SORTCERY-A High-Throughput Method to Affinity Rank Peptide Ligands

Machine learning to predict continuous protein properties from binary cell sorting data and map unseen sequence space

Proteins are a diverse class of biomolecules responsible for wide-ranging cellular functions, from catalyzing reactions to recognizing pathogens. The ability to evolve proteins rapidly and inexpensively toward improved properties is a common objective for protein engineers. Powerful high-throughput methods like fluorescent activated cell sorting and next-generation sequencing have dramatically improved directed evolution experiments. However, it is unclear how to best leverage these data to characterize protein fitness landscapes more completely and identify lead candidates. In this work, we develop a simple yet powerful framework to improve protein optimization by predicting continuous protein properties from simple directed evolution experiments using interpretable, linear machine learning models. Importantly, we find that these models, which use data from simple but imprecise experimental estimates of protein fitness, have predictive capabilities that approach more precise but expensive data. Evaluated across five diverse protein engineering tasks, continuous properties are consistently predicted from readily available deep sequencing data, demonstrating that protein fitness space can be reasonably well modeled by linear relationships among sequence mutations. To prospectively test the utility of this approach, we generated a library of stapled peptides and applied the framework to predict affinity and specificity from simple cell sorting data. We then coupled integer linear programming, a method to optimize protein fitness from linear weights, with mutation scores from machine learning to identify variants in unseen sequence space that have improved and co-optimal properties. This approach represents a versatile tool for improved analysis and identification of protein variants across many domains of protein engineering.

Engineering minimal tissue inhibitors of metalloproteinase targeting MMPs via gene shuffling and yeast surface display

Overexpression of specific matrix metalloproteinases (MMPs) has a key role in development of several diseases, such as cancer, neurological disorders, and cardiovascular diseases due to their critical role in degradation and remodeling of the extracellular matrix (ECM). Tissue inhibitors of metalloproteinases (TIMPs), a family of four in humans, are endogenous inhibitors of MMPs. TIMPs have a high level of sequence and structure homology, with a broad range of binding and inhibition to the family of MMPs. It is important to identify the key motifs of TIMPs responsible for inhibition of MMPs to develop efficient therapeutics targeting specific MMPs. We used DNA shuffling between the human TIMP family to generate a minimal TIMP hybrid library in yeast to identify the dominant minimal MMP inhibitory regions. The minimal TIMP variants screened toward MMP-3 and MMP-9 using fluorescent-activated cell sorting (FACS). Interestingly, several minimal TIMP variants selected after screening toward MMP-3cd or MMP-9cd, with lengths as short as 20 amino acids, maintained or improved binding to MMP-3 and MMP-9. The TIMP-MMP binding dissociation constant (KD ), in the nM range, and MMP inhibition constants (Ki ), in the pM range, of these minimal TIMP variants were similar to the N-terminal domain of TIMP-1 on the yeast surface and in solution indicating the potency of these minimal variants as MMP inhibitors. We further used molecular modeling simulation, and molecular docking of the minimal TIMP variants in complex with MMP-3cd to understand the binding and inhibition mechanism of these variants.

Position-Specific Enrichment Ratio Matrix scores predict antibody variant properties from deep sequencing data

MOTIVATION

Deep sequencing of antibody and related protein libraries after phage or yeast-surface display sorting is widely used to identify variants with increased affinity, specificity, and/or improvements in key biophysical properties. Conventional approaches for identifying optimal variants typically use the frequencies of observation in enriched libraries or the corresponding enrichment ratios. However, these approaches disregard the vast majority of deep sequencing data and often fail to identify the best variants in the libraries.

RESULTS

Here, we present a method, Position-Specific Enrichment Ratio Matrix (PSERM) scoring, that uses entire deep sequencing datasets from pre- and post-selections to score each observed protein variant. The PSERM scores are the sum of the site-specific enrichment ratios observed at each mutated position. We find that PSERM scores are much more reproducible and correlate more strongly with experimentally measured properties than frequencies or enrichment ratios, including for multiple antibody properties (affinity and non-specific binding) for a clinical-stage antibody (emibetuzumab). We expect that this method will be broadly applicable to diverse protein engineering campaigns.

AVAILABILITY AND IMPLEMENTATION

All deep sequencing datasets and code to perform the analyses presented within are available via https://github.com/Tessier-Lab-UMich/PSERM_paper.

Protein engineering via sequence-performance mapping

Discovery and evolution of new and improved proteins has empowered molecular therapeutics, diagnostics, and industrial biotechnology. Discovery and evolution both require efficient screens and effective libraries, although they differ in their challenges because of the absence or presence, respectively, of an initial protein variant with the desired function. A host of high-throughput technologies-experimental and computational-enable efficient screens to identify performant protein variants. In partnership, an informed search of sequence space is needed to overcome the immensity, sparsity, and complexity of the sequence-performance landscape. Early in the historical trajectory of protein engineering, these elements aligned with distinct approaches to identify the most performant sequence: selection from large, randomized combinatorial libraries versus rational computational design. Substantial advances have now emerged from the synergy of these perspectives. Rational design of combinatorial libraries aids the experimental search of sequence space, and high-throughput, high-integrity experimental data inform computational design. At the core of the collaborative interface, efficient protein characterization (rather than mere selection of optimal variants) maps sequence-performance landscapes. Such quantitative maps elucidate the complex relationships between protein sequence and performance-e.g., binding, catalytic efficiency, biological activity, and developability-thereby advancing fundamental protein science and facilitating protein discovery and evolution.

Position-Specific Enrichment Ratio Matrix scores predict antibody variant properties from deep sequencing data

Motivation

Deep sequencing of antibody and related protein libraries after phage or yeast-surface display sorting is widely used to identify variants with increased affinity, specificity and/or improvements in key biophysical properties. Conventional approaches for identifying optimal variants typically use the frequencies of observation in enriched libraries or the corresponding enrichment ratios. However, these approaches disregard the vast majority of deep sequencing data and often fail to identify the best variants in the libraries.

Results

Here, we present a method, Position-Specific Enrichment Ratio Matrix (PSERM) scoring, that uses entire deep sequencing datasets from pre- and post-selections to score each observed protein variant. The PSERM scores are the sum of the site-specific enrichment ratios observed at each mutated position. We find that PSERM scores are much more reproducible and correlate more strongly with experimentally measured properties than frequencies or enrichment ratios, including for multiple antibody properties (affinity and non-specific binding) for a clinical-stage antibody (emibetuzumab). We expect that this method will be broadly applicable to diverse protein engineering campaigns.

Availability

All deep sequencing datasets and code to do the analyses presented within are available via GitHub.

Contact

Peter Tessier, ptessier@umich.edu.

Supplementary information

Supplementary data are available at Bioinformatics online.

The next wave of interactomics: Mapping the SLiM-based interactions of the intrinsically disordered proteome

Short linear motifs (SLiMs) are a unique and ubiquitous class of protein interaction modules that perform key regulatory functions and drive dynamic complex formation. For decades, interactions mediated by SLiMs have accumulated through detailed low-throughput experiments. Recent methodological advances have opened this previously underexplored area of the human interactome to high-throughput protein-protein interaction discovery. In this article, we discuss that SLiM-based interactions represent a significant blind spot in the current interactomics data, introduce the key methods that are illuminating the elusive SLiM-mediated interactome of the human cell on a large scale, and discuss the implications for the field.

Mutational scan inferred binding energetics and structure in intrinsically disordered protein CcdA

Unlike globular proteins, mutational effects on the function of Intrinsically Disordered Proteins (IDPs) are not well-studied. Deep Mutational Scanning of a yeast surface displayed mutant library yields insights into sequence-function relationships in the CcdA IDP. The approach enables facile prediction of interface residues and local structural signatures of the bound conformation. In contrast to previous titration-based approaches which use a number of ligand concentrations, we show that use of a single rationally chosen ligand concentration can provide quantitative estimates of relative binding constants for large numbers of protein variants. This is because the extended interface of IDP ensures that energetic effects of point mutations are spread over a much smaller range than for globular proteins. Our data also provides insights into the much-debated role of helicity and disorder in partner binding of IDPs. Based on this exhaustive mutational sensitivity dataset, a rudimentary model was developed in an attempt to predict mutational effects on binding affinity of IDPs that form alpha-helical structures upon binding.

Rational design of Harakiri (HRK)-derived constrained peptides as BCL-xL inhibitors

Using the HRK BH3 domain, sequence hybridization and in silico methods we show dibromomaleimide staple scanning can be used to inform the design of BCL-x_L selective peptidomimetic ligands. These HRK-inspired reagents may serve as starting points for the discovery of therapeutics to target BCL-x_L-overexpressed cancers.

Titrating Avidity of Yeast-Displayed Proteins Using a Transcriptional Regulator

Yeast surface display is a valuable tool for protein engineering and directed evolution; however, significant variability in the copy number (i.e., avidity) of displayed variants on the yeast cell wall complicates screening and selection campaigns. Here, we report an engineered titratable display platform that modulates the avidity of Aga2-fusion proteins on the yeast cell wall dependent on the concentration of the anhydrotetracycline (aTc) inducer. Our design is based on a genomic Aga1 gene copy and an episomal Aga2-fusion construct both under the control of an aTc-dependent transcriptional regulator that enables stoichiometric and titratable expression, secretion, and display of Aga2-fusion proteins. We demonstrate tunable display levels over 2-3 orders of magnitude for various model proteins, including glucose oxidase enzyme variants, mechanostable dockerin-binding domains, and anti-PDL1 affibody domains. By regulating the copy number of displayed proteins, we demonstrate the effects of titratable avidity levels on several specific phenotypic activities, including enzyme activity and cell adhesion to surfaces under shear flow. Finally, we show that titrating down the display level allows yeast-based binding affinity measurements to be performed in a regime that avoids ligand depletion effects while maintaining small sample volumes, avoiding a well-known artifact in yeast-based binding assays. The ability to titrate the multivalency of proteins on the yeast cell wall through simple inducer control will benefit protein engineering and directed evolution methodology relying on yeast display for broad classes of therapeutic and diagnostic proteins of interest.

Mutational fitness landscape and drug resistance

Robust technology has been developed to systematically quantify fitness landscapes that provide valuable opportunities to improve our understanding of drug resistance and define new avenues to develop drugs with reduced resistance susceptibility. We outline the critical importance of drug resistance studies and the potential for fitness landscape approaches to contribute to this effort. We describe the major technical advancements in mutational scanning, which is the primary approach used to quantify protein fitness landscapes. There are many complex steps to consider in planning and executing mutational scanning projects including developing a selection scheme, generating mutant libraries, tracking the frequency of variants using next-generation sequencing, and processing and interpreting the data. Key experimental parameters impacting each of these steps are discussed to aid in planning fitness landscape studies. There is a strong need for improved understanding of drug resistance, and fitness landscapes provide a promising new approach.

Ter-Seq: A high-throughput method to stabilize transient ternary complexes and measure associated kinetics

Regulation of biological processes by proteins often involves the formation of transient, multimeric complexes whose characterization is mechanistically important but challenging. The bacterial toxin CcdB binds and poisons DNA Gyrase. The corresponding antitoxin CcdA extracts CcdB from its complex with Gyrase through the formation of a transient ternary complex, thus rejuvenating Gyrase. We describe a high throughput methodology called Ter-Seq to stabilize probable ternary complexes and measure associated kinetics using the CcdA-CcdB-GyrA14 ternary complex as a model system. The method involves screening a yeast surface display (YSD) saturation mutagenesis library of one partner (CcdB) for mutants that show enhanced ternary complex formation. We also isolated CcdB mutants that were either resistant or sensitive to rejuvenation, and used surface plasmon resonance (SPR) with purified proteins to validate the kinetics measured using the surface display. Positions, where CcdB mutations lead to slower rejuvenation rates, are largely involved in CcdA-binding, though there were several notable exceptions suggesting allostery. Mutations at these positions reduce the affinity towards CcdA, thereby slowing down the rejuvenation process. Mutations at GyrA14-interacting positions significantly enhanced rejuvenation rates, either due to reduced affinity or complete loss of CcdB binding to GyrA14. We examined the effect of different parameters (CcdA affinity, GyrA14 affinity, surface accessibilities, evolutionary conservation) on the rate of rejuvenation. Finally, we further validated the Ter-Seq results by monitoring the kinetics of ternary complex formation for individual CcdB mutants in solution by fluorescence resonance energy transfer (FRET) studies.

Genetically Encoded Whole Cell Biosensor for Drug Discovery of HIF-1 Interaction Inhibitors

The heterodimeric transcription factor, hypoxia inducible factor-1 (HIF-1), is an important anticancer target as it supports the adaptation and response of tumors to hypoxia. Here, we optimized the repressed transactivator yeast two-hybrid system to further develop it as part of a versatile yeast-based drug discovery platform and validated it using HIF-1. We demonstrate both fluorescence-based and auxotrophy-based selections that could detect HIF-1α/HIF-1β dimerization inhibition. The engineered genetic selection is tunable and able to differentiate between strong and weak interactions, shows a large dynamic range, and is stable over different growth phases. Furthermore, we engineered mechanisms to control for cellular activity and off-target drug effects. We thoroughly characterized all parts of the biosensor system and argue this tool will be generally applicable to a wide array of protein-protein interaction targets. We anticipate this biosensor will be useful as part of a drug discovery platform, particularly when screening DNA-encoded new modality drugs.

Randomized gates eliminate bias in sort‐seq assays

Sort‐seq assays are a staple of the biological engineering toolkit, allowing researchers to profile many groups of cells based on any characteristic that can be tied to fluorescence. However, current approaches, which segregate cells into bins deterministically based on their measured fluorescence, introduce systematic bias. We describe a surprising result: one can obtain unbiased estimates by incorporating randomness into sorting. We validate this approach in simulation and experimentally, and describe extensions for both estimating group level variances and for using multi‐bin sorters.

Co-optimization of therapeutic antibody affinity and specificity using machine learning models that generalize to novel mutational space

Therapeutic antibody development requires selection and engineering of molecules with high affinity and other drug-like biophysical properties. Co-optimization of multiple antibody properties remains a difficult and time-consuming process that impedes drug development. Here we evaluate the use of machine learning to simplify antibody co-optimization for a clinical-stage antibody (emibetuzumab) that displays high levels of both on-target (antigen) and off-target (non-specific) binding. We mutate sites in the antibody complementarity-determining regions, sort the antibody libraries for high and low levels of affinity and non-specific binding, and deep sequence the enriched libraries. Interestingly, machine learning models trained on datasets with binary labels enable predictions of continuous metrics that are strongly correlated with antibody affinity and non-specific binding. These models illustrate strong tradeoffs between these two properties, as increases in affinity along the co-optimal (Pareto) frontier require progressive reductions in specificity. Notably, models trained with deep learning features enable prediction of novel antibody mutations that co-optimize affinity and specificity beyond what is possible for the original antibody library. These findings demonstrate the power of machine learning models to greatly expand the exploration of novel antibody sequence space and accelerate the development of highly potent, drug-like antibodies.

Mutational fitness landscapes reveal genetic and structural improvement pathways for a vaccine-elicited HIV-1 broadly neutralizing antibody

Vaccine-based elicitation of broadly neutralizing antibodies holds great promise for preventing HIV-1 transmission. However, the key biophysical markers of improved antibody recognition remain uncertain in the diverse landscape of potential antibody mutation pathways, and a more complete understanding of anti-HIV-1 fusion peptide (FP) antibody development will accelerate rational vaccine designs. Here we survey the mutational landscape of the vaccine-elicited anti-FP antibody, vFP16.02, to determine the genetic, structural, and functional features associated with antibody improvement or fitness. Using site-saturation mutagenesis and yeast display functional screening, we found that 1% of possible single mutations improved HIV-1 envelope trimer (Env) affinity, but generally comprised rare somatic hypermutations that may not arise frequently in vivo. We observed that many single mutations in the vFP16.02 Fab could enhance affinity >1,000-fold against soluble FP, although affinity improvements against the HIV-1 trimer were more measured and rare. The most potent variants enhanced affinity to both soluble FP and Env, had mutations concentrated in antibody framework regions, and achieved up to 37% neutralization breadth compared to 28% neutralization of the template antibody. Altered heavy- and light-chain interface angles and conformational dynamics, as well as reduced Fab thermal stability, were associated with improved HIV-1 neutralization breadth and potency. We also observed parallel sets of mutations that enhanced viral neutralization through similar structural mechanisms. These data provide a quantitative understanding of the mutational landscape for vaccine-elicited FP-directed broadly neutralizing antibody and demonstrate that numerous antigen-distal framework mutations can improve antibody function by enhancing affinity simultaneously toward HIV-1 Env and FP.

Functional Profiling of Antibody Immune Repertoires in Convalescent Zika Virus Disease Patients

The re-emergence of Zika virus (ZIKV) caused widespread infections that were linked to Guillain-Barré syndrome in adults and congenital malformation in fetuses, and epidemiological data suggest that ZIKV infection can induce protective antibody responses. A more detailed understanding of anti-ZIKV antibody responses may lead to enhanced antibody discovery and improved vaccine designs against ZIKV and related flaviviruses. Here, we applied recently-invented library-scale antibody screening technologies to determine comprehensive functional molecular and genetic profiles of naturally elicited human anti-ZIKV antibodies in three convalescent individuals. We leveraged natively paired antibody yeast display and NGS to predict antibody cross-reactivities and coarse-grain antibody affinities, to perform in-depth immune profiling of IgM, IgG, and IgA antibody repertoires in peripheral blood, and to reveal virus maturation state-dependent antibody interactions. Repertoire-scale comparison of ZIKV VLP-specific and non-specific antibodies in the same individuals also showed that mean antibody somatic hypermutation levels were substantially influenced by donor-intrinsic characteristics. These data provide insights into antiviral antibody responses to ZIKV disease and outline systems-level strategies to track human antibody immune responses to emergent viral infections.

The covalent SNAP tag for protein display quantification and low-pH protein engineering

Yeast display has become an important tool for modern biotechnology with many advantages for eukaryotic protein engineering. Antibody-based peptide interactions are often used to quantify yeast surface expression (e.g., by fusing a target protein to a FLAG, Myc, polyhistidine, or other peptide tag). However, antibody-antigen interactions require high stability for accurate quantification, and conventional tag systems based on such interactions may not be compatible with a low pH environment. In this study, a SNAP tag was introduced to a yeast display platform to circumvent disadvantages of conventional antibody display tags at low pH. SNAP forms a covalent bond with its small-molecule substrate, enabling precise and pH-independent protein display tagging. We compared the SNAP tag to conventional antibody-based peptide fusion and to direct fluorescent domain fusion using antibody fragment crystallizable (Fc) gene libraries as a case study in low pH protein engineering. Our results demonstrated that covalent SNAP tags can effectively quantify protein-surface expression at low pH, enabling the enrichment of Fc variants with increased affinity at pH 6.0 to the neonatal Fc receptor (FcRn). Incorporation of a covalent SNAP tag thus overcomes disadvantages of conventional antibody-based expression tags and enables protein-engineering applications outside of physiological pH.

Generating quantitative binding landscapes through fractional binding selections combined with deep sequencing and data normalization

Quantifying the effects of various mutations on binding free energy is crucial for understanding the evolution of protein-protein interactions and would greatly facilitate protein engineering studies. Yet, measuring changes in binding free energy (ΔΔG_bind) remains a tedious task that requires expression of each mutant, its purification, and affinity measurements. We developed an attractive approach that allows us to quantify ΔΔG_bind for thousands of protein mutants in one experiment. Our protocol combines protein randomization, Yeast Surface Display technology, deep sequencing, and a few experimental ΔΔG_bind data points on purified proteins to generate ΔΔG_bind values for the remaining numerous mutants of the same protein complex. Using this methodology, we comprehensively map the single-mutant binding landscape of one of the highest-affinity interaction between BPTI and Bovine Trypsin (BT). We show that ΔΔG_bind for this interaction could be quantified with high accuracy over the range of 12 kcal mol⁻¹ displayed by various BPTI single mutants.

Quantifying the effect of mutations on binding free energy is important to understand protein-protein interaction (PPI). Here the authors develop a method based on yeast display and next-generation sequencing to generate quantitative binding landscapes for any PPI regardless of their Kd value.

Parallelized identification of on- and off-target protein interactions

Genetic selection combined with next-generation sequencing enables the simultaneous interrogation of the functionality and stability of large numbers of naturally occurring, engineered, or computationally designed protein variants in parallel. We display hundreds of engineered proteins on the surface of yeast cells, select for binding to a set of target molecules by flow cytometry, and sequence the starting pool as well as selected pools to obtain enrichment values for each displayed protein with each target. We show that this high-throughput workflow of multiplex genetic selections followed by large-scale sequencing and comparative analysis allows not only the determination of relative affinities, but also the monitoring of specificity profiles for hundreds to thousands of protein-protein and protein-small molecule interactions in parallel. The approach not only identifies new interactions of designed proteins, but also detects unintended and undesirable off-target interactions. This provides a general framework for screening of engineered protein binders, which often have no negative selection or design step as part of their development pipelines. Hence, this method will be generally useful in the development of protein-based therapeutics.

MaveDB: an open-source platform to distribute and interpret data from multiplexed assays of variant effect

Multiplex assays of variant effect (MAVEs), such as deep mutational scans and massively parallel reporter assays, test thousands of sequence variants in a single experiment. Despite the importance of MAVE data for basic and clinical research, there is no standard resource for their discovery and distribution. Here, we present MaveDB ( https://www.mavedb.org ), a public repository for large-scale measurements of sequence variant impact, designed for interoperability with applications to interpret these datasets. We also describe the first such application, MaveVis, which retrieves, visualizes, and contextualizes variant effect maps. Together, the database and applications will empower the community to mine these powerful datasets.

A Computational Pipeline for the Extraction of Actionable Biological Information From NGS-Phage Display Experiments

Phage Display is a powerful method for the identification of peptide binding to targets of variable complexities and tissues, from unique molecules to the internal surfaces of vessels of living organisms. Particularly for in vivo screenings, the resulting repertoires can be very complex and difficult to study with traditional approaches. Next Generation Sequencing (NGS) opened the possibility to acquire high resolution overviews of such repertoires and thus facilitates the identification of binders of interest. Additionally, the ever-increasing amount of available genome/proteome information became satisfactory regarding the identification of putative mimicked proteins, due to the large scale on which partial sequence homology is assessed. However, the subsequent production of massive data stresses the need for high-performance computational approaches in order to perform standardized and insightful molecular network analysis. Systems-level analysis is essential for efficient resolution of the underlying molecular complexity and the extraction of actionable interpretation, in terms of systemic biological processes and pathways that are systematically perturbed. In this work we introduce PepSimili, an integrated workflow tool, which performs mapping of massive peptide repertoires on whole proteomes and delivers a streamlined, systems-level biological interpretation. The tool employs modules for modeling and filtering of background noise due to random mappings and amplifies the biologically meaningful signal through coupling with BioInfoMiner, a systems interpretation tool that employs graph-theoretic methods for prioritization of systemic processes and corresponding driver genes. The current implementation exploits the Galaxy environment and is available online. A case study using public data is presented, with and without a control selection.

Molecular evolution of peptides by yeast surface display technology

Genetically encoded peptides possess unique properties, such as a small molecular weight and ease of synthesis and modification, that make them suitable to a large variety of applications. However, despite these favorable qualities, naturally occurring peptides are often limited by intrinsic weak binding affinities, poor selectivity and low stability that ultimately restrain their final use. To overcome these limitations, a large variety of in vitro display methodologies have been developed over the past few decades to evolve genetically encoded peptide molecules with superior properties. Phage display, mRNA display, ribosome display, bacteria display, and yeast display are among the most commonly used methods to engineer peptides. While most of these in vitro methodologies have already been described in detail elsewhere, this review describes solely the yeast surface display technology and its valuable use for the evolution of a wide range of peptide formats.

Massively Parallel Assays and Quantitative Sequence-Function Relationships

Over the last decade, a rich variety of massively parallel assays have revolutionized our understanding of how biological sequences encode quantitative molecular phenotypes. These assays include deep mutational scanning, high-throughput SELEX, and massively parallel reporter assays. Here, we review these experimental methods and how the data they produce can be used to quantitatively model sequence-function relationships. In doing so, we touch on a diverse range of topics, including the identification of clinically relevant genomic variants, the modeling of transcription factor binding to DNA, the functional and evolutionary landscapes of proteins, and cis-regulatory mechanisms in both transcription and mRNA splicing. We further describe a unified conceptual framework and a core set of mathematical modeling strategies that studies in these diverse areas can make use of. Finally, we highlight key aspects of experimental design and mathematical modeling that are important for the results of such studies to be interpretable and reproducible.

Tertiary Structural Motif Sequence Statistics Enable Facile Prediction and Design of Peptides that Bind Anti-apoptotic Bfl-1 and Mcl-1

Understanding the relationship between protein sequence and structure well enough to design new proteins with desired functions is a longstanding goal in protein science. Here, we show that recurring tertiary structural motifs (TERMs) in the PDB provide rich information for protein-peptide interaction prediction and design. TERM statistics can be used to predict peptide binding energies for Bcl-2 family proteins as accurately as widely used structure-based tools. Furthermore, design using TERM energies (dTERMen) rapidly and reliably generates high-affinity peptide binders of anti-apoptotic proteins Bfl-1 and Mcl-1 with just 15%-38% sequence identity to any known native Bcl-2 family protein ligand. High-resolution structures of four designed peptides bound to their targets provide opportunities to analyze the strengths and limitations of the computational design method. Our results support dTERMen as a powerful approach that can complement existing tools for protein engineering.

Selection of Protein-Protein Interactions of Desired Affinities with a Bandpass Circuit

We have developed a genetic circuit in Escherichia coli that can be used to select for protein-protein interactions of different strengths by changing antibiotic concentrations in the media. The genetic circuit links protein-protein interaction strength to β-lactamase activity while simultaneously imposing tuneable positive and negative selection pressure for β-lactamase activity. Cells only survive if they express interacting proteins with affinities that fall within set high- and low-pass thresholds; i.e. the circuit therefore acts as a bandpass filter for protein-protein interactions. We show that the circuit can be used to recover protein-protein interactions of desired affinity from a mixed population with a range of affinities. The circuit can also be used to select for inhibitors of protein-protein interactions of defined strength.

Protein model discrimination attempts using mutational sensitivity, predicted secondary structure, and model quality information

Structure prediction methods often generate a large number of models for a target sequence. Even if the correct fold for the target sequence is sampled in this dataset, it is difficult to distinguish it from other decoy structures. An attempt to solve this problem using experimental mutational sensitivity data for the CcdB protein was described previously by exploiting the correlation of residue depth with mutational sensitivity (r ~ 0.6). We now show that such a correlation extends to four other proteins with localized active sites, and for which saturation mutagenesis datasets exist. We also examine whether incorporation of predicted secondary structure information and the DOPE model quality assessment score, in addition to mutational sensitivity, improves the accuracy of model discrimination using a decoy dataset of 163 targets from CASP. Although most CASP models would have been subjected to model quality assessment prior to submission, we find that the DOPE score makes a substantial contribution to the observed improvement. We therefore also applied the approach to CcdB and four other proteins for which reliable experimental mutational data exist and observe that inclusion of experimental mutational data results in a small qualitative improvement in model discrimination relative to that seen with just the DOPE score. This is largely because of our limited ability to quantitatively predict effects of point mutations on in vivo protein activity. Further improvements in the methodology are required to facilitate improved utilization of single mutant data.

Emerging Themes in PDZ Domain Signaling: Structure, Function, and Inhibition

Post-synaptic density-95, disks-large and zonula occludens-1 (PDZ) domains are small globular protein-protein interaction domains widely conserved from yeast to humans. They are composed of ∼90 amino acids and form a classical two α-helical/six β-strand structure. The prototypical ligand is the C-terminus of partner proteins; however, they also bind internal peptide sequences. Recent findings indicate that PDZ domains also bind phosphatidylinositides and cholesterol. Through their ligand interactions, PDZ domain proteins are critical for cellular trafficking and the surface retention of various ion channels. In addition, PDZ proteins are essential for neuronal signaling, memory, and learning. PDZ proteins also contribute to cytoskeletal dynamics by mediating interactions critical for maintaining cell-cell junctions, cell polarity, and cell migration. Given their important biological roles, it is not surprising that their dysfunction can lead to multiple disease states. As such, PDZ domain-containing proteins have emerged as potential targets for the development of small molecular inhibitors as therapeutic agents. Recent data suggest that the critical binding function of PDZ domains in cell signaling is more than just glue, and their binding function can be regulated by phosphorylation or allosterically by other binding partners. These studies also provide a wealth of structural and biophysical data that are beginning to reveal the physical features that endow this small modular domain with a central role in cell signaling.

Data-driven supervised learning of a viral protease specificity landscape from deep sequencing and molecular simulations

Biophysical interactions between proteins and peptides are key determinants of molecular recognition specificity landscapes. However, an understanding of how molecular structure and residue-level energetics at protein-peptide interfaces shape these landscapes remains elusive. We combine information from yeast-based library screening, next-generation sequencing, and structure-based modeling in a supervised machine learning approach to report the comprehensive sequence-energetics-function mapping of the specificity landscape of the hepatitis C virus (HCV) NS3/4A protease, whose function-site-specific cleavages of the viral polyprotein-is a key determinant of viral fitness. We screened a library of substrates in which five residue positions were randomized and measured cleavability of ∼30,000 substrates (∼1% of the library) using yeast display and fluorescence-activated cell sorting followed by deep sequencing. Structure-based models of a subset of experimentally derived sequences were used in a supervised learning procedure to train a support vector machine to predict the cleavability of 3.2 million substrate variants by the HCV protease. The resulting landscape allows identification of previously unidentified HCV protease substrates, and graph-theoretic analyses reveal extensive clustering of cleavable and uncleavable motifs in sequence space. Specificity landscapes of known drug-resistant variants are similarly clustered. The described approach should enable the elucidation and redesign of specificity landscapes of a wide variety of proteases, including human-origin enzymes. Our results also suggest a possible role for residue-level energetics in shaping plateau-like functional landscapes predicted from viral quasispecies theory.

Affinity and specificity of motif-based protein-protein interactions

It is becoming increasingly clear that eukaryotic cell physiology is largely controlled by protein-protein interactions involving disordered protein regions, which usually interact with globular domains in a coupled binding and folding reaction. Several protein recognition domains are part of large families where members can interact with similar peptide ligands. Because of this, much research has been devoted to understanding how specificity can be achieved. A combination of interface complementarity, interactions outside of the core binding site, avidity from multidomain architecture and spatial and temporal regulation of expression resolves the conundrum. Here, we review recent advances in molecular aspects of affinity and specificity in such protein-protein interactions.

Peptide design by optimization on a data-parameterized protein interaction landscape

Many applications in protein engineering require optimizing multiple protein properties simultaneously, such as binding one target but not others or binding a target while maintaining stability. Such multistate design problems require navigating a high-dimensional space to find proteins with desired characteristics. A model that relates protein sequence to functional attributes can guide design to solutions that would be hard to discover via screening. In this work, we measured thousands of protein-peptide binding affinities with the high-throughput interaction assay amped SORTCERY and used the data to parameterize a model of the alpha-helical peptide-binding landscape for three members of the Bcl-2 family of proteins: Bcl-x_L, Mcl-1, and Bfl-1. We applied optimization protocols to explore extremes in this landscape to discover peptides with desired interaction profiles. Computational design generated 36 peptides, all of which bound with high affinity and specificity to just one of Bcl-x_L, Mcl-1, or Bfl-1, as intended. We designed additional peptides that bound selectively to two out of three of these proteins. The designed peptides were dissimilar to known Bcl-2-binding peptides, and high-resolution crystal structures confirmed that they engaged their targets as expected. Excellent results on this challenging problem demonstrate the power of a landscape modeling approach, and the designed peptides have potential uses as diagnostic tools or cancer therapeutics.

Nonproteinogenic deep mutational scanning of linear and cyclic peptides

The 20 proteinogenic amino acids have physicochemical properties that allow peptides and proteins to fold and bind. However, there are numerous unnatural, nonproteinogenic amino acids that may be equally good, or even better, at folding and binding. Exploration of these alternative peptide building blocks has been limited by slow, one-at-a-time synthesis and testing. We describe how, in a single experiment, multiple nonproteinogenic amino acids can be trialed at all positions in a peptide sequence, with thousands of modifications tested in parallel. This permits detailed analysis of how chemical structure relates to function and allows for systematic comparisons of proteinogenic and nonproteinogenic chemistry. Such analysis can guide the improvement of drug-candidate peptides, including the therapeutically promising class of cyclic peptides.

High-resolution structure–activity analysis of polypeptides requires amino acid structures that are not present in the universal genetic code. Examination of peptide and protein interactions with this resolution has been limited by the need to individually synthesize and test peptides containing nonproteinogenic amino acids. We describe a method to scan entire peptide sequences with multiple nonproteinogenic amino acids and, in parallel, determine the thermodynamics of binding to a partner protein. By coupling genetic code reprogramming to deep mutational scanning, any number of amino acids can be exhaustively substituted into peptides, and single experiments can return all free energy changes of binding. We validate this approach by scanning two model protein-binding peptides with 21 diverse nonproteinogenic amino acids. Dense structure–activity maps were produced at the resolution of single aliphatic atom insertions and deletions. This permits rapid interrogation of interaction interfaces, as well as optimization of affinity, fine-tuning of physical properties, and systematic assessment of nonproteinogenic amino acids in binding and folding.

Multiplexed assays of variant effects contribute to a growing genotype-phenotype atlas

Given the constantly improving cost and speed of genome sequencing, it is reasonable to expect that personal genomes will soon be known for many millions of humans. This stands in stark contrast with our limited ability to interpret the sequence variants which we find. Although it is, perhaps, easiest to interpret variants in coding regions, knowledge of functional impact is unknown for the vast majority of missense variants. While many computational approaches can predict the impact of coding variants, they are given a little weight in the current guidelines for interpreting clinical variants. Laboratory assays produce comparatively more trustworthy results, but until recently did not scale to the space of all possible mutations. The development of deep mutational scanning and other multiplexed assays of variant effect has now brought feasibility of this endeavour within view. Here, we review progress in this field over the last decade, break down the different approaches into their components, and compare methodological differences.

Insights into protein structure, stability and function from saturation mutagenesis

Where convenient phenotypic readouts are available, saturation mutagenesis coupled to deep sequencing provides a rapid and facile method to infer sequence determinants of protein structure, stability and function. We provide brief descriptions and currently available options for the various steps involved, and mention limitations of current implementations. We also highlight recent applications such as estimating relative stabilities and affinities of protein variants, mapping epitopes, protein model discrimination and prediction of mutant phenotypes. Most mutational scans have so far been applied to single genes and proteins. Additional methodological improvements are required to expand the scope to study intergenic epistasis and intermolecular interactions in macromolecular complexes.

Functional interrogation and mining of natively paired human VH:VL antibody repertoires

We present a technology to screen natively-paired human antibody repertoires from millions of B cells. Libraries of natively-paired variable region heavy and light (V_H:V_L) amplicons are expressed in a yeast display platform that is optimized for human Fab surface expression. Using our method we identify HIV-1 broadly neutralizing antibodies (bNAbs) from an HIV-1 slow progressor and high-affinity neutralizing antibodies against Ebola virus glycoprotein and influenza hemagglutinin.

A statistical framework for analyzing deep mutational scanning data

Deep mutational scanning is a widely used method for multiplex measurement of functional consequences of protein variants. We developed a new deep mutational scanning statistical model that generates error estimates for each measurement, capturing both sampling error and consistency between replicates. We apply our model to one novel and five published datasets comprising 243,732 variants and demonstrate its superiority in removing noisy variants and conducting hypothesis testing. Simulations show our model applies to scans based on cell growth or binding and handles common experimental errors. We implemented our model in Enrich2, software that can empower researchers analyzing deep mutational scanning data.

Measuring the sequence-affinity landscape of antibodies with massively parallel titration curves

Despite the central role that antibodies play in the adaptive immune system and in biotechnology, much remains unknown about the quantitative relationship between an antibody’s amino acid sequence and its antigen binding affinity. Here we describe a new experimental approach, called Tite-Seq, that is capable of measuring binding titration curves and corresponding affinities for thousands of variant antibodies in parallel. The measurement of titration curves eliminates the confounding effects of antibody expression and stability that arise in standard deep mutational scanning assays. We demonstrate Tite-Seq on the CDR1H and CDR3H regions of a well-studied scFv antibody. Our data shed light on the structural basis for antigen binding affinity and suggests a role for secondary CDR loops in establishing antibody stability. Tite-Seq fills a large gap in the ability to measure critical aspects of the adaptive immune system, and can be readily used for studying sequence-affinity landscapes in other protein systems.

DOI:http://dx.doi.org/10.7554/eLife.23156.001

Antibodies are proteins produced by cells of the immune system to tag or neutralize potential threats to the body, such as foreign substances and disease-causing microbes. Antibodies do this by binding to target molecules called antigens. An antibody’s ability to bind to an antigen depends on the sequence of amino acids – the building blocks of proteins – that make up the antibody. Through a process that randomizes this sequence of amino acids, the immune system generates a vast pool of antibodies that are able to target almost any foreign antigen that exists in nature.

Currently, little is understood about how the sequence of amino acids in an antibody determines how strongly that antibody binds to its antigen target – a property referred to as the antibody’s binding affinity. Answering this fundamental question requires techniques that can measure the affinities of many different antibodies at the same time. However, previous high-throughput methods have been unable to provide quantitative measurements of binding affinities. These kinds of measurements are difficult because an antibody’s amino acid sequence governs more than just binding affinity: it also affects how easy it is to produce that antibody, and what fraction of antibody molecules work properly.

Adams et al. now describe a new method, named “Tite-Seq”, that overcomes these issues. First, thousands of different antibodies are displayed on the surface of yeast cells, with each cell carrying a single kind of antibody. These cells are then incubated with fluorescently labeled antigen at a wide range of different concentrations. Next, the yeast cells are sorted based on how brightly they glow; brighter cells have more antigen bound to them, and so it is possible to calculate how much of the antigen is bound to each kind of antibody at each concentration. Plotting these data provides a “binding curve” for each antibody, which is then used to read off the antibody’s binding affinity in a way that is not affected by the factors that have plagued other high-throughput methods.

Tite-Seq is thus able to measure the binding affinities for thousands of different antibodies at the same time. This will potentially allow researchers to address many fundamental and yet unanswered questions about how the immune system works. Tite-Seq can also be used to measure how amino acid sequence affects the binding affinity of proteins other than antibodies.

DOI:http://dx.doi.org/10.7554/eLife.23156.002

Affinity Maturation of a Cyclic Peptide Handle for Therapeutic Antibodies Using Deep Mutational Scanning

Meditopes are cyclic peptides that bind in a specific pocket in the antigen-binding fragment of a therapeutic antibody such as cetuximab. Provided their moderate affinity can be enhanced, meditope peptides could be used as specific non-covalent and paratope-independent handles in targeted drug delivery, molecular imaging, and therapeutic drug monitoring. Here we show that the affinity of a recently reported meditope for cetuximab can be substantially enhanced using a combination of yeast display and deep mutational scanning. Deep sequencing was used to construct a fitness landscape of this protein-peptide interaction, and four mutations were identified that together improved the affinity for cetuximab 10-fold to 15 nm Importantly, the increased affinity translated into enhanced cetuximab-mediated recruitment to EGF receptor-overexpressing cancer cells. Although in silico Rosetta simulations correctly identified positions that were tolerant to mutation, modeling did not accurately predict the affinity-enhancing mutations. The experimental approach reported here should be generally applicable and could be used to develop meditope peptides with low nanomolar affinity for other therapeutic antibodies.

Determination of binding affinity upon mutation for type I dockerin-cohesin complexes from Clostridium thermocellum and Clostridium cellulolyticum using deep sequencing

The comprehensive sequence determinants of binding affinity for type I cohesin toward dockerin from Clostridium thermocellum and Clostridium cellulolyticum was evaluated using deep mutational scanning coupled to yeast surface display. We measured the relative binding affinity to dockerin for 2970 and 2778 single point mutants of C. thermocellum and C. cellulolyticum, respectively, representing over 96% of all possible single point mutants. The interface ΔΔG for each variant was reconstructed from sequencing counts and compared with the three independent experimental methods. This reconstruction results in a narrow dynamic range of -0.8-0.5 kcal/mol. The computational software packages FoldX and Rosetta were used to predict mutations that disrupt binding by more than 0.4 kcal/mol. The area under the curve of receiver operator curves was 0.82 for FoldX and 0.77 for Rosetta, showing reasonable agreements between predictions and experimental results. Destabilizing mutations to core and rim positions were predicted with higher accuracy than support positions. This benchmark dataset may be useful for developing new computational prediction tools for the prediction of the mutational effect on binding affinities for protein-protein interactions. Experimental considerations to improve precision and range of the reconstruction method are discussed. Proteins 2016; 84:1914-1928. © 2016 Wiley Periodicals, Inc.

Protein stability: computation, sequence statistics, and new experimental methods

Calculating protein stability and predicting stabilizing mutations remain exceedingly difficult tasks, largely due to the inadequacy of potential functions, the difficulty of modeling entropy and the unfolded state, and challenges of sampling, particularly of backbone conformations. Yet, computational design has produced some remarkably stable proteins in recent years, apparently owing to near ideality in structure and sequence features. With caveats, computational prediction of stability can be used to guide mutation, and mutations derived from consensus sequence analysis, especially improved by recent co-variation filters, are very likely to stabilize without sacrificing function. The combination of computational and statistical approaches with library approaches, including new technologies such as deep sequencing and high throughput stability measurements, point to a very exciting near term future for stability engineering, even with difficult computational issues remaining.

Comparison of the peptide binding preferences of three closely related TRAF paralogs: TRAF2, TRAF3, and TRAF5

Tumor necrosis factor receptor-associated factors (TRAFs) constitute a family of adapter proteins that act in numerous signaling pathways important in human biology and disease. The MATH domain of TRAF proteins binds peptides found in the cytoplasmic domains of signaling receptors, thereby connecting extracellular signals to downstream effectors. Beyond several very general motifs, the peptide binding preferences of TRAFs have not been extensively characterized, and differences between the binding preferences of TRAF paralogs are poorly understood. Here we report a screening system that we established to explore TRAF peptide-binding specificity using deep mutational scanning of TRAF-peptide ligands. We displayed single- and double-mutant peptide libraries based on the TRAF-binding sites of CD40 or TANK on the surface of Escherichia coli and screened them for binding to TRAF2, TRAF3, and TRAF5. Enrichment analysis of the library sequencing results showed differences in the permitted substitution patterns in the TANK versus CD40 backgrounds. The three TRAF proteins also demonstrated different preferences for binding to members of the CD40 library, and three peptides from that library that were analyzed individually showed striking differences in affinity for the three TRAFs. These results illustrate a previously unappreciated level of binding specificity between these close paralogs and demonstrate that established motifs are overly simplistic. The results from this work begin to outline differences between TRAF family members, and the experimental approach established herein will enable future efforts to investigate and redesign TRAF peptide-binding specificity.

Rapid Optimization of Mcl-1 Inhibitors using Stapled Peptide Libraries Including Non-Natural Side Chains

Alpha helices form a critical part of the binding interface for many protein-protein interactions, and chemically stabilized synthetic helical peptides can be effective inhibitors of such helix-mediated complexes. In particular, hydrocarbon stapling of peptides to generate constrained helices can improve binding affinity and other peptide properties, but determining the best stapled peptide variant often requires laborious trial and error. Here, we describe the rapid discovery and optimization of a stapled-helix peptide that binds to Mcl-1, an antiapoptotic protein that is overexpressed in many chemoresistant cancers. To accelerate discovery, we developed a peptide library synthesis and screening scheme capable of identifying subtle affinity differences among Mcl-1-binding stapled peptides. We used our method to sample combinations of non-natural amino-acid substitutions that we introduced into Mcl-1 inhibitors in the context of a fixed helix-stabilizing hydrocarbon staple that increased peptide helical content and reduced proteolysis. Peptides discovered in our screen contained surprising substitutions at sites that are conserved in natural binding partners. Library-identified peptide M3d is the most potent molecule yet tested for selectively triggering mitochondrial permeabilization in Mcl-1 dependent cell lines. Our library approach for optimizing helical peptide inhibitors can be readily applied to the study of other biomedically important targets.

PHASTpep: Analysis Software for Discovery of Cell-Selective Peptides via Phage Display and Next-Generation Sequencing

Next-generation sequencing has enhanced the phage display process, allowing for the quantification of millions of sequences resulting from the biopanning process. In response, many valuable analysis programs focused on specificity and finding targeted motifs or consensus sequences were developed. For targeted drug delivery and molecular imaging, it is also necessary to find peptides that are selective—targeting only the cell type or tissue of interest. We present a new analysis strategy and accompanying software, PHage Analysis for Selective Targeted PEPtides (PHASTpep), which identifies highly specific and selective peptides. Using this process, we discovered and validated, both in vitro and in vivo in mice, two sequences (HTTIPKV and APPIMSV) targeted to pancreatic cancer-associated fibroblasts that escaped identification using previously existing software. Our selectivity analysis makes it possible to discover peptides that target a specific cell type and avoid other cell types, enhancing clinical translatability by circumventing complications with systemic use.

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.

Generating High-Accuracy Peptide-Binding Data in High Throughput with Yeast Surface Display and SORTCERY

Library methods are widely used to study protein-protein interactions, and high-throughput screening or selection followed by sequencing can identify a large number of peptide ligands for a protein target. In this chapter, we describe a procedure called "SORTCERY" that can rank the affinities of library members for a target with high accuracy. SORTCERY follows a three-step protocol. First, fluorescence-activated cell sorting (FACS) is used to sort a library of yeast-displayed peptide ligands according to their affinities for a target. Second, all sorted pools are deep sequenced. Third, the resulting data are analyzed to create a ranking. We demonstrate an application of SORTCERY to the problem of ranking peptide ligands for the anti-apoptotic regulator Bcl-xL.

High-throughput methods for identification of protein-protein interactions involving short linear motifs

Interactions between modular domains and short linear motifs (3–10 amino acids peptide stretches) are crucial for cell signaling. The motifs typically reside in the disordered regions of the proteome and the interactions are often transient, allowing for rapid changes in response to changing stimuli. The properties that make domain-motif interactions suitable for cell signaling also make them difficult to capture experimentally and they are therefore largely underrepresented in the known protein-protein interaction networks. Most of the knowledge on domain-motif interactions is derived from low-throughput studies, although there exist dedicated high-throughput methods for the identification of domain-motif interactions. The methods include arrays of peptides or proteins, display of peptides on phage or yeast, and yeast-two-hybrid experiments. We here provide a survey of scalable methods for domain-motif interaction profiling. These methods have frequently been applied to a limited number of ubiquitous domain families. It is now time to apply them to a broader set of peptide binding proteins, to provide a comprehensive picture of the linear motifs in the human proteome and to link them to their potential binding partners. Despite the plethora of methods, it is still a challenge for most approaches to identify interactions that rely on post-translational modification or context dependent or conditional interactions, suggesting directions for further method development.