CRISPECTOR provides accurate estimation of genome editing translocation and off-target activity from comparative NGS data

Abstract

Controlling off-target editing activity is one of the central challenges in making CRISPR technology accurate and applicable in medical practice. Current algorithms for analyzing off-target activity do not provide statistical quantification, are not sufficiently sensitive in separating signal from noise in experiments with low editing rates, and do not address the detection of translocations. Here we present CRISPECTOR, a software tool that supports the detection and quantification of on- and off-target genome-editing activity from NGS data using paired treatment/control CRISPR experiments. In particular, CRISPECTOR facilitates the statistical analysis of NGS data from multiplex-PCR comparative experiments to detect and quantify adverse translocation events. We validate the observed results and show independent evidence of the occurrence of translocations in human cell lines, after genome editing. Our methodology is based on a statistical model comparison approach leading to better false-negative rates in sites with weak yet significant off-target activity.

Fig. 1. Quantifying off-target activity by CRISPECTOR.

a, b Hypothetical indels at the expected cut-site in the M experiments. a Out of 226 off-target sites that we have examined in our experimental data, we have found 31 sites to have indel frequency higher than 0.1% (fraction = 0.001) in the M experiments. M indel frequencies at the cut-site were measured by a direct calculation, not using a comparative set-up. b These 31 noisy off-target sites come from different gRNAs examined, as depicted. (c) CRISPECTOR workflow. CRISPECTOR assigns each read in the Tx and M FASTQ files to a specific locus of interest or a putative translocation. Then, a Bayesian inference classifier accurately estimates the indel editing activity, and a hypergeometric test is performed to detect translocation reads.

Fig. 2. Off-target activity levels estimates by CRISPECTOR.

a Site 51 of RAG1, XT 2p in HEK293-Cas9. High rate of deletions of length 1 at the expected cut-site masks the real editing signal (Tx=17, M = 41, solid circle) if only subtraction is used. Also, the PAM site is GGG. Thus, we can find an alternative cut-site 1-bp to the right of the expected cut-site. CRISPECTOR points to five reads with deletions of length 1 on this alternative cut-site (dashed circle). Editing activity estimation—CRISPECTOR = 0.113%, CRISPResso2 = 0.068% and ampliCan = 0.070%. Note that reference is depicted 3’→5’ since the gRNA cuts the negative strand in this locus. The notation Del k stands for deletions of length k. b Editing activity by CRISPECTOR, ampliCan and CRISPResso2 for RAG1, XT 2p, HEK293-Cas9 sample. For many instances with low editing activity levels, there is a disagreement between tools with respect to the active (≥0.1%) or inactive (<0.1%) classification. In this experiment, CRISPECTOR reported four sites as ≥0.1% (84, 61, 43, and 51, denoted with asterisks), where both ampliCan and CRISPResso2 determined the sites as ≤0.1%. Error bars represent confidence intervals of CRISPECTOR results. This activity was validated by human examination of the individual actual reads (see “Discussion” and Supplementary Note 2). c Tool comparison for the active off-target site classification. Off-target site classification results for CRISPECTOR, CRISPResso2 (CS2), and ampliCan (Amp). In the left column squares, the number of instances with agreement and disagreement in the classification task are depicted. Each square refers to a comparison of two tools. P (Positive) and N (Negative) stand for active and inactive sites, respectively. The two pie charts depict CRISPECTOR and CRISPResso2 classification results on the 19 validated sites. TP, TN, FP, and FN denote true-positive, true-negative, false-positive, and false-negative, respectively. True state determined by the actual reads.

Fig. 3. CRISPECTOR detects and quantifies adverse translocation events.

a Schematic description of all four translocation types. Schematic description of all possible translocations for chromosomes 3 and 10 p arms. The expected cut-sites are colored red and ⊕ denotes the concatenation/fusion of the sequences. The four possible fusion types give rise to structures that are either single-centromeric (A & B), centromere-free (C), or double-centromeric (D). b Translocation reads heatmap for RAG2, XT 2p, HEK293-Cas9. 19 translocations with a Hypergeometric (FDR corrected) p-value < 0.05 are presented in the heatmap with the associated read counts. For example: 225 reads were found between RAG2_5 and RAG2_1, yielding a p-value of 1.53 × 10⁻⁶⁵. c ddPCR Validation of translocation events. Specific translocation or structural variation events were individually experimentally measured using event-specific ddPCR primers and probes. The means of 8 technical repeats are also depicted above (black lines). WRS (Wilcoxon Rank Sum) one-sided p-values to support the higher Tx values measured are: 7.7 × 10⁻⁴, 0.29, 0.027, 0.074, 0.09, and 0.17, ordered as in the figure. Source data is provided as a source data file. d Validation of CRISPECTOR translocation detection process based on synthetic spike-in standards. Edited RAG2, XT 2p, HEK293-Cas9 genomic DNA was spiked-in with serial dilutions (0, 0.016, 0.16, 1.6 and 16%) of RAG2_10/RAG2_7 synthetic fusion construct (see text). Observed numbers of fusion reads for the spiked-in RAG2_10/RAG2_7 construct (in orange) and for the naturally occurring RAG2_1/RAG2_5 (in green) are presented.