From R-loops to Genomes

Abstract

Editing genes has become much easier since the discovery of CRISPRCas proteins such as Cas9. Cas9 can be directed to cut almost any location on the genome by supplying it with a short guide RNA (gRNA) with a nucleotide sequence that matches the DNA target. With tools like Cas9, scientists can study disease, improve crops and develop therapies. Despite the apparent simplicity of Cas9based gene editing, it remains challenging to predict if a chosen gRNA causes Cas9 to cut DNA at the intended location, and if it risks cutting other locations too. A common strategy for Cas9 activity prediction is to collect large datasets describing which gRNAs cause which edits in a particular cell, and to train machine learning models on the observed patterns. However, such empirical models rapidly lose predictive power when applied to other cellular contexts or other experimental conditions. An alternative approach is to isolate Cas9, gRNA, and DNA from a cell, measure their dynamics with high precision, and thus learn about the molecular interactions that govern Cas9 activity in any context. With this molecular knowledge, one can formulate rational and general predictions about activity across cells and experiments.

This thesis uses biophysical modeling to translate understanding of Cas9’s molecular mechanisms to prediction of its genome editing activity in cells. Chapter 1 introduces CRISPR, outlining its origin as an immune system in bacteria and archaea, its transformation into a DNA editing toolkit, and the wide range of genetic engineering applications it has enabled. It discusses the main challenges to successful CRISPR applications, and explains how biophysical models can support gRNA selection and other design choices. Chapter 2 describes mathematically how Cas9 dynamics differ between cellfree experiments (in vitro) and living cells (in vivo) due to differences in Cas9 availability and the processes of target search and recognition. By connecting the two settings, the framework allows the findings of later chapters to be generalized across experimental and cellular conditions. Chapter 3 introduces CRISPRzip, a model of Cas9 target recognition that integrates the physics of gRNADNA interactions, and thus explains the variation in activity across gRNA and DNA sequences. Moreover, CRISPRzip incorporates the effects of Cas9 concentration and DNA twisting (supercoiling) on how Cas9 binds and cuts DNA. Because CRISPRzip adapts to gRNA sequence and to environment conditions like concentration and supercoiling, it provides a flexible basis for gRNA design even when the application setting deviates from the experiments used for training. Chapter 4 presents an experimental approach to quantify the effects of DNA supercoiling across thousands of DNA sequences. While these results can largely be explained by CRISPRzip, they also reveal a complex interplay between supercoiling and sequence that calls for further study. Chapter 5 extends the molecular description of Cas9 to the scale of the full genome. Using analytical models of target search and numerical predictions from CRISPRzip, it shows that prediction of ontarget and major offtarget activity does not require explicit consideration of the surrounding genome sequence under typical conditions. Also, it formulates a theory that connects different Cas9 delivery strategies and shows how to balance precision and efficiency in gene editing applications. Chapter 6 concludes the thesis by summarizing the main results and presenting an outlook towards biophysicssupported CRISPR activity prediction in cells