Team:SYSU-CHINA/Poster

Semi-rational Evolution of ADAR1 inhibitor
Presented by Team SYSU-CHINA 2020

Yuansi He¹, Jiaxin Wang¹, Wenjie Lu¹, Yating Yu¹, Yechang Liu¹, Rui Zhang²

¹iGEM Student Team Member, ²iGEM Team Primary PI

Abstract
RNA binding proteins (RBPs) play an essential role in tumors and neurodegenerative diseases, while most of them lack effective inhibitors. Since directed evolution has shown its high efficiency in selecting new products, this year we provided a sample combining rational design and directed evolution to obtain dsRNA inhibitors of RBPs and took ADAR1, a dsRNA adenosine deaminase, as an example. In our project, an algorithm-guided model was trained from natural substrates of ADAR1 and used to established a candidate dsRNA library. In cells, dsRNA is forced to compete for binding ADAR1 with an editable stem-loop which has a toxic gene downstream of it. The cells will only survive when endogenous ADAR1 is inhibited by transferred dsRNAs. The intensity of the competition is regulated by IFN-α and Tet-on system. As the experiment progresses, efficient substrates are extracted and used to train the model above for the next round. Through the continuous cycles of this screening process, we can obtain high-efficient inhibitors of ADAR1 efficiently, while this model can also be extended to other RBPs.

Project Goals
  • Explore a new paradigm and a general method of dsRNA directed evolution
  • Build a mathematical modeling for experimental dose analysis
  • Develop an algorithm for RNA feature extraction and analysis using deep learning algorithms as the framework
  • Provide a new idea for the development and screening of RNA drugs
    • Motivation

    Why RBPs?
    RNA-binding proteins (RBPs) are related to many diseases so it is significant to find these RBP's inhibitors. Currently, the way is synthesis or chemical modification which are usually blind and random.

    Why ADAR1?
    ADAR catalyzes A to I editing in human body, among which includes ADAR1. The high expression of it promotes the growth and metastasis of many cancers. The loss of it help overcome resistance to checkpoint blockade, sensitizing tumors to immunotherapy.[1]

    Why directed evolution?
    It is an emerging technology that has been successfully used in many areas. Unlike methods above, it is “directed”. While there is currently a lack of directed evolution strategy for dsRNA.
    Overview

    Establish dsRNA library with the help of algorithm
    Sum up the characteristics of ADAR1 substrate, determine the affinity of specific sequences and predict mutations that can change the affinity of sequences

    Link the result of ADAR1 editing with a selectable trait
    By combining the stem-loop with a toxic gene to block the translation of it. When the stem-loop is edited, it fails and the gene will be expressed, resulting the death of the cells.

    Introduce the dsRNA into the cell to compete with stem-loop for binding ADAR1
    If the affinity of dsRNA were higher, the stem-loop would not be edited so the cells survive, allowing us to pick out the effective substrates. On the contrary, the stem-loop would be broken so the cells died, and the corresponding dsRNA were eliminated.
    Change the selection pressure by exogenous chemical inducers
    By adding exogenous chemical inducers, the affinity requirement for dsRNA increases as the editing activity increases, through which we realize the directed evolution of dsRNA
    Experiment
    Stem-loop and toxic gene (apoptin)
  • Leak expression
    • Positive control: simulated 100% editing, amber termination in stem loop: A→G
    Negative control: simulated 0% editing,18nt miss at the edited part of the stem-loop
    Experimental group: normal [2]
  • Tet-on system
    • Regulate the transcription of the entire stem-loop and toxic gene part


    dsRNA
  • Stability
    • To improve the stability of dsRNA, cyclize dsRNA with the help of endogenous RNA ligase by adding self-cleaving ribozyme and ligation sequence to both its ends[3].
  • Update
    • Error-prone PCR and semi-rational design guided by algorithm.


    Selective Pressure
  • IFN-α
    • Affect the editing level of ADAR1 in a dose-dependent manner [2].
  • DOX
    • Affect the transcription of the stem-loop and toxic gene part.

    Results
    Part: pTRE-GFP, pTRE-GFP-STEM-apoptin
    Fluorescence detection was performed after transfection. The fluorescence intensity of both increased with the increase of DOX concentration, but the introduction of stem-loop seemed to reduce the background expression of tet-on system.
    Model
    Experiment Model
  • Simulate the models of toxin gene expression and the number of dead cells changes over time.
  • Evaluate the evolution process by defining a value called "evolution percentage" as the ratio of the current dsRNA's inhibition capability to the ideal inhibition capability.


    • Feature Extraction
  • Use the K-MERS method and statistical analysis to obtain sequence features with positive and negative training sets respectively.
  • Determine how to mutate dsRNA to obtain sequences with higher affinity.


    • Deep Learnig
  • Build Convolutional Neural Network to extract useful patterns from dsRNA sequences.
  • Provide prior knowledge of the affinity degree between a dsRNA sequence and the substrate.
    • Human Practices
    Interview
  • Researchers from biological companies.
  • Professors in the field of bioethics, cancer treatment, RNA and bioinformatics.
    • Professor Rui Zhang
    Professor Tienming Li
    Professor Jianhua Yang


    Partnership
  • With SJTU-BioX-Shanghai in science popularization and modeling work.

    • Science Popularization
  • Publish articles on Wechat public number.
  • Science popularization lectures & videos.
    • Contributions
  • A design for semi-rational evolution of dsRNA.
  • Construction of components.
  • Development of algorithms for feature extraction and deep learning.
  • Modeling work in assisting experiment.
  • Exploring possibilities of implementing our design in the real world.
    • Future Work
    Experiment
  • Verification and characterization.
  • Close the loop between experiment and algorithm.

    • Modeling Work
  • Substitute the parameters.
  • Explore the prediction model of the secondary structure.
  • Collect more data and compare the performance of neural networks with different structures.

    • Implementation
  • Further connect with potential stakeholders and professionals
    • Acknowledgement&Sponsors
    Acknowledgement
    Prof Rui Zhang
    Prof Tienming Lee
    Prof Jianhua Yang
    Wenbing Yang
    Yulong Song

    Sponsors
    References
    [1]Ishizuka JJ, et al. Loss of ADAR1 in tumours overcomes resistance to immune checkpoint blockade. Nature. 2019. 565(7737): 43-48.
    [2]Fritzell, K., et al. Sensitive ADAR editing reporter in cancer cells enables high-throughput screening of small molecule libraries. Nucleic Acids Research, 2019. 47(4): p. e22-e22.
    [3]Litke, J.L., S.R. Jaffrey. Highly efficient expression of circular RNA aptamers in cells using autocatalytic transcripts. Nature Biotechnology, 2019. 37(6): p. 667-675.