Generation of inheritable and “transgene clean” targeted genome-modified rice in later generations using the CRISPR/Cas9 system

The CRISPR/Cas9 system is becoming an important genome editing tool for crop breeding. Although it has been demonstrated that target mutations can be transmitted to the next generation, their inheritance pattern has not yet been fully elucidated. Here, we describe the CRISPR/Cas9-mediated genome editing of four different rice genes with the help of online target-design tools. High-frequency mutagenesis and a large percentage of putative biallelic mutations were observed in T0 generations. Nonetheless, our results also indicate that the progeny genotypes of biallelic T0 lines are frequently difficult to predict and that the transmission of mutations largely does not conform to classical genetic laws, which suggests that the mutations in T0 transgenic rice are mainly somatic mutations. Next, we followed the inheritance pattern of T1 plants. Regardless of the presence of the CRISPR/Cas9 transgene, the mutations in T1 lines were stably transmitted to later generations, indicating a standard germline transmission pattern. Off-target effects were also evaluated, and our results indicate that with careful target selection, off-target mutations are rare in CRISPR/Cas9-mediated rice gene editing. Taken together, our results indicate the promising production of inheritable and “transgene clean” targeted genome-modified rice in the T1 generation using the CRISPR/Cas9 system.


Results
CRISPR/Cas9-mediated targeted mutagenesis in T 0 transgenic rice. The rice AOX1 family is composed of three members (OsAOX1a, OsAOX1b and OsAOX1c) with high sequence similarity. To introduce individual mutations, we designed specific 20-bp gRNAs with at least a two-base mismatch at potential off-target sites using bioinformatic tools 34,35 . These gRNAs were inserted into a GATEWAY-based vector system using a rice-codon-optimized Cas9 gene and an OsU3 promoter, as previously reported 7 .
To evaluate the off-target effects of the system, a 20-bp region of a P450 gene, OsBEL, was selected and constructed. This region has similarity to a sequence located ~7 kb upstream of OsBEL, with a one base mismatch 13 bp upstream of the PAM.
Through Agrobacterium-mediated stable transformation in Nipponbare, 8, 7, 12 and 14 independent T 0 transgenic events were obtained, carrying constructs targeting OsAOX1a, OsAOX1b, OsAOX1c and OsBEL, respectively. To detect mutations, genomic DNA was isolated from the third leaf from the top of 10-week-old plants. The target regions were analyzed by sequencing the products of the corresponding site-specific genomic PCR and/or further confirmed by sequencing clones of the PCR amplicons. High mutation rates were induced in all tested targets, and more than half of the lines of each transgene carried mutations (Table 1, Supplemental Fig. S1-S4). The highest mutagenesis efficiency was observed at the OsBEL site, in which the target region was modified in 12 lines out of a total of 14 lines (Table 1, Supplemental Fig. S4). To investigate the possible reason for unsuccessful target mutagenesis, the presence of the transgene was determined by amplifying sgRNA, Cas9, and hygromycin phosphotransferase (HPT) in non-mutated lines (designated as WT). Among a total of 12 WT lines, 6  Cas9 and/or sgRNA transgene fragments (Supplemental Table S1), implying that a deficiency in the integrity of the sgRNA/Cas9 expression cassette might be a major reason for the failure of targeted mutagenesis. We also tested the numbers of T-DNA insertions in all transgenic lines by determining the copy number of HPT via real-time PCR analysis. Most lines carrying target mutations contained 1-2 copies ( Table 1), suggesting that it may be possible to segregate out the transgene. Some reports have indicated that 1-bp changes (deletions or insertions) are the main type of mutation induced by stably transformed CRISPR/Cas9 in both Arabidopsis and rice 23,26 . However, a remarkable abundance of deletions of long fragments (≥ 3 bp) were observed in a study of high-efficiency gene editing targeting rice SWEET13 27 . We found that the types of mutations varied among target regions. For OsAOX1c and OsBEL, short changes (≤ 3 bp) were the major type of mutation; conversely, all mutations in OsAOX1a were relatively long deletions. Because short deletions are often associated with classical NHEJ (cNHEJ) and longer deletions may represent the results of microhomology-mediated end-joining (or alternative NHEJ, aNHEJ) 37,38 , the CRISPR/Cas9-induced mutation patterns might be different for specific DSB sites through distinguished NHEJ repair pathway.
The genotypes of the mutants were also analyzed. Consistent with previous reports 26,27 , putative biallelic i.e., homozygous or compound heterozygous mutations mostly occurred in genotypes in the T 0 generation, accounting for 41.4% (12/29) and 31.0% (9/29) of all mutant plants, respectively.
Inheritance and stability of targeted mutagenesis in the T 1 generation. To investigate the pattern of transmission of CRISPR/Cas9-mediated targeted gene modification, several T 1 progeny were obtained by strict self-pollination and used for testing of targeted mutations. For each T 0 line, 8-24 progeny were randomly selected and examined. As shown in Table 2, all of the mutated T 0 lines produced mutated T 1 progeny, whereas targeted sequence changes still could not be detected in the progeny of WT T 0 plants.
Genotypes are thought to be easily predicted in the progeny of biallelic T 0 lines 26 . As expected, all 12 and 16 T 1 progeny of homozygous OsBEL #1 and #3, respectively, exhibited consistent homozygous 12-bp or 1-bp deletion genotypes (Table 2). However, we observed that the transmission of targeted mutations was relatively ruleless in large subsets of putative biallelic T 0 lines. Three unexpected patterns can be summarized as follows. (1) The mutations occurring in T 0 were lost in the T 1 generation. OsAOX1a line #1 was determined to be a putative biallelic mutation with 7-bp and 11-bp deletions, whereas only the 11-bp deletion could be detected in T 1 plants as a homozygous genotype ( Table 2). In OsAOX1a #2 and OsAOX1b #2 and #3, the selective transmission of a single parental allele was also found in the progeny.
(2) New mutations were created in the T 1 generation ( Table 2, Supplemental Fig. S5). For example, the sequencing results indicated a putative homozygous 1-bp insertion genotype in an OsAOX1c #4 T 0 plant, whereas 2 additional different 1-bp insertions were found in the T 1 population (Table 2). Similarly, a number of additional mutations were detected in the T 1 generations of OsAOX1c #9 and OsBEL #2, even though the majority of progeny mutations were already observed in the parental genome. In addition, several putative biallelic mutated T 0 lines, e.g., OsAOX1c #3, #9 and OsBEL #2, could generate progeny carrying the WT allele. This result suggested that some cells of the T 0 plants might not be target modified in these lines. Meanwhile, these cells were also not detected in the genotyping of T 0 plants. (3) The segregation ratio of target mutations in the T 1 generation was distorted. A 1:1 ratio of the parental mutated alleles was anticipated in the progeny of biallelic plants, based on regular segregation laws. In the T 1 plants of the biallelic mutated OsAOX1b #1 lines, although the mutation types did not increase or decrease, the ratio between the two alleles in the T 1 plants did not conform to 1:1, indicating that they were not inherited with equal frequencies.
Because the sgRNA-Cas9 complex has been shown to be active in heterozygous and chimeric plants 5,24,26,27 , WT alleles are likely to be modified continuously. As expected, a number of new mutations were found in the corresponding T 1 lines (e.g., OsAOX1c #6 and #7; OsBEL #7 and #8), whereas most of the mutations detected in the T 0 heterozygotes and chimeras were passed on to the next generation ( Table 2). Whether or not there was additional mutation of T 0 heterozygotes in the T 1 generation, the ratio between the parental mutated allele and other alleles should be expected to be 1:1 in the T 1 population. However, the frequency of the parental mutation in some T 1 generated from T 0 heterozygotes, e.g., OsAOX1c #6 and OsBEL #6, was significantly lower than 50% by the chi-square test, suggesting that additional mutations likely occurred in other undetermined parts of the T 0 plants.
The presence of the transgene (T-DNA) region was also examined in T 1 populations. The absence of the transgene was determined to be concurrent in PCRs negative for Cas9, sgRNA and HPT genes, and the results indicated that the T-DNA region could be segregated out in most lines. Transgene-negative plants were observed in nearly all of the low-copy T 0 progeny ( Table 2).
Segregation of targeted mutagenesis in T 2 generations. As described above, intricate segregation patterns were detected in the T 1 generation. To further investigate the inheritance of targeted mutations in later generations, the genotypes of several T 2 plants were analyzed in detail. Because the sgRNA/ Cas9 complex may still be active in progeny and thus disturb genotype transmission, the segregation of mutations in T-DNA-lacking T 1 progenitors was examined first. A total of 12 T 1 lines carrying 3 genotypes (8 homozygous, 3 compound heterozygous and 1 heterozygous) and lacking the transgene were selected and analyzed. By sequencing targeted genomic regions of an extensive T 2 population derived Scientific RepoRts | 5:11491 | DOi: 10.1038/srep11491 from T 1 homozygotes, all of the descendants were found to exhibit the same homozygous mutations, without exception (Table 3). Similarly, the ratio between the two alleles of the biallelic and heterozygous T 1 plants conformed to the expected 1:1 ratio of classical Mendelian segregation by the chi-square test ( Table 3). All of these results indicate that, in the absence of the transgene, the inheritance of targeted mutations is stable and regular in later generations. Furthermore, the patterns of transmission from T 1 to T 2 were examined in the presence of the transgene. For this assay, 4 T-DNA-positive T 1 homozygous and 2 compound heterozygous T 1 lines were selected, and the genotypes were examined in their progeny. As shown in Table 4, the parental mutations were not modified or revised in the T 2 generation, possibly due to the absence of editable targets of CRISPR/Cas9. We also followed 4 T 1 heterozygotes to the T 2 generation and found that most of the genotypes were inherited normally, with only one additional mutation detected in a single T 2 plant (Table 4).
Off-target analysis. Based on the predictions of the CRISPR-P tool, we first analyzed the off-target effects of the editing of OsAOX1 genes. The two most likely off-target sites of each target were selected and examined in all of the T 0 plants, all of the T-DNA-negative T 1 plants generated from mutated T 0 lines and 24 randomly selected lines of T-DNA-positive T 1 plants with on-target mutation by site-specific genomic PCR based Sanger sequencing. As shown in Table 5, no mutations were found in the putative loci, even though on-target mutations could easily be detected.
of the seed region) with the other PAM ended sequence. The targeted site and the putative off-target site are 7 kb apart; therefore, we first examined the large deletion formed by the re-joining of the two cleavage sites. We did not detect a deletion between the two sites by PCR in any of the T 0 and T 1 generation OsBEL target plants (data not shown). To further evaluate the potential off-target effects of the CRISPR/Cas9 system in rice, we used PCR to amplify a 254-bp region around the putative site and then sequenced that region. It was not changed in any of the T 0 and transgene-negative T 1 plants. We further examined 60 lines of transgene-positive T 1 plants, and mutations were observed at off-target sites in two individual plants derived from different T 0 lines (Table 5). These results suggest that in this system, off-target modifications are rare and occur only in the transgene-positive T 1 generation.

Discussion
The predictable inheritance and segregation of genome modifications in later generations is highly desired in molecular breeding as well as in basic research. In this study, we targeted four different genes using a previously reported Gateway-based CRISPR/Cas9 system 7 . The schematic procedure of generation and analysis of targeted mutated plants was described in Fig. 1A. Our results confirm the high efficiency of this system in the T 0 generation. We found that a part of the un-mutated lines lacked the sgRNA, the Cas9 cassette, or even both, which is consistent with another rice CRISPR/Cas9 application using a different vector system 27 . Interestingly, the left border (LB) of T-DNA is easier to truncate during the integration.   However, the selectable marker was located closer to LB than the transgene fragments of CRISPR/Cas9 39 . These phenomena suggest that the possible recombination of the T-DNA fragment might be a potential reason to restrict the mutagenesis efficiency. According to the sequencing results for genomic DNA isolated from single leaves, a large percentage of edited T 0 generations are biallelically modified. An abundance of biallelic modifications, especially homozygous types, typically indicates that the mutations were generated at a very early developmental stage of the transformed embryogenic cell, suggesting the high possibility of predictable germline transmission 26 . However, we found that T 1 genotypes are not easily predicted. Increases and decreases in types of mutation were frequently found not only in heterozygous and chimeric lines but also in putative biallelic lines, and the segregations were distorted even when the mutation type was stably transmitted. Various lines, e.g., OsAXO1a #2, OsAOX1b #3, OsAOX1c #4 and OsBEL #2, showed that the T-DNA region was segregating according to standard laws by the chi-square test, whereas the transmission of targeted mutations was disrupted ( Table 2). One possible explanation is that the abnormal inheritance was caused by somatic mutations. For example, putative biallelic plants are actually chimeras with different homozygous mutations in separate cells. The mutations thus may have been lost or inherited unequally during germline segregation. Meanwhile, the PCR-based detection method has limitations in the detection of larger deletions because the PCR reaction would fail if the deletion removed the primer-binding sites. Therefore, the mutation frequency might be underestimated, and the failure of detection of the mutated allele might confound the analysis of inheritance patterns in certain lines. Moreover, it has been reported that different mutations can be detected in samples of different tissues 26 . Because we only examined the target sequence in a single leaf, it would not be surprising that some genotypes present in the rest of the plant were overlooked. Therefore, the additional mutations found in T 1 might be transmitted from undetected T 0 somatic mutations. Unexpected inheritance patterns of T 0 plants have also been observed in other studies, but with lower frequency 26,27 . Although there are no significant differences in the mutation rates of the two CRISPR/Cas9 systems 7,10 , the differences in the system construction still might be a reason for the observed variation in the frequency of somatic mutations. Compared to the random and complicated genetic transmission in the first generation, the patterns are stable and easier to predict in later generations. Except for newly occurring mutations, all of the T 2 genotypes were inherited normally from T 1 plants in the presence or in the absence of the transgene (Fig. 1B), showing standard germline transmission. Off-target events are an important concern in the application of CRISPR/Cas9 in plants. A double nicking approach, combining paired sgRNAs with distinct locations adjacent to the target site and a nickase version of mutated Cas9, was reported to effectively avoid off-target effects 28,40 , but it would limit the potential target range compared to the single sgRNA/Cas9 system. The off-target efficiency may vary greatly depending on the construction of CRISPR/Cas9, the organism and the transformation method. For the four different targets in this study, we found low-frequency mutations in only one off-target site, which had a 1-bp mismatch outside of the seed region, with on-target sites in the T 1 generation plant in the presence of the transgene. These results indicate that the off-target effect is indeed quite low in rice targeted gene modification using the vector and transformation system described here. By selecting target sites with the help of bioinformatic tools 34,35 , we successfully generated mutations in three individual AOX1 family members, which is difficult to achieve using standard RNAi methods (data not shown) due to the high sequence similarity. These results demonstrate the reliability of software-aided target selection   methods and also suggest that the off-target events in the highly efficient CRISPR/Cas9 system could be virtually avoided in rice-gene editing through careful sgRNA design. In animals, biallelic mutations can be efficiently generated in one-cell-stage embryos by microinjecting an excessive amount of sgRNA and Cas9 RNA [41][42][43] . This method nearly guarantees reliable germline transmission both in theory and in practice. However, the integration of the CRISPR/Cas9 sequence into the genome is necessary for generating targeted modifications in plants. Agrobacterium-mediated transformation of embryogenic calli is a common method for generating transgenic crops. The transformed cells soon divide, allowing only a short time window for generating the germline mutation. In contrast, the regeneration of transgenic crop plants from embryogenic cells normally requires several weeks or months, and the sgRNA/Cas9 complex should be continuously expressed during this period. This long expression period may give rise to the high risk of somatic mutations in the first generation. The intricate T 0 segregation pattern in this report strongly supports the widespread occurrence of somatic mutations. In contrast, we revealed that T 1 mutations, especially biallelic mutations, were stably transmitted to the next generation through germline transmission. An advantage of the CRISPR/Cas9-mediated gene editing is the potential for transgene-free progeny. Once the desired gene editing has been achieved, the transgene region can be easily segregated out in progeny via simple self-fertilization. In this study, we show that the T-DNA had indeed completely segregated out in the T 1 generation, with the targeted mutation transmitting independently. In addition, our results reveal that off-target mutations were only found in the transgene-positive T 1 plants. Therefore, the off-target effects might be largely reduced by selecting appropriate T 1 progeny. Taken together, our results indicate that stable inheritance and "transgene clean" homozygous targeted gene editing can be produced in the T 1 generation in CRISPR/Cas9-transgenic rice. Therefore, the system can be used as a simple, rapid and powerful molecular tool in crop variety improvement and will greatly advance molecular design in breeding.

Materials and Methods
Plant material and growth conditions. Rice plants (Oryza sativa L. ssp. japonica) were used for plant transformation. Mature, non-dormant seeds were sterilized and germinated in 1/2 MS medium under a light/dark cycle of 16 h/8 h at 28 °C for at least 10 days. Rice seedlings at the trifoliate stage or regenerated rice after 4 weeks of rooting were transferred to plastic buckets in a greenhouse maintained at 30 °C during the day and 28 °C at night.
Vector construction and rice transformation. The oligonucleotides used for targeted mutagenesis were designed with the help of the CRISPR-P and CRISPR-PLANT tools 34,35 and are listed in Supplemental Table S2. The Gateway-based CRISPR/Cas9 plant expression vectors were constructed as previously described 7 . The binary constructs were then introduced into the Agrobacterium tumefaciens strain EHA105. Embryonic calli from mature rice seeds were transformed by co-cultivation, selected with 50 mg/l hygromycin, and used to regenerate transgenic plants as previously described 44 . The numbers of transgene copies were determined using real-time PCR 45 .