Exploring the origins of binding specificity through the computational redesign of calmodulin

Designed Loop Extension Followed by Combinatorial Screening Confers High Specificity to a Broad Matrix MetalloproteinaseInhibitor

Matrix metalloproteinases (MMPs) are key drivers of various diseases, including cancer. Development of probes and drugs capable of selectively inhibiting the individual members of the large MMP family remains a persistent challenge. The inhibitory N-terminal domain of tissue inhibitor of metalloproteinases-2 (N-TIMP2), a natural broad MMP inhibitor, can provide a scaffold for protein engineering to create more selective MMP inhibitors. Here, we pursued a unique approach harnessing both computational design and combinatorial screening to confer high binding specificity toward a target MMP in preference to an anti-target MMP. We designed a loop extension of N-TIMP2 to allow new interactions with the non-conserved MMP surface and generated an efficient focused library for yeast surface display, which was then screened for high binding to the target MMP-14 and low binding to anti-target MMP-3. Deep sequencing analysis identified the most promising variants, which were expressed, purified, and tested for selectivity of inhibition. Our best N-TIMP2 variant exhibited 29 pM binding affinity to MMP-14 and 2.4 µM affinity to MMP-3, revealing 7500-fold greater specificity than WT N-TIMP2. High-confidence structural models were obtained by including NGS data in the AlphaFold multiple sequence alignment. The modeling together with experimental mutagenesis validated our design predictions, demonstrating that the loop extension packs tightly against non-conserved residues on MMP-14 and clashes with MMP-3. This study demonstrates how introduction of loop extensions in a manner guided by target protein conservation data and loop design can offer an attractive strategy to achieve specificity in design of protein ligands.

Connecting Artificial Proteolytic and Electrochemical Signaling Systems with Caged Messenger Peptides

Enzymatic polypeptide proteolysis is a widespread and powerful biological control mechanism. Over the last few years, substantial progress has been made in creating artificial proteolytic systems where an input of choice modulates the protease activity and thereby the activity of its substrates. However, all proteolytic systems developed so far have relied on the direct proteolytic cleavage of their effectors. Here, we propose a new concept where protease biosensors with a tunable input uncage a signaling peptide, which can then transmit a signal to an allosteric protein reporter. We demonstrate that both the cage and the regulatory domain of the reporter can be constructed from the same peptide-binding domain, such as calmodulin. To demonstrate this concept, we constructed a proteolytic rapamycin biosensor and demonstrated its quantitative actuation on fluorescent, luminescent, and electrochemical reporters. Using the latter, we constructed sensitive bioelectrodes that detect the messenger peptide release and quantitatively convert the recognition event into electric current. We discuss the application of such systems for the construction of in vitro sensory arrays and in vivo signaling circuits.

Rosetta:MSF:NN: Boosting performance of multi-state computational protein design with a neural network

Rational protein design aims at the targeted modification of existing proteins. To reach this goal, software suites like Rosetta propose sequences to introduce the desired properties. Challenging design problems necessitate the representation of a protein by means of a structural ensemble. Thus, Rosetta multi-state design (MSD) protocols have been developed wherein each state represents one protein conformation. Computational demands of MSD protocols are high, because for each of the candidate sequences a costly three-dimensional (3D) model has to be created and assessed for all states. Each of these scores contributes one data point to a complex, design-specific energy landscape. As neural networks (NN) proved well-suited to learn such solution spaces, we integrated one into the framework Rosetta:MSF instead of the so far used genetic algorithm with the aim to reduce computational costs. As its predecessor, Rosetta:MSF:NN administers a set of candidate sequences and their scores and scans sequence space iteratively. During each iteration, the union of all candidate sequences and their Rosetta scores are used to re-train NNs that possess a design-specific architecture. The enormous speed of the NNs allows an extensive assessment of alternative sequences, which are ranked on the scores predicted by the NN. Costly 3D models are computed only for a small fraction of best-scoring sequences; these and the corresponding 3D-based scores replace half of the candidate sequences during each iteration. The analysis of two sets of candidate sequences generated for a specific design problem by means of a genetic algorithm confirmed that the NN predicted 3D-based scores quite well; the Pearson correlation coefficient was at least 0.95. Applying Rosetta:MSF:NN:enzdes to a benchmark consisting of 16 ligand-binding problems showed that this protocol converges ten-times faster than the genetic algorithm and finds sequences with comparable scores.

Assessing the Role of Calmodulin's Linker Flexibility in Target Binding

Calmodulin (CaM) is a highly-expressed Ca2+ binding protein known to bind hundreds of protein targets. Its binding selectivity to many of these targets is partially attributed to the protein’s flexible alpha helical linker that connects its N- and C-domains. It is not well established how its linker mediates CaM’s binding to regulatory targets yet. Insights into this would be invaluable to understanding its regulation of diverse cellular signaling pathways. Therefore, we utilized Martini coarse-grained (CG) molecular dynamics simulations to probe CaM/target assembly for a model system: CaM binding to the calcineurin (CaN) regulatory domain. The simulations were conducted assuming a ‘wild-type’ calmodulin with normal flexibility of its linker, as well as a labile, highly-flexible linker variant to emulate structural changes that could be induced, for instance, by post-translational modifications. For the wild-type model, 98% of the 600 simulations across three ionic strengths adopted a bound complex within 2 μs of simulation time; of these, 1.7% sampled the fully-bound state observed in the experimentally-determined crystallographic structure. By calculating the mean-first-passage-time for these simulations, we estimated the association rate to be ka= 8.7 × 108 M−1 s−1, which is similar to the diffusion-limited, experimentally-determined rate of 2.2 × 108 M−1 s−1. Furthermore, our simulations recapitulated its well-known inverse relationship between the association rate and the solution ionic strength. In contrast, although over 97% of the labile linker simulations formed tightly-bound complexes, only 0.3% achieved the fully-bound configuration. This effect appears to stem from a difference in the ensembles of extended and collapsed states which are controlled by the linker flexibility. Therefore, our simulations suggest that variations in the CaM linker’s propensity for alpha helical secondary structure can modulate the kinetics of target binding.

Optimizing the Specificity Window of Biomolecular Receptors Using Structure-Switching and Allostery

To ensure maximum specificity (i.e., minimize cross-reactivity with structurally similar analogues of the desired target), most bioassays invoke "stringency", the careful tuning of the conditions employed (e.g., pH, ionic strength, or temperature). Willingness to control assay conditions will fall, however, as quantitative, single-step biosensors begin to replace multistep analytical processes. This is especially true for sensors deployed in vivo, where the tuning of such parameters is not just inconvenient but impossible. In response, we describe here the rational adaptation of two strategies employed by nature to tune the affinity of biomolecular receptors so as to optimize the placement of their specificity "windows" without the need to alter measurement conditions: structure-switching and allosteric control. We quantitatively validate these approaches using two distinct, DNA-based receptors: a simple, linear-chain DNA suitable for detecting a complementary DNA strand and a structurally complex DNA aptamer used for the detection of a small-molecule drug. Using these models, we show that, without altering assay conditions, structure-switching and allostery can tune the concentration range over which a receptor achieves optimal specificity over orders of magnitude, thus optimally matching the specificity window with the range of target concentrations expected to be seen in a given application.

Ultrahigh specificity in a network of computationally designed protein-interaction pairs

Protein networks in all organisms comprise homologous interacting pairs. In these networks, some proteins are specific, interacting with one or a few binding partners, whereas others are multispecific and bind a range of targets. We describe an algorithm that starts from an interacting pair and designs dozens of new pairs with diverse backbone conformations at the binding site as well as new binding orientations and sequences. Applied to a high-affinity bacterial pair, the algorithm results in 18 new ones, with cognate affinities from pico- to micromolar. Three pairs exhibit 3-5 orders of magnitude switch in specificity relative to the wild type, whereas others are multispecific, collectively forming a protein-interaction network. Crystallographic analysis confirms design accuracy, including in new backbones and polar interactions. Preorganized polar interaction networks are responsible for high specificity, thus defining design principles that can be applied to program synthetic cellular interaction networks of desired affinity and specificity.

The molecular basis of ultrahigh specificity in protein-protein interactions remains obscure. The authors present a computational method to design atomically accurate new pairs exhibiting >100,000-fold specificity switches, generating a large and complex interaction network.

Converting a broad matrix metalloproteinase family inhibitor into a specific inhibitor of MMP-9 and MMP-14

MMP-14 and MMP-9 are two well-established cancer targets for which no specific clinically relevant inhibitor is available. Using a powerful combination of computational design and yeast surface display technology, we engineered such an inhibitor starting from a nonspecific MMP inhibitor, N-TIMP2. The engineered purified N-TIMP2 variants showed enhanced specificity toward MMP-14 and MMP-9 relative to a panel of off-target MMPs. MMP-specific N-TIMP2 sequence signatures were obtained that could be understood from the structural perspective of MMP/N-TIMP2 interactions. Our MMP-9 inhibitor exhibited 1000-fold preference for MMP-9 vs. MMP-14, which is likely to translate into significant differences under physiological conditions. Our results provide new insights regarding evolution of promiscuous proteins and optimization strategies for design of inhibitors with single-target specificities.

Scratching the Surface: Resurfacing Proteins to Endow New Properties and Function

Protein engineering is an emerging discipline that dovetails modern molecular biology techniques with high-throughput screening, laboratory evolution technologies, and computational approaches to modify sequence, structure, and, in some cases, function and properties of proteins. The ultimate goal is to develop new proteins with improved or designer functions for use in biotechnology, medicine, and basic research. One way to engineer proteins is to change their solvent-exposed regions through focused or random "protein resurfacing." In this review we explain what protein resurfacing is, and discuss recent examples of how this strategy is used to generate proteins with altered or broadened recognition profiles, improved stability, solubility, and expression, cell-penetrating ability, and reduced immunogenicity. Additionally we comment on how these properties can be further improved using chemical resurfacing approaches. Protein resurfacing will likely play an increasingly important role as more biologics enter clinical use, and we present some arguments to support this view.

Rosetta:MSF: a modular framework for multi-state computational protein design

Computational protein design (CPD) is a powerful technique to engineer existing proteins or to design novel ones that display desired properties. Rosetta is a software suite including algorithms for computational modeling and analysis of protein structures and offers many elaborate protocols created to solve highly specific tasks of protein engineering. Most of Rosetta’s protocols optimize sequences based on a single conformation (i. e. design state). However, challenging CPD objectives like multi-specificity design or the concurrent consideration of positive and negative design goals demand the simultaneous assessment of multiple states. This is why we have developed the multi-state framework MSF that facilitates the implementation of Rosetta’s single-state protocols in a multi-state environment and made available two frequently used protocols. Utilizing MSF, we demonstrated for one of these protocols that multi-state design yields a 15% higher performance than single-state design on a ligand-binding benchmark consisting of structural conformations. With this protocol, we designed de novo nine retro-aldolases on a conformational ensemble deduced from a (βα)₈-barrel protein. All variants displayed measurable catalytic activity, testifying to a high success rate for this concept of multi-state enzyme design.

Protein engineering, i. e. the targeted modification or design of proteins has tremendous potential for medical and industrial applications. One generally applicable strategy for protein engineering is rational protein design: based on detailed knowledge of structure and function, computer programs like Rosetta propose the sequence of a protein possessing the desired properties. So far, most computer protocols have used rigid structures for design, which is a simplification because a protein’s structure is more accurately specified by a conformational ensemble. We have now implemented a framework for computational protein design that allows certain design protocols of Rosetta to make use of multiple design states like structural ensembles. An in silico assessment simulating ligand-binding design showed that this new approach generates more reliably native-like sequences than a single-state approach. As a proof-of-concept, we introduced de novo retro-aldolase activity into a scaffold protein and characterized nine variants experimentally, all of which were catalytically active.

Computational protein design: a review

Proteins are one of the most versatile modular assembling systems in nature. Experimentally, more than 110 000 protein structures have been identified and more are deposited every day in the Protein Data Bank. Such an enormous structural variety is to a first approximation controlled by the sequence of amino acids along the peptide chain of each protein. Understanding how the structural and functional properties of the target can be encoded in this sequence is the main objective of protein design. Unfortunately, rational protein design remains one of the major challenges across the disciplines of biology, physics and chemistry. The implications of solving this problem are enormous and branch into materials science, drug design, evolution and even cryptography. For instance, in the field of drug design an effective computational method to design protein-based ligands for biological targets such as viruses, bacteria or tumour cells, could give a significant boost to the development of new therapies with reduced side effects. In materials science, self-assembly is a highly desired property and soon artificial proteins could represent a new class of designable self-assembling materials. The scope of this review is to describe the state of the art in computational protein design methods and give the reader an outline of what developments could be expected in the near future.

Achievements and Challenges in Computational Protein Design

Computational protein design (CPD), a yet evolving field, includes computer-aided engineering for partial or full de novo designs of proteins of interest. Designs are defined by a requested structure, function, or working environment. This chapter describes the birth and maturation of the field by presenting 101 CPD examples in a chronological order emphasizing achievements and pending challenges. Integrating these aspects presents the plethora of CPD approaches with the hope of providing a "CPD 101". These reflect on the broader structural bioinformatics and computational biophysics field and include: (1) integration of knowledge-based and energy-based methods, (2) hierarchical designated approach towards local, regional, and global motifs and the integration of high- and low-resolution design schemes that fit each such region, (3) systematic differential approaches towards different protein regions, (4) identification of key hot-spot residues and the relative effect of remote regions, (5) assessment of shape-complementarity, electrostatics and solvation effects, (6) integration of thermal plasticity and functional dynamics, (7) negative design, (8) systematic integration of experimental approaches, (9) objective cross-assessment of methods, and (10) successful ranking of potential designs. Future challenges also include dissemination of CPD software to the general use of life-sciences researchers and the emphasis of success within an in vivo milieu. CPD increases our understanding of protein structure and function and the relationships between the two along with the application of such know-how for the benefit of mankind. Applied aspects range from biological drugs, via healthier and tastier food products to nanotechnology and environmentally friendly enzymes replacing toxic chemicals utilized in the industry.

The Framework of Computational Protein Design

Computational protein design (CPD) has established itself as a leading field in basic and applied science with a strong coupling between the two. Proteins are computationally designed from the level of amino acids to the level of a functional protein complex. Design targets range from increased thermo- (or other) stability to specific requested reactions such as protein-protein binding, enzymatic reactions, or nanotechnology applications. The design scheme may encompass small regions of the proteins or the entire protein. In either case, the design may aim at the side-chains or at the full backbone conformation. Herein, the main framework for the process is outlined highlighting key elements in the CPD iterative cycle. These include the very definition of CPD, the diverse goals of CPD, components of the CPD protocol, methods for searching sequence and structure space, scoring functions, and augmenting the CPD with other optimization tools. Taken together, this chapter aims to introduce the framework of CPD.

An Evolution-Based Approach to De Novo Protein Design

EvoDesign is a computational algorithm that allows the rapid creation of new protein sequences that are compatible with specific protein structures. As such, it can be used to optimize protein stability, to resculpt the protein surface to eliminate undesired protein-protein interactions, and to optimize protein-protein binding. A major distinguishing feature of EvoDesign in comparison to other protein design programs is the use of evolutionary information in the design process to guide the sequence search toward native-like sequences known to adopt structurally similar folds as the target. The observed frequencies of amino acids in specific positions in the structure in the form of structural profiles collected from proteins with similar folds and complexes with similar interfaces can implicitly capture many subtle effects that are essential for correct folding and protein-binding interactions. As a result of the inclusion of evolutionary information, the sequences designed by EvoDesign have native-like folding and binding properties not seen by other physics-based design methods. In this chapter, we describe how EvoDesign can be used to redesign proteins with a focus on the computational and experimental procedures that can be used to validate the designs.

Protein Engineering by Combined Computational and In Vitro Evolution Approaches

Two alternative strategies are commonly used to study protein-protein interactions (PPIs) and to engineer protein-based inhibitors. In one approach, binders are selected experimentally from combinatorial libraries of protein mutants that are displayed on a cell surface. In the other approach, computational modeling is used to explore an astronomically large number of protein sequences to select a small number of sequences for experimental testing. While both approaches have some limitations, their combination produces superior results in various protein engineering applications. Such applications include the design of novel binders and inhibitors, the enhancement of affinity and specificity, and the mapping of binding epitopes. The combination of these approaches also aids in the understanding of the specificity profiles of various PPIs.

Bioorthogonal Chemoenzymatic Functionalization of Calmodulin for Bioconjugation Applications

Calmodulin (CaM) is a widely studied Ca(2+)-binding protein that is highly conserved across species and involved in many biological processes, including vesicle release, cell proliferation, and apoptosis. To facilitate biophysical studies of CaM, researchers have tagged and mutated CaM at various sites, enabling its conjugation to fluorophores, microarrays, and other reactive partners. However, previous attempts to add a reactive label to CaM for downstream studies have generally employed nonselective labeling methods or resulted in diminished CaM function. Here we report the first engineered CaM protein that undergoes site-specific and bioorthogonal labeling while retaining wild-type activity levels. By employing a chemoenzymatic labeling approach, we achieved selective and quantitative labeling of the engineered CaM protein with an N-terminal 12-azidododecanoic acid tag; notably, addition of the tag did not interfere with the ability of CaM to bind Ca(2+) or a partner protein. The specificity of our chemoenzymatic labeling approach also allowed for selective conjugation of CaM to reactive partners in bacterial cell lysates, without intermediate purification of the engineered protein. Additionally, we prepared CaM-affinity resins that were highly effective in purifying a representative CaM-binding protein, demonstrating that the engineered CaM remains active even after surface capture. Beyond studies of CaM and CaM-binding proteins, the protein engineering and surface capture methods described here should be translatable to other proteins and other bioconjugation applications.

Lessons in Protein Design from Combined Evolution and Conformational Dynamics

Protein-protein interactions play important roles in the control of every cellular process. How natural selection has optimized protein design to produce molecules capable of binding to many partner proteins is a fascinating problem but not well understood. Here, we performed a combinatorial analysis of protein sequence evolution and conformational dynamics to study how calmodulin (CaM), which plays essential roles in calcium signaling pathways, has adapted to bind to a large number of partner proteins. We discovered that amino acid residues in CaM can be partitioned into unique classes according to their degree of evolutionary conservation and local stability. Holistically, categorization of CaM residues into these classes reveals enriched physico-chemical interactions required for binding to diverse targets, balanced against the need to maintain the folding and structural modularity of CaM to achieve its overall function. The sequence-structure-function relationship of CaM provides a concrete example of the general principle of protein design. We have demonstrated the synergy between the fields of molecular evolution and protein biophysics and created a generalizable framework broadly applicable to the study of protein-protein interactions.

Computational design and experimental verification of a symmetric protein homodimer

Homodimers are the most common type of protein assembly in nature and have distinct features compared with heterodimers and higher order oligomers. Understanding homodimer interactions at the atomic level is critical both for elucidating their biological mechanisms of action and for accurate modeling of complexes of unknown structure. Computation-based design of novel protein-protein interfaces can serve as a bottom-up method to further our understanding of protein interactions. Previous studies have demonstrated that the de novo design of homodimers can be achieved to atomic-level accuracy by β-strand assembly or through metal-mediated interactions. Here, we report the design and experimental characterization of a α-helix-mediated homodimer with C2 symmetry based on a monomeric Drosophila engrailed homeodomain scaffold. A solution NMR structure shows that the homodimer exhibits parallel helical packing similar to the design model. Because the mutations leading to dimer formation resulted in poor thermostability of the system, design success was facilitated by the introduction of independent thermostabilizing mutations into the scaffold. This two-step design approach, function and stabilization, is likely to be generally applicable, especially if the desired scaffold is of low thermostability.

Optimizing the affinity and specificity of ligand binding with the inclusion of solvation effect

Solvation effect is an important factor for protein-ligand binding in aqueous water. Previous scoring function of protein-ligand interactions rarely incorporates the solvation model into the quantification of protein-ligand interactions, mainly due to the immense computational cost, especially in the structure-based virtual screening, and nontransferable application of independently optimized atomic solvation parameters. In order to overcome these barriers, we effectively combine knowledge-based atom-pair potentials and the atomic solvation energy of charge-independent implicit solvent model in the optimization of binding affinity and specificity. The resulting scoring functions with optimized atomic solvation parameters is named as specificity and affinity with solvation effect (SPA-SE). The performance of SPA-SE is evaluated and compared to 20 other scoring functions, as well as SPA. The comparative results show that SPA-SE outperforms all other scoring functions in binding affinity prediction and "native" pose identification. Our optimization validates that solvation effect is an important regulator to the stability and specificity of protein-ligand binding. The development strategy of SPA-SE sets an example for other scoring function to account for the solvation effect in biomolecular recognitions.

Artificial Diiron Enzymes with a De Novo Designed Four-Helix Bundle Structure

A single polypeptide chain may provide an astronomical number of conformers. Nature selected only a trivial number of them through evolution, composing an alphabet of scaffolds, that can afford the complete set of chemical reactions needed to support life. These structural templates are so stable that they allow several mutations without disruption of the global folding, even having the ability to bind several exogenous cofactors. With this perspective, metal cofactors play a crucial role in the regulation and catalysis of several processes. Nature is able to modulate the chemistry of metals, adopting only a few ligands and slightly different geometries. Several scaffolds and metal-binding motifs are representing the focus of intense interest in the literature. This review discusses the widespread four-helix bundle fold, adopted as a scaffold for metal binding sites in the context of de novo protein design to obtain basic biochemical components for biosensing or catalysis. In particular, we describe the rational refinement of structure/function in diiron-oxo protein models from the due ferri (DF) family. The DF proteins were developed by us through an iterative process of design and rigorous characterization, which has allowed a shift from structural to functional models. The examples reported herein demonstrate the importance of the synergic application of de novo design methods as well as spectroscopic and structural characterization to optimize the catalytic performance of artificial enzymes.

Computational design of selective peptides to discriminate between similar PDZ domains in an oncogenic pathway

Reagents that target protein-protein interactions to rewire signaling are of great relevance in biological research. Computational protein design may offer a means of creating such reagents on demand, but methods for encoding targeting selectivity are sorely needed. This is especially challenging when targeting interactions with ubiquitous recognition modules--for example, PDZ domains, which bind C-terminal sequences of partner proteins. Here we consider the problem of designing selective PDZ inhibitor peptides in the context of an oncogenic signaling pathway, in which two PDZ domains (NHERF-2 PDZ2-N2P2 and MAGI-3 PDZ6-M3P6) compete for a receptor C-terminus to differentially modulate oncogenic activities. Because N2P2 has been shown to increase tumorigenicity and M3P6 to decreases it, we sought to design peptides that inhibit N2P2 without affecting M3P6. We developed a structure-based computational design framework that models peptide flexibility in binding yet is efficient enough to rapidly analyze tradeoffs between affinity and selectivity. Designed peptides showed low-micromolar inhibition constants for N2P2 and no detectable M3P6 binding. Peptides designed for reverse discrimination bound M3P6 tighter than N2P2, further testing our technology. Experimental and computational analysis of selectivity determinants revealed significant indirect energetic coupling in the binding site. Successful discrimination between N2P2 and M3P6, despite their overlapping binding preferences, is highly encouraging for computational approaches to selective PDZ targeting, especially because design relied on a homology model of M3P6. Still, we demonstrate specific deficiencies of structural modeling that must be addressed to enable truly robust design. The presented framework is general and can be applied in many scenarios to engineer selective targeting.

Rational design of topographical helix mimics as potent inhibitors of protein-protein interactions

Protein–protein interactions encompass large surface areas, but often a handful of key residues dominate the binding energy landscape. Rationally designed small molecule scaffolds that reproduce the relative positioning and disposition of important binding residues, termed “hotspot residues”, have been shown to successfully inhibit specific protein complexes. Although this strategy has led to development of novel synthetic inhibitors of protein complexes, often direct mimicry of natural amino acid residues does not lead to potent inhibitors. Experimental screening of focused compound libraries is used to further optimize inhibitors but the number of possible designs that can be efficiently synthesized and experimentally tested in academic settings is limited. We have applied the principles of computational protein design to optimization of nonpeptidic helix mimics as ligands for protein complexes. We describe the development of computational tools to design helix mimetics from canonical and noncanonical residue libraries and their application to two therapeutically important protein–protein interactions: p53-MDM2 and p300-HIF1α. The overall study provides a streamlined approach for discovering potent peptidomimetic inhibitors of protein–protein interactions.

Predicting the Impact of Missense Mutations on Protein-Protein Binding Affinity

The crucial prerequisite for proper biological function is the protein's ability to establish highly selective interactions with macromolecular partners. A missense mutation that alters the protein binding affinity may cause significant perturbations or complete abolishment of the function, potentially leading to diseases. The availability of computational methods to evaluate the impact of mutations on protein-protein binding is critical for a wide range of biomedical applications. Here, we report an efficient computational approach for predicting the effect of single and multiple missense mutations on protein-protein binding affinity. It is based on a well-tested simulation protocol for structure minimization, modified MM-PBSA and statistical scoring energy functions with parameters optimized on experimental sets of several thousands of mutations. Our simulation protocol yields very good agreement between predicted and experimental values with Pearson correlation coefficients of 0.69 and 0.63 and root-mean-square errors of 1.20 and 1.90 kcal mol^-1 for single and multiple mutations, respectively. Compared with other available methods, our approach achieves high speed and prediction accuracy and can be applied to large datasets generated by modern genomics initiatives. In addition, we report a crucial role of water model and the polar solvation energy in estimating the changes in binding affinity. Our analysis also reveals that prediction accuracy and effect of mutations on binding strongly depends on the type of mutation and its location in a protein complex.

Optimizing scoring function of protein-nucleic acid interactions with both affinity and specificity

Protein-nucleic acid (protein-DNA and protein-RNA) recognition is fundamental to the regulation of gene expression. Determination of the structures of the protein-nucleic acid recognition and insight into their interactions at molecular level are vital to understanding the regulation function. Recently, quantitative computational approach has been becoming an alternative of experimental technique for predicting the structures and interactions of biomolecular recognition. However, the progress of protein-nucleic acid structure prediction, especially protein-RNA, is far behind that of the protein-ligand and protein-protein structure predictions due to the lack of reliable and accurate scoring function for quantifying the protein-nucleic acid interactions. In this work, we developed an accurate scoring function (named as SPA-PN, SPecificity and Affinity of the Protein-Nucleic acid interactions) for protein-nucleic acid interactions by incorporating both the specificity and affinity into the optimization strategy. Specificity and affinity are two requirements of highly efficient and specific biomolecular recognition. Previous quantitative descriptions of the biomolecular interactions considered the affinity, but often ignored the specificity owing to the challenge of specificity quantification. We applied our concept of intrinsic specificity to connect the conventional specificity, which circumvents the challenge of specificity quantification. In addition to the affinity optimization, we incorporated the quantified intrinsic specificity into the optimization strategy of SPA-PN. The testing results and comparisons with other scoring functions validated that SPA-PN performs well on both the prediction of binding affinity and identification of native conformation. In terms of its performance, SPA-PN can be widely used to predict the protein-nucleic acid structures and quantify their interactions.

Epitope-guided engineering of monobody binders for in vivo inhibition of Erk-2 signaling

Although the affinity optimization of protein binders is straightforward, engineering epitope specificity is more challenging. Targeting a specific surface patch is important because the biological relevance of protein binders depends on how they interact with the target. They are particularly useful to test hypotheses motivated by biochemical and structural studies. We used yeast display to engineer monobodies that bind a defined surface patch on the mitogen activated protein kinase (MAPK) Erk-2. The targeted area ("CD" domain) is known to control the specificity and catalytic efficiency of phosphorylation by the kinase by binding a linear peptide ("D" peptide) on substrates and regulators. An inhibitor of the interaction should thus be useful for regulating Erk-2 signaling in vivo. Although the CD domain constitutes only a small percentage of the surface area of the enzyme (~5%), sorting a yeast displayed monobody library with wild type (wt) Erk-2 and a rationally designed mutant led to isolation of high affinity clones with desired epitope specificity. The engineered binders inhibited the activity of Erk-2 in vitro and in mammalian cells. Furthermore, they specifically inhibited the activity of Erk-2 orthologs in yeast and suppressed a mutant phenotype in round worms caused by overactive MAPK signaling. The study therefore shows that positive and negative screening can be used to bias the evolution of epitope specificity and predictably design inhibitors of biologically relevant protein-protein interaction.

Multistate approaches in computational protein design

Computational protein design (CPD) is a useful tool for protein engineers. It has been successfully applied towards the creation of proteins with increased thermostability, improved binding affinity, novel enzymatic activity, and altered ligand specificity. Traditionally, CPD calculations search and rank sequences using a single fixed protein backbone template in an approach referred to as single-state design (SSD). While SSD has enjoyed considerable success, certain design objectives require the explicit consideration of multiple conformational and/or chemical states. Cases where a "multistate" approach may be advantageous over the SSD approach include designing conformational changes into proteins, using native ensembles to mimic backbone flexibility, and designing ligand or oligomeric association specificities. These design objectives can be efficiently tackled using multistate design (MSD), an emerging methodology in CPD that considers any number of protein conformational or chemical states as inputs instead of a single protein backbone template, as in SSD. In this review article, recent examples of the successful design of a desired property into proteins using MSD are described. These studies employing MSD are divided into two categories--those that utilized multiple conformational states, and those that utilized multiple chemical states. In addition, the scoring of competing states during negative design is discussed as a current challenge for MSD.

Cascaded walks in protein sequence space: use of artificial sequences in remote homology detection between natural proteins

Over the past two decades, many ingenious efforts have been made in protein remote homology detection. Because homologous proteins often diversify extensively in sequence, it is challenging to demonstrate such relatedness through entirely sequence-driven searches. Here, we describe a computational method for the generation of 'protein-like' sequences that serves to bridge gaps in protein sequence space. Sequence profile information, as embodied in a position-specific scoring matrix of multiply aligned sequences of bona fide family members, serves as the starting point in this algorithm. The observed amino acid propensity and the selection of a random number dictate the selection of a residue for each position in the sequence. In a systematic manner, and by applying a 'roulette-wheel' selection approach at each position, we generate parent family-like sequences and thus facilitate an enlargement of sequence space around the family. When generated for a large number of families, we demonstrate that they expand the utility of natural intermediately related sequences in linking distant proteins. In 91% of the assessed examples, inclusion of designed sequences improved fold coverage by 5-10% over searches made in their absence. Furthermore, with several examples from proteins adopting folds such as TIM, globin, lipocalin and others, we demonstrate that the success of including designed sequences in a database positively sensitized methods such as PSI-BLAST and Cascade PSI-BLAST and is a promising opportunity for enormously improved remote homology recognition using sequence information alone.

Structural, kinetic, and thermodynamic studies of specificity designed HIV-1 protease

HIV-1 protease recognizes and cleaves more than 12 different substrates leading to viral maturation. While these substrates share no conserved motif, they are specifically selected for and cleaved by protease during viral life cycle. Drug resistant mutations evolve within the protease that compromise inhibitor binding but allow the continued recognition of all these substrates. While the substrate envelope defines a general shape for substrate recognition, successfully predicting the determinants of substrate binding specificity would provide additional insights into the mechanism of altered molecular recognition in resistant proteases. We designed a variant of HIV protease with altered specificity using positive computational design methods and validated the design using X-ray crystallography and enzyme biochemistry. The engineered variant, Pr3 (A28S/D30F/G48R), was designed to preferentially bind to one out of three of HIV protease's natural substrates; RT-RH over p2-NC and CA-p2. In kinetic assays, RT-RH binding specificity for Pr3 increased threefold compared to the wild-type (WT), which was further confirmed by isothermal titration calorimetry. Crystal structures of WT protease and the designed variant in complex with RT-RH, CA-p2, and p2-NC were determined. Structural analysis of the designed complexes revealed that one of the engineered substitutions (G48R) potentially stabilized heterogeneous flap conformations, thereby facilitating alternate modes of substrate binding. Our results demonstrate that while substrate specificity could be engineered in HIV protease, the structural pliability of protease restricted the propagation of interactions as predicted. These results offer new insights into the plasticity and structural determinants of substrate binding specificity of the HIV-1 protease.

Rationalization and design of the complementarity determining region sequences in an antibody-antigen recognition interface

Protein-protein interactions are critical determinants in biological systems. Engineered proteins binding to specific areas on protein surfaces could lead to therapeutics or diagnostics for treating diseases in humans. But designing epitope-specific protein-protein interactions with computational atomistic interaction free energy remains a difficult challenge. Here we show that, with the antibody-VEGF (vascular endothelial growth factor) interaction as a model system, the experimentally observed amino acid preferences in the antibody-antigen interface can be rationalized with 3-dimensional distributions of interacting atoms derived from the database of protein structures. Machine learning models established on the rationalization can be generalized to design amino acid preferences in antibody-antigen interfaces, for which the experimental validations are tractable with current high throughput synthetic antibody display technologies. Leave-one-out cross validation on the benchmark system yielded the accuracy, precision, recall (sensitivity) and specificity of the overall binary predictions to be 0.69, 0.45, 0.63, and 0.71 respectively, and the overall Matthews correlation coefficient of the 20 amino acid types in the 24 interface CDR positions was 0.312. The structure-based computational antibody design methodology was further tested with other antibodies binding to VEGF. The results indicate that the methodology could provide alternatives to the current antibody technologies based on animal immune systems in engineering therapeutic and diagnostic antibodies against predetermined antigen epitopes.

Restricted sidechain plasticity in the structures of native proteins and complexes

Protein-design methodology can now generate models of protein structures and interfaces with computed energies in the range of those of naturally occurring structures. Comparison of the properties of native structures and complexes to isoenergetic design models can provide insight into the properties of the former that reflect selection pressure for factors beyond the energy of the native state. We report here that sidechains in native structures and interfaces are significantly more constrained than designed interfaces and structures with equal computed binding energy or stability, which may reflect selection against potentially deleterious non-native interactions.

Anchored design of protein-protein interfaces

Background

Few existing protein-protein interface design methods allow for extensive backbone rearrangements during the design process. There is also a dichotomy between redesign methods, which take advantage of the native interface, and de novo methods, which produce novel binders.

Methodology

Here, we propose a new method for designing novel protein reagents that combines advantages of redesign and de novo methods and allows for extensive backbone motion. This method requires a bound structure of a target and one of its natural binding partners. A key interaction in this interface, the anchor, is computationally grafted out of the partner and into a surface loop on the design scaffold. The design scaffold's surface is then redesigned with backbone flexibility to create a new binding partner for the target. Careful choice of a scaffold will bring experimentally desirable characteristics into the new complex. The use of an anchor both expedites the design process and ensures that binding proceeds against a known location on the target. The use of surface loops on the scaffold allows for flexible-backbone redesign to properly search conformational space.

Conclusions and Significance

This protocol was implemented within the Rosetta3 software suite. To demonstrate and evaluate this protocol, we have developed a benchmarking set of structures from the PDB with loop-mediated interfaces. This protocol can recover the correct loop-mediated interface in 15 out of 16 tested structures, using only a single residue as an anchor.

Computational protein design with a generalized Born solvent model: application to Asparaginyl-tRNA synthetase

Computational Protein Design (CPD) is a promising method for high throughput protein and ligand mutagenesis. Recently, we developed a CPD method that used a polar-hydrogen energy function for protein interactions and a Coulomb/Accessible Surface Area (CASA) model for solvent effects. We applied this method to engineer aspartyl-adenylate (AspAMP) specificity into Asparaginyl-tRNA synthetase (AsnRS), whose substrate is asparaginyl-adenylate (AsnAMP). Here, we implement a more accurate function, with an all-atom energy for protein interactions and a residue-pairwise generalized Born model for solvent effects. As a first test, we compute aminoacid affinities for several point mutants of Aspartyl-tRNA synthetase (AspRS) and Tyrosyl-tRNA synthetase and stability changes for three helical peptides and compare with experiment. As a second test, we readdress the problem of AsnRS aminoacid engineering. We compare three design criteria, which optimize the folding free-energy, the absolute AspAMP affinity, and the relative (AspAMP-AsnAMP) affinity. The sequences and conformations are improved with respect to our previous, polar-hydrogen/CASA study: For several designed complexes, the AspAMP carboxylate forms three interactions with a conserved arginine and a designed lysine, as in the active site of the AspRS:AspAMP complex. The conformations and interactions are well maintained in molecular dynamics simulations and the sequences have an inverted specificity, favoring AspAMP over AsnAMP. The method is not fully successful, since experimental measurements with the seven most promising sequences show that they do not catalyze at a detectable level the adenylation of Asp (or Asn) with ATP. This may be due to weak AspAMP binding and/or disruption of transition-state stabilization.

Optimizing energy functions for protein-protein interface design

Protein design methods have been originally developed for the design of monomeric proteins. When applied to the more challenging task of protein–protein complex design, these methods yield suboptimal results. In particular, they often fail to recapitulate favorable hydrogen bonds and electrostatic interactions across the interface. In this work, we aim to improve the energy function of the protein design program ORBIT to better account for binding interactions between proteins. By using the advanced machine learning framework of conditional random fields, we optimize the relative importance of all the terms in the energy function, attempting to reproduce the native side-chain conformations in protein–protein interfaces. We evaluate the performance of several optimized energy functions, each describes the van der Waals interactions using a different potential. In comparison with the original energy function, our best energy function (a) incorporates a much “softer” repulsive van der Waals potential, suitable for the discrete rotameric representation of amino acid side chains; (b) does not penalize burial of polar atoms, reflecting the frequent occurrence of polar buried residues in protein–protein interfaces; and (c) significantly up-weights the electrostatic term, attesting to the high importance of these interactions for protein–protein complex formation. Using this energy function considerably improves side chain placement accuracy for interface residues in a large test set of protein–protein complexes. Moreover, the optimized energy function recovers the native sequences of protein–protein interface at a higher rate than the default function and performs substantially better in predicting changes in free energy of binding due to mutations.

Triathlon for energy functions: who is the winner for design of protein-protein interactions?

Computational prediction of stabilizing mutations into monomeric proteins has become an almost ordinary task. Yet, computational stabilization of protein–protein complexes remains a challenge. Design of protein–protein interactions (PPIs) is impeded by the absence of an energy function that could reliably reproduce all favorable interactions between the binding partners. In this work, we present three energy functions: one function that was trained on monomeric proteins, while the other two were optimized by different techniques to predict side-chain conformations in a dataset of PPIs. The performances of these energy functions are evaluated in three different tasks related to design of PPIs: predicting side-chain conformations in PPIs, recovering native binding-interface sequences, and predicting changes in free energy of binding due to mutations. Our findings show that both functions optimized on side-chain repacking in PPIs are more suitable for PPI design compared to the function trained on monomeric proteins. Yet, no function performs best at all three tasks. Comparison of the three energy functions and their performances revealed that (1) burial of polar atoms should not be penalized significantly in PPI design as in single-protein design and (2) contribution of electrostatic interactions should be increased several-fold when switching from single-protein to PPI design. In addition, the use of a softer van der Waals potential is beneficial in cases when backbone flexibility is important. All things considered, we define an energy function that captures most of the nuances of the binding energetics and hence, should be used in future for design of PPIs.

Structure-based prediction of the peptide sequence space recognized by natural and synthetic PDZ domains

Protein-protein recognition, frequently mediated by members of large families of interaction domains, is one of the cornerstones of biological function. Here, we present a computational, structure-based method to predict the sequence space of peptides recognized by PDZ domains, one of the largest families of recognition proteins. As a test set, we use a considerable amount of recent phage display data that describe the peptide recognition preferences for 169 naturally occurring and engineered PDZ domains. For both wild-type PDZ domains and single point mutants, we find that 70-80% of the most frequently observed amino acids by phage display are predicted within the top five ranked amino acids. Phage display frequently identified recognition preferences for amino acids different from those present in the original crystal structure. Notably, in about half of these cases, our algorithm correctly captures these preferences, indicating that it can predict mutations that increase binding affinity relative to the starting structure. We also find that we can computationally recapitulate specificity changes upon mutation, a key test for successful forward design of protein-protein interface specificity. Across all evaluated data sets, we find that incorporation backbone sampling improves accuracy substantially, irrespective of using a crystal or NMR structure as the starting conformation. Finally, we report successful prediction of several amino acid specificity changes from blind tests in the DREAM4 peptide recognition domain specificity prediction challenge. Because the foundational methods developed here are structure based, these results suggest that the approach can be more generally applied to specificity prediction and redesign of other protein-protein interfaces that have structural information but lack phage display data.

An efficient algorithm for multistate protein design based on FASTER

Most of the methods that have been developed for computational protein design involve the selection of side-chain conformations in the context of a single, fixed main-chain structure. In contrast, multistate design (MSD) methods allow sequence selection to be driven by the energetic contributions of multiple structural or chemical states simultaneously. This methodology is expected to be useful when the design target is an ensemble of related states rather than a single structure, or when a protein sequence must assume several distinct conformations to function. MSD can also be used with explicit negative design to suggest sequences with altered structural, binding, or catalytic specificity. We report implementation details of an efficient multistate design optimization algorithm based on FASTER (MSD-FASTER). We subjected the algorithm to a battery of computational tests and found it to be generally applicable to various multistate design problems; designs with a large number of states and many designed positions are completely feasible. A direct comparison of MSD-FASTER and multistate design Monte Carlo indicated that MSD-FASTER discovers low-energy sequences much more consistently. MSD-FASTER likely performs better because amino acid substitutions are chosen on an energetic basis rather than randomly, and because multiple substitutions are applied together. Through its greater efficiency, MSD-FASTER should allow protein designers to test experimentally better-scoring sequences, and thus accelerate progress in the development of improved scoring functions and models for computational protein design.

De novo backbone scaffolds for protein design

In recent years, there have been significant advances in the field of computational protein design including the successful computational design of enzymes based on backbone scaffolds from experimentally solved structures. It is likely that large-scale sampling of protein backbone conformations will become necessary as further progress is made on more complicated systems. Removing the constraint of having to use scaffolds based on known protein backbones is a potential method of solving the problem. With this application in mind, we describe a method to systematically construct a large number of de novo backbone structures from idealized topological forms in a top–down hierarchical approach. The structural properties of these novel backbone scaffolds were analyzed and compared with a set of high-resolution experimental structures from the protein data bank (PDB). It was found that the Ramachandran plot distribution and relative γ- and β-turn frequencies were similar to those found in the PDB. The de novo scaffolds were sequence designed with RosettaDesign, and the energy distributions and amino acid compositions were comparable with the results for redesigned experimentally solved backbones. Proteins 2010. © 2009 Wiley-Liss, Inc.

What makes Ras an efficient molecular switch: a computational, biophysical, and structural study of Ras-GDP interactions with mutants of Raf

Ras is a small GTP-binding protein that is an essential molecular switch for a wide variety of signaling pathways including the control of cell proliferation, cell cycle progression and apoptosis. In the GTP-bound state, Ras can interact with its effectors, triggering various signaling cascades in the cell. In the GDP-bound state, Ras looses its ability to bind to known effectors. The interaction of the GTP-bound Ras (Ras(GTP)) with its effectors has been studied intensively. However, very little is known about the much weaker interaction between the GDP-bound Ras (Ras(GDP)) and Ras effectors. We investigated the factors underlying the nucleotide-dependent differences in Ras interactions with one of its effectors, Raf kinase. Using computational protein design, we generated mutants of the Ras-binding domain of Raf kinase (Raf) that stabilize the complex with Ras(GDP). Most of our designed mutations narrow the gap between the affinity of Raf for Ras(GTP) and Ras(GDP), producing the desired shift in binding specificity towards Ras(GDP). A combination of our best designed mutation, N71R, with another mutation, A85K, yielded a Raf mutant with a 100-fold improvement in affinity towards Ras(GDP). The Raf A85K and Raf N71R/A85K mutants were used to obtain the first high-resolution structures of Ras(GDP) bound to its effector. Surprisingly, these structures reveal that the loop on Ras previously termed the switch I region in the Ras(GDP).Raf mutant complex is found in a conformation similar to that of Ras(GTP) and not Ras(GDP). Moreover, the structures indicate an increased mobility of the switch I region. This greater flexibility compared to the same loop in Ras(GTP) is likely to explain the natural low affinity of Raf and other Ras effectors to Ras(GDP). Our findings demonstrate that an accurate balance between a rigid, high-affinity conformation and conformational flexibility is required to create an efficient and stringent molecular switch.

Design of multispecific protein sequences using probabilistic graphical modeling

In nature, proteins partake in numerous protein- protein interactions that mediate their functions. Moreover, proteins have been shown to be physically stable in multiple structures, induced by cellular conditions, small ligands, or covalent modifications. Understanding how protein sequences achieve this structural promiscuity at the atomic level is a fundamental step in the drug design pipeline and a critical question in protein physics. One way to investigate this subject is to computationally predict protein sequences that are compatible with multiple states, i.e., multiple target structures or binding to distinct partners. The goal of engineering such proteins has been termed multispecific protein design. We develop a novel computational framework to efficiently and accurately perform multispecific protein design. This framework utilizes recent advances in probabilistic graphical modeling to predict sequences with low energies in multiple target states. Furthermore, it is also geared to specifically yield positional amino acid probability profiles compatible with these target states. Such profiles can be used as input to randomly bias high-throughput experimental sequence screening techniques, such as phage display, thus providing an alternative avenue for elucidating the multispecificity of natural proteins and the synthesis of novel proteins with specific functionalities. We prove the utility of such multispecific design techniques in better recovering amino acid sequence diversities similar to those resulting from millions of years of evolution. We then compare the approaches of prediction of low energy ensembles and of amino acid profiles and demonstrate their complementarity in providing more robust predictions for protein design.