Proposed three-phenylalanine motif involved in magnetoreception signalling of an Actinopterygii protein expressed in mammalian cells.

Studies at the cellular and molecular level of magnetoreception-sensing and responding to magnetic fields-are a relatively new research area. It appears that different mechanisms of magnetoreception in animals evolved from different origins, and, therefore, many questions about its mechanisms remain left open. Here we present new information regarding the Electromagnetic Perceptive Gene (EPG) from Kryptopterus vitreolus that may serve as part of the foundation to understanding and applying magnetoreception. Using HaloTag coupled with fluorescent ligands and phosphatidylinositol specific phospholipase C we show that EPG is associated with the membrane via glycosylphosphatidylinositol anchor. EPG's function of increasing intracellular calcium was also used to generate an assay using GCaMP6m to observe the function of EPG and to compare its function with that of homologous proteins. It was also revealed that EPG relies on a motif of three phenylalanine residues to function-stably swapping these residues using site directed mutagenesis resulted in a loss of function in EPG. This information not only expands upon our current understanding of magnetoreception but may provide a foundation and template to continue characterizing and discovering more within the emerging field.

Generative models for protein sequence modeling: recent advances and future directions.

The widespread adoption of high-throughput omics technologies has exponentially increased the amount of protein sequence data involved in many salient disease pathways and their respective therapeutics and diagnostics. Despite the availability of large-scale sequence data, the lack of experimental fitness annotations underpins the need for self-supervised and unsupervised machine learning (ML) methods. These techniques leverage the meaningful features encoded in abundant unlabeled sequences to accomplish complex protein engineering tasks. Proficiency in the rapidly evolving fields of protein engineering and generative AI is required to realize the full potential of ML models as a tool for protein fitness landscape navigation. Here, we support this work by (i) providing an overview of the architecture and mathematical details of the most successful ML models applicable to sequence data (e.g. variational autoencoders, autoregressive models, generative adversarial neural networks, and diffusion models), (ii) guiding how to effectively implement these models on protein sequence data to predict fitness or generate high-fitness sequences and (iii) highlighting several successful studies that implement these techniques in protein engineering (from paratope regions and subcellular localization prediction to high-fitness sequences and protein design rules generation). By providing a comprehensive survey of model details, novel architecture developments, comparisons of model applications, and current challenges, this study intends to provide structured guidance and robust framework for delivering a prospective outlook in the ML-driven protein engineering field.

Single-Cell B-Cell Sequencing to Generate Natively Paired scFab Yeast Surface Display Libraries.

The immune cell profiling capabilities of single-cell RNA sequencing (scRNA-seq) are powerful tools that can be applied to the design of theranostic monoclonal antibodies (mAbs). Using scRNA-seq to determine natively paired B-cell receptor (BCR) sequences of immunized mice as a starting point for design, this method outlines a simplified workflow to express single-chain antibody fragments (scFabs) on the surface of yeast for high-throughput characterization and further refinement with directed evolution experiments. While not extensively detailed in this chapter, this method easily accommodates the implementation of a growing body of in silico tools that improve affinity and stability among a range of other developability criteria (e.g., solubility and immunogenicity).

Topological deep learning based deep mutational scanning.

High-throughput deep mutational scanning (DMS) experiments have significantly impacted protein engineering, drug discovery, immunology, cancer biology, and evolutionary biology by enabling the systematic understanding of protein functions. However, the mutational space associated with proteins is astronomically large, making it overwhelming for current experimental capabilities. Therefore, alternative methods for DMS are imperative. We propose a topological deep learning (TDL) paradigm to facilitate in silico DMS. We utilize a new topological data analysis (TDA) technique based on the persistent spectral theory, also known as persistent Laplacian, to capture both topological invariants and the homotopic shape evolution of data. To validate our TDL-DMS model, we use SARS-CoV-2 datasets and show excellent accuracy and reliability for binding interface mutations. This finding is significant for SARS-CoV-2 variant forecasting and designing effective antibodies and vaccines. Our proposed model is expected to have a significant impact on drug discovery, vaccine design, precision medicine, and protein engineering.

Protein Fitness Prediction Is Impacted by the Interplay of Language Models, Ensemble Learning, and Sampling Methods.

Advances in machine learning (ML) and the availability of protein sequences via high-throughput sequencing techniques have transformed the ability to design novel diagnostic and therapeutic proteins. ML allows protein engineers to capture complex trends hidden within protein sequences that would otherwise be difficult to identify in the context of the immense and rugged protein fitness landscape. Despite this potential, there persists a need for guidance during the training and evaluation of ML methods over sequencing data. Two key challenges for training discriminative models and evaluating their performance include handling severely imbalanced datasets (e.g., few high-fitness proteins among an abundance of non-functional proteins) and selecting appropriate protein sequence representations (numerical encodings). Here, we present a framework for applying ML over assay-labeled datasets to elucidate the capacity of sampling techniques and protein encoding methods to improve binding affinity and thermal stability prediction tasks. For protein sequence representations, we incorporate two widely used methods (One-Hot encoding and physiochemical encoding) and two language-based methods (next-token prediction, UniRep; masked-token prediction, ESM). Elaboration on performance is provided over protein fitness, protein size, and sampling techniques. In addition, an ensemble of protein representation methods is generated to discover the contribution of distinct representations and improve the final prediction score. We then implement multiple criteria decision analysis (MCDA; TOPSIS with entropy weighting), using multiple metrics well-suited for imbalanced data, to ensure statistical rigor in ranking our methods. Within the context of these datasets, the synthetic minority oversampling technique (SMOTE) outperformed undersampling while encoding sequences with One-Hot, UniRep, and ESM representations. Moreover, ensemble learning increased the predictive performance of the affinity-based dataset by 4% compared to the best single-encoding candidate (F1-score = 97%), while ESM alone was rigorous enough in stability prediction (F1-score = 92%).

Phytochemical drug discovery for COVID-19 using high-resolution computational docking and machine learning assisted binder prediction.

The COVID-19 pandemic has resulted in millions of deaths around the world. Multiple vaccines are in use, but there are many underserved locations that do not have adequate access to them. Variants may emerge that are highly resistant to existing vaccines, and therefore cheap and readily obtainable therapeutics are needed. Phytochemicals, or plant chemicals, can possibly be such therapeutics. Phytochemicals can be used in a polypharmacological approach, where multiple viral proteins are inhibited and escape mutations are made less likely. Finding the right phytochemicals for viral protein inhibition is challenging, but in-silico screening methods can make this a more tractable problem. In this study, we screen a wide range of natural drug products against a comprehensive set of SARS-CoV-2 proteins using a high-resolution computational workflow. This workflow consists of a structure-based virtual screening (SBVS), where an initial phytochemical library was docked against all selected protein structures. Subsequently, ligand-based virtual screening (LBVS) was employed, where chemical features of 34 lead compounds obtained from the SBVS were used to predict 53 lead compounds from a larger phytochemical library via supervised learning. A computational docking validation of the 53 predicted leads obtained from LBVS revealed that 28 of them elicit strong binding interactions with SARS-CoV-2 proteins. Thus, the inclusion of LBVS resulted in a 4-fold increase in the lead discovery rate. Of the total 62 leads, 18 showed promising pharmacokinetic properties in a computational ADME screening. Collectively, this study demonstrates the advantage of incorporating machine learning elements into a virtual screening workflow.Communicated by Ramaswamy H. Sarma.

Site-wise Diversification of Combinatorial Libraries Using Insights from Structure-guided Stability Calculations.

Many auspicious clinical and industrial accomplishments have improved the human condition by means of protein engineering. Despite these achievements, our incomplete understanding of the sequence-structure-function relationship prevents rapid innovation. To tackle this problem, we must develop and integrate new and existing technologies. To date, directed evolution and rational design have dominated as protein engineering principles. Even so, prior to screening for novel or improved functions, a large collection of variants, within a protein library, exist along an ambiguous mutational terrain. Complicating things further, the choice of where to initialize investigation along a vast sequence space becomes even more difficult given that the majority of any sequence lacks function entirely. Unfortunately, even when considering functionally relevant positions, random substitutions can prove to be destabilizing, causing a hindrance to an otherwise function-inducing, stability-reliant folding process. To enhance productivity in the field, we seek to address this issue of destabilization, and subsequent disfunction, at protein-protein and protein-ligand interacting regions. Herein, the process of choosing amenable positions - and amino acids at those positions - allows for a refined, knowledge-based approach to combinatorial library design. Using structural data, we perform computational stability prediction with FoldX's PositionScan and Rosetta's ddG_monomer in tandem, allowing for the refinement of our thermodynamic stability data through the comparison of results. In turn, we provide a process for selecting in silico predicted mutually stabilizing positions and avoiding overly destabilizing ones that guides the site-wise diversification of combinatorial libraries.

Ancestral Sequence Reconstruction and Alternate Amino Acid States Guide Protein Library Design for Directed Evolution.

Engineered proteins possess nearly limitless possibilities in medical and industrial applications but finding a precise amino acid sequence for these applications is challenging. A robust approach for discovering protein sequences with a desired functionality uses a library design method in which combinations of mutations are applied to a robust starting point. Determining useful mutations can be tortuous, yet rewarding; in this chapter, we present a novel library design method that uses information provided by ancestral sequence reconstruction (ASR) to create a library likely to have stable proteins with diverse function. ASR computational tools use a multi-sequence alignment of homologous proteins and an evolutionary model to estimate the protein sequences of the numerous common ancestors. For all ancestors, these tools calculate the probability of every amino acid occurring at each position within the sequence alignment. The alternate amino acid states at individual positions corelate to a region of stability in sequence space around the ancestral sequence which can inform site-wise diversification within a combinatorial library. The method presented in this chapter balances the quality of results, the computational resources needed, and ease of use.

Machine Learning-driven Protein Library Design: A Path Toward Smarter Libraries.

Proteins are small yet valuable biomolecules that play a versatile role in therapeutics and diagnostics. The intricate sequence-structure-function paradigm in the realm of proteins opens the possibility for directly mapping amino acid sequence to function. However, the rugged nature of the protein fitness landscape and an astronomical number of possible mutations even for small proteins make navigating this system a daunting task. Moreover, the scarcity of functional proteins and the ease with which deleterious mutations are introduced, due to complex epistatic relationships, compound the existing challenges. This highlights the need for auxiliary tools in current techniques such as rational design and directed evolution. To that end, the state-of-the-art machine learning can offer time and cost efficiency in finding high fitness proteins, circumventing unnecessary wet-lab experiments. In the context of improving library design, machine learning provides valuable insights via its unique features such as high adaptation to complex systems, multi-tasking, and parallelism, and the ability to capture hidden trends in input data. Finally, both the advancements in computational resources and the rapidly increasing number of sequences in protein databases will allow more promising and detailed insights delivered from machine learning to protein library design. In this chapter, fundamental concepts and a method for machine learning-driven library design leveraging deep sequencing datasets will be discussed. We elaborate on (1) basic knowledge about machine learning algorithms, (2) the benefit of machine learning in library design, and (3) methodology for implementing machine learning in library design.

Design and Evaluation of Engineered Extracellular Vesicle (EV)-Based Targeting for EGFR-Overexpressing Tumor Cells Using Monobody Display.

BACKGROUND: Extracellular vesicles (EVs) are attracting interest as a new class of drug delivery vehicles due to their intrinsic nature of biomolecular transport in the body. We previously demonstrated that EV surface modification with tissue-specific molecules accomplished targeted EV-mediated DNA delivery. METHODS: Here, we describe reliable methods for (i) generating EGFR tumor-targeting EVs via the display of high-affinity monobodies and (ii) in vitro measurement of EV binding using fluorescence and bioluminescence labeling. Monobodies are a well-suited class of small (10 kDa) non-antibody scaffolds derived from the human fibronectin type III (FN3) domain. RESULTS: The recombinant protein consists of the EGFR-targeting monobody fused to the EV-binding domain of lactadherin (C1C2), enabling the monobody displayed on the surface of the EVs. In addition, the use of bioluminescence or fluorescence molecules on the EV surface allows for the assessment of EV binding to the target cells. CONCLUSIONS: In this paper, we describe methods of EV engineering to generate targeted delivery vehicles using monobodies that will have diverse applications to furnish future EV therapeutic development, including qualitative and quantitative in vitro evaluation for their binding capacity.

A Gradient of Sitewise Diversity Promotes Evolutionary Fitness for Binder Discovery in a Three-Helix Bundle Protein Scaffold.

Engineered proteins provide clinically and industrially impactful molecules and utility within fundamental research, yet inefficiencies in discovering lead variants with new desired functionality, while maintaining stability, hinder progress. Improved function, which can result from a few strategic mutations, is fundamentally separate from discovering novel function, which often requires large leaps in sequence space. While a highly diverse combinatorial library covering immense sequence space would empower protein discovery, the ability to sample only a minor subset of sequence space and the typical destabilization of random mutations preclude this strategy. A balance must be reached. At library scale, compounding several destabilizing mutations renders many variants unable to properly fold and devoid of function. Broadly searching sequence space while reducing the level of destabilization may enhance evolution. We exemplify this balance with affibody, a three-helix bundle protein scaffold. Using natural ligand data sets, stability and structural computations, and deep sequencing of thousands of binding variants, a protein library was designed on a sitewise basis with a gradient of mutational levels across 29% of the protein. In direct competition of biased and uniform libraries, both with 1 × 109 variants, for discovery of 6 × 104 ligands (5 × 103 clusters) toward seven targets, biased amino acid frequency increased ligand discovery 13 ± 3-fold. Evolutionarily favorable amino acids, both globally and site-specifically, are further elucidated. The sitewise amino acid bias aids evolutionary discovery by reducing the level of mutant destabilization as evidenced by a midpoint of denaturation (62 ± 4 °C) 15 °C higher than that of unbiased mutants (47 ± 11 °C; p < 0.001). Sitewise diversification, identified by high-throughput evolution and rational library design, improves discovery efficiency.

Titratable Avidity Reduction Enhances Affinity Discrimination in Mammalian Cellular Selections of Yeast-Displayed Ligands.

Yeast surface display selections against mammalian cell monolayers have proven effective in isolating proteins with novel binding activity. Recent advances in this technique allow for the recovery of clones with even micromolar binding affinities. However, no efficient method has been shown for affinity-based selection in this context. This study demonstrates the effectiveness of titratable avidity reduction using dithiothreitol to achieve this goal. A series of epidermal growth factor receptor binding fibronectin domains with a range of affinities are used to quantitatively identify the number of ligands per yeast cell that yield the strongest selectivity between strong, moderate, and weak affinities. Notably, reduction of ligand display to 3,000-6,000 ligands per yeast cell of a 2 nM binder yields 16-fold better selectivity than that to a 17 nM binder. These lessons are applied to affinity maturation of an EpCAM-binding fibronectin population, yielding an enriched pool of ligands with significantly stronger affinity than that of an analogous pool sorted by standard cellular selection methods. Collectively, this study offers a facile approach for affinity selection of yeast-displayed ligands against full-length cellular targets and demonstrates the effectiveness of this method by generating EpCAM-binding ligands that are promising for further applications.

64Cu-Labeled Gp2 Domain for PET Imaging of Epidermal Growth Factor Receptor.

This purpose of this study is to determine the efficacy of a 45-amino acid Gp2 domain, engineered to bind to epidermal growth factor receptor (EGFR), as a positron emission tomography (PET) probe of EGFR in a xenograft mouse model. The EGFR-targeted Gp2 (Gp2-EGFR) and a nonbinding control were site-specifically labeled with 1,4,7,10-tetraazacyclododecane-1,4,7,10-tetraacetic acid (DOTA) chelator. Binding affinity was tested toward human EGFR and mouse EGFR. Biological activity on downstream EGFR signaling was examined in cell culture. DOTA-Gp2 molecules were labeled with 64Cu and intravenously injected (0.6-2.3 MBq) into mice bearing EGFRhigh (n = 7) and EGFRlow (n = 4) xenografted tumors. PET/computed tomography (CT) images were acquired at 45 min, 2 h, and 24 h. Dynamic PET (25 min) was also acquired. Tomography results were verified with gamma counting of resected tissues. Two-tailed t tests with unequal variances provided statistical comparison. DOTA-Gp2-EGFR bound strongly to human (KD = 7 ± 5 nM) and murine (KD = 29 ± 6 nM) EGFR, and nontargeted Gp2 had no detectable binding. Gp2-EGFR did not agonize EGFR nor antagonize EGF-EGFR. 64Cu-Gp2-EGFR tracer effectively localized to EGFRhigh tumors at 45 min (3.2 ± 0.5%ID/g). High specificity was observed with significantly lower uptake in EGFRlow tumors (0.9 ± 0.3%ID/g, p < 0.001), high tumor-to-background ratios (11 ± 6 tumor/muscle, p < 0.001). Nontargeted Gp2 tracer had low uptake in EGFRhigh tumors (0.5 ± 0.3%ID/g, p < 0.001). Similar data was observed at 2 h, and tumor signal was retained at 24 h (2.9 ± 0.3%ID/g). An engineered Gp2 PET imaging probe exhibited low background and target-specific EGFRhigh tumor uptake at 45 min, with tumor signal retained at 24 h postinjection, and compared favorably with published EGFR PET probes for alternative protein scaffolds. These beneficial in vivo characteristics, combined with thermal stability, efficient evolution, and small size of the Gp2 domain validate its use as a future class of molecular imaging agents.

ScaffoldSeq: Software for characterization of directed evolution populations.

ScaffoldSeq is software designed for the numerous applications-including directed evolution analysis-in which a user generates a population of DNA sequences encoding for partially diverse proteins with related functions and would like to characterize the single site and pairwise amino acid frequencies across the population. A common scenario for enzyme maturation, antibody screening, and alternative scaffold engineering involves naïve and evolved populations that contain diversified regions, varying in both sequence and length, within a conserved framework. Analyzing the diversified regions of such populations is facilitated by high-throughput sequencing platforms; however, length variability within these regions (e.g., antibody CDRs) encumbers the alignment process. To overcome this challenge, the ScaffoldSeq algorithm takes advantage of conserved framework sequences to quickly identify diverse regions. Beyond this, unintended biases in sequence frequency are generated throughout the experimental workflow required to evolve and isolate clones of interest prior to DNA sequencing. ScaffoldSeq software uniquely handles this issue by providing tools to quantify and remove background sequences, cluster similar protein families, and dampen the impact of dominant clones. The software produces graphical and tabular summaries for each region of interest, allowing users to evaluate diversity in a site-specific manner as well as identify epistatic pairwise interactions. The code and detailed information are freely available at http://research.cems.umn.edu/hackel. Proteins 2016; 84:869-874. © 2016 Wiley Periodicals, Inc.

High-Throughput Ligand Discovery Reveals a Sitewise Gradient of Diversity in Broadly Evolved Hydrophilic Fibronectin Domains.

Discovering new binding function via a combinatorial library in small protein scaffolds requires balance between appropriate mutations to introduce favorable intermolecular interactions while maintaining intramolecular integrity. Sitewise constraints exist in a non-spatial gradient from diverse to conserved in evolved antibody repertoires; yet non-antibody scaffolds generally do not implement this strategy in combinatorial libraries. Despite the fact that biased amino acid distributions, typically elevated in tyrosine, serine, and glycine, have gained wider use in synthetic scaffolds, these distributions are still predominantly applied uniformly to diversified sites. While select sites in fibronectin domains and DARPins have shown benefit from sitewise designs, they have not been deeply evaluated. Inspired by this disparity between diversity distributions in natural libraries and synthetic scaffold libraries, we hypothesized that binders resulting from discovery and evolution would exhibit a non-spatial, sitewise gradient of amino acid diversity. To identify sitewise diversities consistent with efficient evolution in the context of a hydrophilic fibronectin domain, >105 binders to six targets were evolved and sequenced. Evolutionarily favorable amino acid distributions at 25 sites reveal Shannon entropies (range: 0.3-3.9; median: 2.1; standard deviation: 1.1) supporting the diversity gradient hypothesis. Sitewise constraints in evolved sequences are consistent with complementarity, stability, and consensus biases. Implementation of sitewise constrained diversity enables direct selection of nanomolar affinity binders validating an efficient strategy to balance inter- and intra-molecular interaction demands at each site.

A 45-Amino-Acid Scaffold Mined from the PDB for High-Affinity Ligand Engineering.

Small protein ligands can provide superior physiological distribution compared with antibodies, and improved stability, production, and specific conjugation. Systematic evaluation of the PDB identified a scaffold to push the limits of small size and robust evolution of stable, high-affinity ligands: 45-residue T7 phage gene 2 protein (Gp2) contains an α helix opposite a ß sheet with two adjacent loops amenable to mutation. De novo ligand discovery from 10(8) mutants and directed evolution toward four targets yielded target-specific binders with affinities as strong as 200 ± 100 pM, Tms from 65 °C ± 3 °C to 80°C ± 1 °C, and retained activity after thermal denaturation. For cancer targeting, a Gp2 domain for epidermal growth factor receptor was evolved with 18 ± 8 nM affinity, receptor-specific binding, and high thermal stability with refolding. The efficiency of evolving new binding function and the size, affinity, specificity, and stability of evolved domains render Gp2 a uniquely effective ligand scaffold.