Targeted gene editing activity and precision of CRISPR/Cas9 and Cas12a in maize

We chose maize glossy2 (gl2) gene, a gene involved in epicuticular wax formation in juvenile leaves, as the target gene. As illustrated in Figure 1a, we targeted two sites in exon 2 of the gl2, where both ‘NGG’ PAM for Cas9 and ‘TTTV’ PAM for Cas12a were present on the two ends of the target site. Thus, at each of these two loci, the target sites of Cas9 and Cas12a are substantially overlapping (Figure 1a), thereby permitting a more direct comparison of Cas9 and Cas12a targeting efficiencies.

We then compared on‐ and off‐target potentials of our gRNAs and crRNAs using available web tools. CRISPR Efficiency Predictor programme (Housden et al., 2015) predicted that the two Cas9‐gRNAs have 6.5 and 10.3 efficiency scores respectively. gRNAs with scores greater than 7.5 are predicted to have high efficiency in target sequence cleavage. For Cas12a, the two crRNAs were identified and designed manually to overlap the sequences of two gRNAs. Because public bioinformatics tools for Cas12a were not available at the time of design, we were not able to predict their targeting efficacies. However, we retrospectively examined efficiencies of the crRNAs using two recently published web tools, CRISPR AsCpf1 INDEL (CINDEL) score (Kim et al., 2017b) and CRISPR DNA Targeting (CRISPR‐DT; Zhu and Liang, 2018). CINDEL score and CRISPR‐DT predicted that crRNA1 has moderate‐high scores, 0.53–0.87 indel frequency, whereas crRNA2 has low scores, 0.05–0.17 indel frequency (Figure 2a). Thus, Cas9 could have a higher efficiency on the second target (6.5 vs. 10.3), whereas Cas12a might have a higher efficiency on the first target (0.53–0.87 vs. 0.05–0.17). Because these tools use different scoring algorithms and the predicted efficiency scores for Cas9 and Cas12a are not directly comparable, we could not predict which Cas system would work better in maize genome editing.

To evaluate off‐target activities of Cas9 and Cas12a in maize, we first examined how specific our intended on‐target sites were relative to the maize genome. To do this, we checked for on‐ and off‐target sites for B73 and B104 using the CRISPR Genome Analysis Tool (CGAT; Brazelton et al., 2015), and confirmed that our gRNAs and crRNAs have unique targeting sequence within the maize genome. Next we performed a more in‐depth in silico identification of closely related sites (off‐targets) using Cas‐OFFinder (Bae et al., 2014). Cas‐OFFinder predicted that Cas12a‐crRNA1 has far fewer sites with six or fewer mismatches in the maize genome than Cas9‐gRNA1: 1193 vs. 19029 (Table S1). Cas12a‐crRNA1 also has only ten sites with four or fewer mismatches, including a gap as a mismatch (Table S1), whereas Cas9‐gRNA1 has 93 such sites. Likewise, Cas12a‐crRNA2 has far fewer off‐target sites compared to Cas9‐gRNA2: 173 versus 71253 (Table S1).

To generate an efficient CRISPR/Cas9 system in maize, we expanded our previous CRISPR/Cas9 toolbox (Lowder et al., 2015) with two Gateway compatible vectors: pYPQ166 is an attL1‐attL5 entry vector carrying a codon‐optimized Cas9 for maize expression (Xing et al., 2014); and pYPQ210 is an attR1–attR2 destination vector that contains the maize ubiquitin promoter (PZmUbi) for Cas9 expression and bialaphos‐resistance gene (bar) for selection of transgenic plants. For CRISPR/Cas12a, we adopted the dual‐Pol II promoter and dual‐ribozyme LbCas12a system that was highly efficient in rice (Tang et al., 2017).

As can be seen in Figure 1b, the Cas12a‐crRNA constructs, A842B and A843B, used the maize ubiquitin promoter (PZmUbi) for the expression of both Cas12a and crRNAs, whereas Cas9‐gRNA constructs, A844B and A845B, used PZmUbi for the expression of Cas9 and rice U3 promoter (POsU3) for the expression of gRNAs (Figure 1b). We chose to use PZmUbi for both Cas12a and crRNAs because this combination was proven highly efficient in rice genome editing resulting in 100% targeting efficiency (Tang et al., 2017), whereas experiments using POsU3 (Xu et al., 2017) or POsU6 (Endo et al., 2016) for crRNA expression showed relatively poor targeting efficiency (Tang et al., 2017).

The resulting four T‐DNA vectors from the Gateway reactions were delivered into maize immature embryo cells via Agrobacterium‐mediated transformation (Frame et al., 2011). Four to nine independent bialaphos‐resistant callus events were obtained and 20–31 T0 plantlets were analysed per construct for on‐target mutagenesis (Table 1). Overall, T0 plants did not exhibit any noticeable morphological difference when compared to transgenic plants from other non‐CRISPR projects. They were acclimated in a growth chamber at 26/22 °C (day/night) for about a week before transferred to the greenhouse which has a 16 : 8 photoperiod with day/night temperature of 28/21 °C (Frame et al., 2011).

We surveyed the targeting efficiency of Cas9 and Cas12a nucleases in the transgenic maize events. Genomic DNA was isolated from leaf tissues of each T0 plant and a 379 bp region of gl2 was amplified by PCR using oligonucleotides Zm‐gl2‐exon2‐F1 and R1 (Table S2). Single‐band PCR products were either directly subjected to Sanger sequencing or cloned into pJET1.2 (cloning; ThermoFisher Scientific) first and then performing colony sequencing. Sequencing trace files were analysed by Tracking of Indels by Decomposition (TIDE; Brinkman et al., 2014) and DSDecode (Liu et al., 2015). Eight individual clones were sequenced for most of the T0 lines (Table S3). T0 and T1 lines were classified into five groups: biallelic, homozygous, heterozygous, mosaic and wild type (Table 1).

On‐target mutagenesis analysis revealed that the Cas9 system tested in this study has a higher efficiency than our Cas12a system for maize genome editing at these two sites (Table 1). Specifically, indel mutation frequency at target 1 was 100% for Cas9 (27/27 T0 plants from seven events) and 60% for Cas12a (12/20 T0 plants from four events; P =0.0003, two‐tailed z‐test). More dramatic contrast was observed on target 2, where Cas9 had a 90% indel mutation frequency (28/31 T0 plants from nine events) and Cas12a showed no indel mutations (0%) in 23 T0 plants from six events (Table 1; P =0.0001, two‐tailed z‐test). These results were consistent with in silico predictions which suggested that crRNA2 has a low targeting efficiency (Figure 2a). Importantly, the frequency of biallelic or homozygous T0 mutants was much higher in Cas9‐edited plants, 63%–77.4%, than that in Cas12a‐edited plants, 0%–5% (Figure 2b), indicating better performance of our Cas9 system in maize genome editing at these two sites. Loss‐of‐function mutant phenotypes were observed in T1 plants (Figure 2c), indicating that biallelic or homozygous mutant alleles were either passed down to the next generation or generated by continuing activities of Cas9 and Cas12a in the next generation.

Short indels were the most commonly observed mutations of the target genes in the Cas9‐ and Cas12a‐mutated T0 plants (Figure 2d‐f). Ten out of twelve Cas12a‐crRNA1 plants had 6–73 bp deletions and two plants had 1 bp insertions (Figure 2d). By contrast, 1 bp insertions were the most commonly observed mutations in the Cas9‐gRNA1 plants (20/27; Figure 2e), and all Cas9‐gRNA2 plants (31/31) showed 1–6 bp deletions (Figure 2f). These observations are in line with previously reported mutations found in Cas9‐ and Cas12a‐mutated plants (Kim et al., 2017a; Li et al., 2018a; Tang et al., 2017; Xu et al., 2017). For Cas12a‐crRNA1 T0 plants, two of them carried mosaic mutations (Figure 2d and Table 1), which is consistent with its relatively low activity. No mosaic mutation was found in all Cas9 T0 lines (Figure 2e and f; Table 1). Notably, most biallelic or homozygous T0 lines originating from the same immature embryo event have identical mutations (Table 2 and Table S3), suggesting early on‐set of mutagenesis during the transformation process, likely at a single‐cell stage before callus induction.

Genotyping of T1 progenies produced by crossing T0 lines with wild type B104 pollen donors showed that mutations observed in the maternal parent lines were readily inherited in the next generation (Table 2 and Table S3). However, new mutations were frequently discovered in Cas12a or Cas9 T1 lines (Table 2 and Table S3), consistent with the idea of desired‐target mutator (DTM) by carrying the CRISPR reagents into the next generation (Li et al., 2017).

The relatively low targeting efficiency of the Cas12a compared to Cas9 used in this study, especially the target 2 construct, prompted us to examine the Cas12a‐crRNA sequences. Cas12a has been successfully applied to plant genome editing, but mutagenesis efficiencies varied greatly depending on crRNA sequences (Kim et al., 2017a; Li et al., 2018a; Tang et al., 2017; Xu et al., 2017). As mentioned earlier, both CINDEL score and CRISPR‐DT predicted that crRNA1 has moderate‐high scores, 0.53–0.87 indel frequency, whereas crRNA2 has low scores, 0.05–0.17 indel frequency (Figure 2a). This prediction seems to be consistent with our results.

It has been suggested that crRNAs with a stable secondary structure within the crRNA targeting region are less efficient for gene editing (Kim et al., 2017b; Zhu and Liang, 2018). We predicted secondary structures of the two crRNAs using the RNAstructure program (Bellaousov et al., 2013) and crRNA2 indeed has a stable stem–loop structure and a lower minimum free energy (MFE) compared to crRNA1, −2.9 versus 2.0 kcal/mol (Figure 2a). Together, these information seems to suggest the possibility that Cas12a crRNAs might be more sensitive than Cas9 gRNAs to structural constraints, such as secondary structures and low MFE, within the guide sequence (Wong et al., 2015; Zhu and Liang, 2018). Therefore, one needs to consider not only the sequence specificity of the guide sequence but also possible intramolecular structure when attempting to design highly efficient crRNAs as has been done previously for SpCas9 (Thyme et al., 2016).

The poor targeting efficiency of Cas12a in maize could also be attributable to low temperatures (20–28 °C) used for Agrobacterium‐mediated maize transformation (Frame et al., 2011). Moreno‐Mateos et al. (2017) showed that Cas12a activity can be substantially affected by temperature and has a higher activity at 34 °C than at 28 °C. Likewise, Li et al. (2018b) also demonstrated temperature‐dependent performance of AsCas12a‐mediated editing in vertebrates. Consistent with these results, we achieved high editing efficiencies with LbCas12a and FnCas12a in rice when higher temperatures (28–30 °C) were applied (Tang et al., 2017; Zhong et al., 2018). Thus, it may be worth developing a sophisticated high temperature treatment regime for maize immature embryos during transformation to attempt to optimize Cas12a‐mediated genome editing as claimed by Cigan et al. (2016) in their patent application.

Another factor is that the Cas9 used in this study was codon‐optimized for maize (Xing et al., 2014), while Cas12a was rice codon‐optimized (Tang et al., 2017). While both crops are monocotyledonous plants and many promoters/genes have been working effectively in both systems, the codon of Cas nucleases may need to be optimized for specific plant species to maximize its activity.

One limitation in this maize study is its small sample size. Many recent large scale CRISPR system studies were carried out in readily transformable plant species, such as Arabidopsis, rice or tomato (Jacobs et al., 2017; Lu et al., 2017; Ma et al., 2015; Meng et al., 2017). Large‐scale analysis comparing efficacies of CRISPR guides and nucleases in recalcitrant crop species such as maize require substantial resources that are not affordable for most academic laboratories. In this study, we used two target sites with overlapping sequences to test Cas9 and Cas12a systems in maize plants. Although the data is limited, the overlapping sequences did allow us to eliminate potential ‘position effect’ of the targets in this comparison. Our Cas9 system appears to be more efficient than our Cas12a system for mutagenizing these two sites, which led to gl2 mutant phenotype in many T1 individuals.

We used two approaches to analyse off‐target effects of Cas9 in the transgenic maize lines. We first identified potential off‐target sites using Cas‐OFFinder (Bae et al., 2014). This bioinformatics approach allows to identify closely matched off‐sites using simple sequence homology alignment. Our second approach was to employ a recently described Circularization for In vitro Reporting of CLeavage Effects by sequencing (CIRCLE‐seq) assay (Tsai et al., 2017). This is an in vitro biochemical method that can identify a list of closely related sites in the maize genome with Cas9 cleavage propensities. The in vitro off‐target list generated by CIRCLE‐seq was then used as a road map to assess the in planta off‐sites in the transgenic mutant maize lines.

One important challenge for maize off‐site analysis is that about 85% of the maize genome consists of highly repetitive sequences (Schnable et al., 2009). The vast majority of the closely matched sites found by Cas‐OFFinder were within highly repetitive regions. This feature presents challenges when designing primers specific for sequencing the off‐site regions. Specific primers could be designed for only one, two and five off‐targets with four or fewer mismatches for Cas12a‐crRNA1, Cas9‐gRNA1 and gRNA2 respectively (Table S4).

We chose to focus on Cas9 and not Cas12a because of the smaller number of closely related sites in the maize genome for Cas12a nucleases as well as their higher genome‐wide specificities reported in human cells (Gao et al., 2017; Kim et al., 2016; Kleinstiver et al., 2016). CIRCLE‐seq analysis identified 18 and 67 off‐target sites cleaved in vitro by Cas9‐gRNA1 and Cas9‐gRNA2 respectively (Figure 3a and b; Table S4).

We used TIDE (Brinkman et al., 2014) to examine whether the off‐target sites identified in vitro by CIRCLE‐seq would show mutations in planta in T1 maize plants originated from T0 lines bearing on‐target mutations (Figure 4; Table S3). We were unable to assess all of these CIRCLE‐seq off‐targets in planta because some of them are present within repetitive regions of the genome. Nevertheless, we were able to design PCR primers to amplify 15 of 18 gRNA1 off‐targets and 18 of 67 gRNA2 off‐targets (Table S4). In addition, we also examined two closely matched sites identified by Cas‐OFFinder for each gRNA (Table S4), thus a total of 17 and 20 off‐targets were examined for gRNA1 and gRNA2 respectively. We screened 32 A844B T1 lines originated from seven T0 lines and 24 A845B T1 lines originated from ten T0 lines (Figure 4; Table S2). Eight A844B and seven A845B T1 lines were T‐DNA negative segregants (Figure 4; Data S1). Most T‐DNA positive T1 lines (21 of 24 A844B and 12 of 17 A845B T1 lines) exhibited gl2 null mutant phenotypes suggesting inherited Cas9 activities, whereas most T‐DNA negative lines (5 of 8 A844B and 5 of 7 A845B T1 lines) had wild type‐like phenotype for the gl2 gene (Data S1). TIDE analysis revealed that none of the tested off‐target sites carried a detectable level of indel mutations (Figure 4; Data S1), suggesting no germ‐line–transmittable mutations at these potential off‐target sites.

The lack of detectable level of off‐target mutations in our study suggests a high specificity of our Cas9‐gRNAs in maize genome editing at the tested two sites. However, it might also be attributed to other factors such as sensitivity of the detection method, discrepancy between the in vitro and in vivo cleavage by Cas9 endonuclease, and incomplete coverage of potential off‐sites due to inability of assessing sequences in the highly repetitive regions. First, TIDE can detect indels with a frequency down to approximately 2% (Brinkman et al., 2014), which is sufficient to detect off‐target mutations inherited from T0 lines or new mutations that had occurred at an early stage during embryogenesis in T1 lines. More sensitive mutation detection methods, such as targeted deep sequencing (Tsai et al., 2017) using the next generation sequencing technology, would be able to detect rare off‐target mutations. Second, unlike the in vitro cleavage reactions which involve naked genomic DNA and a high concentration of Cas9‐gRNA complexes, in vivo cleavages are likely affected by numerous cellular factors. In addition, NHEJ repair pathway is relatively accurate (Bétermier et al., 2014), thus only a small fraction of DSBs result in indel mutations, further reducing the frequency of off‐target mutations in vivo. Lastly, we were unable to test 3 of 18 and 49 of 67 off‐targets identified by CIRCLE‐seq for the two gRNAs, respectively, due to their presence within the repetitive sequences. It is possible that some of the off‐targets within the repetitive sequences might carry mutations, especially for the gRNA2. Nonetheless, recent Cas9‐mediated plant genome editing studies suggest high specificity of Cas9 endonuclease (Feng et al., 2018; Tang et al., 2018). In general, low frequency somatic null mutations would not be as problematic in plants as they can be in animals, because the chance of passing such somatic mutations into the next generation is low and undesirable mutations in plants can be segregated through breeding in the following generations.

Total genomic DNA was isolated from leaf tissues of T0 and T1 maize plants as previously described (Allen et al., 2006). Oligonucleotides for PCR were designed using Primer3 (Untergasser et al., 2012) and primer sequence specificity was checked in MaizeGDB. Primer sequences used for PCR are listed in Table S1. PCR was carried for each locus including a target or off‐target with 50–100 ng of gDNA as template using Phusion high‐fidelity DNA polymerase (ThermoFisher Scientific). A typical PCR reaction contained 1x Phusion high‐fidelity buffer, 125 μm dNTPs, 0.5 μm primers and 0.4 U of Phusion DNA polymerase. PCR reactions were carried out in a thermocycler with the following cycling conditions: 30 s at 98 °C for initial denaturation followed by 35 cycles of [10 s at 98 °C, 20 s at 63 °C, 20 s at 72 °C], and 5 min at 72 °C for final extension. Single‐band amplification was verified by agarose gel electrophoresis. PCR product was cloned into a cloning vector, pJET1.2 (ThermoFisher Scientific), according to the manufacture's instruction or directly subjected to Sanger Sequencing after clean up by ExoSAP‐IT (ThermoFisher Scientific) treatment. The sequence trace files were analysed using the Tracking of Indels by Decomposition (TIDE; Brinkman et al., 2014) and DSDecode (Liu et al., 2015) with the default parameters.