Systematic characterization of genome editing in primary T cells reveals proximal genomic insertions and enables machine learning prediction of CRISPR-Cas9 DNA repair outcomes

Ryan T. Leenay, Amirali Aghazadeh, Joseph Hiatt, David Tse, Judd F. Hultquist, Nevan J. Krogan, Zhenqin Wu, Alexander Marson, Andrew P. May, James ZoubioRxiv 2018 · DOI

TL;DR

This paper introduces SPROUT, a machine learning model that accurately predicts CRISPR-Cas9 DNA repair outcomes in primary human T cells. It reveals that proximal genomic regions can be inserted during repair, suggesting a novel mechanism for DNA insertion.

What Problem Does This Paper Address?

CRISPR-Cas9 is a powerful gene editing tool, but predicting the exact DNA repair outcomes after a double-strand break remains a challenge. While previous studies have explored repair mechanisms in cell lines, a systematic characterization in primary human cells, particularly T cells (a key target for gene therapies), was lacking. Understanding these outcomes is crucial for optimizing gene editing strategies and minimizing unintended consequences.

Methods

The researchers developed a machine learning model called SPROUT (CRISPR Repair OUTcome). SPROUT takes the guide RNA sequence and PAM (Protospacer Adjacent Motif) as input and predicts the fraction of insertions/deletions and their average lengths. They used a gradient boosting approach and trained the model on data from 1,521 unique genomic locations edited in primary human CD4+ T cells. The model's performance was evaluated using cross-validation.

Datasets

  • Primary Dataset: 1,521 unique genomic locations edited in primary human CD4+ T cells isolated from 18 healthy blood donors.
  • Validation Datasets: Published sequencing data from 96 unique cut sites in HEK293, K562, and HCT116 cell lines.

Key Insights and Ideas

  • SPROUT's Predictive Power: The SPROUT model demonstrates significant accuracy in predicting DNA repair outcomes, outperforming previous models, especially in primary T cells.
  • Sequence Context is Key: The nucleotides immediately surrounding the cut site (within three bases) are the most influential features for predicting repair outcomes.
  • Unexpected Large Insertions: A surprising finding is the observation of large insertions (over 25 bp) at a low frequency across many targeted loci.
  • Chromatin Proximity Drives Insertions: These large insertions are enriched for sequences physically proximal to the break site, suggesting a role for 3D genome organization and chromatin interactions in mediating these insertions.
  • Cell-Type Specificity: While SPROUT performs well on T cells, its accuracy is lower on immortalized cell lines, indicating that cell-type specific models may be necessary for precise predictions.

Key Results

  • SPROUT achieved an R² of 0.59 for predicting the fraction of indel mutant reads with an insertion and an R² of 0.40 for the fraction of total reads with an insertion in primary T cells.
  • The model predicted edit efficiency with an accuracy of R² = 0.23 in T cells.
  • Large insertions (≥ 25 bp) were found at 92% of targeted sites, though at a low frequency (0.07% of reads).
  • These long insertions were enriched for sequences from enhancers and transcriptionally active genomic regions.
  • HiC data analysis showed that cut sites had significantly more contacts with the origin of these inserted sequences than with random regions, supporting the role of physical proximity.

Limitations and Future Work

  • Short-read sequencing limitations: The study acknowledges that short-range PCR-based sequencing might miss larger genomic rearrangements.
  • Cell-type specificity: SPROUT's performance is better on primary T cells than on immortalized cell lines, suggesting the need for cell-type specific models.
  • Mechanisms of large insertions: While the study suggests a role for chromatin interactions, the precise molecular mechanisms driving these large insertions remain to be fully elucidated.
  • Future directions: The authors suggest further investigation into the mechanisms and prevalence of these proximal genomic insertions and larger rearrangements. They also highlight the need for improved methods for accurate quantification of these events. The SPROUT software is publicly available for further research.

Summary updated

Abstract

Abstract The Streptococcus pyogenes Cas9 (SpCas9) nuclease has become a ubiquitous genome editing tool due to its ability to target almost any location in DNA and create a double-stranded break 1,2 . After DNA cleavage, the break is fixed with endogenous DNA repair machinery, either by non-templated mechanisms (e.g. non-homologous end joining (NHEJ) or microhomology-mediated end joining (MMEJ)), or homology directed repair (HDR) using a complementary template sequence 3,4 . Previous work has shown that the distribution of repair outcomes within a cell population is non-random and dependent on the targeted sequence, and only recent efforts have begun to investigate this further 5–11 . However, no systematic work to date has been validated in primary human cells 5,7 . Here, we report DNA repair outcomes from 1,521 unique genomic locations edited with SpCas9 ribonucleoprotein complexes (RNPs) in primary human CD4+ T cells isolated from multiple healthy blood donors. We used targeted deep sequencing to measure the frequency distribution of repair outcomes for each guide RNA and discovered distinct features that drive individual repair outcomes after SpCas9 cleavage. Predictive features were combined into a new machine learning model, CRI SP R R epair OUT come (SPROUT), that predicts the length and probability of nucleotide insertions and deletions with R 2 greater than 0.5. Surprisingly, we also observed large insertions at more than 90% of targeted loci, albeit at a low frequency. The inserted sequences aligned to diverse regions in the genome, and are enriched for sequences that are physically proximal to the break site due to chromatin interactions. This suggests a new mechanism where sequences from three-dimensionally neighboring regions of the genome can be inserted during DNA repair after Cas9-induced DNA breaks. Together, these findings provide powerful new predictive tools for Cas9-dependent genome editing and reveal new outcomes that can result from genome editing in primary T cells.
BibTeX
@misc{leenay2018systematic,
  author = {Ryan T. Leenay and Amirali Aghazadeh and Joseph Hiatt and David Tse and Judd F. Hultquist and Nevan J. Krogan and Zhenqin Wu and Alexander Marson and Andrew P. May and James Zou},
  title = {{Systematic characterization of genome editing in primary T cells reveals proximal genomic insertions and enables machine learning prediction of CRISPR-Cas9 DNA repair outcomes}},
  year = {2018},
  doi = {10.1101/404947},
  howpublished = {bioRxiv},
}
Figure