A general model to predict small molecule substrates of enzymes based on machine and deep learning.

For most proteins annotated as enzymes, it is unknown which primary and/or secondary reactions they catalyze. Experimental characterizations of potential substrates are time-consuming and costly. Machine learning predictions could provide an efficient alternative, but are hampered by a lack of information regarding enzyme non-substrates, as available training data comprises mainly positive examples. Here, we present ESP, a general machine-learning model for the prediction of enzyme-substrate pairs with an accuracy of over 91% on independent and diverse test data. ESP can be applied successfully across widely different enzymes and a broad range of metabolites included in the training data, outperforming models designed for individual, well-studied enzyme families. ESP represents enzymes through a modified transformer model, and is trained on data augmented with randomly sampled small molecules assigned as non-substrates. By facilitating easy in silico testing of potential substrates, the ESP web server may support both basic and applied science.

Chiral Alcohols from Alkenes and Water: Directed Evolution of a Styrene Hydratase.

Enantioselective synthesis of chiral alcohols through asymmetric addition of water across an unactivated alkene is a highly sought-after transformation and a big challenge in catalysis. Herein we report the identification and directed evolution of a fatty acid hydratase from Marinitoga hydrogenitolerans for the highly enantioselective hydration of styrenes to yield chiral 1-arylethanols. While directed evolution for styrene hydration was performed in the presence of heptanoic acid to mimic fatty acid binding, the engineered enzyme displayed remarkable asymmetric styrene hydration activity in the absence of the small molecule activator. The evolved styrene hydratase provided access to chiral alcohols from simple alkenes and water with high enantioselectivity (>99 : 1 e.r.) and could be applied on a preparative scale.

Learning deep representations of enzyme thermal adaptation.

Temperature is a fundamental environmental factor that shapes the evolution of organisms. Learning thermal determinants of protein sequences in evolution thus has profound significance for basic biology, drug discovery, and protein engineering. Here, we use a data set of over 3 million BRENDA enzymes labeled with optimal growth temperatures (OGTs) of their source organisms to train a deep neural network model (DeepET). The protein-temperature representations learned by DeepET provide a temperature-related statistical summary of protein sequences and capture structural properties that affect thermal stability. For prediction of enzyme optimal catalytic temperatures and protein melting temperatures via a transfer learning approach, our DeepET model outperforms classical regression models trained on rationally designed features and other deep-learning-based representations. DeepET thus holds promise for understanding enzyme thermal adaptation and guiding the engineering of thermostable enzymes.

Directed Evolution of (R)-2-Hydroxyglutarate Dehydrogenase Improves 2-Oxoadipate Reduction by 2 Orders of Magnitude.

Pathway engineering is commonly employed to improve the production of various metabolites but may incur in bottlenecks due to the low catalytic activity of a particular reaction step. The reduction of 2-oxoadipate to (R)-2-hydroxyadipate is a key reaction in metabolic pathways that exploit 2-oxoadipate conversion via α-reduction to produce adipic acid, an industrially important platform chemical. Here, we engineered (R)-2-hydroxyglutarate dehydrogenase from Acidaminococcus fermentans (Hgdh) with the aim of improving 2-oxoadipate reduction. Using a combination of computational analysis, saturation mutagenesis, and random mutagenesis, three mutant variants with a 100-fold higher catalytic efficiency were obtained. As revealed by rational analysis of the mutations found in the variants, this improvement could be ascribed to a general synergistic effect where mutation A206V played a key role since it boosted the enzyme's activity by 4.8-fold. The Hgdh variants with increased activity toward 2-oxoadipate generated within this study pave the way for the bio-based production of adipic acid.

Dataset for suppressors of amyloid-ß toxicity and their functions in recombinant protein production in yeast.

The production of recombinant proteins at high levels often induces stress-related phenotypes by protein misfolding or aggregation. These are similar to those of the yeast Alzheimer's disease (AD) model in which amyloid-ß peptides (Aß42) were accumulated [1], [2]. We have previously identified suppressors of Aß42 cytotoxicity via the genome-wide synthetic genetic array (SGA) [3] and here we use them as metabolic engineering targets to evaluate their potentiality on recombinant protein production in yeast Saccharomyces cerevisiae. In order to investigate the mechanisms linking the genetic modifications to the improved recombinant protein production, we perform systems biology approaches (transcriptomics and proteomics) on the resulting strain and intermediate strains. The RNAseq data are preprocessed by the nf-core/RNAseq pipeline and analyzed using the Platform for Integrative Analysis of Omics (PIANO) package [4]. The quantitative proteome is analyzed on an Orbitrap Fusion Lumos mass spectrometer interfaced with an Easy-nLC1200 liquid chromatography (LC) system. LC-MS data files are processed by Proteome Discoverer version 2.4 with Mascot 2.5.1 as a database search engine. The original data presented in this work can be found in the research paper titled "Suppressors of Amyloid-ß Toxicity Improve Recombinant Protein Production in yeast by Reducing Oxidative Stress and Tuning Cellular Metabolism", by Chen et al. [5].

Suppressors of amyloid-ß toxicity improve recombinant protein production in yeast by reducing oxidative stress and tuning cellular metabolism.

High-level production of recombinant proteins in industrial microorganisms is often limited by the formation of misfolded proteins or protein aggregates, which consequently induce cellular stress responses. We hypothesized that in a yeast Alzheimer's disease (AD) model overexpression of amyloid-ß peptides (Aß42), one of the main peptides relevant for AD pathologies, induces similar phenotypes of cellular stress. Using this humanized AD model, we previously identified suppressors of Aß42 cytotoxicity. Here we hypothesize that these suppressors could be used as metabolic engineering targets to alleviate cellular stress and improve recombinant protein production in the yeast Saccharomyces cerevisiae. Forty-six candidate genes were individually deleted and twenty were individually overexpressed. The positive targets that increased recombinant α-amylase production were further combined leading to an 18.7-fold increased recombinant protein production. These target genes are involved in multiple cellular networks including RNA processing, transcription, ER-mitochondrial complex, and protein unfolding. By using transcriptomics and proteomics analyses, combined with reverse metabolic engineering, we showed that reduced oxidative stress, increased membrane lipid biosynthesis and repressed arginine and sulfur amino acid biosynthesis are significant pathways for increased recombinant protein production. Our findings provide new insights towards developing synthetic yeast cell factories for biosynthesis of valuable proteins.

Engineering Saccharomyces cerevisiae for the production and secretion of Affibody molecules.

BACKGROUND: Affibody molecules are synthetic peptides with a variety of therapeutic and diagnostic applications. To date, Affibody molecules have mainly been produced by the bacterial production host Escherichia coli. There is an interest in exploring alternative production hosts to identify potential improvements in terms of yield, ease of production and purification advantages. In this study, we evaluated the feasibility of Saccharomyces cerevisiae as a production chassis for this group of proteins. RESULTS: We examined the production of three different Affibody molecules in S. cerevisiae and found that these Affibody molecules were partially degraded. An albumin-binding domain, which may be attached to the Affibody molecules to increase their half-life, was identified to be a substrate for several S. cerevisiae proteases. We tested the removal of three vacuolar proteases, proteinase A, proteinase B and carboxypeptidase Y. Removal of one of these, proteinase A, resulted in intact secretion of one of the targeted Affibody molecules. Removal of either or both of the two additional proteases, carboxypeptidase Y and proteinase B, resulted in intact secretion of the two remaining Affibody molecules. The produced Affibody molecules were verified to bind their target, human HER3, as potently as the corresponding molecules produced in E. coli in an in vitro surface-plasmon resonance binding assay. Finally, we performed a fed-batch fermentation with one of the engineered protease-deficient S. cerevisiae strains and achieved a protein titer of 530 mg Affibody molecule/L. CONCLUSION: This study shows that engineered S. cerevisiae has a great potential as a production host for recombinant Affibody molecules, reaching a high titer, and for proteins where endotoxin removal could be challenging, the use of S. cerevisiae obviates the need for endotoxin removal from protein produced in E. coli.

Basin-scale biogeography of marine phytoplankton reflects cellular-scale optimization of metabolism and physiology.

Extensive microdiversity within Prochlorococcus, the most abundant marine cyanobacterium, occurs at scales from a single droplet of seawater to ocean basins. To interpret the structuring role of variations in genetic potential, as well as metabolic and physiological acclimation, we developed a mechanistic constraint-based modeling framework that incorporates the full suite of genes, proteins, metabolic reactions, pigments, and biochemical compositions of 69 sequenced isolates spanning the Prochlorococcus pangenome. Optimizing each strain to the local, observed physical and chemical environment along an Atlantic Ocean transect, we predicted variations in strain-specific patterns of growth rate, metabolic configuration, and physiological state, defining subtle niche subspaces directly attributable to differences in their encoded metabolic potential. Predicted growth rates covaried with observed ecotype abundances, affirming their significance as a measure of fitness and inferring a nonlinear density dependence of mortality. Our study demonstrates the potential to interpret global-scale ecosystem organization in terms of cellular-scale processes.

Modeling-Assisted Design of Thermostable Benzaldehyde Lyases from Rhodococcus erythropolis for Continuous Production of α-Hydroxy Ketones.

Enantiopure α-hydroxy ketones are important building blocks of active pharmaceutical ingredients (APIs), which can be produced by thiamine-diphosphate-dependent lyases, such as benzaldehyde lyase. Here we report the discovery of a novel thermostable benzaldehyde lyase from Rhodococcus erythropolis R138 (ReBAL). While the overall sequence identity to the only experimentally confirmed benzaldehyde lyase from Pseudomonas fluorescens Biovar I (PfBAL) was only 65 %, comparison of a structural model of ReBAL with the crystal structure of PfBAL revealed only four divergent amino acids in the substrate binding cavity. Based on rational design, we generated two ReBAL variants, which were characterized along with the wild-type enzyme in terms of their substrate spectrum, thermostability and biocatalytic performance in the presence of different co-solvents. We found that the new enzyme variants have a significantly higher thermostability (up to 22 °C increase in T50 ) and a different co-solvent-dependent activity. Using the most stable variant immobilized in packed-bed reactors via the SpyCatcher/SpyTag system, (R)-benzoin was synthesized from benzaldehyde over a period of seven days with a stable space-time-yield of 9.3 mmol ⋅ L-1 ⋅ d-1 . Our work expands the important class of benzaldehyde lyases and therefore contributes to the development of continuous biocatalytic processes for the production of α-hydroxy ketones and APIs.

Discovery of Two Novel Oxidases Using a High-Throughput Activity Screen.

Discovery of novel enzymes is a challenging task, yet a crucial one, due to their increasing relevance as chemical catalysts and biotechnological tools. In our work we present a high-throughput screening approach to discovering novel activities. A screen of 96 putative oxidases with 23 substrates led to the discovery of two new enzymes. The first enzyme, N-acetyl-D-hexosamine oxidase (EC 1.1.3.29) from Ralstonia solanacearum, is a vanillyl alcohol oxidase-like flavoprotein displaying the highest activity with N-acetylglucosamine and N-acetylgalactosamine. Before our discovery of the enzyme, its activity was an orphan one - experimentally characterized but lacking the link to amino acid sequence. The second enzyme, from an uncultured marine euryarchaeota, is a long-chain alcohol oxidase (LCAO, EC 1.1.3.20) active with a range of fatty alcohols, with 1-dodecanol being the preferred substrate. The enzyme displays no sequence similarity to previously characterised LCAOs, and thus is a completely novel representative of a protein with such activity.

Deep learning allows genome-scale prediction of Michaelis constants from structural features.

The Michaelis constant KM describes the affinity of an enzyme for a specific substrate and is a central parameter in studies of enzyme kinetics and cellular physiology. As measurements of KM are often difficult and time-consuming, experimental estimates exist for only a minority of enzyme-substrate combinations even in model organisms. Here, we build and train an organism-independent model that successfully predicts KM values for natural enzyme-substrate combinations using machine and deep learning methods. Predictions are based on a task-specific molecular fingerprint of the substrate, generated using a graph neural network, and on a deep numerical representation of the enzyme's amino acid sequence. We provide genome-scale KM predictions for 47 model organisms, which can be used to approximately relate metabolite concentrations to cellular physiology and to aid in the parameterization of kinetic models of cellular metabolism.

Experimental and computational investigation of enzyme functional annotations uncovers misannotation in the EC 1.1.3.15 enzyme class.

Only a small fraction of genes deposited to databases have been experimentally characterised. The majority of proteins have their function assigned automatically, which can result in erroneous annotations. The reliability of current annotations in public databases is largely unknown; experimental attempts to validate the accuracy within individual enzyme classes are lacking. In this study we performed an overview of functional annotations to the BRENDA enzyme database. We first applied a high-throughput experimental platform to verify functional annotations to an enzyme class of S-2-hydroxyacid oxidases (EC 1.1.3.15). We chose 122 representative sequences of the class and screened them for their predicted function. Based on the experimental results, predicted domain architecture and similarity to previously characterised S-2-hydroxyacid oxidases, we inferred that at least 78% of sequences in the enzyme class are misannotated. We experimentally confirmed four alternative activities among the misannotated sequences and showed that misannotation in the enzyme class increased over time. Finally, we performed a computational analysis of annotations to all enzyme classes in the BRENDA database, and showed that nearly 18% of all sequences are annotated to an enzyme class while sharing no similarity or domain architecture to experimentally characterised representatives. We showed that even well-studied enzyme classes of industrial relevance are affected by the problem of functional misannotation.

CAZyme prediction in ascomycetous yeast genomes guides discovery of novel xylanolytic species with diverse capacities for hemicellulose hydrolysis.

BACKGROUND: Ascomycetous yeasts from the kingdom fungi inhabit every biome in nature. While filamentous fungi have been studied extensively regarding their enzymatic degradation of the complex polymers comprising lignocellulose, yeasts have been largely overlooked. As yeasts are key organisms used in industry, understanding their enzymatic strategies for biomass conversion is an important factor in developing new and more efficient cell factories. The aim of this study was to identify polysaccharide-degrading yeasts by mining CAZymes in 332 yeast genomes from the phylum Ascomycota. Selected CAZyme-rich yeasts were then characterized in more detail through growth and enzymatic activity assays. RESULTS: The CAZyme analysis revealed a large spread in the number of CAZyme-encoding genes in the ascomycetous yeast genomes. We identified a total of 217 predicted CAZyme families, including several CAZymes likely involved in degradation of plant polysaccharides. Growth characterization of 40 CAZyme-rich yeasts revealed no cellulolytic yeasts, but several species from the Trichomonascaceae and CUG-Ser1 clades were able to grow on xylan, mixed-linkage ß-glucan and xyloglucan. Blastobotrys mokoenaii, Sugiyamaella lignohabitans, Spencermartinsiella europaea and several Scheffersomyces species displayed superior growth on xylan and well as high enzymatic activities. These species possess genes for several putative xylanolytic enzymes, including ones from the well-studied xylanase-containing glycoside hydrolase families GH10 and GH30, which appear to be attached to the cell surface. B. mokoenaii was the only species containing a GH11 xylanase, which was shown to be secreted. Surprisingly, no known xylanases were predicted in the xylanolytic species Wickerhamomyces canadensis, suggesting that this yeast possesses novel xylanases. In addition, by examining non-sequenced yeasts closely related to the xylanolytic yeasts, we were able to identify novel species with high xylanolytic capacities. CONCLUSIONS: Our approach of combining high-throughput bioinformatic CAZyme-prediction with growth and enzyme characterization proved to be a powerful pipeline for discovery of novel xylan-degrading yeasts and enzymes. The identified yeasts display diverse profiles in terms of growth, enzymatic activities and xylan substrate preferences, pointing towards different strategies for degradation and utilization of xylan. Together, the results provide novel insights into how yeast degrade xylan, which can be used to improve cell factory design and industrial bioconversion processes.

Adaptation of a Microfluidic qPCR System for Enzyme Kinetic Studies.

Microfluidic platforms offer a drastic increase in throughput while minimizing sample usage and hands-on time, which make them important tools for large-scale biological studies. A range of such systems have been developed for enzyme activity studies, although their complexity largely hinders their application to the wider scientific community. Here, we present adaptation of an easy-to-use commercial microfluidic qPCR system for performing enzyme kinetic studies. We demonstrate the functionality of the Fluidigm Biomark HD system (the Fluidigm system) by determining the kinetic properties of three oxidases in a resorufin-based fluorescence assay. The results obtained in the microfluidic system proved reproducible and comparable to the ones obtained in a standard microplate-based assay. With a wide range of easy-to-use, off-the-shelf components, the microfluidic system presents itself as a simple and customizable platform for high-throughput enzyme activity studies.

Different Routes of Protein Folding Contribute to Improved Protein Production in Saccharomyces cerevisiae.

Protein folding is often considered the flux controlling process in protein synthesis and secretion. Here, two previously isolated Saccharomyces cerevisiae strains with increased α-amylase productivity were analyzed in chemostat cultures at different dilution rates using multi-omics data. Based on the analysis, we identified different routes of the protein folding pathway to improve protein production. In the first strain, the increased abundance of proteins working on the folding process, coordinated with upregulated glycogen metabolism and trehalose metabolism, helped increase α-amylase productivity 1.95-fold compared to the level in the original strain in chemostat culture at a dilution rate of 0.2/h. The second strain further strengthened the folding precision to improve protein production. More precise folding helps the cell improve protein production efficiency and reduce the expenditure of energy on the handling of misfolded proteins. As calculated using an enzyme-constrained genome-scale metabolic model, the second strain had an increased productivity of 2.36-fold with lower energy expenditure than that of the original under the same condition. Further study revealed that the regulation of N-glycans played an important role in the folding precision control and that overexpression of the glucosidase Cwh41p can significantly improve protein production, especially for the strains with improved folding capacity but lower folding precision. Our findings elucidated in detail the mechanisms in two strains having improved protein productivity and thereby provided novel insights for industrial recombinant protein production as well as demonstrating how multi-omics analysis can be used for identification of novel strain-engineering targets.IMPORTANCE Protein folding plays an important role in protein maturation and secretion. In recombinant protein production, many studies have focused on the folding pathway to improve productivity. Here, we identified two different routes for improving protein production by yeast. We found that improving folding precision is a better strategy. Dysfunction of this process is also associated with several aberrant protein-associated human diseases. Here, our findings about the role of glucosidase Cwh41p in the precision control system and the characterization of the strain with a more precise folding process could contribute to the development of novel therapeutic strategies.

A synthesis of bacterial and archaeal phenotypic trait data.

A synthesis of phenotypic and quantitative genomic traits is provided for bacteria and archaea, in the form of a scripted, reproducible workflow that standardizes and merges 26 sources. The resulting unified dataset covers 14 phenotypic traits, 5 quantitative genomic traits, and 4 environmental characteristics for approximately 170,000 strain-level and 15,000 species-aggregated records. It spans all habitats including soils, marine and fresh waters and sediments, host-associated and thermal. Trait data can find use in clarifying major dimensions of ecological strategy variation across species. They can also be used in conjunction with species and abundance sampling to characterize trait mixtures in communities and responses of traits along environmental gradients.

Elimination of rNMPs from mitochondrial DNA has no effect on its stability.

Ribonucleotides (rNMPs) incorporated in the nuclear genome are a well-established threat to genome stability and can result in DNA strand breaks when not removed in a timely manner. However, the presence of a certain level of rNMPs is tolerated in mitochondrial DNA (mtDNA) although aberrant mtDNA rNMP content has been identified in disease models. We investigated the effect of incorporated rNMPs on mtDNA stability over the mouse life span and found that the mtDNA rNMP content increased during early life. The rNMP content of mtDNA varied greatly across different tissues and was defined by the rNTP/dNTP ratio of the tissue. Accordingly, mtDNA rNMPs were nearly absent in SAMHD1-/- mice that have increased dNTP pools. The near absence of rNMPs did not, however, appreciably affect mtDNA copy number or the levels of mtDNA molecules with deletions or strand breaks in aged animals near the end of their life span. The physiological rNMP load therefore does not contribute to the progressive loss of mtDNA quality that occurs as mice age.

3D-Printed Phenacrylate Decarboxylase Flow Reactors for the Chemoenzymatic Synthesis of 4-Hydroxystilbene.

Continuous flow systems for chemical synthesis are becoming a major focus in organic chemistry and there is a growing interest in the integration of biocatalysts due to their high regio- and stereoselectivity. Methods established for 3D bioprinting enable the fast and simple production of agarose-based modules for biocatalytic reactors if thermally stable enzymes are available. We report here on the characterization of four different cofactor-free phenacrylate decarboxylase enzymes suitable for the production of 4-vinylphenol and test their applicability for the encapsulation and direct 3D printing of disk-shaped agarose-based modules that can be used for compartmentalized flow microreactors. Using the most active and stable phenacrylate decarboxylase from Enterobacter spec. in a setup with four parallel reactors and a subsequent palladium(II) acetate-catalysed Heck reaction, 4-hydroxystilbene was synthesized from p-coumaric acid with a total yield of 14.7 % on a milligram scale. We believe that, due to the convenient direct immobilization of any thermostable enzyme and straightforward tuning of the reaction sequence by stacking of modules with different catalytic activities, this simple process will facilitate the establishment and use of cascade reactions and will therefore be of great advantage for many research approaches.

Machine Learning Applied to Predicting Microorganism Growth Temperatures and Enzyme Catalytic Optima.

Enzymes that catalyze chemical reactions at high temperatures are used for industrial biocatalysis, applications in molecular biology, and as highly evolvable starting points for protein engineering. The optimal growth temperature (OGT) of organisms is commonly used to estimate the stability of enzymes encoded in their genomes, but the number of experimentally determined OGT values are limited, particularly for thermophilic organisms. Here, we report on the development of a machine learning model that can accurately predict OGT for bacteria, archaea, and microbial eukaryotes directly from their proteome-wide 2-mer amino acid composition. The trained model is made freely available for reuse. In a subsequent step we use OGT data in combination with amino acid composition of individual enzymes to develop a second machine learning model-for prediction of enzyme catalytic temperature optima ( Topt). The resulting model generates enzyme Topt estimates that are far superior to using OGT alone. Finally, we predict Topt for 6.5 million enzymes, covering 4447 enzyme classes, and make the resulting data set available to researchers. This work enables simple and rapid identification of enzymes that are potentially functional at extreme temperatures.