Engineering the kinetic stability of a ß-trefoil protein by tuning its topological complexity.

Kinetic stability, defined as the rate of protein unfolding, is central to determining the functional lifetime of proteins, both in nature and in wide-ranging medical and biotechnological applications. Further, high kinetic stability is generally correlated with high resistance against chemical and thermal denaturation, as well as proteolytic degradation. Despite its significance, specific mechanisms governing kinetic stability remain largely unknown, and few studies address the rational design of kinetic stability. Here, we describe a method for designing protein kinetic stability that uses protein long-range order, absolute contact order, and simulated free energy barriers of unfolding to quantitatively analyze and predict unfolding kinetics. We analyze two ß-trefoil proteins: hisactophilin, a quasi-three-fold symmetric natural protein with moderate stability, and ThreeFoil, a designed three-fold symmetric protein with extremely high kinetic stability. The quantitative analysis identifies marked differences in long-range interactions across the protein hydrophobic cores that partially account for the differences in kinetic stability. Swapping the core interactions of ThreeFoil into hisactophilin increases kinetic stability with close agreement between predicted and experimentally measured unfolding rates. These results demonstrate the predictive power of readily applied measures of protein topology for altering kinetic stability and recommend core engineering as a tractable target for rationally designing kinetic stability that may be widely applicable.

Potential Self-Peptide Inhibitors of the SARS-CoV-2 Main Protease.

The SARS-CoV-2 main protease (Mpro) plays an essential role in viral replication, cleaving viral polyproteins into functional proteins. This makes Mpro an important drug target. Mpro consists of an N-terminal catalytic domain and a C-terminal α-helical domain (MproC). Previous studies have shown that peptides derived from a given protein sequence (self-peptides) can affect the folding and, in turn, the function of that protein. Since the SARS-CoV-1 MproC is known to stabilize its Mpro and regulate its function, we hypothesized that SARS-CoV-2 MproC-derived self-peptides may modulate the folding and the function of SARS-CoV-2 Mpro. To test this, we studied the folding of MproC in the presence of various self-peptides using coarse-grained structure-based models and molecular dynamics simulations. In these simulations of MproC and one self-peptide, we found that two self-peptides, the α1-helix and the loop between α4 and α5 (loop4), could replace the equivalent native sequences in the MproC structure. Replacement of either sequence in full-length Mpro should, in principle, be able to perturb Mpro function albeit through different mechanisms. Some general principles for the rational design of self-peptide inhibitors emerge: The simulations show that prefolded self-peptides are more likely to replace native sequences than those which do not possess structure. Additionally, the α1-helix self-peptide is kinetically stable and once inserted rarely exchanges with the native α1-helix, while the loop4 self-peptide is easily replaced by the native loop4, making it less useful for modulating function. In summary, a prefolded α1-derived peptide should be able to inhibit SARS-CoV-2 Mpro function.

The diversity of protein-protein interaction interfaces within T=3 icosahedral viral capsids.

Some non-enveloped virus capsids assemble from multiple copies of a single type of coat-protein (CP). The comparative energetics of the diverse CP-CP interfaces present in such capsids likely govern virus assembly-disassembly mechanisms. The T = 3 icosahedral capsids comprise 180 CP copies arranged about two-, three-, five- and six-fold axes of (quasi-)rotation symmetry. Structurally diverse CPs can assemble into T = 3 capsids. Specifically, the Leviviridae CPs are structurally distinct from the Bromoviridae, Tombusviridae and Tymoviridae CPs which fold into the classic "jelly-roll" fold. However, capsids from across the four families are known to disassemble into dimers. To understand whether the overall symmetry of the capsid or the structural details of the CP determine virus assembly-disassembly mechanisms, we analyze the different CP-CP interfaces that occur in the four virus families. Previous work studied protein homodimer interfaces using interface size (relative to the monomer) and hydrophobicity. Here, we analyze all CP-CP interfaces using these two parameters and find that the dimerization interface (present between two CPs congruent through a two-fold axis of rotation) has a larger relative size in the Leviviridae than in the other viruses. The relative sizes of the other Leviviridae interfaces and all the jelly-roll interfaces are similar. However, the dimerization interfaces across families have slightly higher hydrophobicity, potentially making them stronger than other interfaces. Finally, although the CP-monomers of the jelly-roll viruses are structurally similar, differences in their dimerization interfaces leads to varied dimer flexibility. Overall, differences in CP-structures may induce different modes of swelling and assembly-disassembly in the T = 3 viruses.

A Method for Assessing the Robustness of Protein Structures by Randomizing Packing Interactions.

Many single-domain proteins are not only stable and water-soluble, but they also populate few to no intermediates during folding. This reduces interactions between partially folded proteins, misfolding, and aggregation, and makes the proteins tractable in biotechnological applications. Natural proteins fold thus, not necessarily only because their structures are well-suited for folding, but because their sequences optimize packing and fit their structures well. In contrast, folding experiments on the de novo designed Top7 suggest that it populates several intermediates. Additionally, in de novo protein design, where sequences are designed for natural and new non-natural structures, tens of sequences still need to be tested before success is achieved. Both these issues may be caused by the specific scaffolds used in design, i.e., some protein scaffolds may be more tolerant to packing perturbations and varied sequences. Here, we report a computational method for assessing the response of protein structures to packing perturbations. We then benchmark this method using designed proteins and find that it can identify scaffolds whose folding gets disrupted upon perturbing packing, leading to the population of intermediates. The method can also isolate regions of both natural and designed scaffolds that are sensitive to such perturbations and identify contacts which when present can rescue folding. Overall, this method can be used to identify protein scaffolds that are more amenable to whole protein design as well as to identify protein regions which are sensitive to perturbations and where further mutations should be avoided during protein engineering.

Rational Design of Protein-Specific Folding Modifiers.

Protein-folding can go wrong in vivo and in vitro, with significant consequences for the living organism and the pharmaceutical industry, respectively. Here we propose a design principle for small-peptide-based protein-specific folding modifiers. The principle is based on constructing a "xenonucleus", which is a prefolded peptide that mimics the folding nucleus of a protein. Using stopped-flow kinetics, NMR spectroscopy, Förster resonance energy transfer, single-molecule force measurements, and molecular dynamics simulations, we demonstrate that a xenonucleus can make the refolding of ubiquitin faster by 33 ± 5%, while variants of the same peptide have little or no effect. Our approach provides a novel method for constructing specific, genetically encodable folding catalysts for suitable proteins that have a well-defined contiguous folding nucleus.

Understanding the Folding Mediated Assembly of the Bacteriophage MS2 Coat Protein Dimers.

The capsids of RNA viruses such as MS2 are great models for studying protein self-assembly because they are made almost entirely of multiple copies of a single coat protein (CP). Although CP is the minimal repeating unit of the capsid, previous studies have shown that CP exists as a homodimer (CP2) even in an acid-disassembled system, indicating that CP2 is an obligate dimer. Here, we investigate the molecular basis of this obligate dimerization using coarse-grained structure-based models and molecular dynamics simulations. We find that, unlike monomeric proteins of similar size, CP populates a single partially folded ensemble whose "foldedness" is sensitive to denaturing conditions. In contrast, CP2 folds similarly to single-domain proteins populating only the folded and the unfolded ensembles, separated by a prominent folding free energy barrier. Several intramonomer contacts form early, but the CP2 folding barrier is crossed only when the intermonomer contacts are made. A dissection of the structure of CP2 through mutant folding simulations shows that the folding barrier arises both from the topology of CP and the interface contacts of CP2. Together, our results show that CP2 is an obligate dimer because of kinetic stability, that is, dimerization induces a folding barrier and that makes it difficult for proteins in the dimer minimum to partially unfold and access the monomeric state without completely unfolding. We discuss the advantages of this obligate dimerization in the context of dimer design and virus stability.

The Molecular Mechanism of Domain Swapping of the C-Terminal Domain of the SARS-Coronavirus Main Protease.

In three-dimensional domain swapping, two protein monomers exchange a part of their structures to form an intertwined homodimer, whose subunits resemble the monomer. Several viral proteins domain swap to increase their structural complexity or functional avidity. The main protease (Mpro) of the severe acute respiratory syndrome (SARS) coronavirus proteolyzes viral polyproteins and has been a target for anti-SARS drug design. Domain swapping in the α-helical C-terminal domain of Mpro (MproC) locks Mpro into a hyperactive octameric form that is hypothesized to promote the early stages of viral replication. However, in the absence of a complete molecular understanding of the mechanism of domain swapping, investigations into the biological relevance of this octameric Mpro have stalled. Isolated MproC can exist as a monomer or a domain-swapped dimer. Here, we investigate the mechanism of domain swapping of MproC using coarse-grained structure-based models and molecular dynamics simulations. Our simulations recapitulate several experimental features of MproC folding. Further, we find that a contact between a tryptophan in the MproC domain-swapping hinge and an arginine elsewhere forms early during folding, modulates the folding route, and promotes domain swapping to the native structure. An examination of the sequence and the structure of the tryptophan containing hinge loop shows that it has a propensity to form multiple secondary structures and contacts, indicating that it could be stabilized into either the monomer- or dimer-promoting conformations by mutations or ligand binding. Finally, because all residues in the tryptophan loop are identical in SARS-CoV and SARS-CoV-2, mutations that modulate domain swapping may provide insights into the role of octameric Mpro in the early-stage viral replication of both viruses.

Protein energy landscape exploration with structure-based models.

Exploring the multi-dimensional energy landscape of a large protein in detail is a computational challenge. Such investigations may include analysis of multiple folding pathways, rate constants for important conformational transitions, locating intermediate states populated during folding, estimating energetic and entropic barriers that separate populated basins, and visualising a high-dimensional surface. The complexity of the landscape can be simplified through coarse-grained structure-based models (SBMs). These widely used coarse-grained representations of proteins provide a minimalist approximation to the free energy landscape, which subsumes the folding behaviour of many single-domain proteins. Here we describe the combination of SBMs with discrete path sampling (DPS), and show how this approach can provide details of the landscape and folding pathways. Combining SBMs and DPS provides an efficient framework for sampling the protein free energy landscape and for calculating various kinetic and thermodynamic quantities.

Structure dictates the mechanism of ligand recognition in the histidine and maltose binding proteins.

Two mechanisms, induced fit (IF) and conformational selection (CS), have been proposed to explain ligand recognition coupled conformational changes. The histidine binding protein (HisJ) adopts the CS mechanism, in which a pre-equilibrium is established between the open and the closed states with the ligand binding to the closed state. Despite being structurally similar to HisJ, the maltose binding protein (MBP) adopts the IF mechanism, in which the ligand binds the open state and induces a transition to the closed state. To understand the molecular determinants of this difference, we performed molecular dynamics (MD) simulations of coarse-grained dual structure based models. We find that intra-protein contacts unique to the closed state are sufficient to promote the conformational transition in HisJ, indicating a CS-like mechanism. In contrast, additional ligand-mimicking contacts are required to "induce" the conformational transition in MBP suggesting an IF-like mechanism. In agreement with experiments, destabilizing modifications to two structural features, the spine helix (SH) and the balancing interface (BI), present in MBP but absent in HisJ, reduce the need for ligand-mimicking contacts indicating that SH and BI act as structural restraints that keep MBP in the open state. We introduce an SH like element into HisJ and observe that this can impede the conformational transition increasing the importance of ligand-mimicking contacts. Similarly, simultaneous mutations to BI and SH in MBP reduce the barrier to conformational transitions significantly and promote a CS-like mechanism. Together, our results show that structural restraints present in the protein structure can determine the mechanism of conformational transitions and even simple models that correctly capture such structural features can predict their positions. MD simulations of such models can thus be used, in conjunction with mutational experiments, to regulate protein ligand interactions, and modulate ligand binding affinities.

Successes and challenges in simulating the folding of large proteins.

Computational simulations of protein folding can be used to interpret experimental folding results, to design new folding experiments, and to test the effects of mutations and small molecules on folding. However, whereas major experimental and computational progress has been made in understanding how small proteins fold, research on larger, multidomain proteins, which comprise the majority of proteins, is less advanced. Specifically, large proteins often fold via long-lived partially folded intermediates, whose structures, potentially toxic oligomerization, and interactions with cellular chaperones remain poorly understood. Molecular dynamics based folding simulations that rely on knowledge of the native structure can provide critical, detailed information on folding free energy landscapes, intermediates, and pathways. Further, increases in computational power and methodological advances have made folding simulations of large proteins practical and valuable. Here, using serpins that inhibit proteases as an example, we review native-centric methods for simulating the folding of large proteins. These synergistic approaches range from Go and related structure-based models that can predict the effects of the native structure on folding to all-atom-based methods that include side-chain chemistry and can predict how disease-associated mutations may impact folding. The application of these computational approaches to serpins and other large proteins highlights the successes and limitations of current computational methods and underscores how computational results can be used to inform experiments. These powerful simulation approaches in combination with experiments can provide unique insights into how large proteins fold and misfold, expanding our ability to predict and manipulate protein folding.

Go-Kit: A Tool To Enable Energy Landscape Exploration of Proteins.

Coarse-grained Go̅-like models, based on the principle of minimal frustration, provide valuable insight into fundamental questions in the field of protein folding and dynamics. In conjunction with commonly used molecular dynamics (MD) simulations, energy landscape exploration methods like discrete path sampling (DPS) with Go̅-like models can provide quantitative details of the thermodynamics and kinetics of proteins. Here we present Go-kit, a software that facilitates the setup of MD and DPS simulations of several flavors of Go̅-like models. Go-kit is designed for use with MD (GROMACS) and DPS (PATHSAMPLE) simulation engines that are open source. The Go-kit code is written in python2.7 and is also open source. A case study for the ribosomal protein S6 is discussed to illustrate the utility of the software, which is available at https://github.com/gokit1/gokit .

A five-residue motif for the design of domain swapping in proteins.

Domain swapping is the process by which identical monomeric proteins exchange structural elements to generate dimers/oligomers. Although engineered domain swapping is a compelling strategy for protein assembly, its application has been limited due to the lack of simple and reliable design approaches. Here, we demonstrate that the hydrophobic five-residue 'cystatin motif' (QVVAG) from the domain-swapping protein Stefin B, when engineered into a solvent-exposed, tight surface loop between two ß-strands prevents the loop from folding back upon itself, and drives domain swapping in non-domain-swapping proteins. High-resolution structural studies demonstrate that engineering the QVVAG stretch independently into various surface loops of four structurally distinct non-domain-swapping proteins enabled the design of different modes of domain swapping in these proteins, including single, double and open-ended domain swapping. These results suggest that the introduction of the QVVAG motif can be used as a mutational approach for engineering domain swapping in diverse ß-hairpin proteins.

Energy Landscape of the Designed Protein Top7.

To fold on biologically relevant time scales, proteins have evolved funnelled energy landscapes with minimal energetic trapping. However, the polymeric nature of proteins and the spatial arrangement of secondary structural elements can create topological traps and slow folding. It is challenging to identify, visualize, and quantify such topological trapping. Designed proteins have not had the benefit of evolution, and it has been hypothesized that de novo designed protein topologies may therefore feature more topological trapping. Structure-based models (SBMs) are inherently funnelled, removing most energetic trapping, and can thus be used to isolate the effect of protein topology on the landscape. Here, we compare Top7, a designed protein with a topology unknown in nature, to S6, a naturally occurring ribosomal protein of similar size and topology. Possible kinetic traps and the energetic barriers separating them from the native state are elucidated. We find that, even with an SBM, the potential energy landscape (PEL) of the designed protein is more frustrated than that of the natural protein. We then quantify the effect of adding non-native hydrophobic interactions and coarse-grained side-chains through a frustration density parameter. A clear increase in frustration is observed including side-chains, whereas adding hydrophobic interactions leads to a narrowing of the funnel and a decrease in complexity. The most likely (un)folding routes for all models are derived through the construction of probability contact maps. The ability to quantitatively understand and optimize the organization of the PEL for designed proteins may enable us to design structure-seeking landscapes, mimicking the effect of evolution.

The Sensitivity of Computational Protein Folding to Contact Map Perturbations: The Case of Ubiquitin Folding and Function.

Ubiquitin is a small model protein, commonly used in protein folding experiments and simulations. We simulated ubiquitin using a well-tested structure-based model coarse-grained to a Cα level (Cα-SBM) and found that the simulated folding route did not agree with the experimentally observed one. Simulating the Cα-SBM with a cutoff contact map, instead of a screened contact map, switched the folding route with the new route matching the experimental route. Thus, the simulated folding of ubiquitin is sensitive to contact map definition. The screened contact map, which is used in folding simulations because it captures protein folding cooperativity, removes contacts in which the atoms in contact are occluded by a third atom and is less sensitive to the value of the cutoff distance in well-packed regions of the protein. In sparsely packed regions, the larger cutoff distance creates bridging contacts between atoms which are separated by voids. Such contacts do not seem to affect the folding of most proteins, including those of the ubiquitin fold. However, the surface of ubiquitin has several protruding functional side chains which naturally create bridging contacts. Together, our results show that subtle structural features of a protein that may not be apparent by mere observation can be identified by comparing folding simulations of SBMs in which these features are differently encoded. When such structural features are preserved for functional reasons, differences in computational folding can be leveraged to identify functional features. Notably, such features are accessible to a gradation of SBMs even in commonly studied proteins such as ubiquitin.

Intrinsic Disorder in a Well-Folded Globular Protein.

The folded structure of the heterodimeric sweet protein monellin mimics single-chain proteins with topology ß1-α1-ß2-ß3-ß4-ß5 (chain A: ß3-ß4-ß5; chain B: ß1-α1-ß2). Furthermore, like naturally occurring single-chain proteins of a similar size, monellin folds cooperatively with no detectable intermediates. However, the two monellin chains, A and B, are marginally structured in isolation and fold only upon binding to each other. Thus, monellin presents a unique opportunity to understand the design of intrinsically disordered proteins that fold upon binding. Here, we study the folding of a single-chain variant of monellin (scMn) using simulations of an all heavy-atom structure-based model. These simulations can explain mechanistic details derived from scMn experiments performed using several different structural probes. scMn folds cooperatively in our structure-based simulations, as is also seen in experiments. We find that structure formation near the transition-state ensemble of scMn is not uniformly distributed but is localized to a hairpin-like structure which contains one strand from each chain (ß2, ß3). Thus, the sequence and the underlying energetics of heterodimeric monellin promote the early formation of the interchain interface (ß2-ß3). By studying computational scMn mutants whose "interchain" interactions are deleted, we infer that this energy distribution allows the two protein chains to remain largely disordered when this interface is not folded. From these results, we suggest that cutting the protein backbone of a globular protein between residues which lie within its folding nucleus may be one way to construct two disordered fragments which fold upon binding.

On the folding of a structurally complex protein to its metastable active state.

For successful protease inhibition, the reactive center loop (RCL) of the two-domain serine protease inhibitor, α1-antitrypsin (α1-AT), needs to remain exposed in a metastable active conformation. The α1-AT RCL is sequestered in a ß-sheet in the stable latent conformation. Thus, to be functional, α1-AT must always fold to a metastable conformation while avoiding folding to a stable conformation. We explore the structural basis of this choice using folding simulations of coarse-grained structure-based models of the two α1-AT conformations. Our simulations capture the key features of folding experiments performed on both conformations. The simulations also show that the free energy barrier to fold to the latent conformation is much larger than the barrier to fold to the active conformation. An entropically stabilized on-pathway intermediate lowers the barrier for folding to the active conformation. In this intermediate, the RCL is in an exposed configuration, and only one of the two α1-AT domains is folded. In contrast, early conversion of the RCL into a ß-strand increases the coupling between the two α1-AT domains in the transition state and creates a larger barrier for folding to the latent conformation. Thus, unlike what happens in several proteins, where separate regions promote folding and function, the structure of the RCL, formed early during folding, determines both the conformational and the functional fate of α1-AT. Further, the short 12-residue RCL modulates the free energy barrier and the folding cooperativity of the large 370-residue α1-AT. Finally, we suggest experiments to test the predicted folding mechanism for the latent state.

Exploring the effects of sparse restraints on protein structure prediction.

One of the main barriers to accurate computational protein structure prediction is searching the vast space of protein conformations. Distance restraints or inter-residue contacts have been used to reduce this search space, easing the discovery of the correct folded state. It has been suggested that about 1 contact for every 12 residues may be sufficient to predict structure at fold level accuracy. Here, we use coarse-grained structure-based models in conjunction with molecular dynamics simulations to examine this empirical prediction. We generate sparse contact maps for 15 proteins of varying sequence lengths and topologies and find that given perfect secondary-structural information, a small fraction of the native contact map (5%-10%) suffices to fold proteins to their correct native states. We also find that different sparse maps are not equivalent and we make several observations about the type of maps that are successful at such structure prediction. Long range contacts are found to encode more information than shorter range ones, especially for α and αß-proteins. However, this distinction reduces for ß-proteins. Choosing contacts that are a consensus from successful maps gives predictive sparse maps as does choosing contacts that are well spread out over the protein structure. Additionally, the folding of proteins can also be used to choose predictive sparse maps. Overall, we conclude that structure-based models can be used to understand the efficacy of structure-prediction restraints and could, in future, be tuned to include specific force-field interactions, secondary structure errors and noise in the sparse maps.

Characterizing the conformational landscape of MDM2-binding p53 peptides using Molecular Dynamics simulations.

The conformational landscapes of p53 peptide variants and phage derived peptide (12/1) variants, all known to bind to MDM2, are studied using hamiltonian replica exchange molecular dynamics simulations. Complementing earlier observations, the current study suggests that the p53 peptides largely follow the 'conformational selection' paradigm in their recognition of and complexation by MDM2 while the 12/1 peptides likely undergo some element of conformational selection but are mostly driven by 'binding induced folding'. This hypothesis is further supported by pulling simulations that pull the peptides away from their bound states with MDM2. This data extends the earlier mechanisms proposed to rationalize the entropically driven binding of the p53 set and the enthalpically driven binding of the 12/1 set. Using our hypothesis, we suggest mutations to the 12/1 peptide that increase its helicity in simulations and may, in turn, shift the binding towards conformational selection. In summary, understanding the conformational landscapes of the MDM2-binding peptides may suggest new peptide designs with bespoke binding mechanisms.

Amino-acid composition after loop deletion drives domain swapping.

Rational engineering of a protein to enable domain swapping requires an understanding of the sequence, structural and energetic factors that favor the domain-swapped oligomer over the monomer. While it is known that the deletion of loops between ß-strands can promote domain swapping, the spliced sequence at the position of the loop deletion is thought to have a minimal role to play in such domain swapping. Here, two loop-deletion mutants of the non-domain-swapping protein monellin, frame-shifted by a single residue, were designed. Although the spliced sequence in the two mutants differed by only one residue at the site of the deletion, only one of them (YEIKG) promoted domain swapping. The mutant containing the spliced sequence YENKG was entirely monomeric. This new understanding that the domain swapping propensity after loop deletion may depend critically on the chemical composition of the shortened loop will facilitate the rational design of domain swapping.