In CRISPR/Cas systems, the binding of sgRNA to the PAM site is a critical step in gene editing (Figure 1A). An unsuccessful design of sgRNA will result in lower specificity and higher miss rate (Doench et al., 2016). To avoid this, sgRNA must be accurately designed using computational tools (Liu et al., 2020a), such as CRISPR-P 2.0 (Liu et al., 2017a), E-CRISP (Heigwer et al., 2014), and CasFinder (Abby et al., 2014). On the basis of rational design, further modification of sgRNA can improve the specificity of RNA-guided Cas9 by truncation or addition of nucleotides to the 5′ or 3′ end (Pattanayak et al., 2013; Lin et al., 2014b). The 5′ end-truncated sgRNAs (2-3 bp) considerably reduce off-target mutations, but with the same on-target mutation efficiency as the full-length sequence (Fu et al., 2014). By decreasing the binding affinity of the sgRNA, the binding stringency of Cas9 to the target sequence was increased, and the off-target effect was reduced (Fu et al., 2014). Since truncated sgRNAs can reduce the off-target effect of paired Cas9 nickases without compromising the efficiency of on-target genome editing, their combination results in a much greater target specificity (Guilinger et al., 2014). Contrarily, the 3′ end-truncated sgRNA or 5′ end-added sgRNA (-GG) can decrease the on-target activity. Meanwhile, if they consist of 16 nucleotides or fewer, truncated sgRNAs exhibit lower or undetectable activity compared to matched full-length sgRNAs. Thus, at least 17-nucleotide sgRNAs are required for the CRISPR/Cas9 system to be active during gene editing. Due to the disadvantages of the traditional enzymatic preparation of sgRNAs, such as complexity, time consumption, and safety concerns, the direct chemical synthesis of sgRNAs has been widely accepted, with high sgRNA stability and low off-target effect. Recently, a potential strategy has been reported to reduce off-target editing by DNA–RNA chimera (Yin et al., 2018a). Using the Cas9–sgRNA complex as a guide, the 5′- and 3′-DNA-replaced crRNA enables more efficient genome editing—for example, replacing the crRNA with 10 DNA nucleotides could provide the same level of off-target site indel formation as the truncated sgRNA. Additionally, the synthesis cost of DNA bases is much lower (10-fold cheaper) than that of native crRNA. In light of this, the DNA–RNA chimera could provide a novel approach to reduce the cost and off-target effect of CRISPR/Cas systems.

So far, several highly specific Cas9 variants have been identified, including eSpCas9 (Slaymaker et al., 2016), SpCas9-HF1 (Kleinstiver et al., 2016), HypaCas9 (Chen et al., 2017), xCas9 (Legut et al., 2020), Sniper-Cas9 (Lee et al., 2018), evo Cas9 (evolved Cas9) (Casini et al., 2018), HiFiCas9 (Vakulskas et al., 2018), and HeFSpCas9 (Kulcsar et al., 2017). The main mechanism and characteristics of each variant are comprehensively summarized in Table 1. Among these mechanisms, the most common cause of alteration in Cas9 function is amino acid substitution of the critical domain. Due to the minimum binding energy required to introduce DSBs into the genome, non-specific interactions between Cas9 and target DNA were reduced by decreasing the excess energy of wild-type SpCas9. As shown in Table 1, although they exhibit greater target specificity, each variant has its own limitations, such as low activity (Slaymaker et al., 2016), scope limitation (Kleinstiver et al., 2016), strict dependency on a PAM site (Legut et al., 2020), etc. Future studies should be conducted to increase the efficiency of genome editing using Cas9 variants. For a given target sequence, the optimal variant should be selected based on a comparison of activity, specificity, and PAM compatibility. By comparing 13 SpCas9 variants, the results demonstrated that the overall activity order of high-fidelity variants could be ranked as SpCas9 ≥ Sniper-Cas9 > eSpCas9 (1.1) > SpCas9-HF1 > HypaCas9 ≈ xCas9 > evoCas9, whereas their overall specificity could be ranked as evoCas9 > HypaCas9 ≥ SpCas9-HF1 ≈ eSpCas9 (1.1) > xCas9 > Sniper-Cas9 > SpCas9 (Kim et al., 2020b). Using established computational models, these results provide guidance for the selection of Cas9 variants and offer a more effective exploration of variants for future research (Figure 1B).

In general, the specificity and the activity of enzymes are often highly dependent on reaction conditions. RNP delivery produces at least twofold more colonies than plasmid transfection does (Kim et al., 2014). In the CRISPR/Cas9 system, RNP concentration plays a decisive role in both specificity and activity. After delivery to cells, RNPs almost immediately cleave chromosomal DNA and then degrade rapidly. With a high RNP concentration, the off-target effects of a CRISPR/Cas system may be amplified (Figure 1C). Meanwhile, a low RNP concentration leads to a reduction of on-target cleavage efficiency. Therefore, a suitable concentration of RNP is of paramount importance to minimize nonspecific cleavage (Figure 1E). This can be achieved by using either low concentrations of plasmids or different promoters. The former method directly reduces RNP transcription, while the latter alters the 5′-untranslated region of the target sequence, ultimately affecting translation efficiency (Hsu et al., 2013). Therefore, extensive measurements should be performed with consideration of both Cas9 activity and specificity. Compared with typical RNP concentrations, on-target activity will inevitably be inhibited to some extent. By modifying Cas9 and sgRNA instead (Figure 2A), the intrinsic specificity of Cas9 can be improved without sacrificing cleavage efficiency.

During gene editing, the efficacy, efficiency, and accuracy of CRISPR systems are often limited by temperature and time (Figure 2C). Studies have shown that a longer editing time of the Cas proteins in cells increases the off-target effects and negatively impacts outcomes (Ruan et al., 2017). Therefore, reducing the action time or overexpression of Cas proteins enhances the efficiency of gene editing. Early reports have utilized tissue culture-free systems (Manghwar et al., 2020), DNA-free systems (Kim et al., 2020a), and transient systems (Canto, 2016) to overcome these limitations. Now, however, anti-CRISPR proteins and nuclease-dead Cas proteins can be used to address these limitations. Temperature has been shown to affect Cas protein activity, but the findings are inconsistent. Hoyland-Kroghsbo et al. (2018) reported that a lower temperature is more effective than a higher one in P. aeruginosa PA14 due to the greater stability of the CRISPR/Cas complex. Additionally, low temperatures may enhance interference and adaptation by increasing the annealing efficiency of the crRNA to its target. Another report has shown that exposure to heat stress results in a greater amount of targeted mutations than with exposure to standard temperature (22°C) (LeBlanc et al., 2018). This is likely due to the fact that the activities of Cas9 and Cas12a at 37 and 34°C, respectively, are higher than at 28°C, and the expression level of sgRNA is raised at 39°C (Xiang et al., 2017). The mechanism of the effect of temperature on CRISPR/Cas systems is still unclear and should be further elucidated.

In the course of long-term antagonism between bacteria and phages, the phages have evolved Acr proteins to evade CRISPR/Cas-mediated immunity. Up to now, a total of 44 Acr proteins have been identified and named (Zhang et al., 2019a). Within the CRISPR/Cas system subtypes, these Acr proteins are present in types I, II, and V, but not in other subtypes—for instance, FnCas9 (Green and Hu, 2017) and C2c2 (Zhang et al., 2019a) Acr proteins were not found in the subtype II-B CRISPR/Cas9 system and the type VI CRISPR/Cas13a system, respectively. In view of this, the identification and characterization of a novel Acr is a key focus for future studies. As natural inhibitors, Acrs protect the host genome from destruction by inhibiting Cas nuclease activity. This inhibition can be achieved through the following three mechanisms: (1) inhibition of Cas9 binding to DNA (Malone et al., 2020), (2) interference of Cas9 binding to gRNA (Harrington et al., 2017), and (3) blocking the activity of Cas9 (Harrington et al., 2017). Currently, only AcrIIA2 and AcrIIA4 have been utilized as tools to reduce off-target effects. Through competition with the PAM site and/or other Cas9 sequences, these Acr proteins block the cleavage activity of Cas9, preventing the excessive and prolonged expression of CRISPR/Cas9 and thus decreasing off-target effects (Hoffmann et al., 2019; Liu et al., 2019). Acrs appear to be a new agent to improve the accuracy and safety of CRISPR-based therapies. Other special functions of Acr proteins also deserve attention—for example, AcrII-C3 can precisely regulate gene expression with dCas9-based tools, which is very helpful for the development of versatile genome engineering modulators (Liu et al., 2018c). The optogenetic-controlled AcrIIA4 enables light-mediated genome and epigenome editing. By inserting the AsLOV2 domain into the most C-terminal loop of AcrIIA4, the protein can switch the CRISPR-Cas9 activity according to light/dark conditions (Bubeck et al., 2018). The mechanisms of other Acrs, such as AcrIIA5-10 and AcrVA2-3, have yet to be described (Zhang et al., 2019a).

In theory, because of the competition between the two repair pathways, the efficiency of the HDR pathway can be boosted by inhibiting key factors of the NHEJ pathway. Among different inhibitors of the NHEJ pathway, SCR7 is a key factor that interferes with the affinity of DNA ligase IV to DSBs (Srivastava et al., 2012; Chu et al., 2015; Maruyama et al., 2015; Li et al., 2017; Shao et al., 2017; Hu et al., 2018). Maruyama et al. (2015) reported that using SCR7 increased the efficiency of HDR-mediated genome editing by up to 19-fold with the most significant enhancement effect, primarily due to co-injection of the CRISPR-Cas9 constructs with SCR7 into zygotes rather than other cells. The combination of SCR7 with other factors could significantly improve the efficiency of the HDR pathway by either downregulating KU expression (Chu et al., 2015), optimizing the donor template (Hu et al., 2018), or upregulating Rad52 expression and other small molecules (Li et al., 2017; Shao et al., 2017). Among these methods, the efficiency of the HDR pathway using Rad52 combined with SCR7 is the highest, reaching up to 40% (Shao et al., 2017). However, the effect of SCR7 in enhancing the HDR pathway remains controversial at present (Greco et al., 2016), with some reporting that embryonic stem cells tend to occur intrinsically HDR incident, suggesting that the effect of SCR7 is likely uncertain (Yang et al., 2020). Therefore, the effect of SCR7 in increasing HDR efficiency needs to be further explored. Additionally, the use of SCR7 should be extended to other fields as well instead of being limited to human- and mammalian-related diseases only, whether used alone or in combination with other medicines (Manjunath et al., 2021).

Aside from SCR7, other approaches also improved the efficiency of the HDR pathway—for instance, by downregulating KU protein expression via siRNA silencing, the frequency of the HDR pathway can be increased at least slightly (Li et al., 2018a). This method raised the affinity of downstream NHEJ enzymatic components by attaching to DNA termini upstream of the NHEJ pathway (Mateos-Gomez et al., 2017). The combination of multiple inhibitors can further increase the inhibitory effect on the NHEJ pathway—for example, M3814 combined with trichostatin A inhibits the NHEJ pathway 3-fold (Fu et al., 2021). NU7441 and KU-0060648, inhibitors of a key NHEJ pathway factor, DNA-PK, caused a 2-fold increase in HDR efficiency in HEK-293T cells (Robert et al., 2015). Due to the fact that the NHEJ pathway is crucial for the stabilization of the genome, excessive inhibition of the NHEJ pathway may eventually lead to the accumulation of unrepaired DSBs in cells, inducing cell death or embryonic lethality (Beumer et al., 2013). Therefore, it is worth noting that the safety of these inhibitors needs to be carefully evaluated in future work.

Alternatively, compared with inhibition of the NHEJ pathway, direct regulation of HDR-related factors can result in more precise editing and improve the efficiency of the HDR pathway. Several HDR-related factors have been well studied, including Rad51 (Ma et al., 2020), Rad52 (Shao et al., 2017), C-terminal-binding protein interacting protein (CtIP) (Charpentier et al., 2018), and RS-1 (Song et al., 2016). All of these factors enhance a link in the HDR pathway, thereby boosting repair efficiency. Among these, the overexpression of Rad proteins and the application of CtIP result in superior outcomes. In the HDR pathway, Rad51 proteins search for a DNA donor template to perform reconstitution through the formation of filaments on the DNA (Symington, 2014). As a back-up for Rad51, Rad52 is involved in the formation of Rad51 (Lok and Powell, 2012). When Rad52 fuses with any other factor or factors, HDR efficiency can be boosted at least 2- to 7-folds and sometimes much more (Paulsen et al., 2017; Shao et al., 2017; Tran et al., 2019)—for example, Rad52 fused with Cas9 yielded an approximately 3-fold increase in the efficiency of the HDR pathway, while Rad52 combined with SCR7 increased the HDR efficiency by about 40% (Shao et al., 2017). In the early stage of the HDR pathway, a key protein, CtIP, initiates the resection process and then creates 3′ single-stranded overhangs with exonuclease 1 and bloom syndrome protein complex (Symington, 2014). When combined with Cas9 or the MS2 system, CtIP can shift the ratio of the HDR/NHEJ pathway activities by a factor of 14.9 (Tran et al., 2019). A minimal N-terminal fragment of CtIP can also be used as an HDR enhancer, which is sufficient to stimulate the HDR pathway and improve repair efficiency by at least 2-fold (Charpentier et al., 2018). Other small molecules, such as L755507 and resveratrol, can also increase HDR efficiency 2- to 3-fold (Li et al., 2017). Almost all key factors of the HDR pathway were upregulated after treatment with the small molecules mentioned above, but a high concentration of resveratrol (more than 50 μM) resulted in severe cytotoxicity, significantly reducing cell viability and slightly upregulating the expression of the NHEJ factor. Therefore, the application of these factors needs to be further explored to improve the efficiency of the HDR pathway.

The selection of the NHEJ or HDR pathway plays a pivotal role in DNA repair. As an inhibitor of the HDR pathway, the tumor suppressor p53-binding protein 1 (53BP1) blocks DNA end resection and simultaneously inhibits BRCA1 recruitment to DSB sites (Panier and Boulton, 2014; Zimmermann and de Lange, 2014). By means of fusing, inhibiting, and binding 53BP1, HDR repair frequency can be increased from 20 to 86% (Paulsen et al., 2017; Canny et al., 2018; Jayavaradhan et al., 2019). Compared with a control, the correction frequency of the HDR pathway increased by nearly 20% when RAD52 was ectopically co-expressed with dominant-negative 53BP1 (dn53BP1) (Paulsen et al., 2017). Interestingly, dn53BP1 itself did not improve HDR efficiency unless combined with Rad52, suggesting that this fragment may not effectively promote the HDR pathway. To avoid the unwanted effects of global NHEJ inhibition, the fusion of DN1S and dn53BP1 significantly blocked NHEJ events locally while improving the correction frequency of HDR up to 86% (Jayavaradhan et al., 2019). This effect was likely due to the restrictively suppressive effect of dn53BP1 around the DSB site, which makes the CRISPR-Cas9-DN1S system a far more effective and stable approach in clinical treatments with high HDR frequency. Inhibition of 53BP1 is another indispensable strategy for regulating different repair pathways—for instance, utilization of 53BP1 inhibitor (i53) resulted in up to a 5.6-fold gene conversion and demonstrated an effective impact on the conversion mediated by single-stranded oligodeoxynucleotides (ssODN) compared to canonical modality double-stranded DNA (dsDNA) donors (Canny et al., 2018). However, the unknown toxicities or adverse incidents of i53 need to be carefully considered before its use. Moreover, the direct binding of related proteins to 53BP1 is a new target for enhancing the HDR pathway. A typical example is the TIRR protein, which acts similar to i53 and provides strong suppression by binding directly to 53BP1 (Anuchina et al., 2020). Since the function of TIRR is unclear, more studies should focus on its mechanism and the potential effects on the HDR pathway. TIRR may hold promise as a new target for enhancing the HDR pathway in genome editing.

Since HDR repair activity is restricted to the S and G2 phases of the cell cycle, synchronizing cells in these phases can significantly enhance the repair activity. In terms of a single aspect of cell cycle synchronization, chemicals that maintain the cell cycle in the S and G2 phases containing nocodazole (Yiangou et al., 2019), ABT-751 (Yang et al., 2016), and RO-3306 (Sunada et al., 2021) have been commonly used in CRISPR systems and increased the HDR efficiency by a factor of 1.38–6 (Lin et al., 2014a; Yang et al., 2016; Wienert et al., 2020). ABT-751 and nocodazole arrest the cells in the G2/M phase by inhibiting microtubule polymerization (Vasquez et al., 1997; Hande et al., 2006). Meanwhile, RO-3306 can transiently arrest cells at the transition from G2 to M phase by inhibiting the CDK1 function, enriching the number of cells in the S and G2 phases (Vassilev, 2006). Recently, Lomova et al. reported that the transient suppression of Cas9 activity and synchronization of the HDR-active cell cycle may have a prominent effect on the HDR pathway. RO-3306 combined with Cas9, which nuclease activity is reduced in the G1 phase, can improve the HDR/NHEJ ratio 20-fold, thereby limiting unnecessary NHEJ events (Lomova et al., 2019). In addition, the timed delivery of pre-assembled Cas9 RNP and chemical synchronization agents can also enhance the HDR rates by up to 38% (Lin et al., 2014a). Thus, synchronizing the cell cycle paired with controlled timing of Cas9 activity might be more practical and safer than synchronizing the cell cycle alone. In conclusion, more efficient synchronization agents for in vitro application need to be further studied. Agents with lower toxicity should also be explored for in vivo application in subsequent research. More importantly, there is no doubt that the combination of multiple technologies, such as controlled timing of Cas9 activity and cell cycle synchronization, will result in better outcomes.

At the time of DNA cleavage, if the donor DNA template is in closer proximity to the CRISPR components or has a higher concentration in the nucleus, the efficiency of the HDR pathway can be significantly increased (Devkota, 2018). Based on this hypothesis, many studies have demonstrated its feasibility and potential value in clinical applications (Liang et al., 2017). By attaching the donor DNA template to modified sgRNA, a S1mplex strategy improves the enhancement of the HDR pathway. In this system, the modified S1m-sgRNA adds an aptamer, which binds the streptavidin protein. Biotinylated ssODN linked to the streptavidin then increases proximity. Through this powerful S1mplex strategy, the ratio of HDR increased 18-fold compared with the unlinked components (Carlson-Stevermer et al., 2017). By virtue of the affinity between avidin and biotin, Ma et al. devised a Cas9 variant that was fused to avidin via a flexible linker and bridged with biotin-modified ssDNA to increase the proximity. This system achieved ∼20% HDR frequency in mouse embryos (Ma et al., 2017a). HUH endonuclease is a bridge that is also capable of forming robust covalent bindings with unmodified donor DNA templates (Lovendahl et al., 2017). Utilizing this convenient technique could create a stable Cas9 RNP–ssODN complex (Aird et al., 2018). Additionally, Natasa et al. linked ssODN to Cas9 through SNAP-tag technology, allowing O6-benzylguanine-labeled ssODN to covalently bind to SNAP-tag fusion proteins (Savic et al., 2018). Both SNAP-tag and HUH-tag enable the spatio-temporal co-localization of the donor templates at DSBs, thus achieving 24- and 30-fold enhancement of HDR efficiency, respectively. In addition to ssODN attachment to sgRNA in the S1mplex system, other experiments use a variety of ways to attach ssODN to Cas9 protein. Among them, the HUH-tag strategy displayed a more promising prospect due to the superior ease of use and lower cost associated with modifying donor DNA. As noted above, Cas9 RNP complexes can connect with donor DNA templates through a variety of chemical modifications, all of which significantly enhance the transient expression of the HDR pathway. Furthermore, Cas9 and sgRNA delivered in the RNP format also exhibited a lower frequency of unwanted mutations and off-target effects (Svitashev et al., 2016), suggesting that the RNP format may be a promising approach in the broad field of gene editing.

To a considerable extent, the repair efficiency of the HDR pathway depends on the selection of donor DNA, including DNA modality, length, and flanking homologous sequences (Song and Stieger, 2017; Renaud et al., 2016). Generally, the modality of a DNA template can be divided into three forms: plasmid, ssODN, and linear dsDNA (Figure 1D). As the most common form of genetic material, circular plasmids are widely used in gene editing but will not be discussed in this review (Bosch et al., 2020; Sondergaard et al., 2020; Yoshimi et al., 2021). Compared with other donor DNA templates, ssODNs have the advantages of ease of design, lower time investment, less illegitimate random integration (introducing less than 200 nucleotides), and site-specific modification for precise editing (Yang et al., 2013; Miura et al., 2018). By comparing the modification efficiency of CRISPR/mRNA and CRISPR/nuclease for a target gene, the results have shown that the 36-nt length of ssODN with the CRISPR/nuclease form achieved the optimal condition for modification of the target gene, with a slight advantage over the CRISPR/mRNA approach (Kumita et al., 2019). Currently, ssODNs have become a routine editing tool both in vitro and in vivo, especially in multigene knock-in experiments (Yoshimi et al., 2016; Miura et al., 2018; Lim et al., 2020). For large sequence DNA modifications, linear dsDNA templates (up to 11 kb) were commonly used for CRISPR systems with homology arms of 500–800 bp (Yang et al., 2013). The targeted integration of linearized dsDNA–CRISPR can increase the knock-in efficiency 12-fold by injecting PCR-amplified donor DNA, Cas9 mRNA, and sgRNA (Yao et al., 2018b). Therefore, optimal editing outcomes can be obtained by selecting the suitable donor DNA modality according to experimental requirements.

As an alternative, abiotic vectors may offer tantalizing possibilities for CRISPR/Cas9 delivery systems due to their low immunogenicity, larger delivery gene payload, ease of large-scale production (Li et al., 2015), and absence of endogenous virus recombination. Many excellent delivery systems with new properties have been established in various fields, such as gold nanomaterials (Wang et al., 2018), polymers (Lv et al., 2018), and other systems. The characteristics of each material are described in detail in the following sections.