UNCOVERseq Sets New Benchmark for CRISPR Off-Target Detection with 97.6% Sensitivity
核心洞察
A peer-reviewed Nature Communications study found UNCOVERseq (搜索) achieved 97.6% analytical sensitivity and 78% precision, the strongest combined performance among evaluated CRISPR off-target nomination methods.
The method nominated 60–4,150% more off-targets per guide RNA compared to the original GUIDE-seq protocol while maintaining high precision through a novel tiering and replication framework.
Across six guide RNAs in human hematopoietic stem cells, UNCOVERseq (搜索)-nominated sites successfully translated to confirmed off-target editing for both double-strand break and base editing modalities.
A landmark study published in Nature Communications has established a new analytical performance benchmark for CRISPR off-target nomination, demonstrating that UNCOVERseq (搜索)—a cell-based workflow developed by Integrated DNA Technologies (搜索) (IDT)—delivers the strongest combined sensitivity and precision among evaluated methods. Across two widely used benchmarking guide RNAs (gRNAs), UNCOVERseq achieved 97.6% analytical sensitivity and 78% precision, outperforming both in cellulo methods such as GUIDE-seq (35.9% sensitivity) and in vitro biochemical approaches including CHANGE-seq, CIRCLE-seq, SITE-seq, and AID-seq.
The study, titled "UNCOVERseq (搜索) Enables Sensitive and Controlled Gene Editing Off-Target Nomination Across CRISPR-Cas Modalities and Systems," addresses a critical gap in therapeutic genome editing: the lack of consistent empirical evidence for comparing the analytical performance of off-target nomination methods. "By moving the industry conversation from the number of nominated sites to measurable sensitivity, precision and downstream confirmation burden, this study demonstrates how IDT works alongside scientists to solve one of therapeutic genome editing's hardest challenges," said Gavin Kurgan, Sr. Manager of Bioinformatics Applications Development at IDT and corresponding author of the study.
Methodological Innovations Drive Performance Gains
UNCOVERseq (搜索) builds upon the original GUIDE-seq protocol with several key innovations. The method incorporates an orthogonal double-stranded DNA sequence and modified rhPCR to multiplex primers in close proximity within a single reaction while avoiding primer-dimers. A critical improvement was the introduction of a blocking oligonucleotide targeting the adapter:dsDNA junction, which reduced artifact reads from an average of 37–67% to just 0.3–0.5%, ensuring more than 99% of sequencing reads represented genuine gDNA:dsDNA junctions.
On the computational side, the team implemented a glocal Needleman-Wunsch alignment algorithm, which yielded a median of 30% and 150% more qualified off-target locations than the current and historical GUIDE-seq analysis approaches, respectively. In an end-to-end comparison across 48 gRNAs targeting genes including PDCD1 (搜索), LAG3 (搜索), CTLA4 (搜索), NRP1, IL2RA, and TIGIT (搜索), UNCOVERseq (搜索) nominated an average of 132 off-targets per gRNA compared to 23 using the GUIDE-seq protocol—a range of 60–4,150% more off-targets per gRNA.
Promiscuous Cell Systems as Sensitive Proxy Models
The researchers demonstrated that performing UNCOVERseq (搜索) in a promiscuous HEK293 cell line stably expressing S.p. Cas9 (搜索) (HEK293-Cas9) captured an average of 99.7–100% of total UMI-corrected off-target events found in other cell types, including K562 cells, iPSCs, and primary T-cells. Comparison of shared nomination frequencies showed high rank-order correlations: r = 0.69 for K562, r = 0.61 for iPSCs, and r = 0.63 for primary T-cells. The promiscuous system generated an average of 196–1,560% more candidate targets per gRNA compared to primary T-cells, supporting its use as a more sensitive model for off-target nomination even in translational contexts.
Reproducibility and Tiering Framework
Biological triplicate analysis across four gRNAs revealed that 98.9–99.7% of off-target instances based on frequency were shared between any two replicates, with frequency rank order highly conserved (R² = 0.997). The team developed a tiering system (Tier 1–3) to prioritize off-targets for confirmation based on frequency, reproducibility, and genomic impact. Without biological triplicates, 30–40% of high-priority off-targets were not captured or appropriately prioritized. Critically, tiering in replicate nominations led to more impactful sites being recommended for interrogation without significantly increasing the total number of off-targets requiring confirmation.
Process Controls and Sensitivity Characterization
Using a promiscuous LAG3 (搜索)-targeting gRNA across 12 biological replicates, the researchers identified 723 off-target sites consistently reproduced across all replicates. These were stratified into five Severity Bins based on nomination frequency relative to the on-target event: Bin 1 (>50%), Bin 2 (10–50%), Bin 3 (1–10%), Bin 4 (0.5–1%), and Bin 5 (<0.5%). Confirmation sequencing at median read depths exceeding 21,000x showed that 100% of sites in Severity Bins 1–3 had confirmed editing, with indel frequencies ranging from 88% to 0.02%. The linear decrease in confirmation success at lower bins suggests UNCOVERseq (搜索) nominates sites with frequencies below 0.01% indels.
The study recommends more than 500,000 reads per sample for greater than 50% analytical sensitivity at Severity Bin 5, and more than 2 million reads per sample for approximately 100% analytical sensitivity across all bins.
Comparative Performance Across Methods
In head-to-head comparisons, UNCOVERseq (搜索) successfully nominated 47 of 47 (100%) confirmed off-target sites for the EMX1 (搜索) gRNA and 22 of 23 (95%) for FANCF (搜索). Other in cellulo and in situ assays—GUIDE-seq, OliTag-seq, INDUCE-seq, and BLISS—each failed to detect between 14 and 37 significantly edited sites per gRNA, with cumulative indel frequencies ranging from 0.90% to 4.51%. Among in vitro methods, SITE-seq (83.8%) and AID-seq (80.4%) showed the next-highest sensitivity after UNCOVERseq, though with substantially lower precision (32.1% and 29.7%, respectively).
Translation to Clinically Relevant Systems
The study evaluated six gRNAs spanning a specificity ratio range from 0.00063 to 1.0 in human CD34⁺ hematopoietic stem and progenitor cells (HSPCs) using high-fidelity Cas9 (搜索), adenine base editors (ABE8e), and cytosine base editors (AncBE4max). For double-strand break editing with HiFi Cas9, 2.4–34.5% of UNCOVERseq (搜索)-nominated targets were confirmed per gRNA, with indel editing ranging from 0.06% to 88%. For ABE treatments, 2.4–29% of nominated targets had confirmed base editing (0.53–75.9% cumulative editing), while CBE treatments confirmed 2.4–14.5% of targets (0.51–32.9% cumulative editing).
Notably, off-target DSB indel editing frequencies from HiFi Cas9 (搜索) demonstrated rank-order correlation with off-target base editing frequencies (r = 0.77–0.78), suggesting that DSB-nominated sites are meaningful for interrogation across both indel and base editing modalities.
Translocation Analysis and Editing Burden
Translocation analysis across the six gRNAs in HSPCs revealed detectable translocations only for the PDCD1 (搜索) gRNA (specificity ratio = 0.001), with two of three translocations shared between S.p. Cas9 (搜索) and ABE conditions. The overall estimated translocation burden averaged 1.4% for S.p. Cas9 and 2.2% for ABE conditions. Off-target ratios for the PDCD1 gRNA ranged from 9.4 to 89.0 off-target events per one on-target event, while the higher-specificity CYP2C18 gRNA (specificity ratio = 0.290) showed ratios of 0.07–0.76.
The study found that the overall off-target burden of high-fidelity DSB editors was decreased compared to single-strand break base editors for the cumulative frequency of monitored event types, highlighting that even higher-specificity gRNAs may generate observable off-targets in clinically relevant cell types.
Regulatory Context
The publication arrives as the U.S. Food and Drug Administration considers draft guidance on next-generation sequencing and bioinformatics in non-clinical safety studies for human genome editing products. Although conducted independently, the study's empirical comparison of analytical performance and operating conditions provides timely evidence for developers evaluating how to respond to emerging regulatory expectations regarding off-target assessment in support of IND applications and BLAs.
"The future of gene editing will be advanced by generating clear, reliable evidence that helps developers identify and evaluate potential off-target sites with confidence," Kurgan said. "By working across disciplines to measure sensitivity and precision against empirically confirmed editing—and by establishing the operating parameters that influence those measurements—we hope this work provides scientists, developers and regulators with a practical framework for evaluating off-target nomination strategies with greater confidence."