Making ends meet: Targeted integration of DNA fragments by genome editing

Abstract

Targeted insertion of large pieces of DNA is an important goal of genetic engineering. However, this goal has been elusive since classical methods for homology-directed repair are inefficient and often not feasible in many systems. Recent advances are described here that enable site-specific genomic insertion of relatively large DNA with much improved efficiency. Using the preferred repair pathway in the cell of nonhomologous end-joining, DNA of up to several kb could be introduced with remarkably good precision by the methods of HITI and ObLiGaRe with an efficiency up to 30-40%. Recent advances utilizing homology-directed repair (methods of PITCh; short homology arms including ssODN; 2H2OP) have significantly increased the efficiency for DNA insertion, often to 40-50% or even more depending on the method and length of DNA. The remaining challenges of integration precision and off-target site insertions are summarized. Overall, current advances provide major steps forward for site-specific insertion of large DNA into genomes from a broad range of cells and organisms.

Introduction

In order to determine gene function, the initial approach in the field of genetics was to use X-irradiation or chemical mutagenesis to create mutations genome-wide (or to study spontaneous mutations), and then visualize phenotypic changes and map the mutated genes on the chromosomes. With the advent of molecular approaches, it became possible to mutate a desired stretch of DNA and then insert it into the genome for fine-scale mapping of functional regions of the gene or its regulatory elements. Initially, integration of foreign DNA into eukaryotic genomes occurred at ectopic locations beyond the control of the investigator. T-DNA was the first vector used for genetic engineering in plants [Bevan and Chilton 1982]. At about the same time, the P-element transposon was the first vector used for transformation of a multicellular animal, accomplished in Drosophila [Rubin and Spradling 1982; Spradling and Rubin 1982]. Subsequently, several other transposons were used as transformation vectors for a variety of other insects [O’Brochta and Atkinson 1996]. One of the most versatile of these is piggyBac [Lobo et al. 1999; Horn and Wimmer 2000; Handler 2002]. Generally, the size of integrated DNA was <20 Kb. Subsequently, integration of several kb DNA could be introduced at 55% efficiency (five-fold higher than with P-elements) using phiC31, but the placement of the phage att site for integration was still random with regard to the genome [Groth et al. 2004]. Depending on the site of genomic insertion, position effects influenced the extent of gene expression. To overcome this problem, many transformants at various genomic locations were analyzed to obtain their average level of gene expression. For comparative studies of several different mutations in the same gene, techniques were developed to always introduce the mutated DNA into the same ectopic location with cassette exchange methods such as Cre-Lox or FLP-FRT. However, unlike phiC31, Cre/loxP and Flp/FRT are reversible. A major advance in Drosophila was the development of P[acman], which is a BAC transgenic platform for targeted insertion of large DNA fragments over 100 kb [Venken et al. 2006; Venken et al. 2009]. Similarly, inserts of 80 kb into an att site using phiC31 was demonstrated in Drosophila [Wesolowska and Rong 2013].

Although the advances above were suitable for molecular dissection of gene function by mutation, they did not satisfy the need for site-specific insertion of DNA for use in genome editing to cure genetic diseases. Moreover, site-specific insertions into the endogenous locus overcome position effects of random integration and allow the investigator to analyze gene expression at levels dictated by its normal genomic location. This goal became a reality with the advent of programmable nucleases as molecular scissors to cleave a single desired site in the genome into which DNA could be inserted. This technology allows insertion of DNA to study its function by engineered mutations. In addition, it allows locus-specific insertion of reporter genes as well as insertion of tagged genes. Finally, these approaches can be used for gene replacement in gene therapy. The focus of this review is on recent advances for site-specific insertion of relatively large DNA.

Programmable site-specific nucleases

Zinc Finger Nucleases

A major advance for site-specific insertions was the development of zinc finger nucleases (ZFNs) [Urnov et al. 2005; Carroll et al. 2010; Urnov et al. 2010; Carroll 2011]. Each zinc finger recognizes 3 bases and usually 3-6 fingers are used per ZFN to recognize 9-18 bases (Fig. 1a). The individual zinc fingers or two finger modules are chosen from a library so that in principle the large variety of sequence combinations makes it possible to target any sequence of interest in the genome. Because ZFNs work as a heterodimer, the number of bases recognized is doubled to 18-36, thereby enhancing target specificity. For each ZFN, the array of zinc fingers is coupled with the cleavage domain of the restriction enzyme FokI [Li et al. 1992; Kim et al. 1996]. Thus, in the ZFN chimera the zinc fingers provide targeting to the desired site in the genome and the FokI nuclease cleaves at that site, giving rise to overhanging ends (Fig. 1a). Despite the length of the target sequence, off-target cleavages were sometimes seen, often reflecting the erroneous formation of a ZFN homodimer. This problem was minimized by alteration of the amino acids at the ZFN interface so that only heterodimers were permissible, thus lowering off-target insertions and giving increased insertion site specificity [Miller et al. 2007; Szczepek et al. 2007; Söllü et al. 2010; Doyon et al. 2011; Ramalingam et al. 2011]. The introduction of ZFN methodology ushered in the era of programmable nucleases to target specific sequences for genome engineering.

Fig. 1. Programmable nucleases for site-specific cleavage in the genome.

(a) ZFNs: Each zinc finger (ZF) binds to 3 nucleotides of DNA. An array of 3-4 ZFs is coupled with the FokI nuclease for cleavage. (b) TALENs: Each TALE domain binds to a single nucleotide. An array of TALEs is coupled with the FokI nuclease. For both (a) and (b), FokI nuclease acts as an obligate heterodimer where the nuclease domain cleaves the spacer sequence to create a DSB with cohesive ends, (c) The sgRNA contains a 19–20 base sequence (red) within the CRISPR RNA (crRNA) that is complementary to the chromosomal target site, and this is combined with a transactivating CRISPR RNA (tracrRNA) (yellow) to guide the Cas9 nuclease (blue oval) to the target site where it creates DSBs with blunt ends 3 bases upstream of the PAM sequence (green).

An unanticipated problem with the design of ZFNs is that the specificity of individual zinc fingers can be influenced by the neighboring fingers [reviewed by Chandrasegaran and Carroll 2016]. This necessitated further work and validation that the final array of zinc fingers did indeed cut the desired sequence.

TALENs

Even though the problems in ZFN construction could be overcome, their high cost led to a search for alternative programmable nucleases. Transcription activator-like effector nucleases (TALENs) provided such an alternative. The TALE DNA binding module in plant virulence factors contains repeating units of 33-35 amino acids that are virtually identical except for amino acids 12 and 13 that impart specific nucleotide recognition for that TALE motif. Thus, the tandem array of amino acids recognizes the DNA sequence composed of nucleotides specified by each individual TALE motif [Boch et al. 2009; Deng et al. 2012; Mak et al. 2012] (Fig. 1b). This means that the tandem array of TALE motifs can be customized to recognize a desired DNA sequence target binding site. This is fairly straight-forward [Reyon et al. 2012], unlike the custom design of ZFNs. Moreover, the binding of each module to a single base pair is not influenced by context. Like ZFNs, the TALE that determines the target DNA binding site is coupled to FokI nuclease to direct DNA cleavage to a unique site in the genome [Christian et al. 2010; Miller et al. 2011; Mussolino et al. 2011]. However, the large size of the genes encoding TALENs and their repetitive nature introduced some difficulties into their production [reviewed by Chandrasegaran and Carroll 2016].

CRISPR-Cas

The CRISPR-Cas system used by bacteria and archaea to degrade invading DNA was harnessed as a powerful tool for genetic engineering [Jinek et al. 2012 and reviewed by Barrangou and Marraffini 2014; Doudna and Charpentier 2014; Hsu et al. 2014; Sander et al. 2014]. Unlike ZFNs and TALENs that are both protein-targeted nucleases, CRISPR-Cas9 utilizes RNA that guides the associated nuclease to its target site for cleavage. The guide RNA contains a short 19–20 base element within the CRISPR RNA (crRNA) that is complementary to the chromosomal target site, and this acts with a trans-activating CRISPR RNA (tracrRNA) to guide the Cas9 nuclease to the target site (Fig. 1c). The crRNA/tracRNA can be combined into a single guide RNA (sgRNA) that targets Cas9 to its site of cleavage [Jinek et al. 2012]. Cleavage of genomic targets with the CRISPR-Cas9 system requires a PAM sequence (eg, NGG for S. pyogenes Cas9) in the chromosome directly adjacent to the protospacer element of the target site. The guide RNA (gRNA) directs Cas9 to create a double-strand break (DSB) at the target site, 3 bp upstream of the PAM site. However, there can be some sequence-dependent variability of the efficiency of CRISR/Cas9 cleavage [Xu et al. 2015; Liu et al. 2016].

Overall, CRISPR-Cas9 has become the favored site-specific nuclease, with applications in a broad range of organisms (summarized by Chandrasegaran and Carroll 2016]. Off-target cleavage was an initial problem, but several approaches have been designed to minimize this (reviewed by [Barrangou and Doudna 2016; Chandrasegaran and Carroll 2016] and discussed further below). Creating the guide RNA is much simpler and less costly than the designer proteins to target ZFNs or TALENs to the desired genomic sequence for cleavage.

Cellular pathways for repair of DNA double-strand breaks (DSBs)

Success with programmable nucleases as site-specific molecular scissors allowed the development of genetic engineering (“genome editing”) that has now been successfully accomplished in several kinds of animals and plants [reviewed by Urnov et al. 2010; Chandrasegaran and Carroll 2016; Gaj et al. 2013; Segal and Meckler 2013; Carroll 2014; Peng et al. 2014; Salsman and Dellaire 2017]. Cells utilize two main pathways to repair DNA breaks: homology-directed repair (HDR) and classical nonhomologous end-joining (c-NHEJ) (Fig. 2). The mechanisms for the repair pathways have been summarized recently [Ceccaldi et al. 2016; Jasin and Haber 2016; Rodgers and McVey 2016; Danner et al. 2017] and are beyond the scope of this review. Briefly, HDR utilizes several hundred bases of homology on each side of the DSB in contrast to NHEJ where homology is not needed. HDR is maximal in late S phase and G2 whereas NHEJ occurs throughout the cell cycle and is the preferred pathway in G1 when resection activity is low [Sfeir and Symington 2015].

Fig. 2. The major repair pathways.

DNA DSBs are repaired by nonhomologous end-joining (NHEJ) or by homology-dependent repair (HDR). When a DNA donor is present, it can be integrated, as shown for HDR. In the absence of a DNA donor, small insertions and deletions (indels) are created by NHEJ.

Another pathway is termed alternative NHEJ (a-NHEJ, also called alt-EJ). a-NHEJ involves PARP1, the MRN complex and its partner CtIP, whereas c-NHEJ uses the Ku70/80 heterodimer and DNA-PK catalytic subunit [Dueva and Iliakis 2013]. DSB repair occurs by a-NHEJ when the c-NHEJ pathway is compromised by the absence of Ku [Fattah et al. 2010].

The DNA ends are resected in HDR but not in NHEJ. The resected 3’ single-strand DNA (ssDNA) overhangs can be repaired by any of three mechanisms: (i) HDR, (ii) alternative NHEJ (a-NHEJ) including microhomology-mediated end-joining (MMEJ)[Sfeir and Symington 2015), and (iii) single-strand annealing (SSA)[Bhargava et al. 2016]. The DNA ends at the DSB are bound by proteins: Ku70/Ku80 (in c-NHEJ) or MRN (in HDR and a-NHEJ). The other proteins downstream of MRN in the repair pathway differ for HDR and a-NHEJ. We focus on how all these various pathways have been employed for use in genome editing, especially with regard to insertion of relatively large DNA into cells from multicellular eukaryotes.

Homology-directed repair

Insertion of DNA that is several kb long is readily accomplished in yeast by homology-directed repair (HDR) [Scherer and Davis 1979; Orr-Weaver et al. 1981]. Although this is also possible in higher eukaryotes as first demonstrated in mouse cells [Smithies et al. 1985; Thomas et al. 1986; Mansour et al. 1988; Capecchi 1989], the efficiency is low, though the efficiency can be improved to one in ten thousand by increased homology [Deng and Capecchi 1992] or by induced DSBs [Rouet et al. 1994; Smih et al. 1995]. HDR occurs during S and G2 of the cell cycle reflecting the cell cycle regulated interaction of BRCA1–PALB2 that recruits BRCA2 [Orthwein et al. 2015]. Moreover, in late S and G2 a sister chromatid is present for repair utilizing hundreds of bases of homologous DNA sequences that flank the DSB [Branzei and Foiani 2008; Mladenov et al. 2016]. For genome editing, artificially introduced donor DNA with 0.5-2 kb homology arms is used instead of the sister chromatid for HDR at targeted DSBs created by site-specific molecular scissors (Fig. 2). Early experiments demonstrated that the rare-cutting endonuclease I-SceI [Rouet et al. 1994; Smih et al. 1995] or a ZFN-induced DSB [Bibikova et al. 2001] provided the entry point for HDR, thus ushering in the possibility of genome editing. In another early experiment, ZFNs were used to create site-specific cleavage in Drosophila embryos that was repaired by NHEJ in the absence of donor DNA or repaired by HDR in the presence of donor DNA [Bibikova et al. 2002; Bibikova et al. 2003]. Soon thereafter, ZFNs were used to repair a reporter insert [Porteus and Baltimore 2003] and endogenous genes in human cells [Urnov et al. 2005].

Using long stretches of homology (500-800 bp homology arms and ranging from 200 bp to several kb in other experiments), programmable nucleases coupled with the HDR pathway have been used for precise integration of transgenes from 8-15 kb in cultured cells [Moehle et al. 2007; Jiang et al. 2013]. However, the efficiency of DNA insertion is usually low, creating problems for targeted gene insertion within genomes of whole organisms, especially where large-scale screening can be difficult.

So far, targeted DNA insertion by HDR has only been reported in a few whole organisms. Instead, in most systems, DSBs are repaired by nonhomologous end-joining (NHEJ) which is the preferred pathway (e.g., by ~10-fold in Drosophila [Beumer et al. 2008; Carroll et al. 2010]). In order to increase the frequency of HDR, some investigators have disabled the NHEJ preferred pathway by knocking out DNA ligase 4 in C. elegans [Morton et al. 2006] or in Drosophila [Beumer et al. 2008; Bozas et al. 2009; Beumer et al. 2013], or by using the NHEJ inhibitor SCR7 in mammalian cells or embryos [Chu et al. 2015; Maruyama et al. 2015; Singh et al. 2015] or reducing the NHEJ pathway by other means such as gene silencing of Ku70, Ku80 or DNA ligase I [Chu et al. 2015]. Similarly, HDR efficiency can be increased by timed delivery into synchronized cells [Lin et al. 2014], since HDR is maximal in late S and G2. These and other strategies have been reviewed recently [Salsman and Dellaire 2017], but overall there is only a 5 to 10-fold increase in HDR. Thus, even with these approaches, HDR in its classical form may occur at too low a frequency to be practical for introduction of large DNA fragments into intact organisms. Promising recent advances are described later in this article for genomic insertion of DNA by homology.

Nonhomologous end-joining

The nonhomologous end-joining (NHEJ) repair pathway occurs throughout the cell cycle. Since it is preferentially used rather than HDR by most cells and organisms [Sedivy and Sharp 1989; Bozas et al. 2009; Iyama and Wilson 2013], it should be more efficient for targeted gene insertion in whole organisms. Typically the error-prone nature of NHEJ in the absence of donor DNA results in small insertions or deletions (indels) that cause gene disruption at the target site, and this has been the primary use so far of NHEJ (Fig. 2). Indeed, CRISPR-Cas9 technology has become one of the most popular ways to inactivate a gene due to disruptions created by indels.

Previously, the integration of large DNA fragments via NHEJ has been inefficient as well as imprecise and not widely used. One route that has been implemented to improve efficiency is to couple NHEJ with ligation in vivo (“end-capture”) of compatible overhangs created in the donor DNA and the genomic target site by molecular scissors such as ZFNs (Fig. 3) [Orlando et al. 2010; Cristea et al. 2013; Weinthal et al. 2013]. The cohesive ends of the donor and target can pair, resulting in a product with the donor inserted at a single orientation. However, since the cleavage target sites have thus been recreated in the product, they can be re-cleaved until mutations (such as deletions by exonucleases) occur to destroy the target sites. As a result, the donor DNA is then integrated into the genomic target site by NHEJ with imprecise junctions. Potentially, exonucleolytic deletions can also occur in the donor DNA upon its initial introduction into the cell in the short window of time when the ends are not protected by the ZFNs [Cristea et al. 2013]. The donor DNA cannot be inserted readily in the opposite orientation since the cohesive ends would not be complementary between the donor and the genomic target sites (pathway at bottom right of Fig. 3). However, it is possible that exonuclease digestion could resect the ends of the donor until areas of microhomology with the genomic target region are encountered and integration could occur by microhomology-mediated end-joining (MMEJ, discussed below). Both pathways shown in Fig. 3 to produce stable products after end-capture would occur at relatively low frequency and both would result in imprecise junctions.

Fig. 3. End-capture of cohesive ends.

Diagram of “end capture” of a piece of donor DNA at the genomic site of a nuclease-induced (red arrows) DSB. Two pathways could result in stable products, but at low efficiency and with imprecise junctions. Left: The overhanging ends of the donor DNA are complementary to the overhanging ends of the genomic target site, since both the donor and genomic DNAs were cleaved by the same ZFN. However, end capture recreates the target sites that can be re-cleaved. Recutting and religation will be repeated until the target site sequence is altered by mutation and can no longer be cleaved by the programmable nuclease. Small red Xs indicate the mutated sequences. Right: Donor DNA that inserts with an opposite orientation to that shown in the left pathway will have incompatible ends to the genomic target sites, but it can be integrated via microhomology with very low efficiency.

Although the FokI nuclease of ZFNs and TALENs create cohesive ends suitable for end-capture, Cas9 creates blunt ends that are not suitable for end-capture. One recent advance has been to fuse Fok1 to an inactivated Cas9 so that overhanging ends are created by Fok1 after CRISPR direction to the target site [Guilinger et al. 2014; Tsai et al. 2014]. Alternatively, CRISPR can be used with Cpf1 that creates overhanging ends rather than the blunt ends generated by Cas9; Cpf1 is a single RNA-guided endonuclease of a class 2 CRISPR-Cas system [Zetsche et al. 2015]. End-capture works best when the plasmid is cleaved after introduction into the cell, rather than prior linearization [Cristae et al. 2013; Auer et al. 2014; Bachu et al. 2015]. However, the efficiency for site-specific genomic insertion of DNA fragments by NHEJ with end-capture was not much higher than for HDR [Orlando et al. 2010; Cristea et al. 2013; Weinthal et al. 2013; Yamamoto et al. 2015].

CRISPR-Cas9 site-specific cleavage to create blunt ends has been used for DNA insertion by NHEJ without end-capture. Using this approach, insertion of ~5 kb into mammalian cells by NHEJ after CRISPR-Cas9 breaks in both the plasmid and genomic target site only had efficiencies of 0.17% in HEK293 cells and 0.45% in CHO cells, and there were indels at the insertion junctions [Bachu et al. 2015]. In contrast, ~6 kb DNA was inserted into the zebrafish genome by CRISPR-Cas9-mediated NHEJ knock-in with an efficiency of 10% for in-frame germline integrations and transmission to future generations of 31%, but the precision was low and indels were present at the integration junctions [Auer et al. 2014]. A similar approach created >10 kb (and as large as 34 kb) insertions in human cell lines and embryonic stem cells [He et al. 2016]. However, indels were often at the insertion junctions in these reports. The use of CRISPR/Cas9 and NHEJ has been optimized in the HITI methods described below.

Recent advances for site-specific insertion of relatively large DNA by NHEJ

Many shied away from use of NHEJ for site-specific insertion of DNA due to its reputation of imprecision, with indels at the junctions of the insertion. The property of imprecision is a hallmark of the alternative NHEJ pathway (a-NHEJ) and a lesser problem for the canonical NHEJ pathway (c-NHEJ). Although indels occur when the ends of the DNA DSBs are not compatible, c-NHEJ can achieve 75% or more error-free repairs according to Bétermier et al. [2014].

Two new methods are described below that utilize NHEJ with improved efficiency for site-specific insertion of large DNA. One possibility to explain the higher efficiency is that the experimental design prevents reformation of the target site after the insertion has occurred and therefore re-cutting is prevented. The examples described below and depicted in the accompanying figures show site-specific insertion of the donor DNA after creating a single cut at the target site. These methods are readily adaptable to gene replacement when two cuts are made in the genomic DNA, with each cut flanking the gene to be replaced.

Homology-Independent Targeted Integration (HITI)

Recently, extensive optimization of CRISPR-Cas9 with NHEJ (called HITI = homology-independent targeted integration) (Fig. 4) was used for insertion of DNA [Suzuki et al. 2016]. In HITI, the sgRNA of CRISPR cleaves the same target sequence in the genome and flanking the gene of interest in the donor plasmid. Subsequently, the gene of interest is inserted into the genomic target site by NHEJ. The gene of interest can be inserted in either orientation, but if the sgRNA target site is re-created, the donor DNA is presumably excised and re-inserted until ultimately it is in the final stable orientation where the sgRNA target site is no longer present. When used for DNA site-specific integration into proliferating HEK293 cells, the efficiency was 30-40%, and even somewhat better after optimization of the nuclear localization sequence that was used for Cas9. Although the majority of the cells had precise integrations, up to a quarter exhibited imprecise integrations with indels.

Fig. 4. Homology-independent targeted integration (HITI).

The sgRNA of CRISPR cleaves the same target sequence in the genome and flanking the gene of interest in the donor plasmid (the blue box indicates the beginning of the target sequence and the green box indicates the end of the target sequence). The gene of interest is subsequently inserted into the genomic target site by NHEJ in either orientation. If the sgRNA target site is re-created, the inserted DNA is presumably excised and re-inserted until finally it is in the final stable orientation where the sgRNA target site is no longer present

The HITI approach was also successful for DNA site-specific integration in cultured post-mitotic neurons, with an efficiency of 0.58% of all cells and 56% of the transformed cells [Suzuki et al. 2016]. Moreover, the precision of integration was very high (~90%). When used for insertion into rodent neurons in whole animals where HITI with adeno-associated viral (AAV) vectors were injected into the visual cortex of adult mouse brains, the efficiency of site-specific integration was 3.5% of all cells and 10.6% of infected cells [Suzuki et al. 2016]. Intravenous injection of HITI-AAV vectors into mice resulted in nuclear GFP expression in 3-10% of cells in heart, liver, muscle and other tissues, with good precision comparable to the HEK293 experiments (Suzuki et al. 2016). Note that the HITI-mediated DNA insertions above using whole animals were somatic and not germline events.

ObLiGaRe (Obligate Ligation-Gated Recombination)

Another approach using NHEJ for site-specific genomic introduction of large DNA is obligate ligation-gated recombination (ObLiGaRe). The ObLiGaRe method was initially developed for use with ZFN site-specific cleavage of genomic DNA from cultured cells [Maresca et al. 2013] and has been adapted for whole organisms [Yamamoto et al. 2015]. The pair of ZFNs binds to the target DNA sites on opposite strands and forms heterodimers through the modified FokI nuclease domain. In the ObLiGaRe approach, the donor plasmid contains the same ZFN binding sites as the genomic target site to allow end-capture of the overhanging ends after FokI digestion, but the trick is that the ZFN binding sites in the donor plasmid are inverted relative to the host target sequence (Fig. 5). The new junction formed after the integration event prevents subsequent ZFN excision since the ZFN heterodimer target sites are destroyed [Doyon et al. 2011]. This increases the efficiency of formation of the final product. By contrast, in the HITI method, only half of the products have the nuclease cleavage sites destroyed, and the other half are continually re-cleaved until the donor DNA is inserted in the opposite orientation to destroy the cleavage sites of the product (Fig. 4). An improved situation occurs with the ObLiGaRe method where the cleavage sites have been destroyed in all of the products initially (Fig. 5).

Fig. 5. Obligate ligation-gated recombination (ObLiGaRe).

A pair of ZFNs cleaves both the wild type (wt) target site of genomic DNA and the inverted target sites (iv) of the donor construct. Both the wt and iv target sites have the identical spacer sequence between the left and right ZFN binding sites, thus creating complementary overhanging ends between the donor DNA and genomic target for end-capture. Once the linearized donor has been ligated into the genomic target site, the same pair of ZFNs cannot cut (x over red arrow) the newly formed junctions since homodimer formation is prevented by the use of an obligatory heterodimeric FokI nuclease domain. Moreover, the inverted target sequence in the donor plasmid prevents adjacent placement of the green and blue sequences. Thus, the efficiency of formation of the final product of the integrated donor DNA at the genomic target site is increased as re-cutting cannot occur. The single dotted or solid lines represent dsDNA of genomic or plasmid DNA, respectively. In the example shown here, the entire donor plasmid is integrated into the target site.

Therefore, the basis of ObLiGaRe is that two separate DNA binding sites are required to make the inverted target site, thus preventing re-cleavage after genomic integration. [Maresca et al. 2013]. To achieve this, ObLiGaRe uses the FokI endonuclease domain, which has to form a dimer to be active. The heterodimeric nature of ZFNs makes possible the inversion of the target sites in the donor relative to the target. CRISPR-Cas9 in its original form is not compatible with ObLiGaRe because Cas9 is not a heterodimer. However, the newly developed dimeric CRISPR-FokI nuclease where Fok1 is fused to a catalytically inactive Cas9 (dCas9) could be very useful as molecular scissors, providing an alternative to ZFNs or TALENs for future applications [Qi et al. 2013; Guilinger et al. 2014; Tsai et al. 2014].

The transformation efficiency with ObLiGaRe of ~2% in cultured cells was about 5 times higher than using classical approaches of HDR [Maresca et al. 2013]. With ObLiGaRe, site-specific integration of 15 kb DNA into the genome of HCT116 human cells had 75-88% precision at the junctions [Maresca et al. 2013]. When ObLiGaRe was used for site-specific integration in flies, efficient insertion (~27% without any selection) of a 2.4 kb piece of DNA into somatic cell genomes was observed 3 days after injection of embryos and had 87% precision at the insertion junctions [Yamamoto et al. 2015]. Furthermore, ObLiGaRe using a 6.5 kb fragment had 6.9% efficiency for germline site-specific integration, and sequencing the insertion junctions of several fly lines revealed no sequence changes at all, nor was there any evidence for off-target insertions [Yamamoto et al. 2015]. Therefore, ObLiGaRe with NHEJ has quite good efficiency with precise integration and minimal off-target insertions.

Recent advances for site-specific insertion of relatively large DNA by homology

Although classical HDR uses several hundred bases of homology flanking the site for insertion of DNA, some recent approaches have had good success using much smaller length of homology, as described below. Although these recent approaches employ homology between donor and target sequences, they do not always use HDR.

PITCh (Precise Integration into Target Chromosome)

PITCh (Precise Integration into Target Chromosome) (Fig. 6) is mediated by 5-25 bases of microhomology-mediated end-joining (MMEJ). MMEJ is a subset of the a-NHEJ pathway and occurs during M and early S and is independent of Ku and ligase 4, unlike c-NHEJ [Sfeir and Symington 2015]. PITCh has been reported for insertion of relatively large DNA into genomic sites from cultured cells, zebrafish, silkworms and frogs cleaved by TALENs or CRISPR-Cas [Nakade et al. 2014; Hisano et al. 2015; Sakuma et al. 2016]. Moreover, MMEJ appears to be even more efficient than c-NHEJ (79% MMEJ vs. 53% c-NHEJ for genomic integration in zebrafish embryos [Hisano et al. 2015]). For the PITCh system, a programmable nuclease creates a DSB in the donor DNA and the genomic target site with subsequent DNA insertion stimulated by MMEJ. When a single DSB occurs in the plasmid donor, then the plasmid carrying the DNA of interest is inserted at the genomic target site. In contrast, just the DNA of interest (without the plasmid) is inserted at the genomic target site when two DSBs flank the DNA of interest in the plasmid. In version 2 of CRIS-PITCh [Sakuma et al. 2016], targets for the CRISPR guide RNA are in the plasmid, adjacent to the 20 base microhomologous sequences flanking the DNA of interest. Using CRISPR-Cas9 for PITCh in HEK293T cells, the precision of DNA insertion was good but indels sometimes occurred at the junctions (80% perfect 5’ junction; 50% perfect 3’ junction) [Sakuma et al. 2016], as might be expected since the a-NHEJ pathway can be imprecise. The difference in precision of integration at the two junctions might be explained by MMEJ occurring at just one junction and NHEJ occurring at the other junction.

Fig. 6. Precise integration into target chromosome (PITCh).

Two regions of 5-25 base homology at the genomic target site and flanking the gene of interest in the donor plasmid are shown (blue, green) In version 2 of CRIS-PITCh, CRISPR sgRNA cuts adjacent to the microhomology regions in the plasmid, whereas a different sgRNA cuts between the two blocks of microhomology in the genome. The MMEJ pathway (a subset of a-NHEJ) uses the blocks of microhomology to insert the gene of interest into the desired genomic site. Note that the two blocks of microhomology are separated in the end product of PITCh so that the guide RNA that initially cleaved the genomic DNA cannot subsequently cleave the genomic end product. Similarly, the target sites for the PITCh guide RNAs are destroyed in the integrated product. The single dotted or solid lines represent dsDNA of genomic or plasmid DNA, respectively.

HMEJ (Homology-Mediated End-Joining)

The method of HMEJ has a similar strategy to MMEJ (Fig. 6) except that homology arms of 800 bases are used rather than 20 base microhomology arms [Yao et al. 2017; Yao et al. 2018. In an exhaustive set of experiments, HMEJ was compared to MMEJ, NHEJ and classical HDR for DNA insertion in numerous cell types in vitro and in vivo as well as mouse and monkey embryos. In almost all cases, HMEJ outperformed the other methods or in a few cases had the same efficiency as MMEJ. The efficiency of DNA insertion by HMEJ ranged from 5-50% depending on the cell type and from 15-30% in mouse embryos. It is unknown how HMEJ will perform for insertions larger than the few hundred bp of the mCherry reporter that was used. Impressively, sequencing revealed precise integrations occurred in almost all cases. Depending on the recipient cells, HMEJ seems to act by homologous recombination, NHEJ or perhaps a novel HMEJ pathway; further research is needed to clarify the mechanism. A similar approach to HMEJ with comparable results of improved efficiency of DNA insertion has also been reported by others who obtained insertion efficiency of up to 30% in stem cells [Zhang et al. 2017].

HDR with short homology arms

Although classical HDR uses long homology arms of 0.5-2 kb, it has been reported that HDR can work with only ca. 50 bases of homology. Specifically, donor homology length is relatively unimportant when a small piece of DNA is integrated into the target site, but the donor homology length becomes more important as the length of the donor DNA increases [Orlando et al. 2010]. HDR efficiency is enhanced when using linear DNAs as repair templates. The genomic target site is cleaved by CRISPR-Cas9 and homology arms of 30-100 bases facilitate the introduction of the donor DNA. The donor DNA can be a double-stranded PCR fragment [Paix et al. 2014 and 2016] or a long ssODN (single-strand oligodeoxynucleotide) (Fig. 7a); in both cases the donor DNA has the homology arms at its ends. It has been reported that the HDR efficiency for ssODNs can be increased to almost 60% in cultured human cells when the homology arms are asymmetric (e.g., 91 nt PAM proximal and 36 nt PAM distal) [Richardson et al. 2016]. The use of ssODN rather than double-stranded DNA (dsDNA) (as used for classical homologous recombination) can increase the insertion by as much as 500-fold, thus making genome editing possible in organisms that were previously intractable [Ferenczi et al. 2017]. Note that DNA cloning is not needed for either ssODN or double-stranded PCR fragments, thus simplifying the procedure.

Fig. 7. Genomic integration with short homology arms.

(a) A single-stranded DNA donor (long ssODN) containing homology arms 30-100 bases (blue, green) is administered together with the CRISPR guide RNA and Cas9 which cuts the desired target site in the genome. Alternatively (not shown) the donor can be dsDNA with two short ssODNs providing the bridging homology arms. (b) For future conditional knockouts (as by Cre-recombinase in mice), two guide RNAs and two short ssODNs (each with a LoxP site) are introduced to result in LoxP sites (triangles) at the genomic sites where Cas9 cleavage had occurred. The LoxP sites allow future deletion for conditional knock-out. (c) A more efficient strategy for future conditional knock-outs is to use a single long ssODN donor with LoxP sites (triangles) flanking the homology arms at the ends (blue, green). As in panel (b), the genomic DNA is cut by Cas9 that is targeted by two guide RNAs. Note that cutting the genome with two gRNAs (panel c) can also be employed for DNA replacement in the genome, in contrast to a simple DNA insertion after cutting the genome with just one gRNA (panel a). Easi-CRISPR employs the method shown here in panel (c): A long single-stranded DNA donor with short homology arms is injected with the pre-assembled Cas9 ribonucleoprotein (ctRNP) complex containing two guide RNAs to create targeted insertion. The ctRNP:complex contains crRNA + tracRNA +Cas9 protein). The dashed line indicates the inserted DNA sequence in the single-stranded donor DNA and in the double-stranded recombinant DNA. The sequence indicated by the dashed line can be deleted readily in future experiments by use of the LoxP sites that flank it. (d) The “two-hit by sgRNA and two oligos with a targeting plasmid” (2H2OP) portrayed here introduces a double-stranded donor plasmid DNA and two short bridging ssODNs (providing homology to the junctions) together with two sgRNAs to direct Cas9 cleavage of the chromosomal target site and of the donor plasmid, respectively.

HDR with these relatively short homology arms is highly efficient in many systems and also can be quite precise (e.g., >95% precision in C. elegans; Paix et al. 2016). Perhaps surprisingly, the efficiency of site-specific integration in C. elegans decreases as the length of homology increases (75% edits for 35 base homology, 47% edits for 60 base homology, 33% edits for 100 base homology) [Paix et al. 2016]. Optimal homology arms of 30-40 bases for ssODN have also been reported in mammalian cells by others [Liang et al. 2017]. Short homology arms also worked to introduce a double-stranded PCR fragment with a 739 bp mCherry insert and flanking homology arms of 36 bases into mouse zygotes with 31% editing efficiency and 100% precision [Paix et al. 2017a], similar to the efficiency found for long ssODNs [Quadros et al. 2017]. However, the efficiency decreased for DNA insertions > ~1 kb and 33 base homology arms [Paix et al. 2017a]. Interestingly, the efficiency of insertion of 714 bp in human cells was the same regardless if the homology arms were 33 bases or 518 bases [Paix et al. 2017a]. This is in contrast to the HMEJ experiments described above [Yao et al. 2017] where long homology arms perform better than short homology arms. A detailed protocol on performing edits of either DNA insertions or deletions using short homology arms has recently been published and contains useful tips [Paix et al. 2017b].

Introduction of DNA with short homology arms appears to work through DNA synthesis-dependent strand annealing (SDSA), rather than direct integration of the linear donor DNA in C. elegans [Paix et al. 2016] and in cultured human cells [Paix et al. 2017a]. Elegant experiments show that the DNA repair that is utilized is consistent with gene conversion [Paix et al. 2017a]. The accuracy of repair is asymmetric where one junction can be precise and the other junction can be imprecise, suggesting different homology requirements to initiate and resolve repair and consistent with SDSA for gene conversion [Paix et al. 2017a].

The experiments above had short homology arms to target DNA to a Cas9-cleaved single genomic site by a single guide RNA. However, a similar approach can be used with two guide RNAs to direct Cas9 cleavage to two genomic sites. This tactic has been employed to introduce two LoxP sites into mice for future conditional knock-outs by Cre recombinase. Use of Cas9 mRNA, two guide RNAs (sgRNAs) and two ssDNAs (each with a LoxP site) flanked by homologies of 40-80 bases had efficiencies of 2-5% [Bishop et al. 2016; Miano et al. 2016] with only one earlier report of 16% [Yang et al. 2013] (Fig. 7b). Presumably, the low efficiency of this approach could be explained by generation of single LoxP insertions, double LoxP insertions in trans and deletions resulting from NHEJ besides the desired two LoxP sites in cis. Although sequential injections of one sgRNA and one ssODN at the one cell stage followed by the second sgRNA and second ssODN at the two cell stage of mouse embryos helped to improve cis-insertions of the two LoxP sites [Horii et al.et al. 2017], it is cumbersome and not amenable to high throughput. The efficiency was dramatically improved to 44% when a 514 base single ssDNA donor containing two LoxP sites flanked by homology arms was introduced together with two sgRNAs and Cas9 mRNA; this approach had 86% precision [Miura et al. 2015] (Fig. 7c). Similarly, a thorough examination of many loci and testing many variables showed that the efficiency of 9% for the expected insertion using two ssODNs was greatly increased to an efficiency of 48% when using a single longer ssODN [Lanza et al. 2017]. In this case, 20-30% of the mice had an off-target insertion, but it was not observed for the same animal to have both an on-target and off-target insertion.

Easi-CRISPR (efficient additions with ssDNA inserts-CRISPR)

Optimization of this approach in a method called Easi-CRISPR (efficient additions with ssDNA inserts-CRISPR) (Fig. 7c) has had excellent results for edits flanked by LoxP sites [Quadros et al. 2017; Miura et al. 2018]. Up to almost 2 kb inserts were achieved with ssODN donor DNA flanked by 50-100 bases of homology at each end when co-injected into mouse zygotes with ctRNP (a complex of pre-assembled crRNA + tracRNA + Cas9 ribonucleoprotein) [Aida et al. 2015]. Specifically, an efficiency of 40% of the desired insert using a single ssODN donor (862 bases flanked by 91 and 93 base homologies) and ctRNP was achieved, which was better than the efficiency of 13% for the same donor and sgRNP (complex of sgRNA with Cas9). The three-fold greater efficiency when using ctRNP than sgRNP has also been noted by others [Aida 2015]. Moreover, efficiencies can be improved by using sgRNP rather than sgRNA [Chen et al. 2016]. When using Easi-CRISPR to create insertions of ssODN with ctRNP at six different loci to insert 527-893 bases, the efficiencies averaged 43%, ranging from 8.5% to an amazing 100%. Using this ssODN and ctRNP Easi-CRISPR approach for additional experiments at six other loci with ssODNs had 779-1368 base insertion frequencies of 25-67% of which 44% had the correct sequence at both junctions. Another group had 35% efficiency when using Easi-CRISPR in mice and found that the average germline transmission was 52.8% and no off-target cleavage [Ma et al. 2017].

When comparing the dsDNA donor PITCh approach [Aida et al.et al. 2016] to ssODN Easi-CRISPR [Quadros et al. 2017] for DNA insertion at the same locus in the mouse genome, PITCh had an efficiency of 33% for the correct insertion compared to 100% efficiency for Easi-CRISPR. In this case, the PITCh experiment used Exo1 nuclease that is an enhancer of targeted insertion [Aida et al. 2016]. A drawback of ssODN is that lengths over 2 kb are not commercially available [Paix et al. 2016; Quadros et al. 2017]. However, the approach of Miura et al. [2015] to create ssDNA by reverse transcription of a 4-5 kb in vitro transcribed RNA template may be promising for introduction of longer DNA, though errors may occur when producing the longer ssDNA.

2H2OP (two-hit by sgRNA and two oligos with a targeting plasmid)

A variant of the ssODN protocol succeeded in generating transformed rats with inserted DNA even as large as 200 kb for a BAC insertion with 1/15 pups (6.7%) insertion efficiency and correct sequences at the junctions [Yoshimi et al. 2016]. This method is called “two-hit by sgRNA and two oligos with a targeting plasmid” (2H2OP) (Fig. 7d). In this case, the donor vector does not have homology ends. Instead, the donor plasmid DNA and two short ssODNs as bridging oligos that provide homology are co-injected with two sgRNAs. One of the sgRNAs directs Cas9 cleavage at the target site in the chromosome and the other sgRNA directs Cas9 cleavage in the donor plasmid. The two ssODNs span the chromosomal target and inserted donor junctions at each end to increase the ligation efficiency. A related method of 3H2OP (4.3% efficiency) uses three sgRNAs for cleavage at two regions of the chromosome to replace the intervening stretch of DNA with the donor DNA that was made linear by the third sgRNA-mediated cleavage.

Despite the relatively modest insertion efficiency, 2H20P surpassed the TAL-PITCh method where no inserts were found for the same target site and the same 4.8 kb insert in contrast to 3/17 pups (17.6%) with 2H2OP. Although long inserts can be obtained with 2H2OP, a disadvantage is the lack of precision and only one of the pups with the 4.8 kb insertion had correct sequences at the insertion junctions. Another disadvantage is that the vector is also inserted at the target site, though presumably this could be overcome by using two sgRNAs to direct cleavage of the desired insert of the plasmid, in addition to a sgRNA for Cas9 cleavage of the chromosomal target site.

Conclusions and Perspectives

DNA insertion by NHEJ or by homology

Great strides are being made to develop genome editing to precisely introduce DNA at a specific location. This is necessary for experiments in basic biology and also for gene therapy to cure genetic diseases [reviewed by Pan et al. 2013]. The classical approach of HDR had excellent precision but low efficiency and is not feasible in several systems.

Table 1 summarizes the recent advances utilizing NHEJ or homology, as described in this review. NHEJ has been used in two approaches to introduce relatively long DNA into targeted genomic sites. For example, genomic insertions of 15 kb in cultured cells and 6.5 kb for fly germline transformation were achieved using the ObLiGaRe method. Although the efficiencies were modest (2-27%), the percentage of precise insertion was very good (75-100%).

Table 1.

Summary of new methods for site-specific insertion of relatively long DNA

Method	organism	size of inserted DNA	efficiency	% of precise insertion
DNA insertion by NHEJ:
HITI	HEK293 cells	700 bp	few to 30-40%	75-90%
	cultured neurons		56% of transformed cells	~90%
	mouse brain (somatic insertion)		10.6%	75-90%
ObLiGaRe	HCT116 cells;	15 kb	2% of all cells	75-88%
	Sciara (fungus fly)	2.4 kb (somatic)	27%	87%
	Sciara (fungus fly)	6.5 kb (germline)	6.9%	100%
DNA insertion by homology:
PITCh (5-25 bases homology)
	zebrafish embryos		79%	77%
	HEK293T cells	1.4 kb	~75%	80% perfect 5’ junction; 50% perfect 3’ junction
	also silkworms (Bombyx) and frogs (Xenopus)
HMEJ (800 bases homology)
	HEK293 cells	700 bp	~20%	almost always perfect
	stem cells	0.7-6.1 kb	2-8%
	mouse brain	700 bp	10%
	mouse embryos	700 bp	23%	100%
35-100 base homology		<1 kb	31-75%	95-100%
	cultured human cells	<1 kb	~60%
	cultured human cells	514 bp	44%	86%
	C. elegans (worm)	<1 kb	33-75%
	mouse embryos	739 bp	31%	100%
	mouse embryos	<1 kb	48%	70-80%
Easi-CRISPR (DNA replacement; ssODN donor with 50-100 bases homology)
	mouse embryos	<1 kb	8.5-100% (average of 43%)
	mouse embryos	779-1368 bp	25-67%	44%
2H2OP or 3H2OP (DNA replacement by ssODN bridging oligos for 60-120 bases homology)
	rat embryos	~1 kb	18%	25%
	rat embryos	6.2 kb	17%	25%
	rat embryos	200 kb	13.3%	50%

(nb – multiple entries for same cell type or organism reflect results from different experiments or from different groups)

Alternative strategies have used homology of 5-120 bases (and 800 bases homology in HMEJ) flanking the sequence to be inserted. These alternative strategies utilize a variety of pathways (not simply HDR) and sometimes the mechanism is not yet known. Usually, the inserted DNAs were ~1 kb or less with efficiencies up to 79% and even 100% and usually very good precision (often 70-100%). Larger DNA of 6.2 kb and even 200 kb was inserted with the 2H2OP method with 13-17% efficiency and 25-50% precision.

As can be seen, the wide ranges of efficiency to introduce DNA into a genomic target site vary both with the targeted locus and with the biological system. Often it is a trade-off of precision vs. efficiency, where 100% precision may have lower efficiency. The efficiency and precision are generally higher for DNA insertion (using a single guide RNA) rather than for DNA replacement (using two guide RNAs).

Off-target insertions

Beyond efficiency, yet another important consideration is that site-specific insertions should not occur elsewhere in the genome at off-target sites [Zhang 2015]. This has been minimized by using ZFNs with point mutations at the ZFN interface so that only heterodimers and not homodimers are permissible, thus lowering off-target insertions [Miller et al. 2007; Szczepek et al. 2007; Söllü et al. 2010; Doyon et al. 2011; Ramalingam et al. 2011].

Off-target insertions have been problematic with CRISPR-Cas9, but can be somewhat reduced by judicious choice of a unique target sequence [Lin et al. 2016]. Various methods have been used to estimate the frequency of cleavage at off-target sites [Stella and Montoya 2016], including the Digenome-Seq [Kim et al. 2016] and GUIDE-Seq [Tsai et al. 2015; Tsai and Joung 2016]. Off-target insertion has been minimized for CRISPR approaches such as by using a truncated sgRNA [Fu et al. 2014] or using a pair of Cas9s modified to act as nickases with staggered cuts on opposite DNA strands [Mali et al. 2013; Ran et al. 2013]. Another method paired a mutant Cas9 without nuclease activity (dCas9) with a FokI nuclease domain, which requires two sgRNAs with defined spacing and orientation before it cuts [Guilinger et al. 2014; Tsai et al. 2014]. It has been reported that use of Cas9 protein rather than Cas9 mRNA also helps to reduce off-target insertions [Wang et al. 2015]. While these techniques have improved the situation, there were still some residual off-target cleavages. Recent developments using a high fidelity Cas9 with a few engineered point mutations have further reduced off-target integrations in CRISPR approaches to barely detectable levels [Slaymaker et al. 2016; Kleinstiver et al.et al. 2016]. Even more recently, it has been reported that off-target insertions can be abolished by inactivation of Pol Θ and c-NHEJ [Saito et al. 2017; Zelensky et al. 2017].

Overall, the advances summarized here for efficient site-specific insertion of large DNA with very good precision and minimal off-target insertions hold promise to achieve gene editing in whole organisms, which are integral to and will facilitate advances in modern biology. These advances in basic science will also have translational applications in medicine and agriculture.

Abbreviations

AAV

adeno-associated viral vector

a-NHEJ

alternative nonhomologous end-joining

Cas

CRISPR associated protein

c-NHEJ

canonical nonhomologous end-joining

CRISPR

clustered regularly interspaced short palindromic repeat

crRNA

CRISPR RNA (the guide RNA)

ctRNP

a complex of pre-assembled crRNA + tracRNA + Cas9 protein

dCas9

catalytically inactive Cas9

DSB

double-strand break

dsDNA

double-stranded DNA

Easi-CRISPR

efficient additions with ssDNA inserts-CRISPR

HITI

homology-independent targeted integration

HDR

homology-directed repair

HMEJ

homology-mediated end-joining

HR

homologous recombination

Indels

small insertions and deletions

MMEJ

microhomology-mediated end-joining

NHEJ

nonhomologous end-joining

ObLiGaRe

obligate ligation-gated recombination

PAM

protospacer-adjacent motif

PITCh

precise integration into target chromosome

SDSA

DNA synthesis-dependent strand annealing

sgRNA

single guide RNA containing a crRNA sequence and tracRNA (for CRISPR-Cas)

sgRNP

complex of pre-assembled guide RNA and Cas9 protein

SSA

single-strand annealing

ssDNA

single-stranded DNA

ssODN

single-strand oligodeoxynucleotide

TALEN

transcription activator-like effector nuclease

tracrRNA

transactivating CRISPR RNA (a structural RNA to recruit Cas9)

2H2OP

two-hit by sgRNA and two oligos with a targeting plasmid

3H2OP

three-hit by sgRNA and two oligos with a targeting plasmid

ZFN

zinc finger nuclease