Designing Allosteric Protein Switches by In Vivo Directed Evolution and Computational Inference

Project Information

DaVinci-Switches

Grant agreement ID: 101041570

DOI

10.3030/101041570

EC signature date 5 April 2022

Start date 1 September 2022

End date 31 August 2027

Funded under

• European Research Council (ERC)

Total cost

€ 1 619 687,00

EU contribution

€ 1 619 687,00

Coordinated by

RUPRECHT-KARLS-UNIVERSITAET HEIDELBERG
Germany

Periodic Reporting for period 1 - DaVinci-Switches (Designing Allosteric Protein Switches by In Vivo Directed Evolution and Computational Inference)

Reporting period: 2022-09-01 to 2025-02-28

Summary of the context and overall objectives of the project

Proteins are molecular machines that drive all major functions in cells. Controlling the activity of proteins in real time via light or
chemicals is a central goal in synthetic biology. The design of switchable proteins, in particular single-chain, allosteric variants, however,
is a challenging engineering problem thus-far mostly addressed by trial-and-error. DaVinci-Switches takes a radically new, data-driven
perspective to fundamentally advance our understanding of protein allostery and accelerate and eventually rationalize the engineering of
switchable proteins by interfacing synthetic biology with machine learning. We will establish a 'design by directed evolution' approach
to create switchable proteins through receptor and effector fusion followed by phage-assisted in vivo directed evolution using synthetic
gene circuits for selection. We will apply this novel pipeline to a diverse set of effector proteins and monitor the evolutionary process by
next-generation sequencing (Objective 1). In parallel, we will perform an in-depth computational analysis of domain insertions within the
natural protein repertoire. The combined, rich datasets will be used to train machine learning models to infer sequence patterns predictive
of domain insertion tolerance and allosteric coupling between receptor-effector pairs (Objective 2). Finally, we will employ this unique
model to design light- and drug-inducible variants of the Yamanaka cell reprogramming factors. These will provide the foundation
of an Adeno-associated virus-based platform for cyclic, partial in vivo reprogramming of somatic cells with enormous potential for
regenerative medicine, which will be evaluated in a murine model of drug-induced liver injury (Objective 3). DaVinci-Switches harnesses
our key competences in protein engineering, synthetic biology and computation to reveal fundamental principles of allostery and enable
transformative advances in the design of switchable proteins for research and medicine.

Work performed from the beginning of the project to the end of the period covered by the report and main results achieved so far

The primary technical focus of the project has been on developing methods for evolving and engineering allosteric protein switches using phage-assisted evolution, a key objective of this ERC project. We successfully created a robust phage-assisted evolution pipeline to evolve switchable proteins, optimizing plasmid-based positive and negative selection circuits to monitor effector protein activity under specific stimuli. In parallel, we established a scalable method to generate and screen receptor insertion libraries in effector proteins in E. coli (Mathony et al., Advanced Science 10:28, 2023; https://doi.org/10.1002/advs.202303496). Building on this groundwork, we performed iterative evolution on effector-receptor hybrid proteins, notably evolving an AraC-LOV2 hybrid transcription factor. These efforts yielded optimized switchable proteins whose activity levels surpassed the parental proteins by orders of magnitude, demonstrating the robustness and efficiency of our evolution strategies (manuscript in preparation).

We significantly expanded the phage-assisted evolution capabilities by integrating a retron-based indel mutagenesis method, allowing exploration of not only point mutations but also insertions and deletions, thereby dramatically broadening the accessible fitness landscape.

On the computational side, we developed ProDomino, a machine learning model that predicts viable insertion sites for domain insertion-based allosteric control. Trained on semi-synthetic datasets and refined using experimental data from Objective 1, ProDomino enables accurate inference of insertion points and allosteric sites across diverse protein families (Wolf et al., bioRxiv, 2024; https://www.biorxiv.org/content/10.1101/2024.12.04.626757v1). Early applications of the model, including successful validation in complex effectors such as CRISPR-Cas9 and Cas12a, have demonstrated its versatility and significant potential for rational, scalable protein engineering.

Finally, we initiated preparatory steps for creating switchable mammalian transcription factors. We engineered highly potent light-switchable synthetic transcription factors (Gal4-VP64 and dCas9-VPR) using LOV2 domain fusions, establishing a strong technological basis for Objective 3 (Münch et al., originally posted on bioRxiv and now published in Nucleic Acids Research; https://pubmed.ncbi.nlm.nih.gov/39676667/). These achievements lay the foundation for engineering switchable Yamanaka factors to enable inducible cellular reprogramming, extending the scope of the project toward therapeutic and regenerative applications.

Progress beyond the state of the art and expected potential impact (including the socio-economic impact and the wider societal implications of the project so far)

Our project’s breakthroughs in engineering switchable, allosteric protein switches through phage-assisted evolution have the potential to significantly advance synthetic biology and biomedicine. We developed a robust phage-assisted evolution platform that combines positive and negative selection under stimulus control, enabling the iterative evolution of proteins with greatly enhanced switchability and activity. This is exemplified by the successful evolution of the AraC-LOV2 hybrid transcription factor, which exhibited activity improvements by orders of magnitude compared to the parent protein. Furthermore, the integration of a retron-based indel mutagenesis method expanded the mutational landscape beyond point mutations, marking a major step forward for protein evolution methodologies.

On the computational side, our deep learning model ProDomino emerged as a powerful tool for domain insertion engineering. Trained on semi-synthetic datasets and experimentally fine-tuned, the model predicts viable allosteric insertion sites across diverse protein families with high accuracy. Early validations on complex effectors such as CRISPR-Cas9 and Cas12a, as well as additional proteins, demonstrated its broad applicability. Its potential to standardize and streamline domain insertion engineering marks a key advancement in protein switch design.