DNA repair pathway choices in CRISPR-Cas9 mediated genome editing

Abstract

Many CRISPR-Cas-based genome editing technologies take advantage of Cas nucleases to induce DNA double-strand breaks (DSBs) at desired locations within a genome. Further processing of the DSBs by the cellular DSB repair machinery is then necessary to introduce desired mutations, sequence insertions, or gene deletions. Thus, the accuracy and efficiency of genome editing are influenced by the cellular DSB repair pathways. DSBs are themselves highly genotoxic lesions and as such cells have evolved multiple mechanisms for their repair. These repair pathways include homologous recombination (HR), classical non-homologous end joining (cNHEJ), microhomology-mediated end joining (MMEJ) and single-strand annealing (SSA). In this review, we briefly highlight the CRISPR-Cas9 system and we then describe the mechanisms of DSB repair. Finally, we summarize recent findings of factors that can influence the choice of DNA repair pathway response to Cas9-induced DSBs.

Introduction

Double-strand break (DSB) repair mechanisms have been harnessed for genome editing, which is based on the use of engineered nucleases that can introduces DSBs at target sites and relies on these DNA repair pathways to insert, delete, or replace pieces of DNA [1]. Examples of engineered nucleases include zinc-finger nucleases (ZFNs) [2], transcription activator-like effector nucleases (TALENS) [3], and the recently discovered clustered regularly interspaced short palindromic repeats (CRISPR)-Cas-based RNA guided DNA endonucleases [4–8]. DNA double-strand breaks (DSBs) are among the most genotoxic DNA lesions, which have the potential, if left unrepaired, to cause chromosomal rearrangements, genomic instability, and cell death [9]. Eukaryotic cells have evolved numerous pathways to repair DSBs [10–13]. The two major pathways are error-free, template-dependent homologous recombination (HR), and the error-prone, template-independent classic non-homologous end joining (cNHEJ) [10, 11]. Additional pathways include microhomology-mediated end joining (MMEJ), also known as alternative end joining (alt-EJ), and single-strand annealing (SSA), both of which are error-prone [12, 13].

CRISPR-Cas9 has emerged as a powerful tool for genome engineering applications [5–7]. The first CRISPR locus, the defining feature of CRISPR-Cas adaptive immune systems, has been discovered more than three decades ago [14–16]. However, the function of these loci was not recognized until 20 years later [17]. In 2007, Barrangou et al. demonstrated that CRISPR loci represent an adaptive immune system, which can “memorize” past viral infections by inserting viral genetic fragments into the host CRISPR array [17]. The CRISPR array is transcribed and processed into mature CRISPR RNAs (crRNAs, also often referred to as guide RNA or gRNA), which assemble with Cas effector proteins [18]. Using the crRNAs as guides, viral genomes can be targeted through sequence-specific DNA-RNA base pairing interactions and then destroyed, but only if the viral sequences are already present within the CRISPR array [17, 19–21]. CRISPR-Cas systems can be divided into two classes (Class1 and Class 2), and each class contains three types [22]. Class 1 CRISPR-Cas systems are further subdivided into types I, III, IV, whereas Class 2 systems are subdivided into types II, V and VI [22]. Importantly, Class II CRISPR-Cas systems, such as Cas9, require just one protein component for interference, thus offering a significant advantage as genome editing tools [22]. The fundamental basis for genome editing using CRISPR-Cas systems is the ability to induce a DSB at a precise location within a genome based upon sequence complementarity with the gRNA [1].

Precise genome editing can be achieved thorough homology-directed repair (HDR), however, mammalian cells often prefer to repair DSBs by simpler, but more error-prone repair pathways, such as cNHEJ, which can generate undesirable mutations. Importantly, in the genome editing field, the frequently used term HDR does not represent a single DNA repair pathway [23]. Instead, HDR is related to the nature of DNA donor templates used to guide the DNA repair reaction, including double-stranded DNA donor templated repair (DSTR) and single-stranded DNA donor templated repair (SSTR) [24].

Here, we briefly describe the CRISPR-Cas9 and key features of Cas9-gRNA genome editing. We then discuss the different DSBs repair pathways in eukaryotic cells, focusing mainly on mammalian DSB repair mechanisms. Finally, we discuss how CRISPR-Cas9-induced DSBs are repaired by the host cell DNA repair pathways and the factors that influence the choice of repair pathway.

OVERVIEW OF THE CAS9 CRISPR-CAS SYSTEM

Cas9 is the effector protein of type II CRISPR-Cas bacterial immune systems and is the first CRISPR-associated effector protein that was repurposed for genome editing (Figure 1A) [1]. In nature, Cas9 requires two RNA molecules, a crRNA and a trans-activating crRNA (tracrRNA), which base pairs with the repeat sequence of the crRNA [4, 8, 25]. The resulting Cas9–RNA complex must first recognize a short DNA sequence motif called the protospacer-adjacent motif (PAM) and then tests the flanking dsDNA for sequence complementarity to the crRNA (Figure 1A) [26]. Importantly, for genome engineering efforts the crRNA and tracrRNA can be combined into a single guide-RNA (sgRNA) to simplify Cas9 applications without affecting Cas9-mediated target binding and cleavage activities (Figure 1A) [4]. When Cas9 binds to the correct dsDNA target its HNH and RuvC nuclease domains cleave the DNA to yield either blunt or staggered ends (Figure 1A) [27, 28]. Genome editing is achieved during the repair of the resulting DSB [29]. Of great concern during Cas9 genome editing is the potential for off-target DNA cleavage [30–32]. Cas9 can potentially recognize and cleave DNA sites that may not perfectly match the sgRNA or tracrRNA-crRNA. Cas9 cleavage at such non-target sites has the potential to cause off-target genome mutations [29, 33].

Figure 1. CRISPR-Cas9 delivery.

(A) Mechanism of CRISPR-Cas9 genome editing. A sgRNA/tracrRNA-crRNA associates with the Cas9 endonuclease to form the Cas9-gRNA complex. The gRNA guides Cas9 to its target site of the genomic DNA by recognizing the protospacer-adjacent motif (PAM). When Cas9 binds to the correct dsDNA target, its HNH and RuvC nuclease domains cleave the DNA to yield either blunt or staggered ends. Genome editing is achieved during the repair of the resulting DSB, resulting in precise mutations, gene deletions, or sequence insertions. (B) Cas9 can be delivered in three forms. First, introducing Cas9-gRNA ribonucleoprotein (RNP) complex and DNA templates directly through nanoparticles, electroporation, or microinjection. Second, delivering a plasmid DNA or viral vector for Cas9 and gRNA production in situ. Third, delivering separate gRNA together with mRNA for Cas9 protein expression inside the cell. With the exception of microinjection, during which Cas9 components can be directly injected into the nucleus, other delivery methods release Cas9 components into the cytosol. DNA sensing receptors in endosome or cytosol, including the Toll-like receptor 9 (TLR9), melanoma 2 (AIM2) and cyclic GFP-AMP synthase-STING (cGAS-STING), drive cell immune responses to foreign DNA. mRNA can also be recognized by TLR7 and TLR8 in endosome causing mRNA degradation and type I interferon α (IFN-α)-mediated immune response. Cas9 in the cytosol can be directed into nucleus through nuclear localization sequence; how plasmid DNA, viral DNA and DNA templates enter the nucleus is unknown.

Guide RNAs

In the type II Cas9 system, the crRNA is 42-nucleotides in length and contains a unique 20-nucleotide sequence at its 5’ end which can form Watson-Crick base pairing interactions with the target sequence (Figure 1A) [4]. The tracrRNA is 80-nucleotides in length and binds to a 22-nt repeat sequence at the 3’ end of the crRNA, forming a unique dual-RNA hybrid structure which is critical for the recruitment of the Cas9 protein to form an active DNA surveillance complex [4, 26, 34]. However, the crRNA and tracrRNA can be provided as a simplified, single chimeric sgRNA that is about 100-nucleotides in length and fulfills all of the functions of the dual crRNA-tracrRNA found in nature; hereafter we will use the general term “gRNA” to represent either the single sgRNA or dual tracrRNA-crRNA (Figure 1A) [4–7].

For genome editing applications, the gRNA can be produced either in situ or ex situ [35]. For instance, the gRNA can be produced intracellularly via transcription from a plasmid template or from a viral DNA (e.g. Adeno-Associated Viruses (AAV), Adenovirus (AdV), and Lentivirus (LV)) and the resulting gRNA will bind to co-expressed Cas9 protein to form the active Cas9-gRNA complex (Figure 1B) [36–39]. The gRNA can also be produced ex situ using in vitro transcription or chemical synthesis and these approaches allow for rigorous control over the delivery of Cas9, which is achieved through directly delivery the Cas9-gRNA or by transient transfection of gRNAs into cells expressing the Cas9 protein (Figure 1B) [40]. It has to be noted that the second method that delivery gRNA alone is not common during genome editing. Another advantage of ex situ gRNA production is that the gRNAs can be synthesized with chemical modifications that can protect the gRNA from digestion by cellular nucleases [41–43]. Chemical modifications within the gRNA can also alter the thermodynamic stability of the gRNA-target dsDNA to improve target recognition efficiency and decrease off-target activity [42, 43]. Compared to gRNAs produced by in vitro transcription, gRNAs produced by chemical synthesis can also include chemical modifications to reduce the potential for cellular toxicity and inhibit type I interferon α (IFN-α) related pro-inflammatory cytokine response, which is induced by ssRNA recognition through receptor-Toll-like receptor 7 (TLR7) (Figure 1B) [44, 45].

Cas9 nuclease activity

Cas9 is a large multidomain protein consisting of a recognition (REC) lobe, including the REC1, REC2, and REC3 domains, and a nuclease (NUC) lobe which is composed of a PAM interacting (PI) domain and two nuclease domains: a RuvC-like domain that cleaves the non-target strand and an HNH-like domain that cleaves the target strand (Figure 1A) [4, 8, 27]. The Cas9-gRNA complex searches for its target site by first binding to PAMs within the dsDNA through a combination of 3D and 1D diffusion [26, 46]. PAM recognition triggers initial unwinding of the adjacent dsDNA (referred to as the seed region), allowing the gRNA to begin undergoing strand invasion to form a RNA-DNA hybrid and concomitant displacement of the non-complementary DNA strand (Figure 1A) [26, 47]. If the DNA sequence is fully complementary to the gRNA, then seed region of the target dsDNA promotes further RNA-DNA hybrid formation, resulting in the formation of a stable R-loop [47]. During R-loop formation, the Cas9 protein undergoes a conformational change from a nuclease inactive to nuclease active conformation resulting in the exposure of the target DNA strand to the HNH nuclease active site [48–50]. While in the active conformation, the HNH-like domain cuts the target DNA strand 3 base pairs upstream from the PAM, while the RuvC-like nuclease domain cuts 3 to 5 base pairs away on the non-target strand (Figure 1A) [28, 51].

Off-target effects during Cas9-mediated genome editing arise from the ability of Cas9 to tolerate mismatches between the gRNA and potential dsDNA targets [30–32]. Although a single base mismatch within the PAM or seed region can reduce Cas9 cleavage activity, multiple mismatches within the PAM-distal region can be tolerated by Cas9 [4, 52]. The potential for off-target effects hinder clinical applications of Cas9-mediated genome editing [33]. To address this issue, several high-fidelity variants of Cas9 have been established through protein engineering [49, 50, 53–57]. For example, eSpCas9(1.1) decreases off-target effects by weaking interactions between the protein and the non-targeted DNA strand, such that partial DNA-RNA pairing interactions become less tolerated [53]. SpCas9-HF1 decreases off-target effects by reducing interactions between the protein and the targeted DNA strand [54], whereas mutations in the REC3 domain of Cas9 (HypaCas9) prevent Cas9 from transitioning into an active conformation when there are mismatches between the gRNA and potential off-target sites [49, 50]. Cas9 mutations in the REC3 domain, which affect the activation of Cas9 after DNA binding, also enhance target recognition fidelity [56, 57]. The development of these higher-fidelity Cas9 variants will be potentially beneficial for clinical applications.

Cas9-gRNA delivery methods

Genome editing requires delivery of the Cas9-gRNA complex to the cell nucleus (Figure 1B). The timing and dosage of Cas9 delivery are important factors that can be modulated to prevent undesirable off-target effects [58, 59]. Currently, there are three major Cas9 delivery methods, and each has its advantages and disadvantages (Figure 1B). First, the mature Cas9-gRNA RNP can be delivered using gold nanoparticles or lipid nanoparticles [59–61]. This delivery method provides rigorous control over the timing and dosage of the Cas9-gRNA. However, delivering such a large RNP can be challenging and its introduction has the potential to trigger immunological responses [62]. Second, genes encoding Cas9 and the gRNA can be delivered with a DNA vector, such as a plasmid or a viral vector (e.g. adenovirus), allowing for Cas9 and gRNA production in situ (Figure 1B) [6, 7, 38, 63]. This method allows for long-term Cas9 and gRNA expression, however, persistent expression of the Cas9 complex might increase potential for off-target effects [40, 64, 65]. In addition, cellular receptors that sense exogenous DNA, including the Toll-like receptor 9 (TLR9), melanoma 2 (AIM2) and cyclic GFP-AMP synthase-STING (cGAS-STING), can cause cytotoxic effects and DNA degradation during delivery and induce IFN-α related immune response (Figure 1B) [66–69]. It has to be noted that the CRISPR field has not been fully concerned about the potential toxicity of exogenous DNA. Third, the gRNA can be delivered along with an mRNA encoding the Cas9 protein [42]. This approach can allow for transient expression of Cas9 within a limited time window, which can help mitigate off-target events [35]. However, this delivery method may have limited efficiency for genome editing due to rapid degradation of the Cas9-encoding mRNA [42]. Several strategies have been developed to allow for conditional activation of Cas9 in situ, and for a more detailed discussion of these methods we refer readers to several recent reviews [35, 70–72].

Although CRISPR-Cas9 has been used in multiple clinical cases, including CTX001 targeting several genetic diseases, CTX110 targeting CD19 positive malignancies, CTX120 targeting multiple myeloma, PD1/TCR modified T cells targeting multiple solid tumors, and UCART019 targeting CD19 positive leukemia and lymphoma (Clinicaltrials.gov) [73], the type V CRISPR-Cas12a (formerly named Cpf1) systems has received widespread attention due to several advantages over CRISPR-Cas9 system [74]. First, Cas12a is smaller than Cas9 and easier to deliver than Cas9 system in clinical applications [74]. Second, Cas12a is guided by a single crRNA, which is around 42–44 nt in length and less than half the size of sgRNA for Cas9 [75]. In addition, the CRISPR-Cas12a system is considered to have lower off-target efficiency than CRISPR-Cas9 system, which may be related to its lower nuclease activities [76, 77]. Moreover, Cas12a cleaves dsDNA with 4–5 bp staggered 5’ overhangs, in contrast to the mixed blunt and 1–2 bp staggered 5’ overhangs of Cas9, which might affect DNA repair outcomes (see below) [75]. However, the newly discovered Cas12a cis (target-dependent) and trans dsDNA (target-independent) nicking and cleavage activities may be detrimental for genome editing and need careful evaluation [78, 79].

DOUBLE STRAND BREAK REPAIR PATHWAYS

All cells must be able to quickly detect and repair DSBs in order to maintain genome integrity [9, 80, 81]. In eukaryotic cells, the most prevalent DSB repair pathways include cNHEJ, MMEJ, SSA and HR (Figure 2) [10–13]. The relative extent to which a particular repair pathway is utilized varies among different species and also depends on the cell cycle (see below) [82–85]. In general, mammalian cells tend to use cNHEJ throughout the cell cycle, although HR is more frequently utilized during S/G2 phase [86]. Here, we briefly discuss these different DSB repair pathways and highlight the advantages and disadvantages of each.

Figure 2. The four major pathways to repair DNA Double-Strand Breaks (DSBs).

(A) Unprocessed DSBs can be repaired through classic non-homologous end joining (cNHEJ) allowing the two ends of the DSB to be re-ligated. (B) DSB ends can also be processed by the MRN complex and its interacting factors to yield short 3’ ssDNA overhangs. (C) The short 3’ ssDNA overhangs can then be channeled into the microhomology-mediated end joining (MMEJ) pathway. (D) Alternatively, the DSB ends can undergo further long-range resection by either EXO1 or BLM/DNA2. These longer ssDNA overhangs are first bound by RPA and can then be channeled into the (E) SSA pathway, which is mediated by the protein RAD52. (F) Alternatively, the RPA-ssDNA can serve as a substrate for the RAD51 filament assembly, allowing the resulting DNA intermediates to be directed towards repair by (G) HR. For HR, both ssDNA and dsDNA templated homology repair (HDR) pathways are shown.

Classical non-homologous end joining (cNHEJ)

cNHEJ is a commonly used pathway for re-ligating broken DNA ends, but is also inaccurate and often leads to new mutations [11]. During cNHEJ, DSBs are repaired by directly ligating the broken ends together with minimal DNA end processing (Figure 2A) [11]. cNHEJ is initiated when the ring-shaped Ku70/Ku80 protein heterodimer binds to the DSB ends, which protects the DNA ends from further resection and also recruits the DNA-dependent protein kinase catalytic subunit (DNA-PKcs) [87–89]. The DNA-PK complex further recruits the DNA ligase complex LigIV (DNA ligase IV)-XRCC4 (X-ray cross complementing Group 4)-XLF (XRCC4-like factor) to ligate the DSB ends [90–92]. When the DSBs ends are not directly ligatable, additional proteins including Artemis nuclease and polynucleotide kinase 3’ phosphatase (PNKP) are required for end processing (Figure 2A) [93–95]. For Cas9-induced DSBs, Cas9 asymmetrically releases the PAM-proximal end of the cleaved DNA within minutes while the PAM-distal end of the DSB remains bound by Cas9 for a longer time period [96, 97]. As a result, the PAM-proximal DSB end is exposed for further processing, whereas the other end may be temporally blocked from further processing by the still bound Cas9 [97]. RNA-polymerase and FACT can remove Cas9 from the PAM-distal side, but how Cas9 is removed during genome editing remains unclear [97, 98]. How the tight binding of Cas9 to the PAM-distal DSB end affects DSB repair pathway choice is also unclear. Given that cNHEJ is an inaccurate repair pathway, several groups are pursuing strategies to inhibit cNHEJ in favor of the much more accurate repair that is achieved through HR. For example, HR-mediated gene editing in mammalian cells can be increased by blocking LigIV activity using small molecules [99–101] or by stimulating proteolytic degradation of LigIV [101]. Similarly, suppression of DNA-PKcs or Ku through genetic methods or chemical inhibition also promotes DSBs repair through HR [101–104].

Microhomology-mediated end joining (MMEJ)

MMEJ is another error-prone DSBs repair pathway, which begins with short-range resection of the DSB (Figure 2B). Like cNHEJ, MMEJ does not require a template for repairing the damaged DNA (Figure 2C) [12]. Instead, the DSB ends are re-aligned using short (5–25 base pairs) microhomologous sequences near the broken ends (Figure 2C) [105]. Any remaining 3’ ssDNA flaps are cleaved off, which results in the loss of sequence information [12, 105]. The remaining gaps are filled in through DNA synthesis, and the remaining nicks are sealed through ligation [106, 107]. Similar to HR, MMEJ functions mainly during S/G2 phase [108].

During MMEJ, some short-range end resection around the DSBs is needed to generate short ssDNA overhangs (~20 bps in mammalian cells; Figure 2C) [108]. This short-range end resection promotes MMEJ, whereas more extensive end resection favors SSA and HR (see below; Figure 2D) [109, 110]. In mammalian cells, the damage-sensing MRN complex (Mre11-RAD50-NBS1), together with its interacting partner CtIP (C-terminal binding protein interacting protein), initiates end resection (Figure 2C) [111, 112]. CtIP phosphorylation during S/G2 phases stimulates the MRE11 endonuclease activity which generates a nick at the 5’ strand near to the DSB end and prevents cNHEJ by removing Ku from the DNA-ends [112–114]. The resulting nick allows the 3’ to 5’ MRE11 exonuclease activity to generate short 3’ overhangs which can be used to initiate MMEJ [108, 115]. Alternatively, these short overhangs can be processed further by long-range 5’ to 3’ end resection, which favors the SSA and HR pathways (see below). Inhibition of MMEJ by blocking the ligase activities of LigI and III (involved in the last step of MMEJ; see below) with small molecule inhibitors can promote DSB repair through the more accurate HR pathway [101, 116].

We currently have a limited understanding of the factors that regulate the annealing of microhomologous sequences during MMEJ [12, 105]. The ssDNA binding-protein replication protein A (RPA) and the protein RAD52 can inhibit MMEJ by preventing annealing [117, 118], whereas poly ADP-ribose polymerase 1 (PARP1) tethers DNA fragments together and promotes the annealing reaction [119, 120]. However, there is conflicting evidence regarding the regulation of PARP1 in MMEJ [120, 121]. Some studies suggest that PARP1 promotes MMEJ by competing with Ku for DSBs end binding [120], while other studies indicate that PARP1 promotes Ku loading [121]. Future studies will be essential to more fully understand how these protein factors affect MMEJ.

If the microhomologous sequences are at the very ends of the DNA, then no trimming is required before gap filling through DNA synthesis [12]. In contrast, for microhomologous sequences located distal to the DNA ends, the resulting heterologous 3’ ssDNA flap must be removed by XPF-ERCC1 endonuclease (called Rad1-Rad10 in S. cerevisiae) or perhaps another unidentified nuclease to allow DNA polymerase to fill in the gap [122]. In mammalian cells, Pol θ stabilizes annealed overhangs and fills any gaps via template-directed DNA synthesis [106, 123]. Once the gaps are filled, the remaining nicks are sealed by DNA ligase III (LigIII) or DNA ligase I (LigI) [107].

A large fraction (up to ~58%) of Cas9-induced DSBs are repaired through MMEJ [124]. The resulting repair outcomes are not totally random and can be predicated at a given DSB site, indicating that the alignment between resected DNA ends at a given DSB site are reproducible and may depend on local sequence context [28, 124–126]. Several lines of evidence suggest that deletion of two or more nucleotides at the Cas9 cut site is the most common outcome in mammalian cells [28, 124, 126, 127]. Inhibition of MMEJ in mammalian cells decreases these nucleotide deletions significantly [128]. By taking advantage of this reproducible behavior, precise template-free genome editing through the MMEJ pathway has been successfully achieved in mammalian cells [124].

Single-strand annealing (SSA)

SSA and MMEJ are similar in several aspects, in particular, they both require 3’ ssDNA overhangs for the annealing of homologous sequences [12, 13]. Similar to MMEJ, SSA requires the removal of heterologous 3’ flaps followed by gap filling [13]. Like HR, SSA is mainly functional during the S/G2 phases of the cell cycle [82–85].

SSA requires long-range DSB end resection to yield long (>1000 nucleotides) ssDNA overhangs (Figure 2D) [129]. These 3’ ssDNA overhangs are generated by the 5’ to 3’ exonuclease activities of either EXO1 or BLM-DNA2 [109, 130, 131]. The resulting ssDNA overhangs are then coated by RPA, followed by the binding of RAD52, which mediates the annealing of homologous sequences within the two DSB ends (Figure 2E) [132–135]. In S. cerevisiae, Rad52 also stimulates the assembly of Rad51 filaments onto the ssDNA to promote HR (see below) [136]. In contrast, mammalian RAD52 does not stimulate RAD51 filament assembly [135, 137]. Indeed, proteins called RAD51 paralogs promote RAD51 filament assembly in mammals, thus inhibiting SSA and favoring HR [138]. In S. cerevisiae, the Rad52 paralog Rad59 helps to channel repair intermediates through Rad52-mediated SSA by alleviating the inhibition of Rad51 on the strand annealing activity of Rad52 [139]. Although there is no known mammalian homolog of Rad59, other factors that function similarly may exist in mammals [140]. Like MMEJ, the heterologous 3’ flap is removed by XPF-ERCC1 endonuclease during SSA [141]. However, details regarding how the gaps are filled and ligated remain to be determined [13, 140].

Homologous recombination (HR)

HR is a high-fidelity repair pathway that uses a homologous DNA template to guide DSB repair (Figure 2F–G) [10, 110, 140] and as such is preferred during many Cas9-mediated genome editing approaches [1, 24]. HR involves long-range end resection to form long 3’ ssDNA overhangs that are then coated by RPA (Figure 2D) [142]. The bound RPA is then replaced by the ATP-dependent DNA recombinase RAD51, which forms long helical filaments on the ssDNA overhangs (Figure 2F) [86, 143]. The RAD51 filaments then mediate alignment and pairing of the ssDNA overhangs with homologous dsDNA sequences from elsewhere in the genome (Figure 2G) [144]. The 3’ invading end of the RAD51-coated ssDNA strand is used to prime DNA synthesis and the resulting intermediates can be used to complete repair of the DSB [10, 110, 140, 145]. In notable contrast to cNHEJ, HR is restricted to the S/G2 phases of the cell cycle [82–85].

End resection

End resection during HR requires two steps, beginning with MRN-CtIP mediated short-range resection (Figure 2B), followed by long-range end resection involving EXO1 and/or BLM/DNA2 (Figure 2D) [112, 113, 130, 131]. Activation of key DNA end resection factors is strictly regulated by cell cycle and restricted mainly to the S/G2 phases of the cell cycle [85, 111]. In addition, several regulatory factors affect these DNA end resection steps.

Short-range end resection

MRN is crucial for short-range end resection and is controlled by numerous positive and negative regulatory factors (Figure 2B) [81, 146]. PARP1 is recruited to new DSBs within milliseconds and can inhibit DNA resection and recruit Ku to favor cNHEJ [121].However, PARP1 can also facilitate MRN recruitment, which may favor MMEJ, SSA or HR [147]. Further studies are required to more fully understand these seemingly conflicting roles of PARP1. DNA-PK also binds to free DNA ends and may commit cells to the cNHEJ repair pathway [11, 140]. DNA-PK can be removed from DNA ends by several mechanisms, including phosphorylation of Ku70 to reduce its DNA-binding affinity [148], degradation of Ku through ubiquitination [149], and MRN-mediated DNA cleavage [150]. Removal of DNA-PK favors MMEJ, SSA and HR [10, 151]. End-bound DNA-PK can also promote MRN-mediated end processing, indicating that DNA-PK not only plays a key role in cNHEJ but also in initiating short-range end resection [150]. Phosphorylated CtIP and the BRCA1-BARD1 complex both stimulate short-range end resection by MRN [84, 113]. In addition, the recently discovered protein DYNLL1 (dynein light chain 1 protein) inhibits MRN nuclease activity [152] and ZPET (zinc figure protein proximal to RAD18) delays MRN-CtIP recruitment through an unknown mechanism to inhibit the short-range end resection [153, 154].

Long-range end resection

The short overhangs generated by MRN serve as an entry site for enzymes such as BLM/DNA2 or EXO1, which are necessary for long-range end resection (Figure 2D) [130, 131]. The exonuclease EXO1 generates long 3’ ssDNA overhangs through its processive 5’ to 3’ exonuclease activity [131]. MRN stimulates EXO1 exonuclease activity, whereas CtIP inhibits EXO1 [131, 155]. In mammalian cells, RPA can both stimulate and inhibit EXO1 activity [156], while SOSS1, the human SSB homologue 1, can stimulate resection by EXO1 [157].

BLM is a member of the RecQ helicase family that acts in concert with DNA2, which possesses both 5’ and 3’ endonuclease activities [156]. BLM separates the DNA strands, allowing DNA2 to cleave resulting ssDNA [131, 158]. During long-range end resection, RPA inhibits the 3’ endonuclease activity of DNA2 while stimulating its 5’ endonuclease activity to ensure the 5’ to 3’ polarity of resection (Figure 2D) [159]. The BLM/DNA2 pathway is regulated by many factors. For example, BLM/DNA2 loading at DNA ends is mediated by the TOPIIIα-RMI1-RMI2 complex and MRN [160]. CtIP also interacts with BLM and DNA2 and promotes end resection by stimulating BLM helicase activity and DNA2 5’ endonuclease activity [161].

Recent discoveries have revealed a regulatory mechanism involving 53BP1, PTIP, RIF1, and Shieldin-CST which inhibits end resection and a competing regulatory mechanism involving BRCA1-BARD1 that promotes end resection [162–164]. In S. cerevisiae, Rad9 (a homolog of 53BP1) limits long-range resection [165]. Activated ATM at DSBs leads to histone H2AX phosphorylation, referred to as γH2AX, leading to MDC1 recruitment and phosphorylation [166]. Phosphorylated MDC1 recruits two E3 ubiquitin ligases, RNF8 and RNF168, to ubiquitinate histone H2A [167]. Ubiquitylated H2A together with H4K20me2 recruit 53BP1 to chromatin adjacent to the DSBs. Phosphorylated 53BP1 interacts with either PTIP or RIF1, both of which inhibit end resection albeit by different mechanisms [167]. The function of PTIP remains elusive [167], whereas RIF1 recruits the Shieldin complex (comprised of SHLD1, SHLD2, SHLD3 and REV7) [162–164]. One model suggests that Shieldin binds at the DNA end to protect the 5’ end from end resection [162, 164]. An alternative model suggests Shieldin recruits CST, Polα, and Primase to DSBs [163]. CST binds at the dsDNA-ssDNA junction to protect the 5’ end and Polα and primase execute a fill-in reaction, which are stimulated by CST to counteract end resection [163].

The regulatory interplay between BRCA1-BARD1 and 53BP1 remain unclear, but evidence suggests they may act in opposition to one another [168]. Early studies showed that loss of 53BP1 in BRCA1-deficient cells restored HR, suggesting the existence of an antagonistic relationship between BRCA1 and 53BP1 [169]. Current evidence suggests that BRCA1 may offer two layers of regulation by preventing 53BP1 from inhibiting long-range resection and by directly promoting long-range end resection [168]. 53BP1 and RIF1 block BRCA1 accumulation at DSBs during G1 whereas BRCA1 prevents 53BP1 and RIF1 foci formation at DSBs during S/G2 [154, 168]. A recent paper suggested that the antagonistic relationship between BRCA1-BARD1 and 53BP1 may be related to two histone (H2A and H4) post-translational modification states. The recruitment of BRCA1-BARD1 and 53BP1 are controlled by lysine 15 ubiquitylation of H2A and lysine 20 methylation of H4 to effect DSB repair pathway choice [170]. In addition, BRCA1 can interact with MRN and CtIP to stimulate end resection during S/G2 [171].

RAD51 filament formation

The 3’ ssDNA overhang generated by long-range resection is bound by RPA, which prevents formation of secondary structure and protects the ssDNA (Figure 2D) [142]. RPA must be replaced by the protein RAD51, which is the recombinase that catalyzes key DNA transactions during HR (Figure 2F) [86, 143]. However, RPA presents a barrier to RAD51 filament assembly [138, 140]. During S/G2, RAD51 is phosphorylated by CDK1, which enhances its ability to compete with RPA for ssDNA [172]. However, RAD51 alone cannot efficiently replace RPA from ssDNA in vivo [138]. In S. cerevisiae, Rad52 stimulates Rad51 filament formation on RPA-bound ssDNA [136]. However human RAD52 does not fulfill a similar role [135, 137]. Instead, human RAD51 filament assembly is facilitated by several other protein complexes, including the Rad51 paralog-containing complexes RAD51B-RAD51C-RAD51D-XRCC2 complex (BCDX2) and the RAD51C-XRCC3 complex (CX3), the SWS1-SWSAP1-SPIDR complex (otherwise known as the Shu complex), the SWI5-SFR1 complex, as well as the BRCA2-DSS1 complex [138]. BRCA1-BARD1 maybe also be involved in this process as part of a large complex with PALB2 and BRCA2–DSS1 [168, 173].

Loss of BCDX2 or CX3 significantly reduces RAD51 foci formation in cells, indicating that both are required to promote RAD51 filament formation or stability [174]. Loss of SWS1 or SWSAP1 also reduces RAD51 foci formation, although HR efficiency in these cells is comparable to wild-type [175]. These data all suggest a role for BCDX2, CX3, and the Shu complex during RAD51 filament formation, but underlying mechanisms remain unclear. DSS1 is a highly acidic protein that mimics DNA and directly interacts with RPA to weaken the affinity of RPA for ssDNA [176]. BRCA2 is comprised by a N-terminal DNA binding domain (NTD) which interacts with PALB2 [177], eight BRC repeats domain (BRC) that recruit RAD51 [178], a DNA binding domain (DBD) comprised of three OB folds to bind ssDNA [179], and an additional C-terminal RAD51-binding domain (CTRB) [180]. The SWI5-SFR1 complex stabilizes RAD51 filaments bound to ssDNA [181].

There are also negative regulatory factors that remove RAD51 from ssDNA [138]. In human cells, the helicases RECQ5, FBH1, and FANCJ downregulate HR by using the energy derived from ATP hydrolysis to translocate along ssDNA while stripping RAD51 from the ssDNA [138]. Similarly, the helicase FBH1 acts together with SCF ubiquitin ligase complex to ubiquitylate RAD51 causing RAD51 to relocate from nuclear to cytoplasm to reduce RAD51 filament formation [182].

RAD51-mediated pairing of homologous sequences

The RAD51 filament, together with additional accessary factors, must align the ssDNA overhang with a homologous sequence that can be used to guide repair (Figure 2G). It must then invade the homologous dsDNA to form base-pairing interactions with the homologous template strand, resulting in the formation of a heteroduplex DNA joint (Figure 2G) [10, 110, 140, 145]. Accessory factors that participate in this process, include RAD54, BRCA1-BARD1, PALB2, RAD51AP1-UAF1 and HOP2-MND1 [183–189]. Recently, it has been shown that S. cerevisiae Rad54 acts as a molecular motor to drive rapid ATP-dependent translocation of the RAD51 presynaptic filament along the dsDNA to facilitate sequence alignment [183] and opens the dsDNA strand to facilitate formation of base pairing interactions [183]. BRCA1-BARD1 and RAD51AP1-UAF1 promote the interaction between the RAD51 filament and the homologous dsDNA [184, 185]. PALB2 stimulates RAD51 strand invasion activity through an unknown mechanism [186, 187], however, it has been shown that the DNA-binding domain in its N-terminus plays an important role [188]. HOP2-MND1 stabilizes the RAD51 filament and stimulates heteroduplex DNA joint formation through its dsDNA binding activity [189].

DNA synthesis and product resolution

After D-loop formation, the 3′ end of the invading strand is engaged by DNA polymerase δ (Polδ), PCNA, and clamp loader complex RFC1–5, allowing the broken DNA end to be extended using the homologous donor dsDNA as a template [10, 110, 140, 145]. The resulting intermediates can be resolved through several mechanisms, including the non-crossover synthesis-dependent DNA strand annealing (SDSA), double Holliday junction (dHJ) crossover and non-crossover pathway and break-induced replication (BIR) (reviewed in references [140], [190] & [191]). BIR is a pathway enabling repair of just one of the two DSB ends and is not applicable to genome editing [191]. In SDSA, the heteroduplex DNA joint is disrupted by BLM, RTEL1, or another helicase [140, 192, 193] and can then be annealed with the other end of the DSB allowing for completion of repair by gap filling and ligation (Figure 2G) [140]. When both ends invade the same template, the interacting DNAs are joined by two Holliday junctions to form a double Holliday junction (dHJ; Figure 2G). dHJ are resolved to yield two intact dsDNA molecules [190]. In mammalian cells, dHJ dissolution is catalyzed either by the topoisomerase IIIα together with BLM-RMI1-RMI2 (BTR complex) resulting in non-crossover products, or by nucleases MUS81-EME1, SLX1-SLX4, and GEN1 which cleave the dHJ substrate to generate cross-over products [190].

FACTORS AFFECTING DNA REPAIR OUTCOMES OF CAS9-INDUCED DSBs

Cas9-mediated genome editing outcomes depend on which DNA repair pathway is utilized. In this section, we summarize factors that can affect DNA repair outcomes of Cas9-induced DSBs and describe DNA repair outcomes of Cas9-induced DSBs.

Cell cycle

The cell cycle is one of the most important factors affecting DSBs repair pathway choice [82–85]. cNHEJ is active across all stages of the cell-cycle, whereas SSA and HR function mainly during S/G2 (Figure 3A) [82–85]. Aphidicolin which arrests cells in S/G2 promotes repair of Cas9-induced DSBs via SSA and HR [194]. Fusing Cas9 to the protein Geminin, which is a substrate for proteasomal degradation in G1, restricts Cas9 activity to the S/G2 phases of the cell cycle when SSA and HR are functional [195]. The recently developed technique of “very fast CRISPR on demand”, in which Cas9 prebinds to its target in genomic DNA in an inactive form and Cas9 is activated by light to cleave the target DNA in seconds, can introduce DSBs when cells enter S/G2 phases without drug treatment [96].

Figure 3. Factors affecting DNA repair outcomes of Cas9-induced DSBs.

(A) DSB repair pathways during the cell cycle. cNHEJ is active throughout the cell cycle (black circle). MMEJ and HDR can only be employed in S/G2 phases (green circle). (B) In the absence of a template guided repair mechanism editing outcomes are significantly affected by the target site sequence. MMEJ efficiency and the pattern of DNA small deletions are dependent upon the presence of microhomologies within the first 10 bp from the DSB end. 1–2 bp insertions/deletions through cNHEJ is affected by the nucleotide at the 4^th position upstream from the PAM. (C) Nucleosomes can potentially block Cas9 access to target site, although the site can be exposed through nucleosome breathing or by a nucleosome remodeler. Compacted heterochromatin promotes HR, MMEJ, and SSA, while cNHEJ is preferred in euchromatin. (D) Homology-directed repair (HDR) with ssDNA or dsDNA donor templates. Double-stranded DNA donor templated repair (DSTR) occurs mainly through HR and single-stranded DNA donor templated repair (SSTR) occurs mainly through SSA and SDSA. Asymmetric HDR can arise when one end is repaired through HR, while the other end is repaired through cNHEJ or MMEJ. With ssDNA, the end that is complementary to the 3’ end of ssDNA templates is repaired through SSA/SDSA, while how the other end is repaired is unknown. The repair of the other end through MMEJ generates asymmetric HDR which displays a bias directionality with respect to the orientation of the ssDNA templates.

Target site sequence and chromatin structure

Although DNA editing outcomes vary considerably among different genomic sites [125, 126], the editing outcomes are not completely arbitrary. For template-independent repairing pathways (e.g. cNHEJ or MMEJ), the editing outcomes are significantly affected by the target site sequence (Figure 3B). A single protospacer targeting different genomic sites yielded similar repair events (Figure 3B) [127]. It has been shown that the nucleotides adjacent to Cas9 cutting site can affect editing outcomes [28, 124–126]. The most common repair outcomes (up to 50%) are small deletions (>3bp) [124]. These small deletions are MMEJ products indicating that microhomology adjacent to the cutting site affects the deletion patterns (Figure 3B). Indeed, MMEJ efficiency and repair outcomes are significantly dependent on the microhomologies within the first 10 bp from the break [118]. Short 1–2 bp insertions/deletions are another dominant repair outcome (up to 42%) and the relative percentage is affected by the nucleotide at the 4^th position upstream from the PAM (see below) (Figure 3B) [124, 196, 197]. These results indicate that the sequence adjacent to the Cas9 cut site affects DSB repair outcomes.

In eukaryotic cells, DNA is bound by histones to form nucleosomes, the basic unit of chromatin [198]. Cas9 cleavage efficiency is highly dependent on DNA accessibility and nucleosomes can pose a barrier to Cas9 (Figure 3C) [199]. Intrinsic nucleosome breathing or changes in nucleosome architecture arising from the action of chromatin remodeling proteins can increase Cas9 cleavage efficiency (Figure 3C) [199]. Local chromatin structure also affects the choice of DSB repair pathways [200]. For example, histone post-translational modifications control the recruitment of DNA repair proteins such as 53BP1 and BRCA1, both of which influence DSB repair (Figure 3C) [198]. Compacted heterochromatin and transcriptionally active euchromatin also have different impacts on DNA repair [198]. Surprisingly, DSB end-resection is not impeded by heterochromatin [200]. Indeed, the H3K9 trimethylated (H3K9me2/3) present in heterochromatin promotes HR, MMEJ, and SSA, while cNHEJ is preferred for euchromatin (Figure 3C) [200]. Detailed mechanisms of how local chromatin structure affects DNA repair pathways remains unclear and current studies are seeking to elucidate how chromatin structure affects Cas9 mediated genome editing [198].

Homology-directed repair (HDR) with ssDNA or dsDNA donor templates

SSA and HR are template-dependent pathways (Figure 2E & 2G) [10, 13]. During CRISPR-Cas9 induced-DSB repair, template-dependent DSB repair pathways use an exogenous donor template to achieve precise gene editing [1]. Double-stranded DNA donor templated repair (DSTR) occurs mainly through the HR pathway, whereas single-stranded DNA donor templated repair (SSTR) occurs mainly through SSA and SDSA (Figure 3D) [24, 201]. In recent years, it has become clear that mis-integration events arise from MMEJ due to microhomologies within the DNA templates (Figure 3D) [201]. Strategies that suppress MMEJ may improve both DSTR and SSTR based HDR.

For DSTR, the donor dsDNA can be either a PCR product or an exogenous plasmid [1]. For PCR products, mis-integration events arise mainly when the PCR products are ligated directly into a DSB through cNHEJ (Figure 3D) [201]. Inhibition of cNHEJ or 5’ covalent modification of the PCR products to prevent direct ligation can significantly increase the precision of gene editing by directing repair through HR [201]. In addition, a chromatin donor template is more efficient in D-loop formation than a naked DNA donor template in vitro [202]. In vivo, dsDNA donor templates coated with nucleosomes increase the frequency of DSTR-mediated HDR indicating that strategies that stimulate HR may be used to improve HDR [203]. For plasmid-based dsDNA templates, cells can incorporate the desired template along with the undesired plasmid backbone sequence into the genome [201]. Integration of the plasmid backbone is probably caused by asymmetric HDR in which precise editing through HR in one side of DSB and direct ligation through cNHEJ or MMEJ at the other side of DSB (Figure 3D) [201]. Integration of the plasmid backbone into the genome might be reduced by transiently blocking cNHEJ or MMEJ [201]. Indeed, HDR can be significantly improved by introducing two Cas9 cleavage sites at both sides of the homologous fragments to excise the desired template from the plasmid backbone [204, 205].

DSTR is more efficient than SSTR in S. cerevisiae, but SSTR- and DSTR-mediated genome editing occur with comparable efficiencies in mammalian cells [24]. SSTR is a more attractive option because short homology arms (around 70–100 bases compared to the 0.6–1.0 kb requirement for DSTR) can achieve efficient HDR and short ssDNA templates can be easily synthesized with multiple modifications to improve transformation efficiency and in vivo stability (Figure 3D) [206, 207]. In S. cerevisiae, SSTR can be mediated through the SSA pathway and depends on Rad52, Rad59, anti-recombinase Srs2, MRX complex [139]. Thus, factors that inhibit SSA impair SSTR including Rad51, Rad57, and Rdh54 [139]. Similarly, SSTR in mammalian cells is also RAD51-independent and requires RAD52 [208]. RAD52 over-expression or fusion of RAD52 to Cas9 promotes SSTR mediated genome editing efficiency [208]. In addition to the SSA pathway, SDSA is also used during SSTR mediated DNA repair. Asymmetric HDR when using ssDNA donor templates is more prevalent than that of the dsDNA donor templates and displays a bias directionality with respect to the orientation of the targeted insertion (Figure 3D) [201]. These results may indicate that one end of the ssDNA donor is annealed to one of the DSB ends and repaired through SDSA or SSA, while the other end of the ssDNA donors may be repaired through non-HDR mechanism (Figure 3D) [201]. Although short ssDNA donor templates are more efficient than the dsDNA donor templates in promoting genome editing, long ssDNA donors induce more error-prone repair events than dsDNA donors because the microhomology sequences within long ssDNA donors generate truncations.

DNA repair outcomes of Cas9-induced DSBs

As summarized above, cell cycle and end resection properties are important factors in determining DNA repair pathway choice between cNHEJ, MMEJ, SSA, and HR (Figure 2, 3A). DNA repair outcomes are dependent on Cas9-induced DSB formation and detection time (Figure 4). The time required for Cas9 to generate a DSB has been studied extensively and multiple strategies have been developed to generate DSBs in S/G2 when HDR pathways are available to enhance HDR. However, the time required for the cell to detect a Cas9-induced DSB is not currently clear and may be different from the time needed to detect a naturally occurring DSB. In vitro, Cas9 remains tightly bound for hours after cutting the DNA [26]. Although Cas9 dissociation from the DSB is much faster in vivo, it takes at least 10 minutes to detect Cas9-induced DSBs, which is much longer than the time needed to detect normal DSBs [96]. Delayed detection of Cas9-induced DSBs may influence the choice of DNA repair pathway (Figure 4) [97, 196]. One study has demonstrated that there may be a delay on the order of 10 minutes between the time of which Cas9 cleaves DNA in vivo and the time at which the cell detects the resulting DSB [96]. In addition, another study has suggested that Cas9 may remain more tightly bound to the PAM-distal DSB end even after the proximal end is released [97]. One might speculate that tight binding of Cas9 on the PAM-distal DSB end may alter repair kinetics and/or mechanism [196]. Factors such as RNA-polymerase and the histone chaperone facilitates chromatin transcription (FACT) that promote Cas9 release may affect DNA repair outcomes [97, 98]. cNHEJ is the major repair pathway to repair ionizing radiation (IR) induced DSBs and MMEJ is less efficient (1% of all DSBs) [209]. However, MMEJ is improved at least 2-fold for Cas9 induced DSBs [196]. Indeed, up to 58% of Cas9-induced DSBs can be repaired through MMEJ [124]. These results indicate the possibility that the asymmetric release of Cas9-induced DSB ends may somehow enhance MMEJ while disfavoring cNHEJ.

Figure 4. The outcomes of Cas9-induced DNA double-strand break repair.

Once Cas9-gRNA complex binds to its target site, the HNH domain accurately cuts at the target strand −3 bp upstream of the PAM, while the RuvC-like domain cuts the −3, −4, or −5 positions of non-target strand. These cleavages can generate DSBs with either blunt ends or 1–2 nt staggered ends. Blunt ends can be ligated directly through cNHEJ without introducing any mutations, while staggered ends need to be filled or cleaved before ligation resulting 1–2 bp insertions or deletions. End resection directs DSB repair through template-independent MMEJ or template-dependent HDR. For MMEJ, small homologous DNA sequences (5–25 bp) within the two resected ends are paired and lead to DNA repair resulting in small deletions or insertions. In the presence of either ssDNA or dsDNA templates, homology directed repair (HDR) compete with cNHEJ and MMEJ for precise DNA repair.

Recently, it has been shown that DSB ends generated by Cas9 are either blunt, or have a 1–2 bp staggered 5’ end (Figure 4) [28]. Blunt ends can be ligated directly by the LigIV-XRCC4-XLF complex through cNHEJ (Figure 4A), while staggered ends are not compatible and require further processing before repair (Figure 4B–C). It has been shown that cNHEJ is a major pathway used to repair the 5’ staggered ends resulting in 1–2 bp insertions or deletions (Figure 4). It has been proposed that the 1–2 bp staggered ends may be filled in by a polymerase or removed by a nuclease to create a blunt end, followed by ligation through cNHEJ (Figure 4A). In addition, the 5’ staggered ends slow cNHEJ repairing kinetics and the delay time may be long enough for cells to initiate short-range end resection to promote the MMEJ, and HDR pathways or both short-range and long-range end resection to promote the HDR pathways (Figure 3D).

Recently, increasing evidence has suggested that large gene deletions can happen at on-target sites during CRISPR-Cas9 editing [210–212]. Unplanned large deletions extending to several kilobases were detected during mouse and human cell genome editing [210]. The large deletions were also detected during embryos genome editing [211]. More recently, Alexey et al observed a size of ~293 kb deletion near the CRIPSR-Cas9 on-target site in mouse zygotes [212]. The mechanism of large deletions remains unclear, but evidence suggests that large deletions are mainly caused by the DSBs generated by CRISPR-Cas9 in a p53-dependnet mechanism [213]. Although the frequency of large deletion events is very low, these events may cause a serious side effects during genome editing. Studies to elucidate the mechanisms of how large deletions are generated will be beneficial for clinical applications.

CONCLUDING REMARKS

CRISPR-Cas-based genome editing has allowed for precise genome manipulation, raising the possibility of correcting gene mutations in patients [1, 214]. However, it is important to recognize that the efficiency and accuracy of genome editing depends upon the four major DSB repair pathways found in eukaryotic cells. Different pathways result in differences in genome editing accuracy and efficiency. Hence, extensive efforts are being made to control which pathway cells utilize to repair DSBs during genome editing. However, the regulatory mechanisms that control DSB repair are highly complex and are not yet fully understood [10–13, 82–85, 198]. Understanding the regulation of DSB repair mechanisms will likely yield fundamental insights for both understanding genome integrity and will help establish better, more precise genome editing strategies.

HIGHLIGHTS:

1. The CRISPR-Cas9 mediated genome editing offers a powerful approach as a potential therapy for monogenic human genetic diseases.

2. Precise template-free base deletions can be achieved through MMEJ repair and depend on local target site sequence.

3. HDR-related DNA repair pathways in response to CRISPR-Cas9 induced-DSB in mammalian cells are complicated and relatively inefficient. Different DNA repair pathways might be used to repair each end at a DSB resulting in the potential for asymmetric repair.

4. The DNA repair pathway choice in CRISPR-Cas9 induced-DSBs is regulated by several key factors including the cell cycle, target site sequence and chromatin structure, and the identity of the donor DNA template.

Outstanding questions:

1. After cleavage, Cas9 asymmetrically releases the PAM-proximal end of the DSB within minutes. How is the PAM-proximal end of the DSB released? How long does Cas9 remain bound to the PAM-distal end of the DSB after the PAM-proximal end is released? How is the PAM-distal end of the DSB released? How might asymmetric release of Cas9 from a DSB affect repair pathway choice?

2. Long-range end resection commits DSB repair to the HR and SSA pathways. How is end resection regulated by BRCA1-BARD1, 53BP1 and the Shieldin complex? What is the temporal recruitment order of these regulators?

3. For SSTR, how do mammalian cells choose between SSA and SDSA pathways? And what is the relative contribution of each pathway to repair? Do other pathways also contribute to SSTR?

4. How do DNA templates in the cytosol enter the nucleus? How do foreign DNA sensing pathways, including TLR9, AIM2, and cGAS, affect genome editing? Is it possible to transiently inhibit foreign DNA sensing pathways during template delivery?