All experiments were conducted using Arabidopsis thaliana Columbia-0 (Col-0) and fad2-1 (Miquel and Browse, 1992) as an experimental control. Arabidopsis seeds were sterilized with 70% EtOH and 0.5% NaOCl and then were washed 7∼8 times with distilled water before sowing. The seeds were subject to stratification at 4°C for 3 days. They were then grown in 1/2 MS media containing 1% sucrose in a culture chamber at 23°C with a 16-h-light/8-h-dark cycle. About 14-day-grown seedlings in MS media were transferred to soil for further growth.

For basal CBE binary vector pJY-RpAID, plant (A. thaliana) codon-optimized coding sequences of PmCDA1 were synthesized by Integrated DNA Technology (Iowa, United States). First, pKI1.1R (Addgene #85808) (Shimada et al., 2010) was used to generate a pJY-RpEmpty vector, in which the original Cas9 sequences were removed, and two enzyme sites XmaJI and XhoI were inserted instead. Nickase Cas9 (D10A) and the codon-optimized AID were PCR amplified to generate pJY-RpAID binary vectors by the infusion cloning method (Takara, Japan) onto an XmaJI/XhoI-treated pJY-RpEmpty vector. The pJY-RpAID vector contains the following elements instead of the original Cas9 expression cassette from pKI1.1R: RPS5A promoter-SV40 NLS-nCas9 (D10A)-SV40 NLS-67aa linker-3X FLAG-PmCDA-SV40 NLS-UGI-heat shock protein 18.2 terminator (Figure 1B and Supplementary Figure 1). AarI-mediated gRNA cloning was performed for plant transformation vectors as previously described (Kim et al., 2016). For base editing the Arabidopsis FAD2 gene, six different 20-mer guide sequences (Figure 1E) were designed using the CRISPR RGEN Tool¹ and cloned into AarI-treated pJY-RpAID or a pJY-RpABE vector.

Agrobacterium strain GV3101 was used to transform A. thaliana via the floral dipping method (Clough and Bent, 1998). The resulting T₁ seeds were germinated and grown in 1/2 MS media containing 1% sucrose and 50 ng⋅μl^–1 hygromycin. Only surviving plants in the media were selected and transferred to soil for further analyses. Leaf genomic DNA was extracted (Edwards et al., 1991) for Sanger sequencing and was analyzed whether base editing had occurred within the FAD2 gene.

The PCR was performed on extracted genomic DNA as a template using Ex Taq (Takara) to generate FAD2 amplicons for subsequent sequencing analyses. A forward primer 5′-GCATTGTTTCAAACGCTCAA and a reverse primer 5′-TCATAACTTATTGTTGTACCAGTAC were used for 30-cycle PCR as following conditions: pre-denaturation (95°C for 5 min); denaturation (95°C for 30 s); annealing (52°C for 30 s); extension (72°C for 1 min); and post-extension (72°C for 10 min). A purification kit (Cosmo Genetech) was used to purify the PCR products. The quality and the quantity of the products had been determined using Nanodrop (Denovix) before the samples were deposited to a sequencing company (Bioneer).

The 100 seeds were used for the gas chromatography (GC) analysis. About 500 μl of 5% sulfuric acid solution was dissolved in methanol, and toluene was added to each sample. To quantify the fatty acid contents, 15:0 fatty acid was dissolved in sulfuric acid. Each sample immersed with the solution was subject to the reaction in a water bath at 85°C for 2 h. Each sample was then added by 1 ml of 0.9% NaCl. After that, 1 ml of hexane was added to the solution three times and then was centrifuged at 330 × g for 2 min. The supernatant was aspirated, transferred to a 6-ml tube, purified fatty acid methyl ester (FAME) using a nitrogen concentrator (Eyela). The extracted FAME was dissolved in 200-μl hexane and then inserted into a GC vial. A DB-23 column (30 m × 0.25 mm, 0.25-μm film, Agilent) was used, and the extracted FAME was analyzed by GC-2030 (Shimadzu). The GC oven temperature was raised from 190 to 230°C at 3°C min^–1.

The germination tests were conducted on the seeds that had been stratified for 3 days at 4°C. The germination events were scored when the roots started to appear in 1/2 MS media, 1/2 MS with 150-mM NaCl, and 1/2 MS with 300-mM Mannitol. All 1/2 MS used in the experiment contained 1% sucrose. The germination tests were carried out every 12 h for 4 days and replicated three times with 40 seeds per each line.

Col-0 and fad2 alleles including fad2-1 were grown under normal or stress conditions to monitor root development. For stress conditions, 4 DAG (days after germination) seedlings were transferred to 1/2 MS media, containing 75-mM NaCl or 200-mM Mannitol. The measurement of the root length was performed four times each for five individuals. The root length was measured on the 5th day of growing the seedling plant perpendicular to the light. The length measurement was carried out using the Image J program. Relative root length (%) was calculated as the average root length in stress MS media/average root length in MS media × 100.

The differences between the wild-type (Col-0) plants, fad2-1 and the CBE lines, were identified by performing a one-way ANOVA using GraphPad Prism. Asterisks indicate significant differences compared to the wild-type plants (^∗p < 0.05; ^∗∗p < 0.01; ^∗∗∗p < 0.001).

We aimed to generate potential “attenuated” fad2 alleles with increased oleic acid at the cost of linoleic acid and linolenic acid, while minimizing adverse “trade-off” effects. We reasoned that certain C-to-T and/or A-to-G base substitutions, achieved by in planta CBE or ABE, within the coding regions of FAD2 might attenuate, rather than abolish, their function (Figure 1A). To test this, we introduced random C-to-T and/or A-to-G base substitutions within several coding regions of FAD2 locus via in planta CBE or ABE implementation. We first constructed a basal CBE binary vector pJY-RpAID in which AIDv2 is under the control of the RPS5A promoter, and the gRNA cloning cassette is under the control of the Arabidopsis U6 promoter (see section “Materials and Methods” for details; Figure 1B and Supplementary Figure 1). For the ABE binary vector, we adapted our pJY-RpABE vector reported previously (Figure 1B; Kang et al., 2018). Six distinct coding regions within the FAD2 locus were selected as ABE and/or CBE targets based on the structural properties of the ER-membrane-bound FAD2 protein, which consists of six transmembrane domains and three histidine box motifs (Okuley et al., 1994; Shanklin et al., 1994; Figures 1C,D). We designed six independent gRNAs denoted as “g1–6” (Figure 1E) for BEs to target the selected coding regions, which correspond to the N-terminal transmembrane domain, tandem His-motif, a membrane-proximal cytosolic domain, cytosolic stretches, and C-terminal domains of the FAD2 protein (Figures 1C,D). Five ABE and three CBE constructs (Figure 1E) were created and transformed into Arabidopsis Col-0. Upon culture on selective media, 26 T₁ plants were collected for analysis of BE transgenes and transferred to soil for later GC analyses of T₂ seeds.

Individual pools of T₂ seeds harvested from each T₁ plant were subjected to parallel GC analyses to quantify their lipid compositions. Four independent T₁ plants (named as g5CBE1-4) from the CBE constructs with g5CBE guide RNA showed increased oleic acid content in their T₂ seeds (Supplementary Table 1). T₂ seeds from g5CBE1, g5CBE2, g5CBE3, and g5CBE4 exhibited increased oleic acid contents up to 36.5, 36.2, 48.9, and 21.6%, respectively, while wild-type Col-0 and other transgenic lines (g1, g4CBEs and g2, g3, g4, g5, g6ABEs) oleic acid contents were no more than 20% (Supplementary Table 1; values with asterisks). Furthermore, g5CBE plants displayed decreased polyunsaturated fatty acid levels and increased eicosenoic acid (20:1^Δ11) (Supplementary Table 1), implying that the desaturase activities of FAD2 protein had been compromised in those transgenic plants. We progressed to the next segregating T₂ generation with g5CBE1 and g5CBE3, which had the highest amounts of oleic acid accumulation at their T₂ seeds, in order to test whether bona fide cytosine base editing, which might be responsible for those increases, had occurred within the targeted FAD2 region in those transgenic plants. Notably, other gRNAs transformants did not show any significant changes in fatty acids in the T1 generation, which may be due to low levels of either base editing or oleic acid changes. We decided to focus on g5CBE lines for the following experiments.

Seven (g5CBE11-17) and five T₂ progenies (g5CBE31-35) from the g5CBE1 and the g5CBE3 T₁ lineages, respectively, were analyzed for FAD2 genotypes (Figure 2A and Supplementary Figure 2). Genomic DNA extracted from leaves of the T₂ plants was subjected to Sanger sequencing covering the CBE target region of the FAD2 gene (Figure 2A and Supplementary Figure 2). Interestingly, segregating C-to-G substitutions were observed at positions C2 and C12 in g5CBE11-16. Given that the protospacer adjacent motif (PAM) NGG sequences are regarded as the positions 21–23, these C-to-G substitutions are expected to result in the amino acid changes A295G and D298E, respectively (Figure 2A and Supplementary Figure 2). Among T₂ plants, g5CBE12 and g5CBE14 appeared to have discrete C-to-G substitution chromatogram signals at the corresponding positions, indicating exclusive (A295G; D298E)-type amino acid changes had occurred (Figure 2A and Supplementary Figure 2). g5CBE11, g5CBE13, g5CBE15, and g5CBE16 showed mixed signals of C-to-G and C-to-T substitutions at the C2 position, and mixed signals of C and C-to-G at the C12 position (Figure 2A and Supplementary Figure 2). These results indicate that the edited bi-alleles at the C2 position and heterozygous C-to-G alleles at the C12 position had been transmitted to T₂ plants. Of note, g5CBE17 showed all wild-type signals without any evidence of base editing events (Figure 2A and Supplementary Figure 2). Among g5CBE3 lineage, g5CBE31 plants also showed the A295G pattern from C-to-G conversion at the C2 position, accompanying with C-to-T conversion at C6 that would, anyway, result in synonymous mutation (Figure 2A and Supplementary Figure 2). In the g5CBE32, exclusive C-to-T substitutions occurred at the C2 position, resulting in A295V, and C-to-T and C-to-G at the C5 and C6 positions, respectively, resulting in T296M (Figure 2A and Supplementary Figure 2). In the g5CBE33 plant, heterozygous or mosaic C-to-T substitution at C2 was observed (Figure 2A and Supplementary Figure 2). The sequencing results for the g5CBE34 and g5CBE35 plants showed a mixed pool of sequencing traces, suggesting that larger genomic changes, such as indels, rather than point mutations might have occurred (Figure 2A and Supplementary Figure 2).

Gas chromatography analyses were then carried out on the corresponding T₃ seeds from each T₂ plant to investigate how the lipid composition had changed according to allelic differences with amino-acid changes. The oleic acid contents increased the most to 64.9, 63.9, and 69.2% in T₃ seeds from g5CBE12, g5CBE14, and g5CBE32 plants, respectively, at the expense of the polyunsaturated fatty acids (Figure 2B; red-colored values). These lines are characterized by two concurrent amino acid changes at different residues resulting from two independent cytosine base-edited alleles (Figure 2A). The oleic acid content in T₃ seeds from g5CBE11, g5CBE13, g5CBE15, and g5CBE16 showed moderate increases of oleic acid ranging from 42.6 to 45.8% (Figure 2A). Segregating C-to-T alleles at the C2 position and/or WT alleles at the C12 position might be responsible for those moderate increases in T₃ seed oleic acid contents as compared to the maximum level increases from g5CBE12 and g5CBE14 (Figure 2B). g5CBE31 also showed an intermediate level of increase in oleic acid, indicating that A295G results in the partial suppression of FAD2 activity. g5CBE33, which contains heterozygous A295V, showed only slight increases in the oleic acid content to 25.7%, suggesting that A295V might affect FAD2 activity in a dosage-dependent manner (Figure 2B). g5CBE34 and g5CBE35 showed modest increases in oleic acid content up to 54% and 49.5%, respectively, but not to the maximum levels observed in g5CBE12 and g5CBE14 (Figures 2A,B).

We proceeded with g5CBE11, g5CBE14, g5CBE31, g5CBE32, g5CBE33, and g5CBE34 to the next T₃ generation in order to isolate homozygous alleles and to verify whether the changes in fatty acid composition depend on possible T₃ segregations (Figures 2A,B; the line numbers with asterisks). g5CBE11 was selected for a representative among the T₂ group containing biallelic A295G/A295V along with heterozygous D298E mutations. g5CBE14 and g5CBE32 were for A295G; D298E and A295V; and T296M allelic groups, respectively. g5CBE31 had the A295G mutation and a synonymous C-to-T mutation. g5CBE33 was chosen for the heterozygous A295V allele. g5CBE34 was also included in this progression group in anticipation of the isolation of an indel allele in their T3 generation. Sanger sequencing analyses were performed in the T₃ progenies of each allelic group. As a result, we successfully isolated all possible base-edited T₃ homozygous alleles for each genotype represented by g5CBE114 (A295G; D298E), g5CBE144 (A295G; D298E), g5CBE316 (A295G), g5CBE321 (A295V; T296M), and g5CBE337 (A295V), hereafter referred to as fad2-114, fad2-144, fad2-316, fad2-321, and fad2-337, respectively (Figures 3A,B and Supplementary Figure 3). For the T₃ progenies of g5CBE34, we isolated a 4-bp insertional mutation allele, named fad2-349 (frameshift mutation), presumably resulting in FAD2 knockout via a premature stop codon (Figures 3A,B and Supplementary Figure 3). Interestingly, segregating T₃ progenies of g5CBE34 also present base-edited alleles (A295G), indicating that our CBE application had generated both base substitution and insertion mutations in a T₂ g5CBE34 plant (Figure 3A and Supplementary Figure 3). The presence of a transgene was analyzed by PCR amplification of the nCas9 part for the T2 and T3 generations of g5CBE1 and g5CBE3 lines. The results indicate that the transgene had been segregated out from T3 in the g5CBE11 line and T2 in the g5CBE3 line (Supplementary Figure 4). Therefore, de novo additional base editing in the T3 generation was unlikely to have occurred.

Gas chromatography analyses carried out on all T₃ genotypes revealed that all defined homozygous base-edited alleles showed significant increases in their oleic acid contents at their T₄ seeds compared to wild-type Col-0 (Figures 3A,C and Supplementary Table 2). fad2-321 (A295V; T296M) showed the highest increase, in oleic acid content, up to 64.7 ± 5.7%, among the base-edited fad2 alleles (Figures 3A,C and Supplementary Table 2), which is consistent with observations in its parental T₂ generation (Figure 2B). This increase is even higher than that of the previously reported fad2-1 (A104T) (Lemieux et al., 1990) (55.9 ± 1.9% in our analyses, Figure 3C and Supplementary Table 2). fad2-114 and fad2-144 (A295G; D298E) showed enhanced oleic acid production, up to 57.9 ± 5.4% and 56.5 ± 2.3%, respectively (Figures 3A,C and Supplementary Table 2), more than their segregating siblings (38 ± 2%; Figure 3A and Supplementary Table 2). The increases are likely due to the homozygous changes in two amino acids at the same time, which had been segregating and thus having shown moderate increases in the previous generation g5CBE11 as well (Figure 2B). Particularly, the linoleic acid (18:2) and the linolenic acid (18:3) contents were dramatically decreased to below 2.4% and to below 2%, respectively, in both fad2-321 and fad2-114, while those contents were 2.4 ± 0.2 and 6.2 ± 0.5% in fad2-1 (Figure 3C and Supplementary Table 2), supporting the notion that the increases are due to “attenuated” desaturase activities of FAD2 in those alleles. Notably, the oleic acid content of the attenuated allele, such as fad2-321, was comparable to that of a fad2-349 knockout allele in which the contents were measured as 59.5 ± 10.4% (Figures 3A,C and Supplementary Table 2).

We hypothesized that certain “attenuated” fad2 alleles in our study might not have as severe growth defects as the fad2 knockout allele, and possibly even less severe than the previously reported fad2-1 allele while maintaining increased oleic acid levels. In this regard, we decided to characterize our newly generated fad2 alleles for phenotypical differences in terms of their growth responses. We analyzed two early growth responses, i.e., germination and root growth rate, for each allele under normal and stress conditions. We used experimental stress conditions of 150-mM NaCl and 300-mM Mannitol, and 75-mM NaCl and 200-mM Mannitol on top of 1/2 MS media for the germination and root growth test, respectively, with the intention of mimicking salt or water-deficit stress as has been described in previous studies (Zhang et al., 2012; Yu et al., 2017). For each of the alleles, germination tests were performed by scoring seed germination events at 12-h intervals during the initial 4 days after planting to assess whether increases in oleic acid content influence germination (Figures 4A–C). Under normal conditions, the germination frequencies of all tested alleles reached similar levels after 4 days, showing >95% germination success. Of note, fad2-349, fad2-321, and fad2-144 alleles did show slightly reduced germination rates (∼72 h; Figure 4A), which coincide with relatively higher oleic acid contents observed in those alleles compared with others (Figure 3C). Upon both experimental stress conditions, fad2-349 and fad2-321 germination rates deteriorated sharply along with fad2-1, which showed significantly lower germination rates at all time points tested (Figures 4B,C). However, fad2-144, fad2-316, and fad2-337 maintained germination rates comparable to that of WT (Figures 4B,C). These results suggest that those fad2 alleles with increased oleic acid content also have enhanced resistance to stress susceptibility during germination in contrast to the case for fad2-1, fad2-349, and fad2-321 (Figures 4B,C). Notably, fad2-144, which showed one of the highest levels of oleic acid content and contains two concurrent amino acid changes (A295G; D298E, Figure 3), was fairly resistant to the stress conditions tested for germination relative to fad2-1 (Figures 4B,C).

Next, we assessed root growth under normal or stress conditions by measuring root lengths of 4-day-old seedlings from each of the fad2 alleles (Figures 4D–F). Under normal growth conditions, all fad2 alleles tested showed a range of the root length reduction compared to wild-type Col-0 (Figures 4D,E). For example, fad2-144 and fad2-321 showed considerable reduction similar to the reduction observed in fad2-1 and knockout fad2-349 (Figures 4D,E), suggesting that the corresponding allelic changes impede normal root development. On the other hand, fad2-316 and fad2-337 appeared to have insignificant levels of root reduction (Figures 4D,E), indicating that their allelic variation does not alter normal growth processes. Stress treatments caused a range of reduction rates of root lengths on the seedlings from each allele tested (Figures 4D,E). Under the salt-stress condition, the knockout allele fad2-349 showed a reduction of root length to 55.8% of its length under the normal unstressed condition (Figure 4F), which is consistent with the reported growth defects in FAD2 knockout mutations under salt stress. On the other hand, Col-0 and fad2-1 showed 66% and 63.6% of their normal growth lengths, respectively (Figure 4F). Notably, fad2-316 (A295G) and fad2-337 (A295V), which display an intermediate increase of oleic acids from single amino-acid substitutions (Figure 3), showed less reduction in root length to 82.9% and 79.7%, respectively (Figure 4F). The result suggests that their growth might be even less affected by the salt stress condition than WT. fad2-144 (A295G; D298E) and fad2-321 (A295V; T296M), which were regarded as higher oleic acid-producing alleles with two concurrent amino-acid substitutions (Figure 3C), displayed 67.8 and 69.2% of relative root length under salt stress, similar to the reduction level observed in Col-0 (Figure 4F). The growth defects seemed more obvious in the water-deficit stress condition with 200-mM Mannitol, as the root length of fad2-349 was reduced to 32.6% while Col-0 was to 45.5% (Figure 4F). Again, fad2-316 and fad2-337 presented 49.1 and 54% of relative root length, respectively, indicating significantly increased tolerance to water-deficit stress compared with knockout fad2-349 (32.6%) or previously reported fad2-1 (35.5%) (Figure 4F). The relative root lengths of fad2-144 and fad2-321 were 40.4 and 41.1%, respectively (Figure 4F). The result suggests that they are still more tolerant than fad2-349 and fad2-1 but have increased susceptibility to stress conditions comparable to Col-0.

Seed weight and fatty acid content of fad2 alleles were measured (Supplementary Figure 5). Compared to wild-type Col-0, the weight of fad2-1 was decreased, while the new fad2 alleles showed an increase. The fatty acid content of the seeds showed some differences between the new fad2 alleles. The fatty acid content per seed of fad2-1 was not significantly different from that of wild type. The allele produced in this experiment had an increase in seed weight compared to the wild type, and there was no difference or a significant increase in the fatty acid content per seed. These results suggest that mutations in the FAD2 gene can affect seed weight and fatty acid content. The correlation between FAD2 gene mutation and fatty acid content needs further study.