Mapping of amino acid substitutions conferring herbicide resistance in wheat glutathione transferase

Genotype sampling for deep-learning assisted experimental mapping of a combinatorially complete fitness landscape

MOTIVATION

Experimental characterization of fitness landscapes, which map genotypes onto fitness, is important for both evolutionary biology and protein engineering. It faces a fundamental obstacle in the astronomical number of genotypes whose fitness needs to be measured for any one protein. Deep learning may help to predict the fitness of many genotypes from a smaller neural network training sample of genotypes with experimentally measured fitness. Here I use a recently published experimentally mapped fitness landscape of more than 260 000 protein genotypes to ask how such sampling is best performed.

RESULTS

I show that multilayer perceptrons, recurrent neural networks, convolutional networks, and transformers, can explain more than 90% of fitness variance in the data. In addition, 90% of this performance is reached with a training sample comprising merely ≈103 sequences. Generalization to unseen test data is best when training data is sampled randomly and uniformly, or sampled to minimize the number of synonymous sequences. In contrast, sampling to maximize sequence diversity or codon usage bias reduces performance substantially. These observations hold for more than one network architecture. Simple sampling strategies may perform best when training deep learning neural networks to map fitness landscapes from experimental data.

AVAILABILITY AND IMPLEMENTATION

The fitness landscape data analyzed here is publicly available as described previously (Papkou et al. 2023). All code used to analyze this landscape is publicly available at https://github.com/andreas-wagner-uzh/fitness_landscape_sampling.

Human GST P1-1 Redesigned for Enhanced Catalytic Activity with the Anticancer Prodrug Telcyta and Improved Thermostability

Protein engineering can be used to tailor enzymes for medical purposes, including antibody-directed enzyme prodrug therapy (ADEPT), which can act as a tumor-targeted alternative to conventional chemotherapy for cancer. In ADEPT, the antibody serves as a vector, delivering a drug-activating enzyme selectively to the tumor site. Glutathione transferases (GSTs) are a family of naturally occurring detoxication enzymes, and the finding that some of them are overexpressed in tumors has been exploited to develop GST-activated prodrugs. The prodrug Telcyta is activated by GST P1-1, which is the GST most commonly elevated in cancer cells, implying that tumors overexpressing GST P1-1 should be particularly vulnerable to Telcyta. Promising antitumor activity has been noted in clinical trials, but the wildtype enzyme has modest activity with Telcyta, and further functional improvement would enhance its usefulness for ADEPT. We utilized protein engineering to construct human GST P1-1 gene variants in the search for enzymes with enhanced activity with Telcyta. The variant Y109H displayed a 2.9-fold higher enzyme activity compared to the wild-type GST P1-1. However, increased catalytic potency was accompanied by decreased thermal stability of the Y109H enzyme, losing 99% of its activity in 8 min at 50 °C. Thermal stability was restored by four additional mutations simultaneously introduced without loss of the enhanced activity with Telcyta. The mutation Q85R was identified as an important contributor to the regained thermostability. These results represent a first step towards a functional ADEPT application for Telcyta.

Repurposing Glutathione Transferases: Directed Evolution Combined with Chemical Modification for the Creation of a Semisynthetic Enzyme with High Hydroperoxidase Activity

Glutathione peroxidases (GPXs) are antioxidant selenoenzymes, which catalyze the reduction of hydroperoxides via glutathione (GSH), providing protection to cells against oxidative stress metabolites. The present study aims to create an efficient semisynthetic GPX based on the scaffold of tau class glutathione transferase (GSTU). A library of GSTs was constructed via DNA shuffling, using three homologue GSTUs from Glycine max as parent sequences. The DNA library of the shuffled genes was expressed in E. coli and the catalytic activity of the shuffled enzymes was screened using cumene hydroperoxide (CuOOH) as substrate. A chimeric enzyme variant (named Sh14) with 4-fold enhanced GPX activity, compared to the wild-type enzyme, was identified and selected for further study. Selenocysteine (Sec) was substituted for the active-site Ser13 residue of the Sh14 variant via chemical modification. The GPX activity (kcat) and the specificity constant (kcat/Κm) of the evolved seleno-Sh14 enzyme (SeSh14) was increased 177- and 2746-fold, respectively, compared to that of the wild-type enzyme for CuOOH. Furthermore, SeSh14 effectively catalyzed the reduction of hydrogen peroxide, an activity that is completely undetectable in all GSTs. Such an engineered GPX-like biocatalyst based on the GSTU scaffold might serve as a catalytic bioscavenger for the detoxification of hazardous hydroperoxides. Furthermore, our results shed light on the evolution of GPXs and their structural and functional link with GSTs.

Facilitating Machine Learning-Guided Protein Engineering with Smart Library Design and Massively Parallel Assays

Protein design plays an important role in recent medical advances from antibody therapy to vaccine design. Typically, exhaustive mutational screens or directed evolution experiments are used for the identification of the best design or for improvements to the wild-type variant. Even with a high-throughput screening on pooled libraries and Next-Generation Sequencing to boost the scale of read-outs, surveying all the variants with combinatorial mutations for their empirical fitness scores is still of magnitudes beyond the capacity of existing experimental settings. To tackle this challenge, in-silico approaches using machine learning to predict the fitness of novel variants based on a subset of empirical measurements are now employed. These machine learning models turn out to be useful in many cases, with the premise that the experimentally determined fitness scores and the amino-acid descriptors of the models are informative. The machine learning models can guide the search for the highest fitness variants, resolve complex epistatic relationships, and highlight bio-physical rules for protein folding. Using machine learning-guided approaches, researchers can build more focused libraries, thus relieving themselves from labor-intensive screens and fast-tracking the optimization process. Here, we describe the current advances in massive-scale variant screens, and how machine learning and mutagenesis strategies can be integrated to accelerate protein engineering. More specifically, we examine strategies to make screens more economical, informative, and effective in discovery of useful variants.

Machine Learning Enables Selection of Epistatic Enzyme Mutants for Stability Against Unfolding and Detrimental Aggregation

Machine learning (ML) has pervaded most areas of protein engineering, including stability and stereoselectivity. Using limonene epoxide hydrolase as the model enzyme and innov'SAR as the ML platform, comprising a digital signal process, we achieved high protein robustness that can resist unfolding with concomitant detrimental aggregation. Fourier transform (FT) allows us to take into account the order of the protein sequence and the nonlinear interactions between positions, and thus to grasp epistatic phenomena. The innov'SAR approach is interpolative, extrapolative and makes outside-the-box, predictions not found in other state-of-the-art ML or deep learning approaches. Equally significant is the finding that our approach to ML in the present context, flanked by advanced molecular dynamics simulations, uncovers the connection between epistatic mutational interactions and protein robustness.

Statistical Design of Experiments for Synthetic Biology

The design and optimization of biological systems is an inherently complex undertaking that requires careful balancing of myriad synergistic and antagonistic variables. However, despite this complexity, much synthetic biology research is predicated on One Factor at A Time (OFAT) experimentation; the genetic and environmental variables affecting the activity of a system of interest are sequentially altered while all other variables are held constant. Beyond being time and resource intensive, OFAT experimentation crucially ignores the effect of interactions between factors. Given the ubiquity of interacting genetic and environmental factors in biology this failure to account for interaction effects in OFAT experimentation can result in the development of suboptimal systems. To address these limitations, an increasing number of studies have turned to Design of Experiments (DoE), a suite of methods that enable efficient, systematic exploration and exploitation of complex design spaces. This review provides an overview of DoE for synthetic biologists. Key concepts and commonly used experimental designs are introduced, and we discuss the advantages of DoE as compared to OFAT experimentation. We dissect the applicability of DoE in the context of synthetic biology and review studies which have successfully employed these methods, illustrating the potential of statistical experimental design to guide the design, characterization, and optimization of biological protocols, pathways, and processes.

Exploiting Single Domain Antibodies as Regulatory Parts to Modulate Monoterpenoid Production in E. coli

Synthetic biology and metabolic engineering offer potentially green and attractive routes to the production of high value compounds. The provision of high-quality parts and pathways is crucial in enabling the biosynthesis of chemicals using synthetic biology. While a number of regulatory parts that provide control at the transcriptional and translational level have been developed, relatively few exist at the protein level. Single domain antibodies (sdAb) such as camelid heavy chain variable fragments (V_HH) possess binding characteristics which could be exploited for their development and use as novel parts for regulating metabolic pathways at the protein level in microbial cell factories. Here, a platform for the use of V_HH as tools in Escherichia coli is developed and subsequently used to modulate linalool production in E. coli. The coproduction of a Design of Experiments (DoE) optimized pBbE8k His₆-V_HHCyDisCo system alongside a heterologous linalool production pathway facilitated the identification of anti-bLinS V_HH that functioned as modulators of bLinS. This resulted in altered product profiles and significant variation in the titers of linalool, geraniol, nerolidol, and indole obtained. The ability to alter the production levels of high value terpenoids, such as linalool, in a tunable manner at the protein level could represent a significant step forward for the development of improved microbial cell factories. This study serves as a proof of principle indicating that V_HH can be used to modulate enzyme activity in engineered pathways within E. coli. Given their almost limitless binding potential, we posit that single domain antibodies could emerge as powerful regulatory parts in synthetic biology applications.

Applying Statistical Design of Experiments To Understanding the Effect of Growth Medium Components on Cupriavidus necator H16 Growth

Chemically defined media (CDM) for cultivation of C. necator vary in components and compositions. This lack of consensus makes it difficult to optimize new processes for the bacterium. This study employed statistical design of experiments (DOE) to understand how basic components of defined media affect C. necator growth. Our growth model predicts that C. necator can be cultivated to high cell density with components held at low concentrations, arguing that CDM for large-scale cultivation of the bacterium for industrial purposes will be economically competitive. Although existing CDM for the bacterium are without amino acids, addition of a few amino acids to growth medium shortened lag phase of growth. The interactions highlighted by our growth model show how factors can interact with each other during a process to positively or negatively affect process output. This approach is efficient, relying on few well-structured experimental runs to gain maximum information on a biological process, growth.

Cupriavidus necator H16 is gaining significant attention as a microbial chassis for range of biotechnological applications. While the bacterium is a major producer of bioplastics, its lithoautotrophic and versatile metabolic capabilities make the bacterium a promising microbial chassis for biofuels and chemicals using renewable resources. It remains necessary to develop appropriate experimental resources to permit controlled bioengineering and system optimization of this microbe. In this study, we employed statistical design of experiments to gain understanding of the impact of components of defined media on C. necator growth and built a model that can predict the bacterium’s cell density based on medium components. This highlighted medium components, and interaction between components, having the most effect on growth: fructose, amino acids, trace elements, CaCl₂, and Na₂HPO₄ contributed significantly to growth (t values of <−1.65 or >1.65); copper and histidine were found to interact and must be balanced for robust growth. Our model was experimentally validated and found to correlate well (r² = 0.85). Model validation at large culture scales showed correlations between our model-predicted growth ranks and experimentally determined ranks at 100 ml in shake flasks (ρ = 0.87) and 1 liter in a bioreactor (ρ = 0.90). Our approach provides valuable and quantifiable insights on the impact of medium components on cell growth and can be applied to model other C. necator responses that are crucial for its deployment as a microbial chassis. This approach can be extended to other nonmodel microbes of medical and industrial biotechnological importance.

IMPORTANCE Chemically defined media (CDM) for cultivation of C. necator vary in components and compositions. This lack of consensus makes it difficult to optimize new processes for the bacterium. This study employed statistical design of experiments (DOE) to understand how basic components of defined media affect C. necator growth. Our growth model predicts that C. necator can be cultivated to high cell density with components held at low concentrations, arguing that CDM for large-scale cultivation of the bacterium for industrial purposes will be economically competitive. Although existing CDM for the bacterium are without amino acids, addition of a few amino acids to growth medium shortened lag phase of growth. The interactions highlighted by our growth model show how factors can interact with each other during a process to positively or negatively affect process output. This approach is efficient, relying on few well-structured experimental runs to gain maximum information on a biological process, growth.

Machine-learning-guided directed evolution for protein engineering

Protein engineering through machine-learning-guided directed evolution enables the optimization of protein functions. Machine-learning approaches predict how sequence maps to function in a data-driven manner without requiring a detailed model of the underlying physics or biological pathways. Such methods accelerate directed evolution by learning from the properties of characterized variants and using that information to select sequences that are likely to exhibit improved properties. Here we introduce the steps required to build machine-learning sequence-function models and to use those models to guide engineering, making recommendations at each stage. This review covers basic concepts relevant to the use of machine learning for protein engineering, as well as the current literature and applications of this engineering paradigm. We illustrate the process with two case studies. Finally, we look to future opportunities for machine learning to enable the discovery of unknown protein functions and uncover the relationship between protein sequence and function.

Systematic Evaluation of Genetic and Environmental Factors Affecting Performance of Translational Riboswitches

Since their discovery, riboswitches have been attractive tools for the user-controlled regulation of gene expression in bacterial systems. Riboswitches facilitate small molecule mediated fine-tuning of protein expression, making these tools of great use to the synthetic biology community. However, the use of riboswitches is often restricted due to context dependent performance and limited dynamic range. Here, we report the drastic improvement of a previously developed orthogonal riboswitch achieved through in vivo functional selection and optimization of flanking coding and noncoding sequences. The behavior of the derived riboswitches was mapped under a wide array of growth and induction conditions, using a structured Design of Experiments approach. This approach successfully improved the maximal protein expression levels 8.2-fold relative to the original riboswitches, and the dynamic range was improved to afford riboswitch dependent control of 80-fold. The optimized orthogonal riboswitch was then integrated downstream of four endogenous stress promoters, responsive to phosphate starvation, hyperosmotic stress, redox stress, and carbon starvation. These responsive stress promoter-riboswitch devices were demonstrated to allow for tuning of protein expression up to ∼650-fold in response to both environmental and cellular stress responses and riboswitch dependent attenuation. We envisage that these riboswitch stress responsive devices will be useful tools for the construction of advanced genetic circuits, bioprocessing, and protein expression.

Design of Experiments Methodology to Build a Multifactorial Statistical Model Describing the Metabolic Interactions of Alcohol Dehydrogenase Isozymes in the Ethanol Biosynthetic Pathway of the Yeast Saccharomyces cerevisiae

Multifactorial approaches can quickly and efficiently model complex, interacting natural or engineered biological systems in a way that traditional one-factor-at-a-time experimentation can fail to do. We applied a Design of Experiments (DOE) approach to model ethanol biosynthesis in yeast, which is well-understood and genetically tractable, yet complex. Six alcohol dehydrogenase (ADH) isozymes catalyze ethanol synthesis, differing in their transcriptional and post-translational regulation, subcellular localization, and enzyme kinetics. We generated a combinatorial library of all ADH gene deletions and measured the impact of gene deletion(s) and environmental context on ethanol production of a subset of this library. The data were used to build a statistical model that described known behaviors of ADH isozymes and identified novel interactions. Importantly, the model described features of ADH metabolic behavior without explicit a priori knowledge. The method is therefore highly suited to understanding and optimizing metabolic pathways in less well-understood systems.

Characterization of equine GST A3-3 as a steroid isomerase

Glutathione transferases (GSTs) comprise a superfamily of enzymes prominently involved in detoxication by making toxic electrophiles more polar and therefore more easily excretable. However some GSTs have developed alternative functions. Thus, a member of the Alpha class GSTs in pig and human tissues is involved in steroid hormone biosynthesis, catalyzing the obligatory double-bond isomerization of Δ⁵-androstene-3,17-dione to Δ⁴-androstene-3,17-dione and of Δ⁵-pregnene-3,20-dione to Δ⁴-pregnene-3,20-dione on the biosynthetic pathways to testosterone and progesterone. The human GST A3-3 is the most efficient steroid double-bond isomerase known so far in mammals. The current work extends discoveries of GST enzymes that act in the steroidogenic pathways in large mammals. The mRNA encoding the steroid isomerase GST A3-3 was cloned from testis of the horse (Equus ferus caballus). The concentrations of GSTA3 mRNA were highest in hormone-producing organs such as ovary, testis and adrenal gland. EcaGST A3-3 produced in E. coli has been characterized and shown to have highly efficient steroid double-bond isomerase activity, exceeding its activities with conventional GST substrates. The enzyme now ranks as one of the most efficient steroid isomerases known in mammals and approaches the activity of the bacterial ketosteroid isomerase, one of the most efficient enzymes of all categories known today. The high efficiency and the tissue distribution of EcaGST A3-3 support the view that the enzyme plays a physiologically significant role in the biosynthesis of steroid hormones.

Protein Engineering: Past, Present, and Future

The last decade has seen a dramatic increase in the utilization of enzymes as green and sustainable (bio)catalysts in pharmaceutical and industrial applications. This trend has to a significant degree been fueled by advances in scientists' and engineers' ability to customize native enzymes by protein engineering. A review of the literature quickly reveals the tremendous success of this approach; protein engineering has generated enzyme variants with improved catalytic activity, broadened or altered substrate specificity, as well as raised or reversed stereoselectivity. Enzymes have been tailored to retain activity at elevated temperatures and to function in the presence of organic solvents, salts and pH values far from physiological conditions. However, readers unfamiliar with the field will soon encounter the confusingly large number of experimental techniques that have been employed to accomplish these engineering feats. Herein, we use history to guide a brief overview of the major strategies for protein engineering-past, present, and future.

Exploring sequence-function space of a poplar glutathione transferase using designed information-rich gene variants

Exploring the vicinity around a locus of a protein in sequence space may identify homologs with enhanced properties, which could become valuable in biotechnical and other applications. A rational approach to this pursuit is the use of ‘infologs’, i.e. synthetic sequences with specific substitutions capturing maximal sequence information derived from the evolutionary history of the protein family. Ninety-five such infolog genes of poplar glutathione transferase were synthesized and expressed in Escherichia coli, and the catalytic activities of the proteins determined with alternative substrates. Sequence–activity relationships derived from the infologs were used to design a second set of 47 infologs in which 90% of the members exceeded wild-type properties. Two mutants, C2 (V55I/E95D/D108E/A160V) and G5 (F13L/C70A/G122E), were further functionally characterized. The activities of the infologs with the alternative substrates 1-chloro-2,4-dinitrobenzene and phenethyl isothiocyanate, subject to different chemistries, were positively correlated, indicating that the examined mutations were affecting the overall catalytic competence without major shift in substrate discrimination. By contrast, the enhanced protein expressivity observed in many of the mutants were not similarly correlated with the activities. In conclusion, small libraries of well-defined infologs can be used to systematically explore sequence space to optimize proteins in multidimensional functional space.

Plant glutathione transferase-mediated stress tolerance: functions and biotechnological applications

Plant glutathione transferases (EC 2.5.1.18, GSTs) are an ancient, multimember and diverse enzyme class. Plant GSTs have diverse roles in plant development, endogenous metabolism, stress tolerance, and xenobiotic detoxification. Their study embodies both fundamental aspects and agricultural interest, because of their ability to confer tolerance against biotic and abiotic stresses and to detoxify herbicides. Here we review the biotechnological applications of GSTs towards developing plants that are resistant to biotic and abiotic stresses. We integrate recent discoveries, highlight, and critically discuss the underlying biochemical and molecular pathways involved. We elaborate that the functions of GSTs in abiotic and biotic stress adaptation are potentially a result of both catalytic and non-catalytic functions. These include conjugation of reactive electrophile species with glutathione and the modulation of cellular redox status, biosynthesis, binding, and transport of secondary metabolites and hormones. Their major universal functions under stress underline the potential in developing climate-resilient cultivars through a combination of molecular and conventional breeding programs. We propose that future GST engineering efforts through rational and combinatorial approaches, would lead to the design of improved isoenzymes with purpose-designed catalytic activities and novel functional properties. Concurrent GST-GSH metabolic engineering can incrementally increase the effectiveness of GST biotechnological deployment.

A Simple Combinatorial Codon Mutagenesis Method for Targeted Protein Engineering

Directed evolution is a powerful tool for optimizing enzymes, and mutagenesis methods that improve enzyme library quality can significantly expedite the evolution process. Here, we report a simple method for targeted combinatorial codon mutagenesis (CCM). To demonstrate the utility of this method for protein engineering, CCM libraries were constructed for cytochrome P450_BM3, pfu prolyl oligopeptidase, and the flavin-dependent halogenase RebH; 10-26 sites were targeted for codon mutagenesis in each of these enzymes, and libraries with a tunable average of 1-7 codon mutations per gene were generated. Each of these libraries provided improved enzymes for their respective transformations, which highlights the generality, simplicity, and tunability of CCM for targeted protein engineering.

Multiple Pesticides Detoxification Function of Maize (Zea mays) GST34

ZmGST34 is a maize Tau class GST gene and was found to be differently expressed between two maize cultivars differing in tolerance to herbicide metolachlor. To explore the possible role of ZmGST34 in maize development, the expression pattern and substrate specificity of ZmGST34 were characterized by quantitative RT-PCR and heterologous expression system, respectively. The results indicated that the expression level of ZmGST34 was increased ∼2-5-fold per day during the second-leaf stage of maize seedling. Chloroacetanilide herbicides or phytohormone treatments had no influence on the expression level of ZmGST34, suggesting that ZmGST34 is a constitutively expressed gene in maize seedling. Heterologous expression in Escherichia coli and in Arabidopsis thaliana proved that ZmGST34 can metabolize most chloroacetanilide herbicides and increase tolerance to these herbicides in transgenic Arabidopsis thaliana. The constitutive expression pattern and broad substrate activity of ZmGST34 suggested that this gene may play an important role in maize development in addition to the detoxification of pesticides.

Kinetic Characterization of 100 Glycoside Hydrolase Mutants Enables the Discovery of Structural Features Correlated with Kinetic Constants

The use of computational modeling algorithms to guide the design of novel enzyme catalysts is a rapidly growing field. Force-field based methods have now been used to engineer both enzyme specificity and activity. However, the proportion of designed mutants with the intended function is often less than ten percent. One potential reason for this is that current force-field based approaches are trained on indirect measures of function rather than direct correlation to experimentally-determined functional effects of mutations. We hypothesize that this is partially due to the lack of data sets for which a large panel of enzyme variants has been produced, purified, and kinetically characterized. Here we report the kcat and KM values of 100 purified mutants of a glycoside hydrolase enzyme. We demonstrate the utility of this data set by using machine learning to train a new algorithm that enables prediction of each kinetic parameter based on readily-modeled structural features. The generated dataset and analyses carried out in this study not only provide insight into how this enzyme functions, they also provide a clear path forward for the improvement of computational enzyme redesign algorithms.

Harnessing QbD, Programming Languages, and Automation for Reproducible Biology

Building robust manufacturing processes from biological components is a task that is highly complex and requires sophisticated tools to describe processes, inputs, and measurements and administrate management of knowledge, data, and materials. We argue that for bioengineering to fully access biological potential, it will require application of statistically designed experiments to derive detailed empirical models of underlying systems. This requires execution of large-scale structured experimentation for which laboratory automation is necessary. This requires development of expressive, high-level languages that allow reusability of protocols, characterization of their reliability, and a change in focus from implementation details to functional properties. We review recent developments in these areas and identify what we believe is an exciting trend that promises to revolutionize biotechnology.

CRISPR-Cas9 Based Genome Engineering: Opportunities in Agri-Food-Nutrition and Healthcare

Recently developed strategies and techniques that make use of the vast amount of genetic information to perform targeted perturbations in the genome of living organisms are collectively referred to as genome engineering. The wide array of applications made possible by the use of this technology range from agriculture to healthcare. This, along with the applications involving basic biological research, has made it a very dynamic and active field of research. This review focuses on the CRISPR system from its discovery and role in bacterial adaptive immunity to the most recent developments, and its possible applications in agriculture and modern medicine.

Plant GSTome: structure and functional role in xenome network and plant stress response

Glutathione transferases (GSTs) represent a major group of detoxification enzymes. All plants possess multiple cytosolic GSTs, each of which displays distinct catalytic as well as non-catalytic binding properties. The progress made in recent years in the fields of genomics, proteomics and protein crystallography of GSTs, coupled with studies on their molecular evolution, diversity and substrate specificity has provided new insights into the function of these enzymes. In plants, GSTs appear to be implicated in an array of different functions, including detoxification of xenobiotics and endobiotics, primary and secondary metabolism, stress tolerance, and cell signalling. This review focuses on plant GSTome and attempts to give an overview of its catalytic and functional role in xenome and plant stress regulatory networks.