Computational design of Periplasmic binding protein biosensors guided by molecular dynamics.

Periplasmic binding proteins (PBPs) are bacterial proteins commonly used as scaffolds for substrate-detecting biosensors. In these biosensors, effector proteins (for example fluorescent proteins) are inserted into a PBP such that the effector protein's output changes upon PBP-substate binding. The insertion site is often determined by comparison of PBP apo/holo crystal structures, but random insertion libraries have shown that this can miss the best sites. Here, we present a PBP biosensor design method based on residue contact analysis from molecular dynamics. This computational method identifies the best previously known insertion sites in the maltose binding PBP, and suggests further previously unknown sites. We experimentally characterise fluorescent protein insertions at these new sites, finding they too give functional biosensors. Furthermore, our method is sufficiently flexible to both suggest insertion sites compatible with a variety of effector proteins, and be applied to binding proteins beyond PBPs.

PDBench: evaluating computational methods for protein-sequence design.

SUMMARY: Ever increasing amounts of protein structure data, combined with advances in machine learning, have led to the rapid proliferation of methods available for protein-sequence design. In order to utilize a design method effectively, it is important to understand the nuances of its performance and how it varies by design target. Here, we present PDBench, a set of proteins and a number of standard tests for assessing the performance of sequence-design methods. PDBench aims to maximize the structural diversity of the benchmark, compared with previous benchmarking sets, in order to provide useful biological insight into the behaviour of sequence-design methods, which is essential for evaluating their performance and practical utility. We believe that these tools are useful for guiding the development of novel sequence design algorithms and will enable users to choose a method that best suits their design target. AVAILABILITY AND IMPLEMENTATION: https://github.com/wells-wood-research/PDBench. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Generation of Photocaged Nanobodies for Intracellular Applications in an Animal Using Genetic Code Expansion and Computationally Guided Protein Engineering.

Nanobodies are becoming increasingly popular as tools for manipulating and visualising proteins in vivo. The ability to control nanobody/antigen interactions using light could provide precise spatiotemporal control over protein function. We develop a general approach to engineer photo-activatable nanobodies using photocaged amino acids that are introduced into the target binding interface by genetic code expansion. Guided by computational alanine scanning and molecular dynamics simulations, we tune nanobody/target binding affinity to eliminate binding before uncaging. Upon photo-activation using 365 nm light, binding is restored. We use this approach to generate improved photocaged variants of two anti-GFP nanobodies that function robustly when directly expressed in a complex intracellular environment together with their antigen. We apply them to control subcellular protein localisation in the nematode worm Caenorhabditis elegans. Our approach applies predictions derived from computational modelling directly in a living animal and demonstrates the importance of accounting for in vivo effects on protein-protein interactions.

De Novo Designed Peptide and Protein Hairpins Self-Assemble into Sheets and Nanoparticles.

The design and assembly of peptide-based materials has advanced considerably, leading to a variety of fibrous, sheet, and nanoparticle structures. A remaining challenge is to account for and control different possible supramolecular outcomes accessible to the same or similar peptide building blocks. Here a de novo peptide system is presented that forms nanoparticles or sheets depending on the strategic placement of a "disulfide pin" between two elements of secondary structure that drive self-assembly. Specifically, homodimerizing and homotrimerizing de novo coiled-coil α-helices are joined with a flexible linker to generate a series of linear peptides. The helices are pinned back-to-back, constraining them as hairpins by a disulfide bond placed either proximal or distal to the linker. Computational modeling indicates, and advanced microscopy shows, that the proximally pinned hairpins self-assemble into nanoparticles, whereas the distally pinned constructs form sheets. These peptides can be made synthetically or recombinantly to allow both chemical modifications and the introduction of whole protein cargoes as required.

BAlaS: fast, interactive and accessible computational alanine-scanning using BudeAlaScan.

MOTIVATION: In experimental protein engineering, alanine-scanning mutagenesis involves the replacement of selected residues with alanine to determine the energetic contribution of each side chain to forming an interaction. For example, it is often used to study protein-protein interactions. However, such experiments can be time-consuming and costly, which has led to the development of programmes for performing computational alanine-scanning mutagenesis (CASM) to guide experiments. While programmes are available for this, there is a need for a real-time web application that is accessible to non-expert users. RESULTS: Here, we present BAlaS, an interactive web application for performing CASM via BudeAlaScan and visualizing its results. BAlaS is interactive and intuitive to use. Results are displayed directly in the browser for the structure being interrogated enabling their rapid inspection. BAlaS has broad applications in areas, such as drug discovery and protein-interface design. AVAILABILITY AND IMPLEMENTATION: BAlaS works on all modern browsers and is available through the following website: https://balas.app. The project is open source, distributed using an MIT license and is available on GitHub (https://github.com/wells-wood-research/balas).

Navigating the Structural Landscape of De Novo α-Helical Bundles.

The association of amphipathic α helices in water leads to α-helical-bundle protein structures. However, the driving force for this-the hydrophobic effect-is not specific and does not define the number or the orientation of helices in the associated state. Rather, this is achieved through deeper sequence-to-structure relationships, which are increasingly being discerned. For example, for one structurally extreme but nevertheless ubiquitous class of bundle-the α-helical coiled coils-relationships have been established that discriminate between all-parallel dimers, trimers, and tetramers. Association states above this are known, as are antiparallel and mixed arrangements of the helices. However, these alternative states are less well understood. Here, we describe a synthetic-peptide system that switches between parallel hexamers and various up-down-up-down tetramers in response to single-amino-acid changes and solution conditions. The main accessible states of each peptide variant are characterized fully in solution and, in most cases, to high resolution with X-ray crystal structures. Analysis and inspection of these structures helps rationalize the different states formed. This navigation of the structural landscape of α-helical coiled coils above the dimers and trimers that dominate in nature has allowed us to design rationally a well-defined and hyperstable antiparallel coiled-coil tetramer (apCC-Tet). This robust de novo protein provides another scaffold for further structural and functional designs in protein engineering and synthetic biology.

Applying graph theory to protein structures: an Atlas of coiled coils.

Motivation: To understand protein structure, folding and function fully and to design proteins de novo reliably, we must learn from natural protein structures that have been characterized experimentally. The number of protein structures available is large and growing exponentially, which makes this task challenging. Indeed, computational resources are becoming increasingly important for classifying and analyzing this resource. Here, we use tools from graph theory to define an Atlas classification scheme for automatically categorizing certain protein substructures. Results: Focusing on the α-helical coiled coils, which are ubiquitous protein-structure and protein-protein interaction motifs, we present a suite of computational resources designed for analyzing these assemblies. iSOCKET enables interactive analysis of side-chain packing within proteins to identify coiled coils automatically and with considerable user control. Applying a graph theory-based Atlas classification scheme to structures identified by iSOCKET gives the Atlas of Coiled Coils, a fully automated, updated overview of extant coiled coils. The utility of this approach is illustrated with the first formal classification of an emerging subclass of coiled coils called α-helical barrels. Furthermore, in the Atlas, the known coiled-coil universe is presented alongside a partial enumeration of the 'dark matter' of coiled-coil structures; i.e. those coiled-coil architectures that are theoretically possible but have not been observed to date, and thus present defined targets for protein design. Availability and implementation: iSOCKET is available as part of the open-source GitHub repository associated with this work (https://github.com/woolfson-group/isocket). This repository also contains all the data generated when classifying the protein graphs. The Atlas of Coiled Coils is available at: http://coiledcoils.chm.bris.ac.uk/atlas/app.

ISAMBARD: an open-source computational environment for biomolecular analysis, modelling and design.

MOTIVATION: The rational design of biomolecules is becoming a reality. However, further computational tools are needed to facilitate and accelerate this, and to make it accessible to more users. RESULTS: Here we introduce ISAMBARD, a tool for structural analysis, model building and rational design of biomolecules. ISAMBARD is open-source, modular, computationally scalable and intuitive to use. These features allow non-experts to explore biomolecular design in silico. ISAMBARD addresses a standing issue in protein design, namely, how to introduce backbone variability in a controlled manner. This is achieved through the generalization of tools for parametric modelling, describing the overall shape of proteins geometrically, and without input from experimentally determined structures. This will allow backbone conformations for entire folds and assemblies not observed in nature to be generated de novo, that is, to access the 'dark matter of protein-fold space'. We anticipate that ISAMBARD will find broad applications in biomolecular design, biotechnology and synthetic biology. AVAILABILITY AND IMPLEMENTATION: A current stable build can be downloaded from the python package index (https://pypi.python.org/pypi/isambard/) with development builds available on GitHub (https://github.com/woolfson-group/) along with documentation, tutorial material and all the scripts used to generate the data described in this paper. CONTACT: d.n.woolfson@bristol.ac.uk or chris.wood@bristol.ac.uk. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Modular Design of Self-Assembling Peptide-Based Nanotubes.

An ability to design peptide-based nanotubes (PNTs) rationally with defined and mutable internal channels would advance understanding of peptide self-assembly, and present new biomaterials for nanotechnology and medicine. PNTs have been made from Fmoc dipeptides, cyclic peptides, and lock-washer helical bundles. Here we show that blunt-ended α-helical barrels, that is, preassembled bundles of α-helices with central channels, can be used as building blocks for PNTs. This approach is general and systematic, and uses a set of de novo helical bundles as standards. One of these bundles, a hexameric α-helical barrel, assembles into highly ordered PNTs, for which we have determined a structure by combining cryo-transmission electron microscopy, X-ray fiber diffraction, and model building. The structure reveals that the overall symmetry of the peptide module plays a critical role in ripening and ordering of the supramolecular assembly. PNTs based on pentameric, hexameric, and heptameric α-helical barrels sequester hydrophobic dye within their lumens.

CCBuilder: an interactive web-based tool for building, designing and assessing coiled-coil protein assemblies.

MOTIVATION: The ability to accurately model protein structures at the atomistic level underpins efforts to understand protein folding, to engineer natural proteins predictably and to design proteins de novo. Homology-based methods are well established and produce impressive results. However, these are limited to structures presented by and resolved for natural proteins. Addressing this problem more widely and deriving truly ab initio models requires mathematical descriptions for protein folds; the means to decorate these with natural, engineered or de novo sequences; and methods to score the resulting models. RESULTS: We present CCBuilder, a web-based application that tackles the problem for a defined but large class of protein structure, the α-helical coiled coils. CCBuilder generates coiled-coil backbones, builds side chains onto these frameworks and provides a range of metrics to measure the quality of the models. Its straightforward graphical user interface provides broad functionality that allows users to build and assess models, in which helix geometry, coiled-coil architecture and topology and protein sequence can be varied rapidly. We demonstrate the utility of CCBuilder by assembling models for 653 coiled-coil structures from the PDB, which cover >96% of the known coiled-coil types, and by generating models for rarer and de novo coiled-coil structures. AVAILABILITY AND IMPLEMENTATION: CCBuilder is freely available, without registration, at http://coiledcoils.chm.bris.ac.uk/app/cc_builder/.

TIMED-Design: flexible and accessible protein sequence design with convolutional neural networks.

Sequence design is a crucial step in the process of designing or engineering proteins. Traditionally, physics-based methods have been used to solve for optimal sequences, with the main disadvantages being that they are computationally intensive for the end user. Deep learning-based methods offer an attractive alternative, outperforming physics-based methods at a significantly lower computational cost. In this paper, we explore the application of Convolutional Neural Networks (CNNs) for sequence design. We describe the development and benchmarking of a range of networks, as well as reimplementations of previously described CNNs. We demonstrate the flexibility of representing proteins in a three-dimensional voxel grid by encoding additional design constraints into the input data. Finally, we describe TIMED-Design, a web application and command line tool for exploring and applying the models described in this paper. The user interface will be available at the URL: https://pragmaticproteindesign.bio.ed.ac.uk/timed. The source code for TIMED-Design is available at https://github.com/wells-wood-research/timed-design.

Molecular insights into LINC complex architecture through the crystal structure of a luminal trimeric coiled-coil domain of SUN1.

The LINC complex, consisting of interacting SUN and KASH proteins, mechanically couples nuclear contents to the cytoskeleton. In meiosis, the LINC complex transmits microtubule-generated forces to chromosome ends, driving the rapid chromosome movements that are necessary for synapsis and crossing over. In somatic cells, it defines nuclear shape and positioning, and has a number of specialised roles, including hearing. Here, we report the X-ray crystal structure of a coiled-coiled domain of SUN1's luminal region, providing an architectural foundation for how SUN1 traverses the nuclear lumen, from the inner nuclear membrane to its interaction with KASH proteins at the outer nuclear membrane. In combination with light and X-ray scattering, molecular dynamics and structure-directed modelling, we present a model of SUN1's entire luminal region. This model highlights inherent flexibility between structured domains, and raises the possibility that domain-swap interactions may establish a LINC complex network for the coordinated transmission of cytoskeletal forces.

Differential sensing with arrays of de novo designed peptide assemblies.

Differential sensing attempts to mimic the mammalian senses of smell and taste to identify analytes and complex mixtures. In place of hundreds of complex, membrane-bound G-protein coupled receptors, differential sensors employ arrays of small molecules. Here we show that arrays of computationally designed de novo peptides provide alternative synthetic receptors for differential sensing. We use self-assembling α-helical barrels (αHBs) with central channels that can be altered predictably to vary their sizes, shapes and chemistries. The channels accommodate environment-sensitive dyes that fluoresce upon binding. Challenging arrays of dye-loaded barrels with analytes causes differential fluorophore displacement. The resulting fluorimetric fingerprints are used to train machine-learning models that relate the patterns to the analytes. We show that this system discriminates between a range of biomolecules, drink, and diagnostically relevant biological samples. As αHBs are robust and chemically diverse, the system has potential to sense many analytes in various settings.

DE-STRESS: a user-friendly web application for the evaluation of protein designs.

De novo protein design is a rapidly growing field, and there are now many interesting and useful examples of designed proteins in the literature. However, most designs could be classed as failures when characterised in the lab, usually as a result of low expression, misfolding, aggregation or lack of function. This high attrition rate makes protein design unreliable and costly. It is possible that some of these failures could be caught earlier in the design process if it were quick and easy to generate information and a set of high-quality metrics regarding designs, which could be used to make reproducible and data-driven decisions about which designs to characterise experimentally. We present DE-STRESS (DEsigned STRucture Evaluation ServiceS), a web application for evaluating structural models of designed and engineered proteins. DE-STRESS has been designed to be simple, intuitive to use and responsive. It provides a wealth of information regarding designs, as well as tools to help contextualise the results and formally describe the properties that a design requires to be fit for purpose.

Peptide Assembly Directed and Quantified Using Megadalton DNA Nanostructures.

In nature, co-assembly of polypeptides, nucleic acids, and polysaccharides is used to create functional supramolecular structures. Here, we show that DNA nanostructures can be used to template interactions between peptides and to enable the quantification of multivalent interactions that would otherwise not be observable. Our functional building blocks are peptide-oligonucleotide conjugates comprising de novo designed dimeric coiled-coil peptides covalently linked to oligonucleotide tags. These conjugates are incorporated in megadalton DNA origami nanostructures and direct nanostructure association through peptide-peptide interactions. Free and bound nanostructures can be counted directly from electron micrographs, allowing estimation of the dissociation constants of the peptides linking them. Results for a single peptide-peptide interaction are consistent with the measured solution-phase free energy; DNA nanostructures displaying multiple peptides allow the effects of polyvalency to be probed. This use of DNA nanostructures as identifiers allows the binding strengths of homo- and heterodimeric peptide combinations to be measured in a single experiment and gives access to dissociation constants that are too low to be quantified by conventional techniques. The work also demonstrates that hybrid biomolecules can be programmed to achieve spatial organization of complex synthetic biomolecular assemblies.

CCBuilder 2.0: Powerful and accessible coiled-coil modeling.

The increased availability of user-friendly and accessible computational tools for biomolecular modeling would expand the reach and application of biomolecular engineering and design. For protein modeling, one key challenge is to reduce the complexities of 3D protein folds to sets of parametric equations that nonetheless capture the salient features of these structures accurately. At present, this is possible for a subset of proteins, namely, repeat proteins. The α-helical coiled coil provides one such example, which represents ≈ 3-5% of all known protein-encoding regions of DNA. Coiled coils are bundles of α helices that can be described by a small set of structural parameters. Here we describe how this parametric description can be implemented in an easy-to-use web application, called CCBuilder 2.0, for modeling and optimizing both α-helical coiled coils and polyproline-based collagen triple helices. This has many applications from providing models to aid molecular replacement for X-ray crystallography, in silico model building and engineering of natural and designed protein assemblies, and through to the creation of completely de novo "dark matter" protein structures. CCBuilder 2.0 is available as a web-based application, the code for which is open-source and can be downloaded freely. http://coiledcoils.chm.bris.ac.uk/ccbuilder2. LAY SUMMARY: We have created CCBuilder 2.0, an easy to use web-based application that can model structures for a whole class of proteins, the α-helical coiled coil, which is estimated to account for 3-5% of all proteins in nature. CCBuilder 2.0 will be of use to a large number of protein scientists engaged in fundamental studies, such as protein structure determination, through to more-applied research including designing and engineering novel proteins that have potential applications in biotechnology.

Maintaining and breaking symmetry in homomeric coiled-coil assemblies.

In coiled-coil (CC) protein structures α-helices wrap around one another to form rope-like assemblies. Most natural and designed CCs have two-four helices and cyclic (Cn) or dihedral (Dn) symmetry. Increasingly, CCs with five or more helices are being reported. A subset of these higher-order CCs is of interest as they have accessible central channels that can be functionalised; they are α-helical barrels. These extended cavities are surprising given the drive to maximise buried hydrophobic surfaces during protein folding and assembly in water. Here, we show that α-helical barrels can be maintained by the strategic placement of ß-branched aliphatic residues lining the lumen. Otherwise, the structures collapse or adjust to give more-complex multi-helix assemblies without Cn or Dn symmetry. Nonetheless, the structural hallmark of CCs-namely, knobs-into-holes packing of side chains between helices-is maintained leading to classes of CCs hitherto unobserved in nature or accessed by design.

De novo protein design: how do we expand into the universe of possible protein structures?

Protein scientists are paving the way to a new phase in protein design and engineering. Approaches and methods are being developed that could allow the design of proteins beyond the confines of natural protein structures. This possibility of designing entirely new proteins opens new questions: What do we build? How do we build into protein-structure space where there are few, if any, natural structures to guide us? To what uses can the resulting proteins be put? And, what, if anything, does this pursuit tell us about how natural proteins fold, function and evolve? We describe the origins of this emerging area of fully de novo protein design, how it could be developed, where it might lead, and what challenges lie ahead.

Computational design of water-soluble α-helical barrels.

The design of protein sequences that fold into prescribed de novo structures is challenging. General solutions to this problem require geometric descriptions of protein folds and methods to fit sequences to these. The α-helical coiled coils present a promising class of protein for this and offer considerable scope for exploring hitherto unseen structures. For α-helical barrels, which have more than four helices and accessible central channels, many of the possible structures remain unobserved. Here, we combine geometrical considerations, knowledge-based scoring, and atomistic modeling to facilitate the design of new channel-containing α-helical barrels. X-ray crystal structures of the resulting designs match predicted in silico models. Furthermore, the observed channels are chemically defined and have diameters related to oligomer state, which present routes to design protein function.