Control over overall shape and size in de novo designed proteins

PB-GPT: An innovative GPT-based model for protein backbone generation

With advanced computational methods, it is now feasible to modify or design proteins for specific functions, a process with significant implications for disease treatment and other medical applications. Protein structures and functions are intrinsically linked to their backbones, making the design of these backbones a pivotal aspect of protein engineering. In this study, we focus on the task of unconditionally generating protein backbones. By means of codebook quantization and compression dictionaries, we convert protein backbone structures into a distinctive coded language and propose a GPT-based protein backbone generation model, PB-GPT. To validate the generalization performance of the model, we trained and evaluated the model on both public datasets and small protein datasets. The results demonstrate that our model has the capability to unconditionally generate elaborate, highly realistic protein backbones with structural patterns resembling those of natural proteins, thus showcasing the significant potential of large language models in protein structure design.

Design of amyloidogenic peptide traps

Segments of proteins with high β-strand propensity can self-associate to form amyloid fibrils implicated in many diseases. We describe a general approach to bind such segments in β-strand and β-hairpin conformations using de novo designed scaffolds that contain deep peptide-binding clefts. The designs bind their cognate peptides in vitro with nanomolar affinities. The crystal structure of a designed protein-peptide complex is close to the design model, and NMR characterization reveals how the peptide-binding cleft is protected in the apo state. We use the approach to design binders to the amyloid-forming proteins transthyretin, tau, serum amyloid A1 and amyloid β1-42 (Aβ42). The Aβ binders block the assembly of Aβ fibrils as effectively as the most potent of the clinically tested antibodies to date and protect cells from toxic Aβ42 species.

An ancestral fold reveals the evolutionary link between RNA polymerase and ribosomal proteins

Numerous molecular machines are required to drive the central dogma of molecular biology. However, the means by which these numerous proteins emerged in the early evolutionary stage of life remains enigmatic. Many of them possess small β-barrel folds with different topologies, represented by double-psi β-barrels (DPBBs) conserved in DNA and RNA polymerases, and similar but topologically distinct six-stranded β-barrel RIFT or five-stranded β-barrel folds such as OB and SH3 in ribosomal proteins. Here, we discover that the previously reconstructed ancient DPBB sequence could also adopt a β-barrel fold named Double-Zeta β-barrel (DZBB), as a metamorphic protein. The DZBB fold is not found in any modern protein, although its structure shares similarities with RIFT and OB. Indeed, DZBB could be transformed into them through simple engineering experiments. Furthermore, the OB designs could be further converted into SH3 by circular-permutation as previously predicted. These results indicate that these β-barrels diversified quickly from a common ancestor at the beginning of the central dogma evolution.

Validation of de novo designed water-soluble and transmembrane β-barrels by in silico folding and melting

In silico validation of de novo designed proteins with deep learning (DL)-based structure prediction algorithms has become mainstream. However, formal evidence of the relationship between a high-quality predicted model and the chance of experimental success is lacking. We used experimentally characterized de novo water-soluble and transmembrane β-barrel designs to show that AlphaFold2 and ESMFold excel at different tasks. ESMFold can efficiently identify designs generated based on high-quality (designable) backbones. However, only AlphaFold2 can predict which sequences have the best chance of experimentally folding among similar designs. We show that ESMFold can generate high-quality structures from just a few predicted contacts and introduce a new approach based on incremental perturbation of the prediction ("in silico melting"), which can reveal differences in the presence of favorable contacts between designs. This study provides a new insight on DL-based structure prediction models explainability and on how they could be leveraged for the design of increasingly complex proteins; in particular membrane proteins which have historically lacked basic in silico validation tools.

Density Function Theory Study on the Energy and Circular Dichroism Spectrum for Methylene-Linked Triazole Diads Depending on the Substitution Position and Conformation

Since the discovery of metal-catalyzed azide-alkyne cycloadditions, 1,2,3-triazoles have been widely used as linkers for various residues. 1,2,3-Triazole is an aromatic five-membered cyclic compound consisting of three nitrogen and two carbon atoms with large dipoles that absorb UV light. In the past decade, we have been working on the synthesis of dense triazole polymers possessing many 1,2,3-triazole residues linked through a carbon atom in their backbone as a new type of functional polymer. Recently, we reported that stereoregular dense triazole uniform oligomers exhibit a circular dichroism signal based on the chiral arrangement of two neighboring 1,2,3-triazole residues. In this study, to investigate the chiral conformation of two neighboring 1,2,3-triazole residues in stereoregular dense triazole uniform oligomers, density functional theory (DFT) calculations were performed using 1,2,3-triazole diads with different substitution positions and conformations as model compounds and compared with our previous results.

AI in cellular engineering and reprogramming

During the last decade, artificial intelligence (AI) has increasingly been applied in biophysics and related fields, including cellular engineering and reprogramming, offering novel approaches to understand, manipulate, and control cellular function. The potential of AI lies in its ability to analyze complex datasets and generate predictive models. AI algorithms can process large amounts of data from single-cell genomics and multiomic technologies, allowing researchers to gain mechanistic insights into the control of cell identity and function. By integrating and interpreting these complex datasets, AI can help identify key molecular events and regulatory pathways involved in cellular reprogramming. This knowledge can inform the design of precision engineering strategies, such as the development of new transcription factor and signaling molecule cocktails, to manipulate cell identity and drive authentic cell fate across lineage boundaries. Furthermore, when used in combination with computational methods, AI can accelerate and improve the analysis and understanding of the intricate relationships between genes, proteins, and cellular processes. In this review article, we explore the current state of AI applications in biophysics with a specific focus on cellular engineering and reprogramming. Then, we showcase a couple of recent applications where we combined machine learning with experimental and computational techniques. Finally, we briefly discuss the challenges and prospects of AI in cellular engineering and reprogramming, emphasizing the potential of these technologies to revolutionize our ability to engineer cells for a variety of applications, from disease modeling and drug discovery to regenerative medicine and biomanufacturing.

Design of complicated all-α protein structures

A wide range of de novo protein structure designs have been achieved, but the complexity of naturally occurring protein structures is still far beyond these designs. Here, to expand the diversity and complexity of de novo designed protein structures, we sought to develop a method for designing 'difficult-to-describe' α-helical protein structures composed of irregularly aligned α-helices like globins. Backbone structure libraries consisting of a myriad of α-helical structures with five or six helices were generated by combining 18 helix-loop-helix motifs and canonical α-helices, and five distinct topologies were selected for de novo design. The designs were found to be monomeric with high thermal stability in solution and fold into the target topologies with atomic accuracy. This study demonstrated that complicated α-helical proteins are created using typical building blocks. The method we developed will enable us to explore the universe of protein structures for designing novel functional proteins.

ProRefiner: an entropy-based refining strategy for inverse protein folding with global graph attention

Inverse Protein Folding (IPF) is an important task of protein design, which aims to design sequences compatible with a given backbone structure. Despite the prosperous development of algorithms for this task, existing methods tend to rely on noisy predicted residues located in the local neighborhood when generating sequences. To address this limitation, we propose an entropy-based residue selection method to remove noise in the input residue context. Additionally, we introduce ProRefiner, a memory-efficient global graph attention model to fully utilize the denoised context. Our proposed method achieves state-of-the-art performance on multiple sequence design benchmarks in different design settings. Furthermore, we demonstrate the applicability of ProRefiner in redesigning Transposon-associated transposase B, where six out of the 20 variants we propose exhibit improved gene editing activity.

Harnessing the power of artificial intelligence to advance cell therapy

Cell therapies are powerful technologies in which human cells are reprogrammed for therapeutic applications such as killing cancer cells or replacing defective cells. The technologies underlying cell therapies are increasing in effectiveness and complexity, making rational engineering of cell therapies more difficult. Creating the next generation of cell therapies will require improved experimental approaches and predictive models. Artificial intelligence (AI) and machine learning (ML) methods have revolutionized several fields in biology including genome annotation, protein structure prediction, and enzyme design. In this review, we discuss the potential of combining experimental library screens and AI to build predictive models for the development of modular cell therapy technologies. Advances in DNA synthesis and high-throughput screening techniques enable the construction and screening of libraries of modular cell therapy constructs. AI and ML models trained on this screening data can accelerate the development of cell therapies by generating predictive models, design rules, and improved designs.

Single-chain dimers from de novo immunoglobulins as robust scaffolds for multiple binding loops

Antibody derivatives have sought to recapitulate the antigen binding properties of antibodies, but with improved biophysical attributes convenient for therapeutic, diagnostic and research applications. However, their success has been limited by the naturally occurring structure of the immunoglobulin dimer displaying hypervariable binding loops, which is hard to modify by traditional engineering approaches. Here, we devise geometrical principles for de novo designing single-chain immunoglobulin dimers, as a tunable two-domain architecture that optimizes biophysical properties through more favorable dimer interfaces. Guided by these principles, we computationally designed protein scaffolds that were hyperstable, structurally accurate and robust for accommodating multiple functional loops, both individually and in combination, as confirmed through biochemical assays and X-ray crystallography. We showcase the modularity of this architecture by deep-learning-based diversification, opening up the possibility for tailoring the number, positioning, and relative orientation of ligand-binding loops targeting one or two distal epitopes. Our results provide a route to custom-design robust protein scaffolds for harboring multiple functional loops.

Design of stimulus-responsive two-state hinge proteins

In nature, proteins that switch between two conformations in response to environmental stimuli structurally transduce biochemical information in a manner analogous to how transistors control information flow in computing devices. Designing proteins with two distinct but fully structured conformations is a challenge for protein design as it requires sculpting an energy landscape with two distinct minima. Here we describe the design of "hinge" proteins that populate one designed state in the absence of ligand and a second designed state in the presence of ligand. X-ray crystallography, electron microscopy, double electron-electron resonance spectroscopy, and binding measurements demonstrate that despite the significant structural differences the two states are designed with atomic level accuracy and that the conformational and binding equilibria are closely coupled.

Exploration of novel αβ-protein folds through de novo design

A fundamental question in protein evolution is whether nature has exhaustively sampled nearly all possible protein folds throughout evolution, or whether a large fraction of the possible folds remains unexplored. To address this question, we defined a set of rules for β-sheet topology to predict novel αβ-folds and carried out a systematic de novo protein design exploration of the novel αβ-folds predicted by the rules. The designs for all eight of the predicted novel αβ-folds with a four-stranded β-sheet, including a knot-forming one, folded into structures close to the design models. Further, the rules predicted more than 10,000 novel αβ-folds with five- to eight-stranded β-sheets; this number far exceeds the number of αβ-folds observed in nature so far. This result suggests that a vast number of αβ-folds are possible, but have not emerged or have become extinct due to evolutionary bias.

Large language models generate functional protein sequences across diverse families

Deep-learning language models have shown promise in various biotechnological applications, including protein design and engineering. Here we describe ProGen, a language model that can generate protein sequences with a predictable function across large protein families, akin to generating grammatically and semantically correct natural language sentences on diverse topics. The model was trained on 280 million protein sequences from >19,000 families and is augmented with control tags specifying protein properties. ProGen can be further fine-tuned to curated sequences and tags to improve controllable generation performance of proteins from families with sufficient homologous samples. Artificial proteins fine-tuned to five distinct lysozyme families showed similar catalytic efficiencies as natural lysozymes, with sequence identity to natural proteins as low as 31.4%. ProGen is readily adapted to diverse protein families, as we demonstrate with chorismate mutase and malate dehydrogenase.

Why we are made of proteins and nucleic acids: Structural biology views on extraterrestrial life

Is it a miracle that life exists on the Earth, or is it a common phenomenon in the universe? If extraterrestrial organisms exist, what are they like? To answer these questions, we must understand what kinds of molecules could evolve into life, or in other words, what properties are generally required to perform biological functions and store genetic information. This review summarizes recent findings on simple ancestral proteins, outlines the basic knowledge in textbooks, and discusses the generally required properties for biological molecules from structural biology viewpoints (e.g., restriction of shapes, and types of intra- and intermolecular interactions), leading to the conclusion that proteins and nucleic acids are at least one of the simplest (and perhaps very common) forms of catalytic and genetic biopolymers in the universe. This review article is an extended version of the Japanese article, On the Origin of Life: Coevolution between RNA and Peptide, published in SEIBUTSU BUTSURI Vol. 61, p. 232-235 (2021).

De novo design of small beta barrel proteins

Small beta barrel proteins are attractive targets for computational design because of their considerable functional diversity despite their very small size (<70 amino acids). However, there are considerable challenges to designing such structures, and there has been little success thus far. Because of the small size, the hydrophobic core stabilizing the fold is necessarily very small, and the conformational strain of barrel closure can oppose folding; also intermolecular aggregation through free beta strand edges can compete with proper monomer folding. Here, we explore the de novo design of small beta barrel topologies using both Rosetta energy-based methods and deep learning approaches to design four small beta barrel folds: Src homology 3 (SH3) and oligonucleotide/oligosaccharide-binding (OB) topologies found in nature and five and six up-and-down-stranded barrels rarely if ever seen in nature. Both approaches yielded successful designs with high thermal stability and experimentally determined structures with less than 2.4 Å rmsd from the designed models. Using deep learning for backbone generation and Rosetta for sequence design yielded higher design success rates and increased structural diversity than Rosetta alone. The ability to design a large and structurally diverse set of small beta barrel proteins greatly increases the protein shape space available for designing binders to protein targets of interest.

De novo protein fold design through sequence-independent fragment assembly simulations

De novo protein design generally consists of two steps, including structure and sequence design. Many protein design studies have focused on sequence design with scaffolds adapted from native structures in the PDB, which renders novel areas of protein structure and function space unexplored. We developed FoldDesign to create novel protein folds from specific secondary structure (SS) assignments through sequence-independent replica-exchange Monte Carlo (REMC) simulations. The method was tested on 354 non-redundant topologies, where FoldDesign consistently created stable structural folds, while recapitulating on average 87.7% of the SS elements. Meanwhile, the FoldDesign scaffolds had well-formed structures with buried residues and solvent-exposed areas closely matching their native counterparts. Despite the high fidelity to the input SS restraints and local structural characteristics of native proteins, a large portion of the designed scaffolds possessed global folds completely different from natural proteins in the PDB, highlighting the ability of FoldDesign to explore novel areas of protein fold space. Detailed data analyses revealed that the major contributions to the successful structure design lay in the optimal energy force field, which contains a balanced set of SS packing terms, and REMC simulations, which were coupled with multiple auxiliary movements to efficiently search the conformational space. Additionally, the ability to recognize and assemble uncommon super-SS geometries, rather than the unique arrangement of common SS motifs, was the key to generating novel folds. These results demonstrate a strong potential to explore both structural and functional spaces through computational design simulations that natural proteins have not reached through evolution.

Sampling of structure and sequence space of small protein folds

Nature only samples a small fraction of the sequence space that can fold into stable proteins. Furthermore, small structural variations in a single fold, sometimes only a few amino acids, can define a protein's molecular function. Hence, to design proteins with novel functionalities, such as molecular recognition, methods to control and sample shape diversity are necessary. To explore this space, we developed and experimentally validated a computational platform that can design a wide variety of small protein folds while sampling shape diversity. We designed and evaluated stability of about 30,000 de novo protein designs of eight different folds. Among these designs, about 6,200 stable proteins were identified, including some predicted to have a first-of-its-kind minimalized thioredoxin fold. Obtained data revealed protein folding rules for structural features such as helix-connecting loops. Beyond serving as a resource for protein engineering, this massive and diverse dataset also provides training data for machine learning. We developed an accurate classifier to predict the stability of our designed proteins. The methods and the wide range of protein shapes provide a basis for designing new protein functions without compromising stability.

A generic framework for hierarchical de novo protein design

De novo protein design enables the exploration of novel sequences and structures absent from the natural protein universe. De novo design also stands as a stringent test for our understanding of the underlying physical principles of protein folding and may lead to the development of proteins with unmatched functional characteristics. The first fundamental challenge of de novo design is to devise "designable" structural templates leading to sequences that will adopt the predicted fold. Here, we built on the TopoBuilder (TB) de novo design method, to automatically assemble structural templates with native-like features starting from string descriptors that capture the overall topology of proteins. Our framework eliminates the dependency of hand-crafted and fold-specific rules through an iterative, data-driven approach that extracts geometrical parameters from structural tertiary motifs. We evaluated the TopoBuilder framework by designing sequences for a set of five protein folds and experimental characterization revealed that several sequences were folded and stable in solution. The TopoBuilder de novo design framework will be broadly useful to guide the generation of artificial proteins with customized geometries, enabling the exploration of the protein universe.

Dissecting the stability determinants of a challenging de novo protein fold using massively parallel design and experimentation

Designing entirely new protein structures remains challenging because we do not fully understand the biophysical determinants of folding stability. Yet, some protein folds are easier to design than others. Previous work identified the 43-residue ɑββɑ fold as especially challenging: The best designs had only a 2% success rate, compared to 39 to 87% success for other simple folds [G. J. Rocklin et al., Science 357, 168-175 (2017)]. This suggested the ɑββɑ fold would be a useful model system for gaining a deeper understanding of folding stability determinants and for testing new protein design methods. Here, we designed over 10,000 new ɑββɑ proteins and found over 3,000 of them to fold into stable structures using a high-throughput protease-based assay. NMR, hydrogen-deuterium exchange, circular dichroism, deep mutational scanning, and scrambled sequence control experiments indicated that our stable designs fold into their designed ɑββɑ structures with exceptional stability for their small size. Our large dataset enabled us to quantify the influence of universal stability determinants including nonpolar burial, helix capping, and buried unsatisfied polar atoms, as well as stability determinants unique to the ɑββɑ topology. Our work demonstrates how large-scale design and test cycles can solve challenging design problems while illuminating the biophysical determinants of folding.

Identifying well-folded de novo proteins in the new era of accurate structure prediction

Computational de novo protein design tailors proteins for target structures and oligomerisation states with high stability, which allows overcoming many limitations of natural proteins when redesigned for new functions. Despite significant advances in the field over the past decade, it remains challenging to predict sequences that will fold as stable monomers in solution or binders to a particular protein target; thereby requiring substantial experimental resources to identify proteins with the desired properties. To overcome this, here we leveraged the large amount of design data accumulated in the last decade, and the breakthrough in protein structure prediction from last year to investigate on improved ways of selecting promising designs before experimental testing. We collected de novo proteins from previous studies, 518 designed as monomers of different folds and 2112 as binders against the Botulinum neurotoxin, and analysed their structures with AlphaFold2, RoseTTAFold and fragment quality descriptors in combination with other properties related to surface interactions. These features showed high complementarity in rationalizing the experimental results, which allowed us to generate quite accurate machine learning models for predicting well-folded monomers and binders with a small set of descriptors. Cross-validating designs with varied orthogonal computational techniques should guide us for identifying design imperfections, rescuing designs and making more robust design selections before experimental testing.

De novo design of immunoglobulin-like domains

Antibodies, and antibody derivatives such as nanobodies, contain immunoglobulin-like (Ig) β-sandwich scaffolds which anchor the hypervariable antigen-binding loops and constitute the largest growing class of drugs. Current engineering strategies for this class of compounds rely on naturally existing Ig frameworks, which can be hard to modify and have limitations in manufacturability, designability and range of action. Here, we develop design rules for the central feature of the Ig fold architecture—the non-local cross-β structure connecting the two β-sheets—and use these to design highly stable Ig domains de novo, confirm their structures through X-ray crystallography, and show they can correctly scaffold functional loops. Our approach opens the door to the design of antibody-like scaffolds with tailored structures and superior biophysical properties.

The immunoglobulin domain framework of antibodies has been a long standing design challenge. Here, the authors describe design rules for tailoring these domains and show they can be accurately designed, de novo, with high stability and the ability to scaffold functional loops.

Active site center redesign increases protein stability preserving catalysis in thioredoxin

The stabilization of natural proteins is a long‐standing desired goal in protein engineering. Optimizing the hydrophobicity of the protein core often results in extensive stability enhancements. However, the presence of totally or partially buried catalytic charged residues, essential for protein function, has limited the applicability of this strategy. Here, focusing on the thioredoxin, we aimed to augment protein stability by removing buried charged residues in the active site without loss of catalytic activity. To this end, we performed a charged‐to‐hydrophobic substitution of a buried and functional group, resulting in a significant stability increase yet abolishing catalytic activity. Then, to simulate the catalytic role of the buried ionizable group, we designed a combinatorial library of variants targeting a set of seven surface residues adjacent to the active site. Notably, more than 50% of the library variants restored, to some extent, the catalytic activity. The combination of experimental study of 2% of the library with the prediction of the whole mutational space by partial least squares regression revealed that a single point mutation at the protein surface is sufficient to fully restore the catalytic activity without thermostability cost. As a result, we engineered one of the highest thermal stabilities reported for a protein with a natural occurring fold (137°C). Further, our hyperstable variant preserves the catalytic activity both in vitro and in vivo.

De novo designed protein inhibitors of amyloid aggregation and seeding

Neurodegenerative diseases are characterized by the pathologic accumulation of aggregated proteins. Known as amyloid, these fibrillar aggregates include proteins such as tau and amyloid-β (Aβ) in Alzheimer's disease (AD) and alpha-synuclein (αSyn) in Parkinson's disease (PD). The development and spread of amyloid fibrils within the brain correlates with disease onset and progression, and inhibiting amyloid formation is a possible route toward therapeutic development. Recent advances have enabled the determination of amyloid fibril structures to atomic-level resolution, improving the possibility of structure-based inhibitor design. In this work, we use these amyloid structures to design inhibitors that bind to the ends of fibrils, "capping" them so as to prevent further growth. Using de novo protein design, we develop a library of miniprotein inhibitors of 35 to 48 residues that target the amyloid structures of tau, Aβ, and αSyn. Biophysical characterization of top in silico designed inhibitors shows they form stable folds, have no sequence similarity to naturally occurring proteins, and specifically prevent the aggregation of their targeted amyloid-prone proteins in vitro. The inhibitors also prevent the seeded aggregation and toxicity of fibrils in cells. In vivo evaluation reveals their ability to reduce aggregation and rescue motor deficits in Caenorhabditis elegans models of PD and AD.

Design and Engineering of Miniproteins

The potential of miniproteins in the biological and chemical sciences is constantly increasing. Significant progress in the design methodologies has been achieved over the last 30 years. Early approaches based on propensities of individual amino acid residues to form individual secondary structures were subsequently improved by structural analyses using NMR spectroscopy and crystallography. Consequently, computational algorithms were developed, which are now highly successful in designing structures with accuracy often close to atomic range. Further perspectives include construction of miniproteins incorporating non-native secondary structures derived from sequences with units other than α-amino acids. Noteworthy, miniproteins with extended structures, which are now feasibly accessible, are excellent scaffolds for construction of functional molecules.

The Structural Rule Distinguishing a Superfold: A Case Study of Ferredoxin Fold and the Reverse Ferredoxin Fold

Superfolds are folds commonly observed among evolutionarily unrelated multiple superfamilies of proteins. Since discovering superfolds almost two decades ago, structural rules distinguishing superfolds from the other ordinary folds have been explored but remained elusive. Here, we analyzed a typical superfold, the ferredoxin fold, and the fold which reverses the N to C terminus direction from the ferredoxin fold as a case study to find the rule to distinguish superfolds from the other folds. Though all the known structural characteristics for superfolds apply to both the ferredoxin fold and the reverse ferredoxin fold, the reverse fold has been found only in a single superfamily. The database analyses in the present study revealed the structural preferences of αβ- and βα-units; the preferences separate two α-helices in the ferredoxin fold, preventing their collision and stabilizing the fold. In contrast, in the reverse ferredoxin fold, the preferences bring two helices near each other, inducing structural conflict. The Rosetta folding simulations suggested that the ferredoxin fold is physically much more realizable than the reverse ferredoxin fold. Therefore, we propose that minimal structural conflict or minimal frustration among secondary structures is the rule to distinguish a superfold from ordinary folds. Intriguingly, the database analyses revealed that a most stringent structural rule in proteins, the right-handedness of the βαβ-unit, is broken in a set of structures to prevent the frustration, suggesting the proposed rule of minimum frustration among secondary structural units is comparably strong as the right-handedness rule of the βαβ-unit.

Multivalent designed proteins neutralize SARS-CoV-2 variants of concern and confer protection against infection in mice

New variants of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) continue to arise and prolong the coronavirus disease 2019 (COVID-19) pandemic. Here, we used a cell-free expression workflow to rapidly screen and optimize constructs containing multiple computationally designed miniprotein inhibitors of SARS-CoV-2. We found the broadest efficacy was achieved with a homotrimeric version of the 75-residue angiotensin-converting enzyme 2 (ACE2) mimic AHB2 (TRI2-2) designed to geometrically match the trimeric spike architecture. Consistent with the design model, in the cryo-electron microscopy structure TRI2-2 forms a tripod at the apex of the spike protein that engaged all three receptor binding domains simultaneously. TRI2-2 neutralized Omicron (B.1.1.529), Delta (B.1.617.2), and all other variants tested with greater potency than the monoclonal antibodies used clinically for the treatment of COVID-19. TRI2-2 also conferred prophylactic and therapeutic protection against SARS-CoV-2 challenge when administered intranasally in mice. Designed miniprotein receptor mimics geometrically arrayed to match pathogen receptor binding sites could be a widely applicable antiviral therapeutic strategy with advantages over antibodies in greater resistance to viral escape and antigenic drift, and advantages over native receptor traps in lower chances of autoimmune responses.

Protein design via deep learning

Proteins with desired functions and properties are important in fields like nanotechnology and biomedicine. De novo protein design enables the production of previously unseen proteins from the ground up and is believed as a key point for handling real social challenges. Recent introduction of deep learning into design methods exhibits a transformative influence and is expected to represent a promising and exciting future direction. In this review, we retrospect the major aspects of current advances in deep-learning-based design procedures and illustrate their novelty in comparison with conventional knowledge-based approaches through noticeable cases. We not only describe deep learning developments in structure-based protein design and direct sequence design, but also highlight recent applications of deep reinforcement learning in protein design. The future perspectives on design goals, challenges and opportunities are also comprehensively discussed.

Reconfigurable asymmetric protein assemblies through implicit negative design

Asymmetric multiprotein complexes that undergo subunit exchange play central roles in biology but present a challenge for design because the components must not only contain interfaces that enable reversible association but also be stable and well behaved in isolation. We use implicit negative design to generate β sheet-mediated heterodimers that can be assembled into a wide variety of complexes. The designs are stable, folded, and soluble in isolation and rapidly assemble upon mixing, and crystal structures are close to the computational models. We construct linearly arranged hetero-oligomers with up to six different components, branched hetero-oligomers, closed C4-symmetric two-component rings, and hetero-oligomers assembled on a cyclic homo-oligomeric central hub and demonstrate that such complexes can readily reconfigure through subunit exchange. Our approach provides a general route to designing asymmetric reconfigurable protein systems.

Structure-based protein design with deep learning

Since the first revelation of proteins functioning as macromolecular machines through their three dimensional structures, researchers have been intrigued by the marvelous ways the biochemical processes are carried out by proteins. The aspiration to understand protein structures has fueled extensive efforts across different scientific disciplines. In recent years, it has been demonstrated that proteins with new functionality or shapes can be designed via structure-based modeling methods, and the design strategies have combined all available information - but largely piece-by-piece - from sequence derived statistics to the detailed atomic-level modeling of chemical interactions. Despite the significant progress, incorporating data-derived approaches through the use of deep learning methods can be a game changer. In this review, we summarize current progress, compare the arc of developing the deep learning approaches with the conventional methods, and describe the motivation and concepts behind current strategies that may lead to potential future opportunities.

State-Targeting Stabilization of Adenosine A2A Receptor by Fusing a Custom-Made De Novo Designed α-Helical Protein

G-protein coupled receptors (GPCRs) are known for their low stability and large conformational changes upon transitions between multiple states. A widely used method for stabilizing these receptors is to make chimeric receptors by fusing soluble proteins (i.e., fusion partner proteins) into the intracellular loop 3 (ICL3) connecting the transmembrane helices 5 and 6 (TM5 and TM6). However, this fusion approach requires experimental trial and error to identify appropriate soluble proteins, residue positions, and linker lengths for making the fusion. Moreover, this approach has not provided state-targeting stabilization of GPCRs. Here, to rationally stabilize a class A GPCR, adenosine A2A receptor (A2AR) in a target state, we carried out the custom-made de novo design of α-helical fusion partner proteins, which can fix the conformation of TM5 and TM6 to that in an inactive state of A2AR through straight helical connections without any kinks or intervening loops. The chimeric A2AR fused with one of the designs (FiX1) exhibited increased thermal stability. Moreover, compared with the wild type, the binding affinity of the chimera against the agonist NECA was significantly decreased, whereas that against the inverse agonist ZM241385 was similar, indicating that the inactive state was selectively stabilized. Our strategy contributes to the rational state-targeting stabilization of GPCRs.

Enumeration and comprehensive in-silico modeling of three-helix bundle structures composed of typical αα-hairpins

BACKGROUND

The design of protein structures from scratch requires special attention to the combination of the types and lengths of the secondary structures and the loops required to build highly designable backbone structure models. However, it is difficult to predict the combinations that result in globular and protein-like conformations without simulations. In this study, we used single-chain three-helix bundles as simple models of protein tertiary structures and sought to thoroughly investigate the conditions required to construct them, starting from the identification of the typical αα-hairpin motifs.

RESULTS

First, by statistical analysis of naturally occurring protein structures, we identified three αα-hairpins motifs that were specifically related to the left- and right-handedness of helix-helix packing. Second, specifying these αα-hairpins motifs as junctions, we performed sequence-independent backbone-building simulations to comparatively build single-chain three-helix bundle structures and identified the promising combinations of the length of the α-helix and αα-hairpins types that results in tight packing between the first and third α-helices. Third, using those single-chain three-helix bundle backbone structures as template structures, we designed amino acid sequences that were predicted to fold into the target topologies, which supports that the compact single-chain three-helix bundles structures that we sampled show sufficient quality to allow amino-acid sequence design.

CONCLUSION

The enumeration of the dominant subsets of possible backbone structures for small single-chain three-helical bundle topologies revealed that the compact foldable structures are discontinuously and sparsely distributed in the conformational space. Additionally, although the designs have not been experimentally validated in the present research, the comprehensive set of computational structural models generated also offers protein designers the opportunity to skip building similar structures by themselves and enables them to quickly focus on building specialized designs using the prebuilt structure models. The backbone and best design models in this study are publicly accessible from the following URL: https://doi.org/10.5281/zenodo.4321632 .

The register shift rules for βαβ-motifs for de novo protein design

A wide range of de novo design of αβ-proteins has been achieved based on the design rules, which describe secondary structure lengths and loop torsion patterns favorable for design target topologies. This paper proposes design rules for register shifts in βαβ-motifs, which have not been reported previously, but are necessary for determining a target structure of de novo design of αβ-proteins. By analyzing naturally occurring protein structures in a database, we found preferences for register shifts in βαβ-motifs, and derived the following empirical rules: (1) register shifts must not be negative regardless of torsion types for a constituent loop in βαβ-motifs; (2) preferred register shifts strongly depend on the loop torsion types. To explain these empirical rules by physical interactions, we conducted physics-based simulations for systems mimicking a βαβ-motif that contains the most frequently observed loop type in the database. We performed an exhaustive conformational sampling of the loop region, imposing the exclusion volume and hydrogen bond satisfaction condition. The distributions of register shifts obtained from the simulations agreed well with those of the database analysis, indicating that the empirical rules are a consequence of physical interactions, rather than an evolutionary sampling bias. Our proposed design rules will serve as a guide to making appropriate target structures for the de novo design of αβ-proteins.

Step-by-step design of proteins for small molecule interaction: A review on recent milestones

Protein design is the field of synthetic biology that aims at developing de novo custom-made proteins and peptides for specific applications. Despite exploring an ambitious goal, recent computational advances in both hardware and software technologies have paved the way to high-throughput screening and detailed design of novel folds and improved functionalities. Modern advances in the field of protein design for small molecule targeting are described in this review, organized in a step-by-step fashion: from the conception of a new or upgraded active binding site, to scaffold design, sequence optimization, and experimental expression of the custom protein. In each step, contemporary examples are described, and state-of-the-art software is briefly explored.

Principles and Methods in Computational Membrane Protein Design

After decades of progress in computational protein design, the design of proteins folding and functioning in lipid membranes appears today as the next frontier. Some notable successes in the de novo design of simplified model membrane protein systems have helped articulate fundamental principles of protein folding, architecture and interaction in the hydrophobic lipid environment. These principles are reviewed here, together with the computational methods and approaches that were used to identify them. We provide an overview of the methodological innovations in the generation of new protein structures and functions and in the development of membrane-specific energy functions. We highlight the opportunities offered by new machine learning approaches applied to protein design, and by new experimental characterization techniques applied to membrane proteins. Although membrane protein design is in its infancy, it appears more reachable than previously thought.

Multivalent designed proteins protect against SARS-CoV-2 variants of concern

Escape variants of SARS-CoV-2 are threatening to prolong the COVID-19 pandemic. To address this challenge, we developed multivalent protein-based minibinders as potential prophylactic and therapeutic agents. Homotrimers of single minibinders and fusions of three distinct minibinders were designed to geometrically match the SARS-CoV-2 spike (S) trimer architecture and were optimized by cell-free expression and found to exhibit virtually no measurable dissociation upon binding. Cryo-electron microscopy (cryoEM) showed that these trivalent minibinders engage all three receptor binding domains on a single S trimer. The top candidates neutralize SARS-CoV-2 variants of concern with IC 50 values in the low pM range, resist viral escape, and provide protection in highly vulnerable human ACE2-expressing transgenic mice, both prophylactically and therapeutically. Our integrated workflow promises to accelerate the design of mutationally resilient therapeutics for pandemic preparedness.

ONE-SENTENCE SUMMARY

We designed, developed, and characterized potent, trivalent miniprotein binders that provide prophylactic and therapeutic protection against emerging SARS-CoV-2 variants of concern.

Limitations of the ABEGO-based backbone design: ambiguity between αα-corner and αα-hairpin

ABEGO is a coarse-grained representation for poly-peptide backbone dihedral angles. The Ramachandran map is divided into four segments denoted as A, B, E, and G to represent the local conformation of poly-peptide chains in the character strings. Although the ABEGO representation is widely used in backbone building simulation for de novo protein design, it cannot capture minor differences in backbone dihedral angles, which potentially leads to ambiguity between two structurally distinct fragments. Here, I show a nontrivial example of two local motifs that could not be distinguished by their ABEGO representations. I found that two well-known local motifs αα-hairpins and αα-corners are both represented as α-GBB-α and thus indistinguishable in the ABEGO representation, although they show distinct arrangements of the flanking α-helices. I also found that α-GBB-α motifs caused a loss of efficiency in the ABEGO-based fragment-assembly simulations for de novo protein backbone design. Nevertheless, I was able to design amino-acid sequences that were predicted to fold into the target topologies that contained these α-GBB-α motifs, which suggests such topologies that are difficult to build by ABEGO-based simulations are designable once the backbone structures are modeled by some means. The finding that certain local motifs bottleneck the ABEGO-based fragment-assembly simulations for construction of backbone structures suggests that finer representations of backbone torsion angles are required for efficiently generating diverse topologies containing such indistinguishable local motifs.

Role of backbone strain in de novo design of complex α/β protein structures

We previously elucidated principles for designing ideal proteins with completely consistent local and non-local interactions which have enabled the design of a wide range of new αβ-proteins with four or fewer β-strands. The principles relate local backbone structures to supersecondary-structure packing arrangements of α-helices and β-strands. Here, we test the generality of the principles by employing them to design larger proteins with five- and six- stranded β-sheets flanked by α-helices. The initial designs were monomeric in solution with high thermal stability, and the nuclear magnetic resonance (NMR) structure of one was close to the design model, but for two others the order of strands in the β-sheet was swapped. Investigation into the origins of this strand swapping suggested that the global structures of the design models were more strained than the NMR structures. We incorporated explicit consideration of global backbone strain into the design methodology, and succeeded in designing proteins with the intended unswapped strand arrangements. These results illustrate the value of experimental structure determination in guiding improvement of de novo design, and the importance of consistency between local, supersecondary, and global tertiary interactions in determining protein topology. The augmented set of principles should inform the design of larger functional proteins.

Structure elements can be predicted using the contact volume among protein residues

Previously, the structure elements of dihydrofolate reductase (DHFR) were determined using comprehen­sive Ala-insertion mutation analysis, which is assumed to be a kind of protein “building blocks.” It is hypo­thesized that our comprehension of the structure elements could lead to understanding how an amino acid sequence dictates its tertiary structure. However, the comprehensive Ala-insertion mutation analysis is a time- and cost-consuming process and only a set of the DHFR structure elements have been reported so far. Therefore, developing a computational method to predict structure elements is an urgent necessity. We focused on intramolecular residue–residue contacts to predict the structure elements. We introduced a simple and effective parameter: the overlapped contact volume (CV) among the residues and calculated the CV along the DHFR sequence using the crystal structure. Our results indicate that the CV profile can recapitulate its precipitate ratio profile, which was used to define the structure elements in the Ala-insertion mutation analysis. The CV profile allowed us to predict structure elements like the experimentally determined structure elements. The strong correlation between the CV and precipitate ratio profiles indicates the importance of the intramolecular residue–residue contact in maintaining the tertiary structure. Additionally, the CVs between the structure elements are considerably more than those between a structure element and a linker or two linkers, indicating that the structure elements play a funda­mental role in increasing the intramolecular adhesion. Thus, we propose that the structure elements can be considered a type of “building blocks” that maintain and dictate the tertiary structures of proteins.

Transferrin receptor targeting by de novo sheet extension

The de novo design of polar protein-protein interactions is challenging because of the thermodynamic cost of stripping water away from the polar groups. Here, we describe a general approach for designing proteins which complement exposed polar backbone groups at the edge of beta sheets with geometrically matched beta strands. We used this approach to computationally design small proteins that bind to an exposed beta sheet on the human transferrin receptor (hTfR), which shuttles interacting proteins across the blood-brain barrier (BBB), opening up avenues for drug delivery into the brain. We describe a design which binds hTfR with a 20 nM Kd, is hyperstable, and crosses an in vitro microfluidic organ-on-a-chip model of the human BBB. Our design approach provides a general strategy for creating binders to protein targets with exposed surface beta edge strands.

Breakthroughs in computational design methods open up new frontiers for de novo protein engineering

Proteins catalyze the majority of chemical reactions in organisms, and harnessing this power has long been the focus of the protein engineering field. Computational protein design aims to create new proteins and functions in silico, and in doing so, accelerate the process, reduce costs and enable more sophisticated engineering goals to be accomplished. Challenges that very recently seemed impossible are now within reach thanks to several landmark advances in computational protein design methods. Here, we summarize these new methods, with a particular emphasis on de novo protein design advancements occurring within the past 5 years.

De novo design of transmembrane β barrels

Transmembrane β-barrel proteins (TMBs) are of great interest for single-molecule analytical technologies because they can spontaneously fold and insert into membranes and form stable pores, but the range of pore properties that can be achieved by repurposing natural TMBs is limited. We leverage the power of de novo computational design coupled with a "hypothesis, design, and test" approach to determine TMB design principles, notably, the importance of negative design to slow β-sheet assembly. We design new eight-stranded TMBs, with no homology to known TMBs, that insert and fold reversibly into synthetic lipid membranes and have nuclear magnetic resonance and x-ray crystal structures very similar to the computational models. These advances should enable the custom design of pores for a wide range of applications.

Incorporation of sensing modalities into de novo designed fluorescence-activating proteins

Through the efforts of many groups, a wide range of fluorescent protein reporters and sensors based on green fluorescent protein and its relatives have been engineered in recent years. Here we explore the incorporation of sensing modalities into de novo designed fluorescence-activating proteins, called mini-fluorescence-activating proteins (mFAPs), that bind and stabilize the fluorescent cis-planar state of the fluorogenic compound DFHBI. We show through further design that the fluorescence intensity and specificity of mFAPs for different chromophores can be tuned, and the fluorescence made sensitive to pH and Ca²⁺ for real-time fluorescence reporting. Bipartite split mFAPs enable real-time monitoring of protein-protein association and (unlike widely used split GFP reporter systems) are fully reversible, allowing direct readout of association and dissociation events. The relative ease with which sensing modalities can be incorporated and advantages in smaller size and photostability make de novo designed fluorescence-activating proteins attractive candidates for optical sensor engineering.

De Novo Protein Design Using the Blueprint Builder in Rosetta

While native proteins cover diverse structural spaces and achieve various biological events, not many of them can directly serve human needs. One reason is that the native proteins usually contain idiosyncrasies evolved for their native functions but disfavoring engineering requirements. To overcome this issue, one strategy is to create de novo proteins which are designed to possess improved stability, high environmental tolerance, and enhanced engineering potential. Compared to other protein engineering strategies, in silico design of de novo proteins has significantly expanded the protein structural and sequence spaces, reduced wet lab workload, and incorporated engineered features in a guided and efficient manner. In the Baker laboratory we have been applying a design pipeline that uses the blueprint builder to design different folds of de novo proteins, and have successfully obtained libraries of de novo proteins with improved stability and engineering potential. In this article, we will use the design of de novo β-barrel proteins as an example to describe the principles and basic procedures of the blueprint builder-based design pipeline. © 2020 Wiley Periodicals LLC. Basic Protocol 1: The construction of blueprints Alternate Protocol: Build blueprints based on existing protein .pdb files Basic Protocol 2: De novo protein design pipeline using the blueprint builder.

Recent advances in de novo protein design: Principles, methods, and applications

The computational de novo protein design is increasingly applied to address a number of key challenges in biomedicine and biological engineering. Successes in expanding applications are driven by advances in design principles and methods over several decades. Here, we review recent innovations in major aspects of the de novo protein design and include how these advances were informed by principles of protein architecture and interactions derived from the wealth of structures in the Protein Data Bank. We describe developments in de novo generation of designable backbone structures, optimization of sequences, design scoring functions, and the design of the function. The advances not only highlight design goals reachable now but also point to the challenges and opportunities for the future of the field.

Robust folding of a de novo designed ideal protein even with most of the core mutated to valine

De novo designed proteins exhibit a remarkable property of extremely high thermal stability compared with naturally occurring proteins. The designed proteins are completely optimized for folding; the backbone structures are created by using a set of rules that relate local backbone structures to preferred tertiary motifs and the side chains are designed to favor both the local backbone structures and the entire tertiary structures. Here, we found that one of the de novo designed proteins, which was mutated to fill the core with mostly valine residues, still has the folding ability and shows high stability (T_m = 106 °C) even with its reduced and loosened core packing. This result supports the importance of local backbone structures to protein folding.

Protein design provides a stringent test for our understanding of protein folding. We previously described principles for designing ideal protein structures stabilized by consistent local and nonlocal interactions, based on a set of rules relating local backbone structures to tertiary packing motifs. The principles have made possible the design of protein structures having various topologies with high thermal stability. Whereas nonlocal interactions such as tight hydrophobic core packing have traditionally been considered to be crucial for protein folding and stability, the rules proposed by our previous studies suggest the importance of local backbone structures to protein folding. In this study, we investigated the robustness of folding of de novo designed proteins to the reduction of the hydrophobic core, by extensive mutation of large hydrophobic residues (Leu, Ile) to smaller ones (Val) for one of the designs. Surprisingly, even after 10 Leu and Ile residues were mutated to Val, this mutant with the core mostly filled with Val was found to not be in a molten globule state and fold into the same backbone structure as the original design, with high stability. These results indicate the importance of local backbone structures to the folding ability and high thermal stability of designed proteins and suggest a method for engineering thermally stabilized natural proteins.

Expanding the space of protein geometries by computational design of de novo fold families

Naturally occurring proteins vary the precise geometries of structural elements to create distinct shapes optimal for function. We present a computational design method, loop-helix-loop unit combinatorial sampling (LUCS), that mimics nature's ability to create families of proteins with the same overall fold but precisely tunable geometries. Through near-exhaustive sampling of loop-helix-loop elements, LUCS generates highly diverse geometries encompassing those found in nature but also surpassing known structure space. Biophysical characterization showed that 17 (38%) of 45 tested LUCS designs encompassing two different structural topologies were well folded, including 16 with designed non-native geometries. Four experimentally solved structures closely matched the designs. LUCS greatly expands the designable structure space and offers a new paradigm for designing proteins with tunable geometries that may be customizable for novel functions.

An enumerative algorithm for de novo design of proteins with diverse pocket structures

To create new enzymes and biosensors from scratch, precise control over the structure of small-molecule binding sites is of paramount importance, but systematically designing arbitrary protein pocket shapes and sizes remains an outstanding challenge. Using the NTF2-like structural superfamily as a model system, we developed an enumerative algorithm for creating a virtually unlimited number of de novo proteins supporting diverse pocket structures. The enumerative algorithm was tested and refined through feedback from two rounds of large-scale experimental testing, involving in total the assembly of synthetic genes encoding 7,896 designs and assessment of their stability on yeast cell surface, detailed biophysical characterization of 64 designs, and crystal structures of 5 designs. The refined algorithm generates proteins that remain folded at high temperatures and exhibit more pocket diversity than naturally occurring NTF2-like proteins. We expect this approach to transform the design of small-molecule sensors and enzymes by enabling the creation of binding and active site geometries much more optimal for specific design challenges than is accessible by repurposing the limited number of naturally occurring NTF2-like proteins.

Pair Potentials as Machine Learning Features

Atom pairwise potential functions make up an essential part of many scoring functions for protein decoy detection. With the development of machine learning (ML) tools, there are multiple ways to combine potential functions to create novel ML models and methods. Potential function parameters can be easily extracted; however, it is usually hard to directly obtain the calculated atom pairwise energies from scoring functions. Amber, as one of the most popular suites of modeling programs, has an extensive history and library of force field potential functions. In this work, we directly used the force field parameters in ff94 and ff14SB from Amber and encoded them to calculate atom pairwise energies for different interactions. Two sets of structures (single amino acid set and a dipeptide set) were used to evaluate the performance of our encoded Amber potentials. From the comparison results between energy terms obtained from our encoding and Amber, we find energy difference within ±0.06 kcal/mol for all tested structures. Previously we have shown that the Random Forest (RF) model can help to emphasize more important atom pairwise interactions and ignore insignificant ones [Pei, J.; Zheng, Z.; Merz, K. M. J. Chem. Inf. Model. 2019, 59, 1919-1929]. Here, as an example of combining ML methods with traditional potential functions, we followed the same work flow to combine the RF models with force field potential functions from Amber. To determine the performance of our RF models with force field potential functions, 224 different protein native-decoy systems were used as our training and testing sets We find that the RF models with ff94 and ff14SB force field parameters outperformed all other scoring functions (RF models with KECSA2, RWplus, DFIRE, dDFIRE, and GOAP) considered in this work for native structure detection, and they performed similarly in detecting the best decoy. Through inclusion of best decoy to decoy comparisons in building our RF models, we were able to generate models that outperformed the score functions tested herein both on accuracy and best decoy detection, again showing the performance and flexibility of our RF models to tackle this problem. Finally, the importance of the RF algorithm and force field parameters were also tested and the comparison results suggest that both the RF algorithm and force field potentials are important with the ML scoring function achieving its best performance only by combining them together. All code and data used in this work are available at https://github.com/JunPei000/FFENCODER_for_Protein_Folding_Pose_Selection.

Synthetic Biology: A Solution for Tackling Nanomaterial Challenges

Bioengineers have mastered practical techniques for tuning a biomaterial's properties with only limited information on the relationship between the material's structure and function. These techniques have been quintessential to engineering proteins, which are most often riddled with ill-defined structure-function relationships. In this Perspective, we review bioengineering approaches aimed at overcoming the elusive protein structure-function relation. We extend these principles to engineering synthetic nanomaterials, specifically applying the underlying theory to optical sensors based on single-stranded DNA-wrapped single-walled carbon nanotubes (ssDNA-SWCNTs). Bioengineering techniques such as directed evolution, computational design, and noncanonical synthesis are reviewed in the broader context of nanomaterials engineering. We further provide an order-of-magnitude analysis of empirical approaches that rely on random or guided searches for designing new nanomaterials. The underlying concepts presented in these approaches can be further extended to a broad range of engineering fields confronted with empirical design strategies, including catalysis, metal-organic frameworks (MOFs), pharmaceutical dosing, and optimization algorithms.