Genome-wide structural variant landscape following HDR-enhanced CRISPR editing in human hematopoietic stem and progenitor cells
Curation statements for this article:-
Curated by eLife
eLife Assessment
This study provides valuable insights into the structural consequences of CRISPR-Cas9 editing in human hematopoietic stem and progenitor cells, identifying large deletions and genome-wide structural variants using nanopore sequencing and Strand-seq, while demonstrating PolQ inhibition as a potential strategy to reduce large deletions. Although the evidence is generally solid and supports the conclusions, several aspects remain incomplete. In particular, further validation of genome-wide structural alterations, larger single-cell datasets, and functional studies would strengthen the findings and their implications for gene therapy safety and DNA repair mechanisms.
This article has been Reviewed by the following groups
Listed in
- Evaluated articles (eLife)
Abstract
DNA-PK inhibitors (DNA-PKi) have emerged as potent enhancers of homology-directed repair (HDR) in genome editing, however, recent studies have reported that such inhibitors can induce large-scale structural variations (SVs) raising safety concerns. To investigate factors influencing SV induction in human hematopoietic stem and progenitor cells (HSPCs), we employed a combination of on-target long-read nanopore sequencing and unbiased genome wide detection of SVs using Strand-seq. Our results show that editing alone results in a high frequency of large on-target deletions which were modestly increased by DNA-PK inhibition and could be mitigated using DNA polǪ inhibitors. Genome-wide SVs were also induced at low frequencies by editing but were not substantially increased with DNA-PKi. SVs including translocations were, however, increased with multiple simultaneous editing. Our results highlight that edit-induced SVs are not limited to DNA-PKi enhanced protocols, thus careful assessment will be critical to ensure the safety of any edited therapeutic cell product.
Article activity feed
-
eLife Assessment
This study provides valuable insights into the structural consequences of CRISPR-Cas9 editing in human hematopoietic stem and progenitor cells, identifying large deletions and genome-wide structural variants using nanopore sequencing and Strand-seq, while demonstrating PolQ inhibition as a potential strategy to reduce large deletions. Although the evidence is generally solid and supports the conclusions, several aspects remain incomplete. In particular, further validation of genome-wide structural alterations, larger single-cell datasets, and functional studies would strengthen the findings and their implications for gene therapy safety and DNA repair mechanisms.
-
Reviewer #1 (Public review):
Summary:
The authors investigate genomic structural alterations following CRISPR-Cas9 editing of DNMT3A and SRSF2 in human HSPCs, with particular attention to DNA-PK inhibition. They report that CRISPR editing induces large on-target deletions and genome-wide structural variants, while AZD7648 has a modest additional effect. PolQ inhibition reduces large deletions, and simultaneous editing of two loci is associated with increased genomic rearrangements.
Strengths:
The combination of nanopore sequencing and single-cell Strand-seq provides a comprehensive approach to investigating genomic alterations. The improved analytical pipeline for detecting large deletions is a methodological strength. The identification of PolQ inhibition as a potential strategy to reduce undesirable editing outcomes is particularly …
Reviewer #1 (Public review):
Summary:
The authors investigate genomic structural alterations following CRISPR-Cas9 editing of DNMT3A and SRSF2 in human HSPCs, with particular attention to DNA-PK inhibition. They report that CRISPR editing induces large on-target deletions and genome-wide structural variants, while AZD7648 has a modest additional effect. PolQ inhibition reduces large deletions, and simultaneous editing of two loci is associated with increased genomic rearrangements.
Strengths:
The combination of nanopore sequencing and single-cell Strand-seq provides a comprehensive approach to investigating genomic alterations. The improved analytical pipeline for detecting large deletions is a methodological strength. The identification of PolQ inhibition as a potential strategy to reduce undesirable editing outcomes is particularly valuable.
Weaknesses:
The genome-wide structural variant findings would benefit from independent validation, particularly the reported translocations outside the targeted loci. The analysis was performed on cultured HSC-enriched cells rather than functionally validated long-term HSCs, limiting conclusions about therapeutic safety. Additionally, the mechanistic interpretation of PolQ-dependent repair requires further support.
-
Reviewer #2 (Public review):
Summary:
In this work, the authors aim to explore in detail the effects of the DNA-PK inhibitor AZD7648 (AZD), previously applied by themselves and others as a strategy to enhance homology-directed repair in genome editing by shifting the balance of repair events away from non-homologous end joining (NHEJ), which is mediated by DNA-PKcs. While successful at promoting NHEJ and thus desired targeted editing, various reports in the field have described DNA-PK inhibitors as also promoting large-scale structural variants at the target site and elsewhere in the genome
In this study, Cloarec-Ung and team use two approaches, namely bulk long-read nanopore sequencing and, at the single-cell level, STRAND-seq, an approach to map strand-specific read alignments for each chromosome, to define the nature of such SVs in …
Reviewer #2 (Public review):
Summary:
In this work, the authors aim to explore in detail the effects of the DNA-PK inhibitor AZD7648 (AZD), previously applied by themselves and others as a strategy to enhance homology-directed repair in genome editing by shifting the balance of repair events away from non-homologous end joining (NHEJ), which is mediated by DNA-PKcs. While successful at promoting NHEJ and thus desired targeted editing, various reports in the field have described DNA-PK inhibitors as also promoting large-scale structural variants at the target site and elsewhere in the genome
In this study, Cloarec-Ung and team use two approaches, namely bulk long-read nanopore sequencing and, at the single-cell level, STRAND-seq, an approach to map strand-specific read alignments for each chromosome, to define the nature of such SVs in singly (SRSF2 or DNMT3a) and doubly edited hematopoietic stem and progenitor cells (HSPCs) with AZD treatment. For both sequencing approaches, the authors developed custom analysis pipelines to improve/expand SV calling. In the case of long-read sequencing, this pertains to a strategy to discriminate true deletions from sequencing artifact truncated reads, while in the STRAND-seq analysis they developed a strategy to call inter-chromosomal events in addition to the intrachromosomal events previous pipelines have been optimized for.
The authors describe two main findings:
(1) DNA-PK inhibition modestly increases the frequency of on-target large deletions above CRISPR editing, which they also describe as inducing a high frequency of editing on its own.
(2) They show that in the single-edit scenario, genome-wide off-target editing is low-frequency in nature with editing alone and not substantially increased with AZD. Multiple editing events did however, exhibit increased SVs, including translocations, with AZD treatment resolving this. Pursuing the potential to ameliorate the elevation of large on-target deletions via DNA-PK, the authors also explored the inhibition of PolQ-dependent theta-mediated end joining (TMEJ), a process anticipated to increase in response to the repression of NHEJ and thus potentially drive these elevated SVs. They describe the addition of a DNA PolQ inhibitor, showing through long-read sequencing a mitigation of the AZD-dependent and independent large on-target deletions.
Altogether, this work provides validation that CRISPR cutting coupled with HDR repair carries risks of a variety of SV events, that AZD addition can promote these (in particular large on-target deletions and inversions) but modestly, and offers a strategy to mitigate some of these SVs, namely large on-target deletions. This work will be of interest to those in the gene therapy space and working to understand the fundamental control of gene editing strategies and DNA repair mechanisms.
Strengths:
The bilateral anchoring strategy the authors develop represents a generally useful new approach for calling a large category of deletion events on long-read data.
Similarly, the inter-chromosomal chromatin exchange calling pipeline provides a helpful advance in measuring these changes in single-cell STRAND-seq data.
The study provides a comprehensive mapping of all manner of SVs spanning on- and off-target deletions, duplications, inversions, and translocations through two orthogonal approaches and in the same, highly relevant HSPC context.
The use of STRAND-seq provides an unbiased assessment of SVs across the genome and can also capture the maternal vs paternal chromosome contributions and the combinations of SVs that occur at the individual cell level, which offers new insights not previously provided with bulk-level analyses or approaches that are "bait-dependent" and linked to changes that involve the target site.
The PolQ inhibitor (PolQi) studies provide good evidence that some of the unwanted SVs enacted in CRISPR HDR strategies can be offset by preventing TMEJ with this approach. They further convincingly indicate that the combined effects of AZD and PolQi could be an advantageous approach to both achieve highly efficient on-target editing while minimizing detrimental genomic changes, particularly large deletions.
Weaknesses:
Although STRAND-seq provides several unique advantages, certain questions could not be rigorously addressed in the present study. In particular, assessing statistically significant differences between conditions was complicated by the inherent cellular heterogeneity captured by this single-cell technology. Robust quantification of SV patterns generally requires profiling large numbers of cells; however, only modest numbers of cells were available for each editing condition in this study. As a result, the number of detected SV events was insufficient to support robust statistical testing across groups, necessitating more qualitative comparisons of SV patterns and frequencies.
As the authors discuss, another caveat of STRAND-seq is the requirement of carrying out sequencing after HSPCs have been cultured for 7-9 days, which could underestimate the editing changes as cells are lost both by selection and random attrition over this time.
While a bulk sequencing approach, CAST-seq has been used by others in the field for mapping SVs and, if tested here, could have offered a direct comparison to the author's STRAND-seq outcomes as well as a means to benchmark both techniques and cross-compare results to other studies. This strategy could also have allowed for the more direct testing of the full scope of SVs occurring directly following editing.
Relating to the long-read sequencing done, mixing of HSPCs from female and male was performed, and the total number of male vs. female HSPC samples individually edited was small. While there was an intriguing observation of a potential increased rate of large deletions in the HSPCs of male cord blood donors, the study was underpowered and thus could not verify this with statistical significance.
One caveat to the bilateral anchoring strategy used in the long-read pipeline would seem to be the possibility of undercalling telomeric deletions, which may be of particular importance in this study given that SRSF2 is subtelomeric on Chromosome 17.
-
Reviewer #3 (Public review):
Summary:
This work uses the Strand-seq method to evaluate the structural variant consequences of gene editing. It also compares the effects of end-joining inhibitors that may be used to enrich intended homology-directed repair gene editing on-target outcomes.
In this manuscript, Cloarec-Ung et al characterize structural variants following CRISPR gene editing using long-read sequencing and Strand-seq methods. The long-read sequencing results mainly confirm what is expected about long deletions after editing and their modulation by AZD7648 and PolQi2. The main novelty here is the use of Strand-seq to study large structural variants. This is a powerful method to look for SVs in an unbiased manner, albeit at limited scale and resolution. This work promises to provide added perspective as to the possible …
Reviewer #3 (Public review):
Summary:
This work uses the Strand-seq method to evaluate the structural variant consequences of gene editing. It also compares the effects of end-joining inhibitors that may be used to enrich intended homology-directed repair gene editing on-target outcomes.
In this manuscript, Cloarec-Ung et al characterize structural variants following CRISPR gene editing using long-read sequencing and Strand-seq methods. The long-read sequencing results mainly confirm what is expected about long deletions after editing and their modulation by AZD7648 and PolQi2. The main novelty here is the use of Strand-seq to study large structural variants. This is a powerful method to look for SVs in an unbiased manner, albeit at limited scale and resolution. This work promises to provide added perspective as to the possible genotoxicities that might be associated with gene editing. However, the authors claim that gene editing introduces SVs not overlapping the target loci, which is a provocative claim that would benefit from more experimental validation and also contextualization (such as these studies appear to be only with nuclease editing along with p53 inhibition under conditions of enforced cell cycling and then under selection for likely rare cells that can survive enforced proliferation in vitro).
Summary of strengths:
The study uses the novel method Strand-seq to evaluate gene editing SV outcomes. It identifies a burden of SVs (many away from target loci) associated with gene editing.
Summary of weaknesses:
There is a lack of orthogonal validation of the SVs not directly related to target loci. Also, there is limited contextualization of the narrow experimental conditions studied, which may limit generalization.
Major Points:
(1) The results of the Strand-seq are somewhat surprising-89 of 90 translocations identified did NOT involve the target loci. This is a provocative finding and demands orthogonal evidence to support it to ensure it does not merely reflect some technical artifact of this complex procedure and analysis. How would a Cas9 DSB at one locus lead to a translocation not involving the target locus but rather at a separate locus? Could these specific translocations be validated by junction PCR from populations of cells from which these single cells were isolated? Could these or similar translocations be observed with a method that has no PCR or ligation steps (such as ONT Rapid kit WGS from single-cell-derived colonies) to avoid artifacts of PCR- and ligation-mediated library preparation? If verifiable, are these translocations directly related to the editing or perhaps indirectly based on the greater selection pressure for edited cells to survive and proliferate after the editing procedure?
(2. For the genome-wide SV analysis in Figures 4 and 5, it appeared that many of the del/dup and inv SV events were found away from the target loci, some involving "fragile" loci. This finding seems provocative, but again some orthogonal evidence must be presented to substantiate as mentioned above such as targeted sequencing from pools or ideally PCR-free WGS. Again, is this SV burden related to the editing itself or perhaps to the selection pressure for cells to survive and proliferate despite the procedure, which may be especially toxic in the editing conditions?
(3) Why is there a strong signal for "dup_hom" category in a few cells in SRSF2 and Dual+AZD7648 conditions, and the remainder of Figure 4A is essentially white? What does "dup_hom" mean? I couldn't find this defined. To non-technical expert readers, this figure will likely be very difficult to understand. Likewise, in Figure 5A, only one inv event and no del/dup events appear to overlap the target loci, yet in 5B there appear to be more inv and del/dup events at the target loci. Why the disparity between panels? Please consider clarifying these data representations so that non-technical experts can understand the results.
(4) The claim is that dual editing (of two loci, DNMT3A and SRSF2, on two different chromosomes) raises the editing, but not the AZD7648 inhibitor; however, the key control of dual editing without inhibitor is not included, making direct assessment of this effect not possible. For the "dual editing" analysis, it looks like some combinations of conditions were experimentally tested (SRSF2 editing AND DNMT3A editing AND AZD7648 presence), but not others (SRSF2 editing AND DNMT3A editing AND AZD7648 absence). If so, how can the three-way interaction effect be estimated between SRSF2/DNMT3A/AZD7648? It looks like some, but not all, of the models include this three-way interaction effect. Please clarify.
(5) The main risk of editing-induced SVs seems to be in cycling cells (see Boutin 2026 and references therein). It would have been very important to test editing in noncycling cells to compare the SV landscape with the Strand-seq method. At least in the abstract and discussion, the caveat should be clearly made that SV outcomes were measured in "cycling HSPCs" and that these SVs may be an editing liability limited to dividing cells. None of the data presented here informs about the SV risk of editing nondividing cells. Likewise, it would have been informative to study engrafted HSPCs since the SVs in HSPCs that fail to engraft may be of limited clinical significance. This caveat should be more clearly described.
(6) Cell recovery of HSPCs after editing under the various conditions is not reported throughout but is very important to interpreting results. E.g., if the fraction of desired editing may go up with combined end-joining inhibitors (AZD7648 and PolQi2), it may be offset if recovery of viable/clonogenic/engraftable HSPCs goes down. If the cell recovery varies across conditions, the findings on SV burden may reflect the selection pressure as much as the editing condition itself.
(7) It is unclear why p53 inhibition was used in all the experiments. This seems to be an important confounder and must be more clearly acknowledged throughout, including the discussion section. The SV landscape of gene editing with intact p53 would be very important information for the field and seems to be missing here.
(8) Base and prime editing harnessing nickase versions of Cas9 are now widely used for gene editing. But this work is only about nuclease editing with p53 inhibition, which seems like a narrow use case for gene editing. This limitation should be clarified in the introduction and discussion.
-
-
-
-