2-3 LD motifs displace FAK from FAs and block kinase-dependent and independent functions of FAK, including downstream integrin signalling and FA localization of the protein p130Cas. In the present work we study by all-atom molecular dynamics simulations the recognition of peptides with the Paxillin and Leupaxin LD motifs by the FAK-FAT and PYK2-FAT domains. Our simulations and free-energy analysis interpret experimental data on binding of Paxillin and Leupaxin LD motifs at FAK-FAT and PYK2-FAT binding sites, and assess the roles of consensus LD regions and flanking residues. Our results can assist in the design of effective inhibitory peptides of the FAK-FAT:Paxillin and PYK2-FAT:Leupaxin complexes and the construction of pharmacophore models for the discovery of potential small-molecule inhibitors of the FAK-FAT and PYK2-FAT focal adhesion based functions.

Natural products and natural product-derived compounds have been widely used for pharmaceuticals for many years, and the search for new natural products that may have interesting activity is on going. Abyssomicins are natural product molecules that have antibiotic activity via inhibition of the folate synthesis pathway in microbiota. These compounds also appear to undergo a required [4+2] cycloaddition in their biosynthetic pathway. Here we report the structure of an FAD-dependent reductase, AbsH3, from the biosynthetic gene cluster of novel abyssomicins found in Streptomyces sp. LC-6-2.

Isoflavonoid is one of the groups of flavonoids that play pivotal roles in the survival of land plants. Chalcone synthase (CHS), the first enzyme of the isoflavonoid biosynthetic pathway, catalyzes the formation of a common isoflavonoid precursor. We have previously reported that an isozyme of soybean CHS (termed GmCHS1) is a key component of the isoflavonoid metabolon, a protein complex to enhance efficiency of isoflavonoid production. Here, we determined the crystal structure of GmCHS1 as a first step of understanding the metabolon structure, as well as to better understand the catalytic mechanism of GmCHS1.

The M42 aminopeptidases are a family of dinuclear aminopeptidases widely distributed in Prokaryotes. They are potentially associated to the proteasome, achieving complete peptide destruction. Their most peculiar characteristic is their quaternary structure, a tetrahedron-shaped particle made of twelve subunits. The catalytic site of M42 aminopeptidases is defined by seven conserved residues. Five of them are involved in metal ion binding which is important to maintain both the activity and the oligomeric state. The sixth conserved residue, a glutamate, is the catalytic base deprotonating the water molecule during peptide bond hydrolysis. The seventh residue is an aspartate whose function remains poorly understood. This aspartate residue, however, must have a critical role as it is strictly conserved in all MH clan enzymes. It forms some kind of catalytic triad with the histidine residue and the metal ion of the M2 binding site. We assess its role in TmPep1050, an M42 aminopeptidase of Thermotoga maritima, through a mutational approach. Asp-62 was substituted with alanine, asparagine, or glutamate residue. The three Asp-62 substitutions completely abolished TmPep1050 activity and impeded dodecamer formation. They also interfered with metal ion binding as only one cobalt ion is bound per subunit instead of two. The structural data showed that the Asp62Ala substitution has an impact on the active site folds becoming similar to TmPep1050 dimer. We propose a structural role for Asp-62, helping to stabilize a crucial loop in the active site and to position correctly the catalytic base and a metal ion ligand of the M1 site.

Protein-protein interactions (PPIs) are ubiquitous and functionally of great importance in biological systems. Hence, the ac-curate prediction of PPIs by protein-protein docking and scoring tools is highly desirable in order to characterize their structure and biological function. Ab initio docking protocols are divided into the sampling of docking poses to produce at least one near-native structure, then to evaluate the vast candidate structures by scoring. Concurrent development in both sampling and scoring is crucial for the deployment of protein-protein docking software. In the present work, we apply a machine learning model on pairwise potentials to refine the task of protein quaternary structure native structure detection among decoys. A decoy set was featurized using the Knowledge and Empirical Combined Scoring Algorithm 2 (KECSA2) pairwise potential. The highly unbalanced decoy set was then balanced using a comparison concept between native and decoy structures. The resultant comparison descriptors were used to train a logistic regression (LR) classifier. The LR model yielded the optimal performance for native detection among decoys compared to conventional scoring functions, while exhibiting lesser performance for the detection of low root mean square deviation (RMSD) decoy structures. Its deployment on an independent benchmark set confirms that the scoring function performs competitively relative to other scoring functions. All data and scripts used are available at: https://github.com/TanemuraKiyoto/PPI-native-detection-via-LR .

Structural characterization of alternatively folded and partially disordered protein conformations remains challenging. Outer surface protein A (OspA) is a pivotal protein in Borrelia infection, which is the etiological agent of Lyme disease. OspA exists in equilibrium with intermediate conformations, in which the central and the C-terminal regions of the protein have lower stabilities than the N-terminal. Here, we characterize pressure- and temperature-stabilized intermediates of OspA by nuclear magnetic resonance spectroscopy combined with paramagnetic relaxation enhancement (PRE). We found that the C-terminal region of the intermediate was partially disordered; however, it retains weak specific contact with the N-terminal region, owing to a twist of the central β-sheet and increased flexibility in the polypeptide chain. The disordered C-terminal region of the pressure-stabilized intermediate was more compact than that of the temperature-stabilized form. Further, molecular dynamics simulation demonstrated that temperature-induced disordering of the β-sheet was initiated at the C-terminal region and continued through to the central region. An ensemble of simulation snapshots qualitatively described the PRE data from the intermediate and indicated that the intermediate structures of OspA may expose tick receptor-binding sites more readily than does the basic folded conformation.