Optimizing Retron Gene Editors Using High-Throughput Libraries and Machine Learning

Retron gene editors are a novel tool in biotechnology, capable of producing single-stranded DNA (ssDNA) within cells. This capability has vast applications in genome editing, molecular barcoding, and more. The efficiency of retron-based technologies, however, hinges on the production of reverse-transcribed DNA (RT-DNA). This article explores how high-throughput variant libraries and machine learning can optimize retron gene editors, particularly focusing on the retron-Eco1 ncRNA system.

Understanding Retrons and Their Role

Retron systems consist of three primary components:

• Reverse Transcriptase (RT): Enzyme that converts RNA into DNA.
• Non-coding RNA (ncRNA): Template for DNA production.
• Effector Protein: Works with the RT and ncRNA to produce DNA.

In biotechnology, retrons are modified to create custom DNA sequences, which can be used for precise genome editing, molecular recording, and other applications. However, the efficiency of DNA production is a limiting factor, necessitating a detailed understanding of how different modifications affect this process.

High-Throughput Variant Libraries

To improve retron technology, researchers developed a library of 3,443 ncRNA variants, each with different modifications. These variants included single-nucleotide substitutions, deletions, and insertions. By testing these variants, the researchers identified which parts of the ncRNA are tolerant to changes and which are critical for RT-DNA production.

Key Findings:

• Single-Nucleotide Substitutions: Most positions showed flexibility, but critical regions around the priming guanosine and the UUU loop required specific sequences.
• Deletions and Insertions: Small modifications were tolerated in some regions, but larger changes often reduced RT-DNA production significantly.
• Structural Requirements: Certain stem-loop structures within the ncRNA are crucial for efficient RT-DNA production.

Machine Learning for Optimization

Using the data from high-throughput experiments, a machine learning model (retDNN) was trained to predict RT-DNA production based on ncRNA sequences. This model allowed the exploration of a broader sequence space than would be feasible experimentally.

Predictions and Validation:

• GC Content: The model predicted that lower GC content in certain stem-loops would increase RT-DNA production. Experimental validation confirmed this, providing a new parameter for optimizing retrons.

Editing Performance in Yeast

The researchers extended their findings to genome editing in yeast, using the retron-Eco1 system combined with CRISPR-Cas9 (called an "editron"). They tested various parameters, including donor length, guide RNA (gRNA) arrangement, and ncRNA structure.

Key Parameters for Optimization:

• Donor Length: Longer donors generally improved editing efficiency.
• Donor Symmetry: Donors centered on the cut site or shifted towards the PAM-proximal side were more effective.
• ncRNA Variants: Certain modifications to the ncRNA, such as extending the a1/a2 region, improved editing outcomes.

Application to Human Cells

To validate their findings in human cells, the researchers tested optimized retron variants in HEK293T cells. They found that the principles derived from yeast experiments applied broadly, improving editing efficiency in human cells as well.

Human Cell Editing Insights:

• Donor Length and Strand: Longer donors and those complementary to the non-target strand yielded better results.
• ncRNA Modifications: Extensions and machine learning-derived variants maintained high editing efficiency.

Conclusion

This study demonstrates that high-throughput variant libraries combined with machine learning can significantly enhance the design and efficiency of retron gene editors. The findings provide a comprehensive set of design principles for optimizing retron-based technologies across different organisms, paving the way for more effective genome editing and other biotechnological applications.

FAQs

Q1. What is a retron system in biotechnology?
A retron system is a tool that produces single-stranded DNA (ssDNA) within cells, consisting of a reverse transcriptase (RT), non-coding RNA (ncRNA), and an effector protein. It has applications in genome editing and molecular recording.

Q2. How does machine learning improve retron gene editors?
Machine learning models, like retDNN, predict RT-DNA production based on ncRNA sequences. This allows researchers to explore a broader sequence space and identify modifications that enhance DNA production efficiency.

Q3: What are the practical applications of optimized retron systems?
Optimized retron systems can be used for precise genome editing, creating custom DNA sequences, molecular barcoding, and recording molecular events within cells.

Q4: What is the importance of donor length in genome editing using retrons?
Longer donor lengths have been shown to improve the efficiency of precise genome editing, as demonstrated in experiments with both yeast and human cells.

Q5: How do retron systems differ from traditional genome editing tools?
Unlike traditional genome editing tools, retron systems produce ssDNA inside cells, which can be used for various applications without the need for external DNA synthesis and transfection.