DisTAL-Seq: A TALEN-specific adaptation of DISCOVER-Seq for off-target profiling

Abstract

Programmable guided nucleases have revolutionized genome editing and biomedical research, with transformative potential for gene and cell therapy. Although the widespread adoption of the CRISPR-Cas system has provided deep insights into target recognition and specificity, the behavior of clinically relevant tools like transcription activator-like effector nucleases (TALENs) remains poorly characterized in human cells. To address this gap, we implemented DisTAL-Seq, a TALEN-specific adaptation of the DISCOVER-Seq pipeline, which detects MRE11 recruitment to double-strand breaks (DSBs). Based on the DISCOVER-Seq principle, DisTAL-Seq incorporates alignment logic tailored to TALEN-binding properties, including variable RVD specificity, cleavage offset, and dimerization behavior. Using DisTAL-Seq, we identified and validated on- and off-target sites across diverse TALENs and T cell donors. This unbiased approach revealed key features of TALEN activity in human cells, including number of tolerated mismatches to a target site and relative location of the induced DSB. DisTAL-Seq thus extends DISCOVER-Seq to the TALEN family and provides a robust platform for assessing modifications in enzyme architecture and application contexts on a genome-wide scale, supporting the development of safer and more effective genome editing tools.

Graphical abstract

DisTAL-Seq extends the DISCOVER-Seq platform to enable genome-wide detection of TALEN-induced double-strand breaks in human cells. By incorporating TALEN-specific dimer architecture and binding-score logic, it refines off-target identification and supports safer, more precise therapeutic genome engineering.

Introduction

The clustered regularly interspaced short palindromic repeats (CRISPR)-Cas system and transcription activator-like effector nucleases (TALENs) have emerged as powerful tools for targeted genome editing.^{1,2,3,4,5,6,7} While both techniques have their advantages, the longer DNA recognition of TALENs theoretically offers a potential for increased specificity.^8,9 This feature would make them a suitable option for clinical applications, such as CAR-T cell generation, where precise genome editing is of paramount importance. Additionally, TALEN inspired architectures of CRISPR-Cas have also emerged as a strategy to improve specificity of Cas-based DNA recognition modules.^9,10

TALENs represent a significant improvement over zinc finger nucleases (ZFNs) due to the modular architecture of their DNA recognition compartment that simplifies their design.^2,11,12 Two protein molecules (arms), each recognizing anywhere between 12 and 27 nts of DNA and carrying a monomer of the FokI nuclease domain, can form an active TALEN if brought together by binding on a DNA template.^13,14 Furthermore, the development of FokI mutant variants that are obligate heterodimers (i.e., are inactive as homodimers) restricts the cleavage to positions where both arms bind, greatly improving specificity.^14,15 Combined, these features allow the targeting and cleavage of a defined 24–54 bp sequence in the genome with minimal sequence restrictions in contrast to the 18–20 bp targeting length and motif restrictions commonly imposed by the CRISPR-Cas system.^{5,14,16,17,18,19} Thus, TALENs are particularly suitable for therapeutic genome editing applications, where minimal targeting restrictions and high precision are essential.^20,21,22

Over the past decade, the assessment of engineered nuclease specificity has been a major research focus, leading to the development of several complementary approaches.^23,24 These methods fall into three broad categories: in silico, in vitro, and in cellulo/in vivo. In silico methods rely on computational models to identify sequences similar to the target, often within a defined number of mismatches or leverage machine learning to predict target sites based on large experimental datasets.^{25,26,27,28,29} In vitro methods evaluate nuclease activity on purified templates, offering high sensitivity but often yielding false positives due to the lack of cellular context.^14,30,31 In cellulo and in vivo approaches overcome this limitation by capturing events in cells or animal models, thereby providing more biologically relevant insights into genome editing outcomes. However, these methods are typically more laborious, costly, and may still be biased by our understanding of intracellular processes during editing. Despite these challenges, in cellulo and in vivo approaches remain the most reliable method to detect nuclease double-strand break (DSB) sites. Among those, DISCOVER-Seq has shown the highest positive predictive value, i.e., enabling precise detection of on- and off-target sites with minimal false positives.³²

The DISCOVER-Seq protocol relies on the principle that all DSBs need to be repaired by the cell’s endogenous DNA repair machinery. The MRN complex, consisting of the MRE11, RAD50, and NBS1 proteins, plays an important role in the early processing of double-strand DNA breaks prior to repair by homology-directed recombination (HDR) or non-homologous end joining (NHEJ).³³ By leveraging MRE11, the MRN component important for DNA end resection, as a target for chromatin immunoprecipitation, DISCOVER-Seq can capture DNA breaks at single-base pair resolution. Although originally developed to map blunt DNA breaks caused by SpyCas9, this approach can also determine the cut sites of other nucleases, such as Cas12a, which cause DNA overhangs. Moreover, DISCOVER-Seq does not require the co-delivery of a double-stranded oligodeoxynucleotide (dsODN) tag, used in GUIDE-Seq, which can limit applicability in difficult to reach organs or cells sensitive to such components.³⁴

Improving, optimizing, and adapting existing OT identification pipelines have been essential for addressing emerging needs in genome editing, such as improving specificity, expanding applicability to diverse cell types, and meeting different scale of demands for therapeutic applications. For example, iGUIDE was developed to filter out mispriming artifacts present in GUIDE-Seq, and the DISCOVER-Seq+ pipeline was introduced to enhance DISCOVER-Seq detection sensitivity by inhibiting NHEJ.^35,36 On the other hand, the recent release of an optimized TALEN extension of CAST-Seq, called T-CAST-Seq, underscores the growing recognition of the need for dedicated tools to comprehensively profile the TALEN activity.^37,38

In the present study, we hypothesized that DISCOVER-Seq, when coupled with tailored TALEN-specific adaptations, could serve as a robust tool to profile TALEN on- and OT sites. The resulting discovery of in situ TALEN lesions by sequencing (DisTAL-Seq) pipeline, allows the genome wide and unbiased identification of TALEN off-targets with high accuracy. These adaptations are essential given the distinct DNA-binding properties and dimerization requirements of TALENs compared to CRISPR-based systems. We further show that such datasets can be leveraged to reveal critical insights into TALEN DNA binding and activity in a cell. Our results demonstrate the potential of DisTAL-Seq to advance the development of more precise TALEN-based genome-editing tools compatible with clinical applications.

Results

MRE11 ChIP-seq precisely profiles TALEN-induced DNA cleavage in human cells

DISCOVER-Seq relies on the capture of DSBs during their repair in cells by tracking the recruitment of DNA repair protein MRE11 to DSB sites. Hence, the timing of sample analysis post-DSB induction plays a crucial role in identifying nuclease activity sites, as already repaired sites would go undetected. On the other hand, dynamics of DSB induction, and thus repair, can vary drastically depending on delivery format, with light inducible CRISPR-Cas9 exhibiting MRE11 recruitment within 5 min of induction, and classical ribonucleoprotein (RNP) as early as 4 h after delivery.^39,40 It is thus recommended to establish an optimal time point for analysis.⁴¹ We evaluated on-target (OnT) site enrichment after MRE11 pull-down in cells transfected with mRNA of a TALEN targeting the T cell receptor alpha (TRAC) locus (TRAC-TAL3). Considering that mRNA delivery into cells and subsequent translation might delay the onset of editing events, we collected and analyzed samples at 8, 12, 18, and 24 h post-delivery of TALEN mRNA. OnT enrichment over a control region reached its peak at 12 h and returned to single-digit levels by 24 h (Figure S1A). Based on these findings, we selected the 12-h time point for all subsequent analyses.

MRE11 chromatin immunoprecipitation sequencing (ChIP-seq) analysis in HEK293T cells edited with TALEN mRNA targeting TRAC or the inositol 1,4,5-trisphosphate receptor type 2 (IP3R2) locus revealed next generation sequencing (NGS) reads accumulating at the OnT site and exhibiting the characteristic profile observed in cells edited using Cas9 (Figures 1A and S1B). A closer examination revealed that the forward and reverse reads partially overlapped, similar to the signal observed in Cas12a-edited cells (Figure S1C).³⁹ This profile confirms that the TALEN-induced cut produces 5′ overhanging strands, aligning with the expected DNA products resulting from FokI nuclease cleavage in ZFN and TALEN applications.^13,16

Figure 1.

Genome-wide DNA repair signal profiling after TALEN editing

(A) Genomic alignments from MRE11 ChIP-seq data in T cells. The enrichment of reads at the on-target (OnT) site generates a characteristic DSB signature. TRAC-TAL2 and TRAC-TAL3 target sites are only a few base pairs apart on the genome, as also defined by the alignment of forward and reverse reads in each sample. (B) Venn diagrams comparing nominated genomic sites for nuclease activity, including the always shared OnT site, between HEK293 and T cells. (C) Amplicon sequencing of OnT and top3 nominated OT sites for the IP3R2 TALEN in HEK293T and T cells. Site 3, not identified in T cells, does not exhibit indels. (D) Indel analysis of all IP3R2-TAL and TRAC-TAL2 sites in HEK293T cells. Unedited cells are used as controls. (E) Sites showing less than 0.1% indels in the IP3R2-TAL and TRAC-TAL2 panels are considered false-positives.

In addition to the expected read pile-up, we also observed reads spanning the predicted cleavage site. Sequence inspection of these reads revealed a variety of small insertions and deletions (indels) (Figure S1D). Because the samples were harvested 12 h after electroporation, these indels most likely reflect early repair of TALEN-induced DSBs. The short indel signatures are consistent with classical NHEJ, which is the predominant repair pathway during this time window. Similar transient retention on re-ligated DNA has been reported in both the original DISCOVER-Seq and DISCOVER-Seq+ studies across multiple loci and cell types.^39,40 Together, these results validate MRE11 ChIP-seq as a robust method to identify TALEN activity and capturing the initial repair signatures at DSB sites.

Genome-wide profiling reveals TALEN-induced DSB signatures

To comprehensively profile TALEN-induced DSBs at OT sites, we analyzed sequencing datasets for signals similar to those observed at OnT sites. Our goal was to prioritize high-confidence loci. Therefore, we excluded reads with low mapping quality (MQ), filtered datasets against the Encode ChIP blacklist,⁴² and applied a scoring method to retain genomic sites with high signal to noise ratio (see materials and methods). Using this approach, we identified 55 genomic sites for the IP3R2 TALEN in HEK293T cells, including the OnT site (Figures 1A, 1B, and S2A). For the TRAC targeting TALENs, 161 and 4 target sites were identified for TRAC-TAL2 and TRAC-TAL3, respectively (Figure 1B). Furthermore, analysis of edited T cells revealed overlapping and unique DSB sites compared to HEK293T cells, potentially reflecting cell-type-specific differences in TALEN activity. Among the three OT sites nominated for IP3R2, all corresponded to the OnT and top-ranked OT sites in the HEK293T dataset (Figures 1B and S2A). Amplicon sequencing confirmed a correlation between indels and ranking of these sites across HEK293T and T cells (Figures 1C and S2B). Notably, the relative rank of certain off-targets differed between the two cell types, but their respective positions still aligned with their indel frequencies when corrected for cell-type-specific naming. Additionally, site 3, uniquely identified in HEK293T cells only shows indels in HEK293T cells and not in T cells (Figures 1C and S2B). These results indicate that TALEN-induced DSB signatures can be effectively identified across different cell types, offering valuable insights into the specificity and variability of TALEN activity in varying cellular contexts.

To validate these findings and further refine our understanding of TALEN-induced DSBs, we analyzed all identified loci through targeted experiments in HEK293T cells. Primer panels were designed for each TALEN pair to amplify DSB sites identified in HEK293T and T cells (Figure 1B). Following NGS and bioinformatic analysis, we classified as true nuclease activity sites those exhibiting more than 0.1% indels following background subtraction from the unedited control condition. For TRAC-TAL2, out of 100 sites, 86 showed indels above background levels in cells provided with the TALEN pair mRNA (Figures 1D and 1E). For the IP3R2 TALEN, we observed indels in 47/49 sites (Figures 1D and 1E). In total, approximately 90% (131/147) of the HEK293T identified DSB loci were validated as true OT sites across these two TALENs (Figure 1E). These results highlight a robust initial framework for profiling TALEN activity, offering a valuable tool for assessing OnT and OT effects across various cell types and experimental conditions.

Mismatch scoring and proximity analyses refine TALEN activity predictions

Guided nucleases recognize specific genomic sequences and cleave DNA near their target sites. Therefore, TALEN-binding sites are expected to coincide with the identified TALEN-induced DSBs. However, the impact of mismatches on binding affinity and the acceptable distance of the TALE arms from one another to still yield efficient FokI cleavage are not well defined. To better understand these parameters, we leveraged our NGS datasets capturing the enriched DNA repair signature from HEK293T cells treated with distinct TALEN pairs. We first implemented a scoring system that differentially penalizes mismatches according to their type.^38,43,44 Each dataset was then analyzed for predicted binding sites of the cognate TALEN or a control TALEN (see materials and methods for details). When multiple sequences received equal scores, proximity to the repair site was used to resolve ties.

We first plotted the ratio of total binding sites identified for IP3R2 compared to the control TALEN, based on mismatch number and scoring (Figure 2A). As expected, the number of predicted binding sites allowing up to 12 mismatches was comparable between IP3R2 and control TALENs, consistent with random matches under these relaxed parameters. In contrast, IP3R2 sites were significantly enriched when considering only sites beyond a defined mismatch threshold, typically corresponding to ≤5 mismatches or ≥27 binding score. These results suggest that incorporating a mismatch or binding score filter can improve the discrimination of true TALEN off-targets from background.

Figure 2.

Analysis of TALEN-induced DSB sites

(A) Frequency of IP3R2 and control binding sites found within 100 bps of DNA repair sites identified in IP3R2-edited cells, as a function of max mismatches and minimum binding score. After 8 maximum mismatches or below a binding score of 24, it is equally likely to find the TALEN used for editing (IP3R2) or a control TALEN site. Control TALEN is the number of TRAC-TAL2, TRAC-TAL3, AAVS1-TAL, and CCR5-TAL sites found (ratio 4:1). (B) Fraction of the identified sites within every 10 bp (starting position) window away from the DSB for IP3R2 or control TALEN (averaged for four TALEN controls). (C) Boxplots showing the distribution of the binding predictions away from the cut site. Sequences with a high-binding score (>28) can be found close to the cut site. Sites with a score between 26 and 28 are enriched for proximal to the DSB. The next bin (score of 24–26) includes sites distributed throughout the search window (here 100 bp). Distance is measured from the cut site to the 3′ end of the TALEN target. Control TALEN (AAVS1-TALEN) binding sites do not show the same pattern. (D) Total number of target sites for IP3R2 identified and percentage of NGS validated (true OT) sites for IP3R2 identified. Effect of a binding score filter (left) or distance from DSB (right) on the total and NGS validated sites. (E) OnT editing at the TRAC locus using the TRAC-TAL2, TRAC-TAL3 pairs or combination of their left and right arms. The L2/R3 is a TALEN pair with a spacer of 57 bp.

To analyze spatial distribution, we examined the distance (starting position) of all predicted binding sites relative to the IP3R2 TALEN DSB loci. As expected, the distribution was slightly biased toward sites closer to the DSB, reflecting our prioritization of proximal sites during search. However, enrichment within 20 bp of the cut site was evident for the IP3R2 (cognate) TALEN compared to the control TALEN (Figures 2B and S3). Moreover, predicted sites with high similarity (i.e., ≤5 mismatches or ≥26 were predominantly located near the DSB, with a median distance of 7.5 and 10 bp, respectively [Figures 2C and S3]). This enrichment was not observed in the control dataset. We observed a similar pattern for samples analyzed with TRAC-TAL2 and TRAC-TAL3 (Figures S6 and S7). Consequently, we conclude that a TALEN-binding site located within 40 bp of a DSB (i.e., a 20 bp window from the cut site plus the length of a TALEN arm) is a strong predictor of a true TALEN target site.

Finally, we evaluated the predictive power of mismatch score and distance on the NGS panel. The unfiltered output for IP3R2 included 187 nominated sites, all containing the true positives. Applying a stringent binding score filter (≥26) reduced the number to 102 predictive sites, but led to some false negatives, as approximately 64% of validated off-targets passed this filter (Figure 2D). A more modest filtering step (≥23) resulted in 183 nominations, including all validated OTs. In contrast, applying a simple mismatch filter (e.g., allowing fewer than 8 mismatches) has a more pronounced effect, as approximately 82% of true OT sites are retained under these conditions. Similarly, applying a proximity filter can reduce site nominations, with more than 80% of validated off-targets falling within 15–20 bp from the DSB, out of 142 and 132 nominations, respectively (Figure 2D). These results support a search window of ±40 bp (including the TALEN-binding sites) around the DSB site and underscore that TALENs can tolerate substantial sequence divergence, reinforcing the value of nuanced scoring rather than simple mismatch counting.

Spacer size and orientation determine TALEN-editing efficiency

Dimerization of the two FokI monomers requires their physical proximity, and typically 12–30 bps of separation are used when designing TALENs for genome editing experiments. We exploited the close genomic positioning of the TRAC-TAL2 and TRAC-TAL3 target sites (Figure 1A) to assess the effect of spacer size and TALEN arm orientation in editing efficiency. Specifically, by delivering combinations of TRAC-TAL2 (in short L2, R2) and TRAC-TAL3 (in short L3, R3) arms we evaluated: the impact of increased spacer distance (L2R3), tandem orientations (L2L3 and R2R3), overlapping binding (L3R2) and canonical editing (L2R2 and L3R3). Amplicon sequencing analysis of HEK293T cells edited with these different combinations showed that delivering the canonical TALEN pairs resulted in robust editing efficiency (Figure 2E). However, the distantly positioned L2R3 combination, with a spacer length of 57 bp exhibited a dramatic reduction in editing efficiency, highlighting the critical role of spacer size, even in scenarios of perfect binding sites (Figure 2E). Moreover, no detectable editing occurred when TALENs were delivered in non-canonical orientation, reinforcing the importance of proper arm alignment for effective FokI dimerization (Figure 2E). These results indicate that presence of TALEN homologous sequences outside the ±40 bp window around the DSB site should not be considered for calling a true OT site.

DisTAL-Seq recapitulates known TALEN homodimer contributions to OT activity

TALENs function through dimerization of their FokI nuclease domains to induce DSBs. While heterodimerization of a left and a right TALEN arm extends the targeted sequence and is favored for specificity, TALENs encoding wild-type FokI domains can also form homodimers, potentially expanding the OT space.

To assess whether DisTAL-Seq captures this known feature, we initially classified identified DSBs based on predicted dimer pairing. Within ±50 bp of each target site, we searched for TALEN-binding sequences allowing up to 7 mismatches per arm and a combined mismatch cap of 13. Each site was then classified as (1) heterodimeric (canonical pairing), (2) homodimeric (same arm twice), or (3) single-arm (when only one arm was found) (Figures 3A and S4). Homodimeric configurations were predicted for 75% (6/8) and 44% (16/36) of sites in the IP3R2 and TRAC-TAL2 datasets, respectively.

Figure 3.

Classification of TALEN activity based on heterodimer or homodimer binding

(A) Schematic depictions of predicted binding sites for the IP3R2 TALEN across nominated genomic loci. A canonical distance and orientation of the two sites corresponding to left and right arm recognition is identified at the OnT. Examples of identified sites in nominated off-targets for each four classifications and their number in parentheses; homodimer binding in OT1, heterodimer binding in OT2, single binding site in OT3, or no identified site in OT7. Number of mismatches and spacer length also varies. (B) Top 20 sites, exhibiting high indel % after LR editing (TRAC-TAL2), compared to results after single arm (L or R) editing. Some sites show high background across samples and cannot be assessed. Examples of editing attribution to both arms or single arm effect are shown. Sites exhibiting single arm editing corresponding to <50% LR editing were classified as TALEN heterodimer effect (LR effect). Sites exhibiting single arm editing corresponding to >80% LR editing were classified as TALEN homodimer activity sites (L or R effect). (C) Classification of TRAC-TAL2 and IP3R2-TAL sites. For TRAC-TAL2, 100 out of 142 sites exhibited more than 0.1% indels after editing, and 84 of those could be assigned an LR, LL, or RR mode editing. For IP3R2-TAL most OT sites are due to homodimer effects.

As this in silico analysis did not assign a TALEN pair for approximately 80% of sites (Figure S4), we turned to experimental validation. HEK293T cells were transfected with either both TALEN arms or each arm individually, and indel frequencies were measured across all nominated OT sites. Comparison of indel levels between conditions (left+right vs. left-only vs. right-only) enabled attribution of each site to heterodimeric or homodimeric activity (Figures 3B and S5). For TRAC-TAL2, 50 of the 100 validated OT sites were attributable to heterodimers, 34 to homodimers (9 left, 25 right), and 16 remained ambiguous (Figure 3C and materials and methods). For IP3R2, 15 sites matched heterodimeric pairing, 28 to homodimers (16 left, 12 right), and 2 were undetermined.

These results confirm that DisTAL-Seq can identify contributions from both hetero- and homodimeric activity in TALEN-edited cells, consistent with established knowledge about wild-type FokI cleavage domains. This analysis supports the robustness of our method in capturing known enzyme behaviors and reinforces the importance of dimer architecture in OT prediction.

DisTAL-Seq identifies TALEN OTs in HEK293T and T cells

Having optimized the key parameters and validated its performance across multiple datasets, we next applied DisTAL-Seq as a practical case study to demonstrate its utility in profiling TALEN-induced DSBs at the genome-wide level. We implemented a combined experimental and computational pipeline that generates a consolidated list of genomic positions by prioritizing sites based on predicted TALEN-binding site proximity to the DSB and a binding score. For each nominated target, the editing configuration (e.g., heterodimer or homodimer) can also be inferred supporting systematic classification of TALEN-binding interactions. This approach enables a simplified and interpretable analysis of genome-wide TALEN activity.

Because TALENs act as dimers, requiring two arms to bind adjacent sequences, the diversity of potential configurations poses visualization challenges. To address this, we independently plotted LR (heterodimer) and LL or RR (homodimer) combinations to represent each scenario. For instance, Figure 4A depicts binding score relative to the preferred repeat variable diresidue (RVD) in each TALEN arm, using color coding and positional annotation. Additional features such as spacer length, DISCO score (see materials and methods), MACS score, and strand polarity (“+” for forward strand, “˗” for reverse strand) are included to improve interpretability. By integrating these parameters, DisTAL-Seq provides a structured framework for dissecting TALEN binding across the genome.

Figure 4.

DisTAL-Seq discovery pipeline across cell types and donors

(A) Example of DisTAL-Seq results. At the top are the TALEN RVDs separated by the spacer sequence. Mismatches to each RVD’s preferred nucleotide are color coded. Distance between the binding sites, DISCO score, MACS score, binding score for each TALEN ARM and genomic coordinates are listed for each site. (B) Heatmap showing the number of shared OT sites between HEK293T cells and primary T cells from different donors, edited with AAVS1, IP3R2, TRAC-TAL2, TRAC-TAL3, or CCR5 TALEN pairs. Numbers indicate overlapping sites; darker shades represent higher counts. Dotted boxes group samples edited with the same TALEN pair. (C) Scatterplots showing the percentage of indels (y axis, log scale) as a function of DISCO score (x axis, log scale) for target sites in HEK293T (gray) and primary T cells (orange) for IP3R2-TAL (top) and TRAC-TAL2 (bottom). Each point represents one site.

We applied DisTAL-Seq across different TALEN pairs to compare TALEN-induced DSBs in HEK293T and primary T cells from different donors, identifying both shared and cell type-specific OTs (Figure 4B). HEK293T cells generally exhibited longer nomination lists, which may reflect increased sensitivity in OT detection or greater editing susceptibility. To assess functional relevance in T cells, we validated editing at selected sites using our targeted NGS panels for IP3R2-TAL and TRAC-TAL2. As expected, shared sites showed indel formation in both cell types, confirming their identity as true targets. Some sites initially classified as unique to one cell type also yielded detectable indels in the other, highlighting the importance of replicate experiments and multi-cell-type profiling to compile comprehensive OT repertoires (Figure 4C).

Overall, DisTAL-Seq provides a robust and adaptable framework for characterizing TALEN-induced DSBs. By integrating binding/mismatch scoring, spacer constraints, and donor-specific variation, it enables comprehensive profiling of TALEN activity across diverse biological contexts. These insights can support the refinement of TALEN design and contribute to safer applications in therapeutic genome editing.

Discussion

In this study, we introduced DisTAL-Seq, a specialized adaptation of the DISCOVER-Seq pipeline, designed to profile TALEN-induced DSBs across the genome. This method provides a comprehensive, unbiased approach for identifying both OnT and OT activity, offering a valuable tool for characterizing TALEN specificity in diverse cellular contexts, such as HEK293T and primary T cells. By leveraging MRE11 recruitment as a marker of DSBs, DisTAL-Seq achieves high-resolution mapping of TALEN activity. Combined, our results underscore the potential of DisTAL-Seq as a platform to enhance genome editing safety and efficacy.

By generating NGS datasets from TALEN edited cells, we identified critical parameters influencing TALEN activity in cellulo, including mismatch tolerance and spacer size. Our analysis showed that true OT sites are significantly enriched when mismatch tolerance is restricted to fewer than seven mismatches and a binding score of >24. Moreover, binding sites contained within 40 bp of the identified DSB were more predictive of true OTs. The importance of spacer size was further corroborated by experiments showing that canonical spacer lengths yielded robust editing, while non-canonical configurations or an extended spacer resulted in reduced or undetectable editing. Our in cellulo findings are in agreement with recommendations regarding TALEN design, where a spacer length between 6 and 40 bp is regarded optimal.^14,45 These insights provide a refined framework for evaluating TALEN activity and can guide the development of more accurate binding predictions and true OT prioritization.

DisTAL-Seq can identify TALEN activity independently of TALEN arm conformation. TALENs, designed to function as heterodimers (a left and right arm), can also form homodimers (two left or two right arms), leading to additional OT cleavage. Using target validation via indel NGS, we demonstrated that approximately 40% of validated TALEN off-targets for TRAC-TAL2 and 65% for IP3R2 were attributed to homodimer activity. This significant contribution of homodimers highlights the importance of using obligate heterodimeric FokI cleavage domains to minimize OT effects, when specificity is essential.

Our comparative analysis between HEK293T and T cells demonstrated that DisTAL-Seq effectively identifies shared and cell-type-specific OT sites. The longer nomination lists in HEK293T cells suggest either greater sensitivity in detecting OT effects or an inherent susceptibility to editing. Some OT sites initially identified as unique to one cell type also appeared as validated off-targets in the other, emphasizing the value of profiling across multiple cell types, to increase sensitivity and confidence. Additionally, differences in nominations between T cell donors, indicates that genetic background could be a critical factor influencing TALEN activity. These findings highlight both biological variability and the potential technical limitations of single-replicate or single-donor analyses.

These analyses across TALENs and cell types have also revealed limitations of our understanding in TALEN binding and activity within cells. Restricting search within a short window away from the DSB and limiting total mismatches per arm to 7 results in validated OT sites without a predicted TALEN pair. This indicates that additional features likely influence activity of TALENs in those sites. Examples include increased number of mismatches for RVDs that exhibit partial affinity to alternative DNA bases, possibility of bulges or gaps within the target sequence, and long range interaction of genomic loci. We observed that the use of a weighted binding score enhances discriminatory power, enabling more effective exclusion of non-specific sites. Ultimately, integrating additional contextual factors, such as chromatin accessibility and conformation, could further improve OT detection. Additionally, while the current pipeline development was based on experimental data from HEK293T and T cells, expanding its validation to other clinically relevant cell types and experimental conditions will broaden its utility.

DisTAL-Seq represents a significant addition in TALEN specificity profiling, combining sensitivity, adaptability, and high-resolution detection of genome-wide activity. By incorporating parameters, such as mismatch tolerance and proximity to DSB, it offers a comprehensive tool for optimizing TALEN designs and applications. This approach is poised to accelerate the development of next-generation genome editing tools, ensuring greater safety and precision in clinical and therapeutic contexts.

Materials and methods

Cell culture

HEK-293T cells were obtained from ATCC and were STR profiled. They were cultured in Dulbecco’s modified Eagle’s medium (DMEM, Gibco), supplemented with 10% FBS (Gibco) and 100 μg/mL penicillin/streptomycin (Gibco). CD3⁺ T cells were purified from frozen human peripheral blood Leukopak (StemCell) by negative selection using the EasySep Human T cell Enrichment Kit (StemCell) according to the manufacturer’s instructions and cryopreserved in CryoStor CS5 (StemCell). Purified T cells were activated using TransAct (Miltenyi Biotec) according to the manufacturer’s instructions and cultured in X-VIVO 15 Media (Lonza) supplemented with 5% human AB serum (GeminiBio) and 100 IU/mL human IL-2 (Miltenyi Biotec). The activated T cells were used in electroporation experiments 6 days post-activation as described in the following text. All cells were incubated at 37°C and 5% CO₂.

Genome editing

TALENs targeting the human AAVS1 locus, and IP3R2 and TRAC genes were purchased from Thermo Fisher Scientific as in vitro transcribed mRNA for each TALEN arm. The mRNA encoding CCR5 was in vitro transcribed from the plasmids 1669_pPIX_TALEN-CCR5_L and 1669_pPIX_TALEN-CCR5_R using the HiScribe T7 ARCA mRNA Kit (New England Biolabs, E2060S) according to the manufacturer’s instructions. The resulting mRNA was purified with the miRNeasy Tissue/Cells Advanced Mini Kit (Qiagen, 217604), and concentrations were quantified using a Qubit fluorometer. HEK-293T cells and activated T cells were electroporated with a 4D-Nucleofector X Unit (Lonza). HEK-293T cells were electroporated using program CM-130, SF Cell Line 4D-Nucleofector solution (Lonza). T cells were electroporated using program DS115, P3 Cell Line 4D-Nucleofector solution (Lonza). Electroporations were conducted with either 10 × 10⁶ cells using a total of 7.5 μg TALEN mRNA (3.75 μg per arm) in 100 μL cuvettes or with 1 × 10⁶ cells using a total of 1.5 μg TALEN mRNA (0.75 μg per arm) in a 20 μL 96-well format. For the AAVS1-targeting TALEN in T cells, electroporations in the 20 μL strip format were performed using a total of 3 μg TALEN mRNA.

RNase H-dependent amplification and sequencing

A union of all sites that received a DISCOVER-score across all analyzed samples, minimum of HEK293T and T cells from a single donor, was used as input for the IDT online tool to design the three RNase H-dependent amplification and sequencing (rhAmpSeq) panels. For TALEN-IP3R2, the design returned 53 out of 55 input sites in Tier1, 1 site as individual assay and no available assay design for 1 site. For TALEN-TRAC2, out of 161 input sites the design returned 156 in Tier1, 4 sites in Tier2, and one site in TierX. For TALEN-TRAC3, the design returned 4 out of 4 input sites in Tier1. We included the OnT sites in the panel design and excluded sites not amenable to pooling with the other loci. NGS data were analyzed using CRISPAltRations with a window parameter of −10 to +2.⁴⁶ The indel % output was used to assess editing, and only sites exhibiting at least 400x coverage across all editing conditions were considered for analysis.

DisTAL-Seq-Cell editing

The DISCOVER-Seq protocol was performed according to Wienert et al., 2020. In brief, 10⁷ cells were electroporated with TALEN mRNA then fixed with PFA 12 h after electroporation and snap frozen. For processing, the cell pellets were thawed and treated with lysis buffer before being sheared with a Covaris S2 resulting in genomic DNA fragments around 300 bps in size. The DNA was then incubated overnight with an MRE11 antibody which was pre-bound to Dynabeads. MRE11 bound DNA fragments were purified and digested with proteinase K to remove any leftover proteins. After a final clean-up using the Qiagen MinElute kit, the samples were prepared for NGS using the NEB Next Ultra II kit according to the manufacturer’s instructions. Sequencing was performed on an Illumina NovaSeq 6000 with a depth of 20 million reads per sample.

DisTAL-Seq-Bioinformatic analysis

Reads were aligned to the hg38 reference genome using bowtie2 (v.2.4.4) and default parameters. The Samtools package (v.1.16.1) was used for all alignments to group reads together by their names (collate), fix mates and add mate score tags (fixmate -m), sort by leftmost read coordinate (sort), mark duplicates with an optical duplicate distance of 2,500 and including marking duplicate supplementary alignments (markdup -S -d 2500), index the final BAM file (index), and generate statistics (stats). Macs3 (v.3.0.0a6) was used to call peaks (callpeak -f BAM -g hs -B -q 0.01). A custom blender version adapted to TALEN NGS data was used to process the BAM files. In detail, reads were filtered for duplicates, insert sizes longer than 1 kb and MQ ≤ 25. Blunt ends were identified genome wide and sites with 5 or more blunt ends in a 10 bp window were analyzed further. Sites overlapping the ENCODE blacklist v2 or with a high background read coverage (ratio of blunt end reads to total reads <0.1) were discarded.⁴² Both TALEN arms were searched forward and as reverse complements, for a total of 4 potential arms, in a 100 bp window up- and downstream of the potential DSB. If both TALEN arms were identical, the best alignments for forward and reverse complement were searched individually upstream and downstream of the DSB. If multiple alignments were found for one arm, the one closest to the potential DSB with the fewest mismatches was reported. TALEN-binding scores were calculated using a substitution matrix based on empirical RVD-base preferences. Perfect matches received a score of 2, and mismatches were penalized based on relative binding affinity. For example, an NN(G)-A mismatch was scored as 0.9, while an HD(C)-T mismatch received 0.4.³⁸ This enabled a more nuanced ranking of candidate sites beyond raw mismatch counting. Sites with total binding scores below 23 were excluded in the unfiltered dataset.

If three arms had valid alignments either upstream or downstream of the DSB, the most likely pair was reported, for example arm1 and arm2 would be reported instead of arm1 and reverse complement of arm1. If all four arms had a valid alignment, the more likely pair, as previously mentioned, with the smaller sum of mismatches or the highest sum of mismatch scores was reported. Adjacent hits, due to the staggered cuts, were clustered and the hit with the higher discover score was reported, defined as the number of blunt-end reads in a 10 bp window up- and downstream of the site. The unfiltered reported hits were limited to fewer than 12 mismatches per TALEN arm and fewer than 18 mismatches between both identified arms. For the mismatch score the unfiltered reported hits were limited to mismatch score higher than 23. The filtered reported hits were limited to fewer than 8 mismatches in at least one arm, the arm pair needed to span the potential DSB and at least one arm needed to be identified up- and downstream of the DSB. Final hits were visualized in three separate figures. One for the combination of arm1 and arm2, and one each for arm1 and reverse complement of arm1 and arm2 and reverse complement of arm2.

Data and code availability

All sequencing data generated in this study have been deposited in NCBI Sequence Read Archive (SRA) under BioProject accession number PRJNA1378902 and will be publicly available upon publication. The custom code used to perform DisTAL-Seq analysis is available at https://github.com/mschroed/distal-seq. Additional data supporting the findings of this study are available in the supplemental information or from the corresponding author upon reasonable request.

Acknowledgments

J.E.C. is supported by the NOMIS Foundation, the Lotte und Adolf Hotz-Sprenger Stiftung, the Swiss National Science Foundation (project grants 310030_188858 and 310030_201160), and the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (grant agreement no. 855741, DDREAMM). This work was partially funded by the European Union under grant agreement no. 101070740, 101057438, and 101057659. Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union or the European Health and Digital Executive Agency (HADEA). Neither the European Union nor the granting authority can be held responsible for them.

Author contributions

Z.K. and J.E.C. conceived the study, supervised the project, and wrote the manuscript with input from all authors. L.K., L.V.d.V., L.V.B., and Z.K. performed experiments and generated the primary datasets. M.S. developed the computational pipeline and performed data analysis with support from Z.K. D.H., Y.A., T.P., and S.G. provided experimental support and technical feedback. J.E.C. provided strategic guidance and contributed to manuscript revision and interpretation of results. All authors read and approved the final manuscript.

Declaration of interests

D.H, Y.A., T.P., and S.G. have been employees and shareholders of Allogene Therapeutics, Inc. at the time of this work. The lab of J.E.C. has previously had a funded collaboration with Allogene Therapeutics, Inc. J.E.C. is a co-founder and SAB member of Serac Biosciences and an SAB member of Mission Therapeutics, Relation Therapeutics, Hornet Bio, and Kano Therapeutics.