Prediction of designer-recombinases for DNA editing with generative deep learning

Design of diverse, functional mitochondrial targeting sequences across eukaryotic organisms using variational autoencoder

Mitochondria play a key role in energy production and metabolism, making them a promising target for metabolic engineering and disease treatment. However, despite the known influence of passenger proteins on localization efficiency, only a few protein-localization tags have been characterized for mitochondrial targeting. To address this limitation, we leverage a Variational Autoencoder to design novel mitochondrial targeting sequences. In silico analysis reveals that a high fraction of the generated peptides (90.14%) are functional and possess features important for mitochondrial targeting. We characterize artificial peptides in four eukaryotic organisms and, as a proof-of-concept, demonstrate their utility in increasing 3-hydroxypropionic acid titers through pathway compartmentalization and improving 5-aminolevulinate synthase delivery by 1.62-fold and 4.76-fold, respectively. Moreover, we employ latent space interpolation to shed light on the evolutionary origins of dual-targeting sequences. Overall, our work demonstrates the potential of generative artificial intelligence for both fundamental research and practical applications in mitochondrial biology.

A data-efficient strategy for building high-performing medical foundation models

Foundation models are pretrained on massive datasets. However, collecting medical datasets is expensive and time-consuming, and raises privacy concerns. Here we show that synthetic data generated via conditioning with disease labels can be leveraged for building high-performing medical foundation models. We pretrained a retinal foundation model, first with approximately one million synthetic retinal images with physiological structures and feature distribution consistent with real counterparts, and then with only 16.7% of the 904,170 real-world colour fundus photography images required in a recently reported retinal foundation model (RETFound). The data-efficient model performed as well or better than RETFound across nine public datasets and four diagnostic tasks; and for diabetic-retinopathy grading, it used only 40% of the expert-annotated training data used by RETFound. We also support the generalizability of the data-efficient strategy by building a classifier for the detection of tuberculosis on chest X-ray images. The text-conditioned generation of synthetic data may enhance the performance and generalization of medical foundation models.

Computational scoring and experimental evaluation of enzymes generated by neural networks

In recent years, generative protein sequence models have been developed to sample novel sequences. However, predicting whether generated proteins will fold and function remains challenging. We evaluate a set of 20 diverse computational metrics to assess the quality of enzyme sequences produced by three contrasting generative models: ancestral sequence reconstruction, a generative adversarial network and a protein language model. Focusing on two enzyme families, we expressed and purified over 500 natural and generated sequences with 70-90% identity to the most similar natural sequences to benchmark computational metrics for predicting in vitro enzyme activity. Over three rounds of experiments, we developed a computational filter that improved the rate of experimental success by 50-150%. The proposed metrics and models will drive protein engineering research by serving as a benchmark for generative protein sequence models and helping to select active variants for experimental testing.

Improving functional protein generation via foundation model-derived latent space likelihood optimization

A variety of deep generative models have been adopted to perform de novo functional protein generation. Compared to 3D protein design, sequence-based generation methods, which aim to generate amino acid sequences with desired functions, remain a major approach for functional protein generation due to the abundance and quality of protein sequence data, as well as the relatively low modeling complexity for training. Although these models are typically trained to match protein sequences from the training data, exact matching of every amino acid is not always essential. Certain amino acid changes (e.g., mismatches, insertions, and deletions) may not necessarily lead to functional changes. This suggests that maximizing the training data likelihood beyond the amino acid sequence space could yield better generative models. Pre-trained protein large language models (PLMs) like ESM2 can encode protein sequences into a latent space, potentially serving as functional validators. We propose training functional protein sequence generative models by simultaneously optimizing the likelihood of training data in both the amino acid sequence space and the latent space derived from a PLM. This training scheme can also be viewed as a knowledge distillation approach that dynamically re-weights samples during training. We applied our method to train GPT-like models (i.e., autoregressive transformers) for antimicrobial peptide (AMP) and malate dehydrogenase (MDH) generation tasks. Computational experiments confirmed that our method outperformed various deep generative models (e.g., generative adversarial net, variational autoencoder, and GPT model without the proposed training strategy) on these tasks, demonstrating the effectiveness of our multi-likelihood optimization strategy.

Bayesian estimation of muscle mechanisms and therapeutic targets using variational autoencoders

Cardiomyopathies, often caused by mutations in genes encoding muscle proteins, are traditionally treated by phenotyping hearts and addressing symptoms post irreversible damage. With advancements in genotyping, early diagnosis is now possible, potentially introducing earlier treatment. However, the intricate structure of muscle and its myriad proteins make treatment predictions challenging. Here, we approach the problem of estimating therapeutic targets for a mutation in mouse muscle using a spatially explicit half sarcomere muscle model. We selected nine rate parameters in our model linked to both small molecules and cardiomyopathy-causing mutations. We then randomly varied these rate parameters and simulated an isometric twitch for each combination to generate a large training data set. We used this data set to train a conditional variational autoencoder, a technique used in Bayesian parameter estimation. Given simulated or experimental isometric twitches, this machine learning model is able to then predict the set of rate parameters that are most likely to yield that result. We then predict the set of rate parameters associated with twitches from control mice with the cardiac troponin C (cTnC) I61Q variant and control twitches treated with the myosin activator Danicamtiv, as well as model parameters that recover the abnormal I61Q cTnC twitches.

Thermostable bacterial L-asparaginase for polyacrylamide inhibition and in silico mutational analysis

The L-asparaginase (ASPN) enzyme has received recognition in various applications including acrylamide degradation in the food industry. The synthesis and application of thermostable ASPN enzymes is required for its use in the food sector, where thermostable enzymes can withstand high temperatures. To achieve this goal, the bacterium Bacillus subtilis was isolated from the hot springs of Tapovan for screening the production of thermostable ASPN enzyme. Thus, ASPN with a maximal specific enzymatic activity of 0.896 U/mg and a molecular weight of 66 kDa was produced from the isolated bacteria. The kinetic study of the enzyme yielded a Km value of 1.579 mM and a Vmax of 5.009 µM/min with thermostability up to 100 min at 75 °C. This may have had a positive indication for employing the enzyme to stop polyacrylamide from being produced. The current study has also been extended to investigate the interaction of native and mutated ASPN enzymes with acrylamide. This concluded that the M10 (with 10 mutations) has the highest protein and thermal stability compared to the wild-type ASPN protein sequence. Therefore, in comparison to a normal ASPN and all other mutant ASPNs, M10 is the most favorable mutation. This research has also demonstrated the usage of ASPN in food industrial applications.

Activation of recombinases at specific DNA loci by zinc-finger domain insertions

Recombinases have several potential advantages as genome editing tools compared to nucleases and other editing enzymes, but the process of engineering them to efficiently recombine predetermined DNA targets demands considerable investment of time and labor. Here we sought to harness zinc-finger DNA-binding domains (ZFDs) to program recombinase binding by developing fusions, in which ZFDs are inserted into recombinase coding sequences. By screening libraries of hybrid proteins, we optimized the insertion site, linker length, spacing and ZFD orientation and generated Cre-type recombinases that remain dormant unless the insertionally fused ZFD binds its target site placed in the vicinity of the recombinase binding site. The developed fusion improved targeted editing efficiencies of recombinases by four-fold and abolished measurable off-target activity in mammalian cells. The ZFD-dependent activity is transferable to a recombinase with relaxed specificity, providing the means for developing fully programmable recombinases. Our engineered recombinases provide improved genome editing tools with increased precision and efficiency.

Bayesian Estimation of Muscle Mechanisms and Therapeutic Targets Using Variational Autoencoders

Cardiomyopathies, often caused by mutations in genes encoding muscle proteins, are traditionally treated by phenotyping hearts and addressing symptoms post irreversible damage. With advancements in genotyping, early diagnosis is now possible, potentially introducing earlier treatment. However, the intricate structure of muscle and its myriad proteins make treatment predictions challenging. Here we approach the problem of estimating therapeutic targets for a mutation in mouse muscle using a spatially explicit half sarcomere muscle model. We selected 9 rate parameters in our model linked to both small molecules and cardiomyopathy-causing mutations. We then randomly varied these rate parameters and simulated an isometric twitch for each combination to generate a large training dataset. We used this dataset to train a Conditional Variational Autoencoder (CVAE), a technique used in Bayesian parameter estimation. Given simulated or experimental isometric twitches, this machine learning model is able to then predict the set of rate parameters which are most likely to yield that result. We then predict the set of rate parameters associated with twitches from control mice with the cardiac Troponin C (cTnC) I61Q variant and control twitches treated with the myosin activator Danicamtiv, as well as model parameters that recover the abnormal I61Q cTnC twitches.

Reaching New Heights in Genetic Code Manipulation with High Throughput Screening

The chemical and physical properties of proteins are limited by the 20 canonical amino acids. Genetic code manipulation allows for the incorporation of noncanonical amino acids (ncAAs) that enhance or alter protein functionality. This review explores advances in the three main strategies for introducing ncAAs into biosynthesized proteins, focusing on the role of high throughput screening in these advancements. The first section discusses engineering aminoacyl-tRNA synthetases (aaRSs) and tRNAs, emphasizing how novel selection methods improve characteristics including ncAA incorporation efficiency and selectivity. The second section examines high-throughput techniques for improving protein translation machinery, enabling accommodation of alternative genetic codes. This includes opportunities to enhance ncAA incorporation through engineering cellular components unrelated to translation. The final section highlights various discovery platforms for high-throughput screening of ncAA-containing proteins, showcasing innovative binding ligands and enzymes that are challenging to create with only canonical amino acids. These advances have led to promising drug leads and biocatalysts. Overall, the ability to discover unexpected functionalities through high-throughput methods significantly influences ncAA incorporation and its applications. Future innovations in experimental techniques, along with advancements in computational protein design and machine learning, are poised to further elevate this field.

Dynamics in Cre-loxP site-specific recombination

Cre recombinase is a phage-derived enzyme that has found utility for precise manipulation of DNA sequences. Cre recognizes and recombines pairs of loxP sequences characterized by an inverted repeat and asymmetric spacer. Cre cleaves and religates its DNA targets such that error-prone repair pathways are not required to generate intact DNA products. Major obstacles to broader applications are lack of knowledge of how Cre recognizes its targets, and how its activity is controlled. The picture emerging from high resolution methods is that the dynamic properties of both the enzyme and its DNA target are important determinants of its activity in both sequence recognition and DNA cleavage. Improved understanding of the role of dynamics in the key steps along the pathway of Cre-loxP recombination should significantly advance our ability to both redirect Cre to new sequences and to control its DNA cleavage activity in the test tube and in cells.

Computational tools for plant genomics and breeding

Plant genomics and crop breeding are at the intersection of biotechnology and information technology. Driven by a combination of high-throughput sequencing, molecular biology and data science, great advances have been made in omics technologies at every step along the central dogma, especially in genome assembling, genome annotation, epigenomic profiling, and transcriptome profiling. These advances further revolutionized three directions of development. One is genetic dissection of complex traits in crops, along with genomic prediction and selection. The second is comparative genomics and evolution, which open up new opportunities to depict the evolutionary constraints of biological sequences for deleterious variant discovery. The third direction is the development of deep learning approaches for the rational design of biological sequences, especially proteins, for synthetic biology. All three directions of development serve as the foundation for a new era of crop breeding where agronomic traits are enhanced by genome design.

Engineering spacer specificity of the Cre/loxP system

Translational research on the Cre/loxP recombination system focuses on enhancing its specificity by modifying Cre/DNA interactions. Despite extensive efforts, the exact mechanisms governing Cre discrimination between substrates remains elusive. Cre recognizes 13 bp inverted repeats, initiating recombination in the 8 bp spacer region. While literature suggests that efficient recombination proceeds between lox sites with non-loxP spacer sequences when both lox sites have matching spacers, experimental validation for this assumption is lacking. To fill this gap, we investigated target site variations of identical pairs of the loxP 8 bp spacer region, screening 6000 unique loxP-like sequences. Approximately 84% of these sites exhibited efficient recombination, affirming the plasticity of spacer sequences for catalysis. However, certain spacers negatively impacted recombination, emphasizing sequence dependence. Directed evolution of Cre on inefficiently recombined spacers not only yielded recombinases with enhanced activity but also mutants with reprogrammed selective activity. Mutations altering spacer specificity were identified, and molecular modelling and dynamics simulations were used to investigate the possible mechanisms behind the specificity switch. Our findings highlight the potential to fine-tune site-specific recombinases for spacer sequence specificity, offering a novel concept to enhance the applied properties of designer-recombinases for genome engineering applications.

Bacteriophage lambda site-specific recombination

The site-specific recombination pathway of bacteriophage λ encompasses isoenergetic but highly directional and tightly regulated integrative and excisive reactions that integrate and excise the vial chromosome into and out of the bacterial chromosome. The reactions require 240 bp of phage DNA and 21 bp of bacterial DNA comprising 16 protein binding sites that are differentially used in each pathway by the phage-encoded Int and Xis proteins and the host-encoded integration host factor and factor for inversion stimulation proteins. Structures of higher-order protein-DNA complexes of the four-way Holliday junction recombination intermediates provided clarifying insights into the mechanisms, directionality, and regulation of these two pathways, which are tightly linked to the physiology of the bacterial host cell. Here we review our current understanding of the mechanisms responsible for regulating and executing λ site-specific recombination, with an emphasis on key studies completed over the last decade.

Machine learning for functional protein design

Recent breakthroughs in AI coupled with the rapid accumulation of protein sequence and structure data have radically transformed computational protein design. New methods promise to escape the constraints of natural and laboratory evolution, accelerating the generation of proteins for applications in biotechnology and medicine. To make sense of the exploding diversity of machine learning approaches, we introduce a unifying framework that classifies models on the basis of their use of three core data modalities: sequences, structures and functional labels. We discuss the new capabilities and outstanding challenges for the practical design of enzymes, antibodies, vaccines, nanomachines and more. We then highlight trends shaping the future of this field, from large-scale assays to more robust benchmarks, multimodal foundation models, enhanced sampling strategies and laboratory automation.

Advances in generative modeling methods and datasets to design novel enzymes for renewable chemicals and fuels

Biotechnology has revolutionized the development of sustainable energy sources by harnessing biomass as a feedstock for energy production. However, challenges such as recalcitrant feedstocks and inefficient metabolic pathways hinder the large-scale integration of renewable energy systems. Enzyme engineering has emerged as a powerful tool to address these challenges by enhancing enzyme activity, specificity, and stability. Generative machine learning (ML) models have shown great promise in accelerating protein design, allowing for the generation of novel protein sequences with desired properties by navigating vast spaces. This review paper aims to summarize the state of the art in generative models for protein design and how they can be applied to bioenergy applications, including the underlying architectures and training strategies. Additionally, it highlights the importance of high-quality datasets for training and evaluating generative models, organizes available datasets for generative protein design, and discusses the potential of applying generative models to strain design for bioenergy production.

Quantification of evolved DNA-editing enzymes at scale with DEQSeq

We introduce DEQSeq, a nanopore sequencing approach that rationalizes the selection of favorable genome editing enzymes from directed molecular evolution experiments. With the ability to capture full-length sequences, editing efficiencies, and specificities from thousands of evolved enzymes simultaneously, DEQSeq streamlines the process of identifying the most valuable variants for further study and application. We apply DEQSeq to evolved libraries of Cas12f-ABEs and designer-recombinases, identifying variants with improved properties for future applications. Our results demonstrate that DEQSeq is a powerful tool for accelerating enzyme discovery and advancing genome editing research.

In vitro continuous protein evolution empowered by machine learning and automation

Directed evolution has become one of the most successful and powerful tools for protein engineering. However, the efforts required for designing, constructing, and screening a large library of variants can be laborious, time-consuming, and costly. With the recent advent of machine learning (ML) in the directed evolution of proteins, researchers can now evaluate variants in silico and guide a more efficient directed evolution campaign. Furthermore, recent advancements in laboratory automation have enabled the rapid execution of long, complex experiments for high-throughput data acquisition in both industrial and academic settings, thus providing the means to collect a large quantity of data required to develop ML models for protein engineering. In this perspective, we propose a closed-loop in vitro continuous protein evolution framework that leverages the best of both worlds, ML and automation, and provide a brief overview of the recent developments in the field.

Precise excision of HTLV-1 provirus with a designer-recombinase

The human T cell leukemia virus type 1 (HTLV-1) is a pathogenic retrovirus that persists as a provirus in the genome of infected cells and can lead to adult T cell leukemia (ATL). Worldwide, more than 10 million people are infected and approximately 5% of these individuals will develop ATL, a highly aggressive cancer that is currently incurable. In the last years, genome editing tools have emerged as promising antiviral agents. In this proof-of-concept study, we use substrate-linked directed evolution (SLiDE) to engineer Cre-derived site-specific recombinases to excise the HTLV-1 proviral genome from infected cells. We identified a conserved loxP-like sequence (loxHTLV) present in the long terminal repeats of the majority of virus isolates. After 181 cycles of SLiDE, we isolated a designer-recombinase (designated RecHTLV), which efficiently recombines the loxHTLV sequence in bacteria and human cells with high specificity. Expression of RecHTLV in human Jurkat T cells resulted in antiviral activity when challenged with an HTLV-1 infection. Moreover, expression of RecHTLV in chronically infected SP cells led to the excision of HTLV-1 proviral DNA. Our data suggest that recombinase-mediated excision of the HTLV-1 provirus represents a promising approach to reduce proviral load in HTLV-1-infected individuals, potentially preventing the development of HTLV-1-associated diseases.

Bioengineered Enzymes and Precision Fermentation in the Food Industry

Enzymes have been used in the food processing industry for many years. However, the use of native enzymes is not conducive to high activity, efficiency, range of substrates, and adaptability to harsh food processing conditions. The advent of enzyme engineering approaches such as rational design, directed evolution, and semi-rational design provided much-needed impetus for tailor-made enzymes with improved or novel catalytic properties. Production of designer enzymes became further refined with the emergence of synthetic biology and gene editing techniques and a plethora of other tools such as artificial intelligence, and computational and bioinformatics analyses which have paved the way for what is referred to as precision fermentation for the production of these designer enzymes more efficiently. With all the technologies available, the bottleneck is now in the scale-up production of these enzymes. There is generally a lack of accessibility thereof of large-scale capabilities and know-how. This review is aimed at highlighting these various enzyme-engineering strategies and the associated scale-up challenges, including safety concerns surrounding genetically modified microorganisms and the use of cell-free systems to circumvent this issue. The use of solid-state fermentation (SSF) is also addressed as a potentially low-cost production system, amenable to customization and employing inexpensive feedstocks as substrate.

Combining different CRISPR nucleases for simultaneous knock-in and base editing prevents translocations in multiplex-edited CAR T cells

BACKGROUND

Multiple genetic modifications may be required to develop potent off-the-shelf chimeric antigen receptor (CAR) T cell therapies. Conventional CRISPR-Cas nucleases install sequence-specific DNA double-strand breaks (DSBs), enabling gene knock-out or targeted transgene knock-in. However, simultaneous DSBs provoke a high rate of genomic rearrangements which may impede the safety of the edited cells.

RESULTS

Here, we combine a non-viral CRISPR-Cas9 nuclease-assisted knock-in and Cas9-derived base editing technology for DSB free knock-outs within a single intervention. We demonstrate efficient insertion of a CAR into the T cell receptor alpha constant (TRAC) gene, along with two knock-outs that silence major histocompatibility complexes (MHC) class I and II expression. This approach reduces translocations to 1.4% of edited cells. Small insertions and deletions at the base editing target sites indicate guide RNA exchange between the editors. This is overcome by using CRISPR enzymes of distinct evolutionary origins. Combining Cas12a Ultra for CAR knock-in and a Cas9-derived base editor enables the efficient generation of triple-edited CAR T cells with a translocation frequency comparable to unedited T cells. Resulting TCR- and MHC-negative CAR T cells resist allogeneic T cell targeting in vitro.

CONCLUSIONS

We outline a solution for non-viral CAR gene transfer and efficient gene silencing using different CRISPR enzymes for knock-in and base editing to prevent translocations. This single-step procedure may enable safer multiplex-edited cell products and demonstrates a path towards off-the-shelf CAR therapeutics.

Thermostability enhancement and insight of L-asparaginase from Mycobacterium sp. via consensus-guided engineering

Acrylamide alleviation in food has represented as a critical issue due to its neurotoxic effect on human health. L-Asparaginase (ASNase, EC 3.5.1.1) is considered a potential additive for acrylamide alleviation in food. However, low thermal stability hinders the application of ASNase in thermal food processing. To obtain highly thermal stable ASNase for its industrial application, a consensus-guided approach combined with site-directed saturation mutation (SSM) was firstly reported to engineer the thermostability of Mycobacterium gordonae L-asparaginase (GmASNase). The key residues Gly97, Asn159, and Glu249 were identified for improving thermostability. The combinatorial triple mutant G97T/N159Y/E249Q (TYQ) displayed significantly superior thermostability with half-life values of 61.65 ± 8.69 min at 50 °C and 5.12 ± 1.66 min at 55 °C, whereas the wild-type was completely inactive at these conditions. Moreover, its T_m value increased by 8.59 °C from parent wild-type. Interestingly, TYQ still maintained excellent catalytic efficiency and specific activity. Further molecular dynamics and structure analysis revealed that the additional hydrogen bonds, increased hydrophobic interactions, and favorable electrostatic potential were essential for TYQ being in a more rigid state for thermostability enhancement. These results suggested that our strategy was an efficient engineering approach for improving fundamental properties of GmASNase and offering GmASNase as a potential agent for efficient acrylamide mitigation in food industry. KEY POINTS: • The thermostability of GmASNase was firstly improved by consensus-guided engineering. • The half-life and T_m value of triple mutant TYQ were significantly increased. • Insight on improved thermostability of TYQ was revealed by MD and structure analysis.