High-resolution genome-wide mapping of chromosome-arm-scale truncations induced by CRISPR-Cas9 editing

Nathan H. Lazar, Safiye Çelik, Lu Chen, Marta M. Fay, Jonathan Irish, James Jensen, Conor A. Tillinghast, John Urbanik, William P. Bone, Genevieve H. L. Roberts, Christopher C. Gibson, Imran S. HaquebioRxiv 2023 · DOI

Abstract

CRISPR-Cas9 editing is a scalable technology for mapping of biological pathways, but it has been reported to cause a variety of undesired large-scale structural changes to the genome. We performed an arrayed CRISPR-Cas9 scan of the genome in primary human cells, targeting 17,065 genes for knockout with 101,029 guides. High-dimensional phenomics reveals a “proximity bias” in which CRISPR knockouts bear unexpected phenotypic similarity to knockouts of biologically-unrelated genes on the same chromosome arm, recapitulating both canonical genome structure and structural variants. Transcriptomics connects proximity bias to chromosome-arm truncations. Analysis of published large-scale knockout and knockdown experiments confirms that this effect is general across cell types, labs, Cas9 delivery mechanisms, and assay modalities, and suggests proximity bias is caused by DNA double-strand-breaks with cell cycle control in a mediating role. Finally, we demonstrate a simple correction for large-scale CRISPR screens to mitigate this pervasive bias while preserving biological relationships.

Introduction

CRISPR-Cas9 has revolutionized functional genomics, enabling researchers to systematically knock out genes to determine their biological roles. While much attention has been paid to off-target effects—where the Cas9 enzyme cuts unintended genomic locations—this study identifies a pervasive on-target problem: large-scale chromosome arm truncations.

When Cas9 induces a double-strand break (DSB) at a target site, the repair process can fail, leading to the loss of the entire distal portion of the chromosome arm. This truncation results in the deletion of all genes between the cut site and the telomere. Consequently, the cell exhibits a "mixed phenotype" reflecting the loss of multiple genes, creating a systematic "proximity bias" in screening data where genes appear functionally related simply due to genomic proximity.

Methodology and Initial Validation

To investigate this phenomenon, the authors employed high-throughput phenomics, combining automated microscopy with deep learning-based feature extraction. The study utilized several massive datasets to confirm the universality of the effect:

  • rxrx3: A newly generated genome-wide phenomics dataset in primary human umbilical vein endothelial cells (HUVEC), targeting 17,065 genes with over 100,000 unique guides.
  • cpg0016: An independent morphological dataset from the JUMP Cell Painting consortium using U2OS (bone cancer) cells.
  • DepMap (Project Achilles): Publicly available CRISPR and shRNA survival screens across hundreds of cancer cell lines.
  • scPerturb: A collection of single-cell RNA-sequencing datasets used to detect physical gene loss at the molecular level.
Figure 1: CRISPR-Cas9 screening workflow and validation of learned features against known biological pathways.

Initial validation showed that the deep-learning model (rxrx3) effectively captured biological signals, outperforming traditional human-engineered features (cpg0016) in recalling known protein complexes from databases like CORUM and HuMAP (Figure 1C).

The Proximity Bias Phenomenon

The primary discovery is "proximity bias"—a "ghost" similarity between genes located on the same chromosome arm. This bias is so pronounced that it can be used to reconstruct the physical map of the human genome and identify known chromosomal rearrangements in cancer cells.

Figure 2: Genome-wide correlation heatmap demonstrating proximity bias and arm-level phenotypic similarity.

The bias follows a centromere-to-telomere gradient: cuts near the centromere delete more genes, producing a stronger phenotypic signal than cuts near the telomere. In the rxrx3 dataset, the probability of a within-arm gene pair being ranked as more similar than a between-arm pair was 0.71, significantly higher than the 0.5 expected by chance. Molecular analysis via single-cell RNA-seq confirmed that 8-33% of targeted genes exhibited physical copy number loss in adjacent genomic regions.

Mechanisms and Mitigation

The study identifies TP53 as a critical gatekeeper. In healthy cells, functional p53 typically triggers cell cycle arrest or apoptosis in response to broken chromosomes. However, in p53-deficient cells—common in cancer research—these truncated chromosomes persist and propagate, exacerbating the proximity bias. Furthermore, the bias appears specific to technologies that create double-strand breaks. Comparisons between standard CRISPR-Cas9, CRISPR interference (CRISPRi), and shRNA revealed that the bias is absent in knockdown technologies that do not physically sever the DNA.

To address this issue, the authors developed a geometric correction method. By calculating the average phenotype of a chromosome arm and subtracting it from the individual gene data, they successfully mitigated the bias. This normalization preserved real biological signals, such as known protein complexes, while removing the artifacts of genomic proximity.

Conclusion

Large-scale chromosome truncations are a pervasive and systematic byproduct of CRISPR-Cas9 editing. This proximity bias affects various assay modalities and cell types, particularly those with compromised p53 function. While this finding raises safety considerations for clinical gene therapies, the proposed arm-based normalization provides a robust path forward for improving the accuracy of large-scale genetic screens. Future research is needed to determine if this effect occurs in vivo, and the authors suggest exploring alternative technologies like base editors or CRISPRoff, which modify genes without creating the double-strand breaks that trigger these deletions.

Summary updated

Abstract

Summary CRISPR-Cas9 editing is a scalable technology for mapping of biological pathways, but it has been reported to cause a variety of undesired large-scale structural changes to the genome. We performed an arrayed CRISPR-Cas9 scan of the genome in primary human cells, targeting 17,065 genes for knockout with 101,029 guides. High-dimensional phenomics reveals a “proximity bias” in which CRISPR knockouts bear unexpected phenotypic similarity to knockouts of biologically-unrelated genes on the same chromosome arm, recapitulating both canonical genome structure and structural variants. Transcriptomics connects proximity bias to chromosome-arm truncations. Analysis of published large-scale knockout and knockdown experiments confirms that this effect is general across cell types, labs, Cas9 delivery mechanisms, and assay modalities, and suggests proximity bias is caused by DNA double-strand-breaks with cell cycle control in a mediating role. Finally, we demonstrate a simple correction for large-scale CRISPR screens to mitigate this pervasive bias while preserving biological relationships.
BibTeX
@misc{lazar2023high,
  author = {Nathan H. Lazar and Safiye Çelik and Lu Chen and Marta M. Fay and Jonathan Irish and James Jensen and Conor A. Tillinghast and John Urbanik and William P. Bone and Genevieve H. L. Roberts and Christopher C. Gibson and Imran S. Haque},
  title = {{High-resolution genome-wide mapping of chromosome-arm-scale truncations induced by CRISPR-Cas9 editing}},
  year = {2023},
  doi = {10.1101/2023.04.15.537038},
  howpublished = {bioRxiv},
}
Figure