IntroductionThousands of genetic diseases could be corrected by precise genomic edits, using tools such as CRISPR-Cas9 to induce perturbations at targeted locations in the genome1,2. However, a fundamental roadblock is our inability to control how those perturbations are repaired3. CRISPR nucleases, base editors, and prime editors perturb DNA in different ways4,5,6,7, but in each case, the editing outcome is ultimately determined by how the cellular DNA repair machinery responds to that perturbation8,9,10. Repair that restores the original sequence instead of editing it is unproductive, and imprecise repair can cause harmful unintended changes3. To ensure that the desired edit occurs in each cell, therapeutic genome editing requires a thorough understanding and control of DNA repair.Surprisingly little is known about DNA repair in postmitotic cells such as neurons, which cannot regenerate yet must withstand an entire lifetime’s worth of DNA damage. This gap in understanding hinders research into many diseases, such as neurodegeneration and aging, and also limits our control over CRISPR editing outcomes. Many neurodegenerative diseases are caused by dominant genetic mutations, making them strong candidates for CRISPR-based gene inactivation11,12,13,14,15,16. Cas9-induced double-strand breaks (DSBs) can disrupt these mutant alleles and reverse disease phenotypes. However, this requires specific DSB repair outcomes that produce the proper insertion/deletion mutations (indels) capable of frameshifting and eliminating the toxic gene product17.Whether the DSB results in a desired indel or not is determined by the competing DSB repair pathways active in the cell (Fig. 1a and Supplementary Fig. 1). In fact, differential expression of even a single DNA repair gene can change a cell’s editing outcome8. DSB repair pathways in nondividing cells likely differ drastically from those in the rapidly-proliferating and transformed cell lines used by most editing studies to date18,19,20,21. Pathways such as homology directed repair (HDR), for example, which are restricted to certain stages of the cell cycle, should be inactive in non-cycling cells22. Furthermore, DSB repair may be particularly unique in neurons, where some early-response genes are activated by the presence of DSBs in their own promoters20, and DSBs have even been implicated in memory formation23. Therefore, the rules of CRISPR editing outcomes may differ in postmitotic neurons compared to the dividing cells that have shaped the literature thus far.Fig. 1: Modeling CRISPR repair outcomes in postmitotic human neurons.a Schematic: Genome editing proteins can perturb DNA, but cellular DNA repair determines the editing outcome. b Timeline of differentiating iPSCs (blue) into neurons (green). After at least 2 weeks of differentiation/maturation, postmitotic neurons are treated with VLPs delivering Cas9 protein (yellow) and sgRNA (orange). c Cas9 VLPs induce DSBs in human iPSC-derived neurons. Representative ICC images of neurons 3 days post-transduction with B2Mg1 VLPs, and age-matched untransduced neurons. Scale bar is 20 µm. Arrows denote examples of DSB foci: yellow puncta co-labeled by γH2AX (red) and 53BP1 (green). Dose: 1 µL VLP (FMLV) per 100 µL media. Additional images shown in Supplementary Fig. 5, with similar results replicated independently in Supplementary Fig. 12. d Genome editing outcomes differ between iPSCs and isogenic neurons. CRISPResso2 analysis of amplicon-NGS, from cells 4 days post-transduction with B2Mg1 VLPs. Dose: 2 µL VLP (HIV) per 100 µL media. Data are averaged across 6 replicate wells per cell type transduced in parallel, and expressed as a percentage of total reads. Thick blue background bars are from iPSCs; thin green foreground bars are from neurons.Full size imageTo test this, in this study, we compare how human induced pluripotent stem cells (iPSCs) and iPSC-derived neurons respond to Cas9-induced DNA damage. Compared to these isogenic dividing cells, neurons accumulate indels over a longer time period and upregulate unexpected DNA repair genes in response to Cas9 exposure. Manipulating this repair response allows us to influence the efficiency and/or precision of genome editing in postmitotic neurons and cardiomyocytes, and in nondividing primary human T cells—adding important new tools to the genome modification toolkit.ResultsVirus-like particles efficiently deliver Cas9 to human iPSC-derived neuronsTo investigate how Cas9-induced DSBs are repaired in neurons, we first needed a platform to deliver controlled amounts of Cas9 into postmitotic human neurons. We used a well-characterized protocol24,25 to differentiate human iPSCs into cortical-like excitatory neurons (Fig. 1b). Immunocytochemistry (ICC) confirmed the purity of these iPSC-derived neurons. Over 99% of cells were Ki67-negative by Day 7 of differentiation, and approximately 95% of cells were NeuN-positive from Day 4 onward (Supplementary Fig. 2). These observations confirm that within one week our cells rapidly become postmitotic, and uniformly express key neuronal markers.While iPSCs and other dividing cells are amenable to electroporation and chemical transfection, transient Cas9 delivery to neurons remains challenging. Recently, virus-like particles (VLPs) inspired by Friend murine leukemia virus (FMLV), human immunodeficiency virus (HIV), and others have been used to successfully deliver CRISPR enzymes to many mouse tissues, including mouse brain26,27,28,29. Unlike viruses, which deliver genomic material into cells, VLPs are engineered to deliver protein cargo such as Cas9. Viruses pseudotyped with the glycoprotein VSVG are known to transduce LDLR-expressing cells, including neurons30, and co-pseudotyping particles with the envelope protein BaEVRless (BRL) has been shown to improve transduction in multiple human cell types31. Therefore, we reasoned that VLPs pseudotyped with VSVG and/or BRL could efficiently transduce human neurons.We produced VLPs containing Cas9 ribonucleoprotein (RNP) to induce DSBs, with or without an mNeonGreen transgene to track transduction. By flow cytometry, we found that multiple types of VLPs effectively delivered cargo to our neurons, with up to 97% efficiency (Supplementary Fig. 3). Additionally, modulating the VLP’s pseudotype or the Cas9’s nuclear localization sequence both greatly impacted delivery efficiency (Supplementary Fig. 4). For subsequent experiments, we proceeded with two particles interchangeably: VSVG-pseudotyped HIV VLPs (also known as enveloped delivery vehicles27), or VSVG/BRL-co-pseudotyped FMLV VLPs. Furthermore, ICC confirmed that Cas9-VLPs successfully induced DSBs in our neurons, co-labeled by markers gamma-H2AX (γH2AX) and 53BP1 (Fig. 1c and Supplementary Fig. 5). This platform to acutely perturb DNA in human neurons enables the study of DNA repair in clinically relevant postmitotic cells.CRISPR repair outcomes differ in neurons compared to dividing cellsTo examine how neurons repair DSBs, we used VLPs to deliver equal doses of Cas9 RNP into human iPSC-derived neurons and genetically identical iPSCs. We selected a single-guide RNA (sgRNA), B2Mg1, that yields a variety of indel types in iPSCs, suggesting it is compatible with multiple DSB repair pathways. End resection-dependent DSB repair pathways, such as microhomology-mediated end joining (MMEJ), are typically restricted to certain stages of the cell cycle (S/G2/M), while nonhomologous end joining (NHEJ) is not22,32,33. Since postmitotic cells have exited the cell cycle, they are predicted to predominantly utilize NHEJ when repairing DSBs.Indeed, while B2Mg1-edited iPSCs displayed a broad range of indels, neurons exhibited a much narrower distribution of outcomes (Fig. 1d). In iPSCs, the most prevalent indel outcomes were larger deletions typically associated with MMEJ, as expected for dividing cells33. In neurons, the most prevalent outcomes were those usually attributed to NHEJ: small indels associated with NHEJ processing, and unedited outcomes caused by either indel-free classical NHEJ (cNHEJ) or lack of Cas9 cutting34,35. This was true for several different sgRNAs tested. Even though each sgRNA had a different intrinsic distribution of available indel types, in each case, the MMEJ-like larger deletions were predominant in iPSCs, and the NHEJ-like smaller indels were predominant in neurons. Therefore, for every sgRNA we tested, the ratio of insertions to deletions was significantly higher in neurons than iPSCs (Supplementary Fig. 6). These results demonstrate that postmitotic neurons employ different DSB repair pathways than dividing cells, yielding different CRISPR editing outcomes.Unresolved DSBs can be lethal to cycling cells, as DNA damage checkpoints trigger cell cycle arrest and/or apoptosis36,37. Therefore, for dividing cells, resolving a DSB mutagenically can be less harmful than leaving it unrepaired. For example, mitotic cells often utilize extremely indel-prone MMEJ repair to avoid progressing through M phase with unresolved DSBs32,33. This is consistent with our observed editing outcomes in iPSCs. On the other hand, postmitotic cells do not face replication checkpoints, and thus might not be subjected to the same pressures. Therefore, we hypothesized that DSBs could be resolved over a longer time scale in postmitotic cells.Cas9-induced indels accumulate slowly in neuronsIn dividing cells, the repair half-life of Cas9-induced DSBs is reportedly between 1 and 10 h; even in the slowest-repaired cut sites, the fraction of unresolved DSBs peaks within just over 1 day38. DSB repair in our iPSCs matched this expected timing, with indels plateauing within a few days. In contrast, indels in neurons continued to increase for up to 2 weeks post-transduction (Fig. 2a).Fig. 2: Cas9-induced indels accumulate over a prolonged time span in neurons.a Cas9-induced indels accumulate more slowly in neurons than in genetically identical iPSCs. Dose: 2 µL B2Mg1 VLP (HIV) per 100 µL media. For (a, b) 6 replicate wells per condition transduced in parallel (some obscured by overlap); curves pass through the mean at each timepoint. CRISPResso2 analysis of amplicon-NGS. b Several sgRNAs show weeks-long accumulation of indels in neurons. Dose: 1 µL VLP (FMLV) per 100 µL media. c–e Cas9-induced DSB foci (γH2AX+) remain detectable in neurons for at least 7 days post-transduction. Quantified in (c) by manual counting across n = 3 wells per condition; center points show means and error bars show SD. Representative ICC images of neurons 1 day (d) and 7 days (e) post-transduction, with age-matched untransduced neurons. Dose: 2 µL VLP (FMLV) per 100 µL media. See Supplementary Fig. 12 for unmerged/uncropped panels and full time course. ICC time course was conducted once; representative images chosen from 3 replicate wells per condition. f MRE11 is bound near the cut site in neurons for at least 8 days post-transduction. Dose: 2 µL B2Mg1 VLP (FMLV) per 100 µL media. Binding quantified by ChIP-qPCR, normalized for amplification efficiency and input chromatin. Average of 3 replicate reactions, normalized to untransduced control for each amplicon. Error bars show SD, centered at the mean. g Schematic: prolonged indel accumulation in neurons could be caused by neurons repairing DSBs more slowly, and/or by neurons undergoing more cycles of indel-free repair and re-cutting before edits arise. Our results do not rule out either model, but the early presence of post-repair products (Supplementary Fig. 13) and the surprising longevity of Cas9 protein in neurons (Supplementary Fig. 14) more strongly support the second model.Full size imageThis extended time course of editing was replicated by both types of VLPs (Supplementary Fig. 7). We tested multiple sgRNAs, including disease-relevant targets. Surprisingly, for every sgRNA, neuron indels continued to increase for at least 16 days post-delivery of transient Cas9 RNP (Fig. 2b and Supplementary Fig. 8). We also observed a similar weeks-long timeline of indel accumulation in postmitotic iPSC-derived cardiomyocytes (Supplementary Fig. 9a), so this prolonged indel accumulation might also apply to other clinically relevant nondividing cells besides neurons.We found no evidence that this prolonged indel accumulation in neurons was influenced by proliferating cells (Supplementary Fig. 2), or by residual VLP in the media (Supplementary Fig. 9b). Furthermore, using the same delivery particle but engaging a different DNA repair pathway than DSBs, VLP-mediated base editing in neurons was comparably efficient to iPSCs—and sometimes even more efficient— even within only three days post transduction (Supplementary Fig. 9c). Therefore, the slower accumulation of indels cannot be attributed solely to a “delivery deficit” in neurons.However, the kinetics of VLP entry and trafficking could still play a role in the prolonged time course of editing. To test this, we used another model of genetically identical dividing and nondividing cells: primary human T cells in the activated vs resting state. While resting T cells are not amenable to VLP delivery, both resting and activated T cells are amenable to electroporation, unlike neurons—enabling Cas9 RNP delivery without encapsulation in a delivery particle. Therefore, we electroporated Cas9 RNP directly into activated or resting primary T cells from multiple human donors, circumventing VLP delivery kinetics entirely. In this model, while activated vs resting T cells reproduced the observed differences in indel types between dividing and nondividing cells, there was not a dramatic difference in the timing of indels (Supplementary Figs. 10 and 11). This suggests that a component of the prolonged indel accumulation observed in postmitotic neurons and cardiomyocytes could be related to delivery kinetics, and/or dependent on cell type.This week-long timeline of editing in postmitotic cells could have major clinical implications. Gene inactivation therapies in nondividing tissues might take longer than anticipated to be effective, and both on-target and off-target editing may accumulate over longer intervals. Additionally, persistent DSBs in neurons have been associated with genomic instability and even neurodegeneration39,40,41, so characterizing the duration of Cas9-induced damage and repair is critical.DSB repair is detectable in neurons for more than 1 week post-Cas9 deliveryTo assess the duration of this damage in neurons, we measured multiple signals of DSB repair over time after delivering transient Cas9 RNP via VLPs. DSB foci (γH2AX/53BP1) were strongly detectable by ICC as early as 1 day post-transduction, confirming efficient delivery and rapid induction of DSBs in neurons. Interestingly, DSB foci remained detectable in neurons for at least 7 days post-transduction (Fig. 2c–e). Persistent DSB repair signal was observed for sgRNAs targeting both lowly-transcribed (B2M) and highly-transcribed (NEFL) genes (Supplementary Fig. 12). This long-lived repair signal is consistent with the prolonged accumulation of indels in neurons. DSB foci in iPSCs cannot be compared over the same span, as proliferating cells replicate many times within a week, diluting any unresolved signal.To more quantitatively measure this repair in neurons, we used chromatin immunoprecipitation with quantitative real-time PCR (ChIP-qPCR) to measure the binding of repair proteins Mre11 and γH2AX near the cut site, at several timepoints post-transduction. Mre11 binding in edited neurons was strongly detected within a few hundred bases of the cut site, and only in transduced samples (Fig. 2f), matching patterns seen in other cell types42. But intriguingly, Mre11 binding near the cut site remained strongly detected in neurons even 8 days post-transduction, decreasing by only ~50% between days 2 and 8.As expected based on previous reports42,43, γH2AX binding was much broader, with maximal signal detected several kilobases away from the cut site. Typically found between 2 and 30 kilobases away from the cut site, γH2AX is thought to coordinate longer-range interactions that facilitate repair, such as damage-induced cohesion to sister chromatids in dividing cells43,44. The role of γH2AX bound closer to the cut site remains unclear. Interestingly, while γH2AX binding farther from the cut site in our neurons decreased to background levels between days 2 and 8, γH2AX binding adjacent to the cut site only decreased by ~50% during this interval (Supplementary Fig. 13a). These week-long analyses cannot be performed in dividing cells like iPSCs, where one locus quickly becomes many due to replication.Some Mre11/γH2AX binding at each timepoint can be attributed to DSBs that had already been repaired (with or without an indel). This is evidenced by their binding to an amplicon that spans across the cut site, and thus should only amplify if the cut was resealed (Supplementary Fig. 13b–d). Cut sites resealed without an indel can be repeatedly recut by any remaining Cas9 RNP, until an indel prevents subsequent Cas9 binding.Altogether, multiple complementary approaches confirm that DSB repair signals at the target site persisted in neurons for much longer than expected, decreasing by only ~50% after 1 week post-delivery of Cas9 RNP. Two models could explain this prolonged timeline: postmitotic neurons might repair DSBs more slowly, or might undergo more cycles of repair and recutting until indels arise—or perhaps both (Fig. 2g).VLP-delivered Cas9 remains detectable in neurons for up to 30 daysWe cannot rule out either model, but our evidence more strongly supports the latter model. First, while Cas9 in dividing cells is rapidly degraded and/or diluted by cell division, we found that Cas9 in nondividing neurons is surprisingly long-lived. Following transient VLP delivery, Cas9 protein was undetectable in iPSCs after 8 days —but remained present in neurons even after 30 days (Supplementary Fig. 14a). Second, we found that this VLP-delivered Cas9 remains functional in neurons for at least a week. Even when we delivered Cas9-only VLP without any sgRNA, then delivered the sgRNA separately 8 days later, we still detected ~15% editing in neurons (Supplementary Fig. 14b). These findings are compatible with the model that DSBs in neurons tend to be repaired without indels, but long-lived Cas9 protein facilitates many cycles of damage and repair until an edit finally arises days or weeks later.Either way, the longevity of Cas9 and Cas9-induced damage could have important consequences for the safety of genome editing in nondividing cells, in particular with regard to cytotoxicity, immunogenicity, and off-target edits.Cas9-VLPs elicit a striking transcription-level response in neuronsGiven this unexpectedly prolonged time scale of editing, we reasoned that neuronal DNA repair might include transcription-level regulation, not only post-translational regulation. To test this, we used bulk RNA sequencing (RNAseq) to characterize differentially expressed genes (DEGs) in iPSCs and neurons transduced with Cas9-VLP, relative to untransduced cells (Supplementary Data 1). Unlike transduced iPSCs, transduced neurons exhibited a skewed transcriptional response, with far more genes upregulated than downregulated (Fig. 3a, b and Supplementary Fig. 15). The top 50 DEGs in neurons, all upregulated, were highly enriched for genes canonically associated with DNA repair and DNA replication (Fig. 3c).Fig. 3: Neuronal response to Cas9 reveals unexpected factors that influence editing outcomes in nondividing cells.Neurons (b), but not iPSCs (a), dramatically upregulate transcription of DNA repair factors upon Cas9-VLP transduction. For a–d: dashed lines show cutoffs for significance (padj 2 or