De novo design of protein interactions with learned surface fingerprints.

Physical interactions between proteins are essential for most biological processes governing life1. However, the molecular determinants of such interactions have been challenging to understand, even as genomic, proteomic and structural data increase. This knowledge gap has been a major obstacle for the comprehensive understanding of cellular protein-protein interaction networks and for the de novo design of protein binders that are crucial for synthetic biology and translational applications2-9. Here we use a geometric deep-learning framework operating on protein surfaces that generates fingerprints to describe geometric and chemical features that are critical to drive protein-protein interactions10. We hypothesized that these fingerprints capture the key aspects of molecular recognition that represent a new paradigm in the computational design of novel protein interactions. As a proof of principle, we computationally designed several de novo protein binders to engage four protein targets: SARS-CoV-2 spike, PD-1, PD-L1 and CTLA-4. Several designs were experimentally optimized, whereas others were generated purely in silico, reaching nanomolar affinity with structural and mutational characterization showing highly accurate predictions. Overall, our surface-centric approach captures the physical and chemical determinants of molecular recognition, enabling an approach for the de novo design of protein interactions and, more broadly, of artificial proteins with function.

From Thalidomide to Rational Molecular Glue Design for Targeted Protein Degradation.

Thalidomide and its derivatives are powerful cancer therapeutics that are among the best-understood molecular glue degraders (MGDs). These drugs selectively reprogram the E3 ubiquitin ligase cereblon (CRBN) to commit target proteins for degradation by the ubiquitin-proteasome system. MGDs create novel recognition interfaces on the surface of the E3 ligase that engage in induced protein-protein interactions with neosubstrates. Molecular insight into their mechanism of action opens exciting opportunities to engage a plethora of targets through a specific recognition motif, the G-loop. Our analysis shows that current CRBN-based MGDs can in principle recognize over 2,500 proteins in the human proteome that contain a G-loop. We review recent advances in tuning the specificity between CRBN and its MGD-induced neosubstrates and deduce a set of simple rules that govern these interactions. We conclude that rational MGD design efforts will enable selective degradation of many more proteins, expanding this therapeutic modality to more disease areas.

Dual ON/OFF-switch chimeric antigen receptor controlled by two clinically approved drugs.

The ability to remotely control the activity of chimeric antigen receptors (CARs) with small molecules can improve the safety and efficacy of gene-modified T cells. Split ON- or OFF-switch CARs involve the dissociation of tumor-antigen binding from T cell activation (i.e., CD3ζ) on the receptor (R-) and signaling (S-) chains, respectively, that either associate or are disrupted in the presence of a small molecule. Here, we have developed an inducible (i)ON-CAR comprising the anti-apoptotic B cell lymphoma protein 2 protein in the ectodomain of both chains which associate in the presence of venetoclax. We showed that inducible ON (iON)-CAR T cells respond to target tumors cells in the presence of venetoclax or the BH3 mimetic navitoclax in a dose-dependent manner, while there is no impact of the drugs on equivalent second generation-CAR T cells. Within 48 h of venetoclax withdrawal, iON-CAR T cells lose the ability to respond to target tumor cells in vitro as evaluated by Interferon-gamma (IFNγ) production, and they are reliant upon the presence of venetoclax for in vivo activity. Finally, by fusing a degron sequence to the endodomain of the iON-CAR S-chain we generated an all-in-one ON/OFF-switch CAR, the iONØ-CAR, down-regulated by lenalidomide within 4 to 6 for functionally inactive T cells (no IFNγ production) within 24 h. We propose that our remote-control CAR designs can reduce toxicity in the clinic. Moreover, the periodic rest of iON and iONØ-CAR T cells may alleviate exhaustion and hence augment persistence and long-term tumor control in patients.

A Synthetic Multivalent Lipopeptide Derived from Pam3CSK4 with Irreversible Influenza Inhibition and Immuno-Stimulating Effects.

The activation of the host adaptive immune system is crucial for eliminating viruses. However, influenza infection often suppresses the innate immune response that precedes adaptive immunity, and the adaptive immune responses are typically delayed. Dendritic cells, serving as professional antigen-presenting cells, have a vital role in initiating the adaptive immune response. In this study, an immuno-stimulating antiviral system (ISAS) is introduced, which is composed of the immuno-stimulating adjuvant lipopeptide Pam3CSK4 that acts as a scaffold onto which it is covalently bound 3 to 4 influenza-inhibiting peptides. The multivalent display of peptides on the scaffold leads to a potent inhibition against H1N1 (EC50 = 20 nM). Importantly, the resulting lipopeptide, Pam3FDA, shows an irreversible inhibition mechanism. The chemical modification of peptides on the scaffold maintains Pam3CSK4's ability to stimulate dendritic cell maturation, thereby rendering Pam3FDA a unique antiviral. This is attributed to its immune activation capability, which also acts in synergy to expedite viral elimination.

Chiral evasion and stereospecific antifolate resistance in Staphylococcus aureus.

Antimicrobial resistance presents a significant health care crisis. The mutation F98Y in Staphylococcus aureus dihydrofolate reductase (SaDHFR) confers resistance to the clinically important antifolate trimethoprim (TMP). Propargyl-linked antifolates (PLAs), next generation DHFR inhibitors, are much more resilient than TMP against this F98Y variant, yet this F98Y substitution still reduces efficacy of these agents. Surprisingly, differences in the enantiomeric configuration at the stereogenic center of PLAs influence the isomeric state of the NADPH cofactor. To understand the molecular basis of F98Y-mediated resistance and how PLAs' inhibition drives NADPH isomeric states, we used protein design algorithms in the osprey protein design software suite to analyze a comprehensive suite of structural, biophysical, biochemical, and computational data. Here, we present a model showing how F98Y SaDHFR exploits a different anomeric configuration of NADPH to evade certain PLAs' inhibition, while other PLAs remain unaffected by this resistance mechanism.

Boosting subdominant neutralizing antibody responses with a computationally designed epitope-focused immunogen.

Throughout the last several decades, vaccination has been key to prevent and eradicate infectious diseases. However, many pathogens (e.g., respiratory syncytial virus [RSV], influenza, dengue, and others) have resisted vaccine development efforts, largely because of the failure to induce potent antibody responses targeting conserved epitopes. Deep profiling of human B cells often reveals potent neutralizing antibodies that emerge from natural infection, but these specificities are generally subdominant (i.e., are present in low titers). A major challenge for next-generation vaccines is to overcome established immunodominance hierarchies and focus antibody responses on crucial neutralization epitopes. Here, we show that a computationally designed epitope-focused immunogen presenting a single RSV neutralization epitope elicits superior epitope-specific responses compared to the viral fusion protein. In addition, the epitope-focused immunogen efficiently boosts antibodies targeting the palivizumab epitope, resulting in enhanced neutralization. Overall, we show that epitope-focused immunogens can boost subdominant neutralizing antibody responses in vivo and reshape established antibody hierarchies.

OSPREY 3.0: Open-source protein redesign for you, with powerful new features.

We present osprey 3.0, a new and greatly improved release of the osprey protein design software. Osprey 3.0 features a convenient new Python interface, which greatly improves its ease of use. It is over two orders of magnitude faster than previous versions of osprey when running the same algorithms on the same hardware. Moreover, osprey 3.0 includes several new algorithms, which introduce substantial speedups as well as improved biophysical modeling. It also includes GPU support, which provides an additional speedup of over an order of magnitude. Like previous versions of osprey, osprey 3.0 offers a unique package of advantages over other design software, including provable design algorithms that account for continuous flexibility during design and model conformational entropy. Finally, we show here empirically that osprey 3.0 accurately predicts the effect of mutations on protein-protein binding. Osprey 3.0 is available at http://www.cs.duke.edu/donaldlab/osprey.php as free and open-source software. © 2018 Wiley Periodicals, Inc.

Protein design algorithms predict viable resistance to an experimental antifolate.

Methods to accurately predict potential drug target mutations in response to early-stage leads could drive the design of more resilient first generation drug candidates. In this study, a structure-based protein design algorithm (K* in the OSPREY suite) was used to prospectively identify single-nucleotide polymorphisms that confer resistance to an experimental inhibitor effective against dihydrofolate reductase (DHFR) from Staphylococcus aureus. Four of the top-ranked mutations in DHFR were found to be catalytically competent and resistant to the inhibitor. Selection of resistant bacteria in vitro reveals that two of the predicted mutations arise in the background of a compensatory mutation. Using enzyme kinetics, microbiology, and crystal structures of the complexes, we determined the fitness of the mutant enzymes and strains, the structural basis of resistance, and the compensatory relationship of the mutations. To our knowledge, this work illustrates the first application of protein design algorithms to prospectively predict viable resistance mutations that arise in bacteria under antibiotic pressure.

Fast gap-free enumeration of conformations and sequences for protein design.

Despite significant successes in structure-based computational protein design in recent years, protein design algorithms must be improved to increase the biological accuracy of new designs. Protein design algorithms search through an exponential number of protein conformations, protein ensembles, and amino acid sequences in an attempt to find globally optimal structures with a desired biological function. To improve the biological accuracy of protein designs, it is necessary to increase both the amount of protein flexibility allowed during the search and the overall size of the design, while guaranteeing that the lowest-energy structures and sequences are found. DEE/A*-based algorithms are the most prevalent provable algorithms in the field of protein design and can provably enumerate a gap-free list of low-energy protein conformations, which is necessary for ensemble-based algorithms that predict protein binding. We present two classes of algorithmic improvements to the A* algorithm that greatly increase the efficiency of A*. First, we analyze the effect of ordering the expansion of mutable residue positions within the A* tree and present a dynamic residue ordering that reduces the number of A* nodes that must be visited during the search. Second, we propose new methods to improve the conformational bounds used to estimate the energies of partial conformations during the A* search. The residue ordering techniques and improved bounds can be combined for additional increases in A* efficiency. Our enhancements enable all A*-based methods to more fully search protein conformation space, which will ultimately improve the accuracy of complex biomedically relevant designs.

Protein design using continuous rotamers.

UNLABELLED: Optimizing amino acid conformation and identity is a central problem in computational protein design. Protein design algorithms must allow realistic protein flexibility to occur during this optimization, or they may fail to find the best sequence with the lowest energy. Most design algorithms implement side-chain flexibility by allowing the side chains to move between a small set of discrete, low-energy states, which we call rigid rotamers. In this work we show that allowing continuous side-chain flexibility (which we call continuous rotamers) greatly improves protein flexibility modeling. We present a large-scale study that compares the sequences and best energy conformations in 69 protein-core redesigns using a rigid-rotamer model versus a continuous-rotamer model. We show that in nearly all of our redesigns the sequence found by the continuous-rotamer model is different and has a lower energy than the one found by the rigid-rotamer model. Moreover, the sequences found by the continuous-rotamer model are more similar to the native sequences. We then show that the seemingly easy solution of sampling more rigid rotamers within the continuous region is not a practical alternative to a continuous-rotamer model: at computationally feasible resolutions, using more rigid rotamers was never better than a continuous-rotamer model and almost always resulted in higher energies. Finally, we present a new protein design algorithm based on the dead-end elimination (DEE) algorithm, which we call iMinDEE, that makes the use of continuous rotamers feasible in larger systems. iMinDEE guarantees finding the optimal answer while pruning the search space with close to the same efficiency of DEE. AVAILABILITY: Software is available under the Lesser GNU Public License v3. Contact the authors for source code.

Rules and mechanisms governing G protein coupling selectivity of GPCRs.

G protein-coupled receptors (GPCRs) convert extracellular stimuli into intracellular signaling by coupling to heterotrimeric G proteins of four classes: Gi/o, Gq, Gs, and G12/13. However, our understanding of the G protein selectivity of GPCRs is incomplete. Here, we quantitatively measure the enzymatic activity of GPCRs in living cells and reveal the G protein selectivity of 124 GPCRs with the exact rank order of their G protein preference. Using this information, we establish a classification of GPCRs by functional selectivity, discover the existence of a G12/13-coupled receptor, G15-coupled receptors, and a variety of subclasses for Gi/o-, Gq-, and Gs-coupled receptors, culminating in development of the predictive algorithm of G protein selectivity. We further identify the structural determinants of G protein selectivity, allowing us to synthesize non-existent GPCRs with de novo G protein selectivity and efficiently identify putative pathogenic variants.

Rational Design of Chemically Controlled Antibodies and Protein Therapeutics.

Protein-based therapeutics, such as monoclonal antibodies and cytokines, are important therapies for various pathophysiological conditions such as oncology, autoimmune disorders, and viral infections. However, the wide application of such protein therapeutics is often hindered by dose-limiting toxicities and adverse effects, namely, cytokine storm syndrome, organ failure, and others. Therefore, spatiotemporal control of the activities of these proteins is crucial to further expand their application. Here, we report the design and application of small-molecule-controlled switchable protein therapeutics by taking advantage of a previously engineered OFF-switch system. We used the Rosetta modeling suite to computationally optimize the affinity between B-cell lymphoma 2 (Bcl-2) protein and a previously developed computationally designed protein partner (LD3) to obtain a fast and efficient heterodimer disruption upon the addition of a competing drug (Venetoclax). The incorporation of the engineered OFF-switch system into anti-CTLA4, anti-HER2 antibodies, or an Fc-fused IL-15 cytokine demonstrated an efficient disruption in vitro, as well as fast clearance in vivo upon the addition of the competing drug Venetoclax. These results provide a proof-of-concept for the rational design of controllable biologics by introducing a drug-induced OFF-switch into existing protein-based therapeutics.

Optimization of therapeutic antibodies by predicting antigen specificity from antibody sequence via deep learning.

The optimization of therapeutic antibodies is time-intensive and resource-demanding, largely because of the low-throughput screening of full-length antibodies (approximately 1 × 103 variants) expressed in mammalian cells, which typically results in few optimized leads. Here we show that optimized antibody variants can be identified by predicting antigen specificity via deep learning from a massively diverse space of antibody sequences. To produce data for training deep neural networks, we deep-sequenced libraries of the therapeutic antibody trastuzumab (about 1 × 104 variants), expressed in a mammalian cell line through site-directed mutagenesis via CRISPR-Cas9-mediated homology-directed repair, and screened the libraries for specificity to human epidermal growth factor receptor 2 (HER2). We then used the trained neural networks to screen a computational library of approximately 1 × 108 trastuzumab variants and predict the HER2-specific subset (approximately 1 × 106 variants), which can then be filtered for viscosity, clearance, solubility and immunogenicity to generate thousands of highly optimized lead candidates. Recombinant expression and experimental testing of 30 randomly selected variants from the unfiltered library showed that all 30 retained specificity for HER2. Deep learning may facilitate antibody engineering and optimization.

A rational blueprint for the design of chemically-controlled protein switches.

Small-molecule responsive protein switches are crucial components to control synthetic cellular activities. However, the repertoire of small-molecule protein switches is insufficient for many applications, including those in the translational spaces, where properties such as safety, immunogenicity, drug half-life, and drug side-effects are critical. Here, we present a computational protein design strategy to repurpose drug-inhibited protein-protein interactions as OFF- and ON-switches. The designed binders and drug-receptors form chemically-disruptable heterodimers (CDH) which dissociate in the presence of small molecules. To design ON-switches, we converted the CDHs into a multi-domain architecture which we refer to as activation by inhibitor release switches (AIR) that incorporate a rationally designed drug-insensitive receptor protein. CDHs and AIRs showed excellent performance as drug responsive switches to control combinations of synthetic circuits in mammalian cells. This approach effectively expands the chemical space and logic responses in living cells and provides a blueprint to develop new ON- and OFF-switches.

A computationally designed chimeric antigen receptor provides a small-molecule safety switch for T-cell therapy.

Approaches to increase the activity of chimeric antigen receptor (CAR)-T cells against solid tumors may also increase the risk of toxicity and other side effects. To improve the safety of CAR-T-cell therapy, we computationally designed a chemically disruptable heterodimer (CDH) based on the binding of two human proteins. The CDH self-assembles, can be disrupted by a small-molecule drug and has a high-affinity protein interface with minimal amino acid deviation from wild-type human proteins. We incorporated the CDH into a synthetic heterodimeric CAR, called STOP-CAR, that has an antigen-recognition chain and a CD3ζ- and CD28-containing endodomain signaling chain. We tested STOP-CAR-T cells specific for two antigens in vitro and in vivo and found similar antitumor activity compared to second-generation (2G) CAR-T cells. Timed administration of the small-molecule drug dynamically inactivated the activity of STOP-CAR-T cells. Our work highlights the potential for structure-based design to add controllable elements to synthetic cellular therapies.

Author Correction: A computationally designed chimeric antigen receptor provides a small-molecule safety switch for T-cell therapy.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

OSPREY Predicts Resistance Mutations Using Positive and Negative Computational Protein Design.

Drug resistance in protein targets is an increasingly common phenomenon that reduces the efficacy of both existing and new antibiotics. However, knowledge of future resistance mutations during pre-clinical phases of drug development would enable the design of novel antibiotics that are robust against not only known resistant mutants, but also against those that have not yet been clinically observed. Computational structure-based protein design (CSPD) is a transformative field that enables the prediction of protein sequences with desired biochemical properties such as binding affinity and specificity to a target. The use of CSPD to predict previously unseen resistance mutations represents one of the frontiers of computational protein design. In a recent study (Reeve et al. Proc Natl Acad Sci U S A 112(3):749-754, 2015), we used our OSPREY (Open Source Protein REdesign for You) suite of CSPD algorithms to prospectively predict resistance mutations that arise in the active site of the dihydrofolate reductase enzyme from methicillin-resistant Staphylococcus aureus (SaDHFR) in response to selective pressure from an experimental competitive inhibitor. We demonstrated that our top predicted candidates are indeed viable resistant mutants. Since that study, we have significantly enhanced the capabilities of OSPREY with not only improved modeling of backbone flexibility, but also efficient multi-state design, fast sparse approximations, partitioned continuous rotamers for more accurate energy bounds, and a computationally efficient representation of molecular-mechanics and quantum-mechanical energy functions. Here, using SaDHFR as an example, we present a protocol for resistance prediction using the latest version of OSPREY. Specifically, we show how to use a combination of positive and negative design to predict active site escape mutations that maintain the enzyme's catalytic function but selectively ablate binding of an inhibitor.

Algorithms for protein design.

Computational structure-based protein design programs are becoming an increasingly important tool in molecular biology. These programs compute protein sequences that are predicted to fold to a target structure and perform a desired function. The success of a program's predictions largely relies on two components: first, the input biophysical model, and second, the algorithm that computes the best sequence(s) and structure(s) according to the biophysical model. Improving both the model and the algorithm in tandem is essential to improving the success rate of current programs, and here we review recent developments in algorithms for protein design, emphasizing how novel algorithms enable the use of more accurate biophysical models. We conclude with a list of algorithmic challenges in computational protein design that we believe will be especially important for the design of therapeutic proteins and protein assemblies.

Compact Representation of Continuous Energy Surfaces for More Efficient Protein Design.

In macromolecular design, conformational energies are sensitive to small changes in atom coordinates; thus, modeling the small, continuous motions of atoms around low-energy wells confers a substantial advantage in structural accuracy. However, modeling these motions comes at the cost of a very large number of energy function calls, which form the bottleneck in the design calculations. In this work, we remove this bottleneck by consolidating all conformational energy evaluations into the pre-computation of a local polynomial expansion of the energy about the "ideal" conformation for each low-energy, "rotameric" state of each residue pair. This expansion is called "energy as polynomials in internal coordinates" (EPIC), where the internal coordinates can be side-chain dihedrals, backrub angles, and/or any other continuous degrees of freedom of a macromolecule, and any energy function can be used without adding any asymptotic complexity to the design. We demonstrate that EPIC efficiently represents the energy surface for both molecular-mechanics and quantum-mechanical energy functions, and apply it specifically to protein design for modeling both side chain and backbone degrees of freedom.

OSPREY: protein design with ensembles, flexibility, and provable algorithms.

UNLABELLED: We have developed a suite of protein redesign algorithms that improves realistic in silico modeling of proteins. These algorithms are based on three characteristics that make them unique: (1) improved flexibility of the protein backbone, protein side-chains, and ligand to accurately capture the conformational changes that are induced by mutations to the protein sequence; (2) modeling of proteins and ligands as ensembles of low-energy structures to better approximate binding affinity; and (3) a globally optimal protein design search, guaranteeing that the computational predictions are optimal with respect to the input model. Here, we illustrate the importance of these three characteristics. We then describe OSPREY, a protein redesign suite that implements our protein design algorithms. OSPREY has been used prospectively, with experimental validation, in several biomedically relevant settings. We show in detail how OSPREY has been used to predict resistance mutations and explain why improved flexibility, ensembles, and provability are essential for this application. AVAILABILITY: OSPREY is free and open source under a Lesser GPL license. The latest version is OSPREY 2.0. The program, user manual, and source code are available at www.cs.duke.edu/donaldlab/software.php. CONTACT: osprey@cs.duke.edu.