Arabidopsis seeds were surface-sterilized with bleach (∼4% sodium hypochlorite) and plated on MS plates. After plating, the plates were kept at 4°C for at least 2 days to obtain germination synchrony before being transferred to 22°C under constant illumination (80–100 μmol m^-2S^-1) and 70% relative humidity for germination and growth. Murashige and Skoog (MS) medium (pH 5.8) was made with full strength MS salts (Caisson Laboratories, United States), 2% sucrose, and 0.3% gelite (Duchefa, Netherlands). For selection plates, hygromycin B was added to a final 25 mg/L concentration to the MS media. For soil growth, the seeds planted on soil (Sungro mixture#5, Canada) were placed in a growth room operating at 22°C with the cycle of 16-h of light and 8-h of darkness (the light intensity of 80–100 μmol m^-2S^-1) after 2 days of cold stratification.

For the selection of multiple targeting sgRNA, multiple sequence alignment software Clustal W¹ was used to align the coding sequence of CBF1, CBF2, and CBF3. Two 19-bp sequences (sgRNA12 and sgRNA23) immediately before a PAM sequence (5′-NGG-3′) were selected and used for DNA oligomer synthesis for sgRNA.

The forward and reverse DNA oligomers for sgRNA targeting CBF1 and CBF2 were CBF12-sgR1-F (5′-GATTGAGCTGCCATCTCAGCGGTT-3′), CBF12-sgR1-R (5′-AAACAACCGCTGAGATGGCAGCTC-3′) and for sgRNA targeting CBF2 and CBF3 were CBF23-sgR1-F (5′-GATTGGAGTCAGCGAAATTGAGAC-3′) and CBF23-sgR1-R (5′-AAACGTCTCAATTTCGCTGACTCC-3′).

Following the protocol suggested by Liu et al. (2015), psgR-Cas9-At was used to generate each single sgRNA-containing vector (i.e., sgRNA12-Cas9 vector and sgRNA23-Cas9 vector). To make a double sgRNA-containing sgRNA12-sgRNA23-Cas9 vector, sgRNA23 module from the sgRNA23-Cas9 plasmid was PCR-amplified using a following primer pair (sgR_U6_Kpn1-F, 5′-GCCGGTACCCATTCGGAGTTTTTGTAT-3′; sgR_end_EcoRI-R, 5′-TATGAATTCGCCATTTGTCTGCAGAATTG-3′). The resultant PCR product was inserted into the KpnI and EcoRI sites of sgRNA12–Cas9 vector. Finally, the whole cassette of sgRNA12-Cas9-sgRNA23 released by HindIII and EcoRI from the double sgRNA containing construct was subcloned to the HindIII-EcoRI sites of pCAMBIA1300, resulting in the pCAMBIA-sgRNA12-sgRNA23-Cas9 construct. A schematic drawing of the construct generation is shown in Supplementary Figure S1.

Arabidopsis thaliana plants (accession Col-0, Ler, and C24RDLUC) were transformed with pCAMBIA-sgRNA12-sgRNA23-Cas9 construct via Agrobacterium tumefaciens strain GV3101 by floral dipping (Clough and Bent, 1998). T1 seeds were collected from the floral dipped plants and then selected on MS plates with hygromycin B 25 μg/mL. The selected T1 plants were transferred to soil and genotyped with 3 different primer pairs (pCAM1300intC-R, 5′-GGCCTCTTCGCTATTACGC-3′ and CBF12-sgR1-R, 5′-AAACAACCGCTGAGATGGCAGCTC-3′ for sgRNA12; doublesgRNA-F, 5′-GATCGACCTGTCTCAGCTGG-3′ and pCAM1300intC-F, 5′-ATTAATGCAGCTGGCACGAC-3′ for sgRNA23; Cas9-F, 5′-CCCAACTTCAAGAGCAACTT-3′ and Cas9-R, 5′-TCACTTTGGTCAGCTCGTTA-3′ for Cas9). For the cbf123 deletion detection, nested PCR was employed with the following primer pairs were used (CBF1-PS-F, 5′-CGTGTGCTCCCCACATATC-3′ and CBF2-PS2-F, 5′-ATTTGTTGCTTATGGGGAGA-3′ for the first round PCR; CBF1_small_R, 5′-AATCCAAAAAGACTGAGAACTCTA-3′ and CBF2q-R, 5′-CTGCACTCAAAAACATTTGCA-3′ for the second round PCR). For the cbf123 deletion confirmation and homozygosity test, 4 different primers were used (CBF1_samll_R, CBF2q-R, CBF1-PS-R, 5′-CCGCTTTTTGGATATCCTTG-3′ and CBF2qRT-F(172), 5′-AACTCCGGTAAGTGGGTGTG-3′).

Luminescence intensities of cbf123LUC-2 were measured by the Lumazone luminescence imaging system (Roper Scientific, United States). Ten-day-old seedlings grown on MS agar plates were incubated at 0°C for the designated times and the luminescence images were taken after luciferin spray. The luminescence intensity of each seedlings was quantified with the WinView32 program.

Total RNA was extracted from 11- to 13-day-old seedlings with or without cold (0°C) treatment using the RNAiso Plus reagent (Takara, Japan). After DNase I treatment (New England Biolabs, United States), cDNA was synthesized with 5 μg total RNA using the TOPScript^TM Reverse Transcriptase Kit (Enzynomics, Korea). cDNA was used as a template for quantitative real-time PCR (qRT-PCR) using SYBR^® FAST (KAPA Biosystems, United States) on the LightCycler^®96 (Roche, Switzerland). The following primer pairs were used for real-time PCR (CBF1q-F, 5′-GCATGTCTCAACTTCGCTGA-3′ and CBF1q-R, 5′-ATCGTCTCCTCCATGTCCAG-3′ for CBF1; CBF2q-F, 5′-TGACGTGTCCTTATGGAGCTA-3′ and CBF2q-R, 5′-CTGCACTCAAAAACATTTGCA-3′ for CBF2; CBF3q-F, 5′-GATGACGACGTATCGTTATGGA-3′ and CBF3q-R, 5′-TACACTCGTTTCTCAGTTTTACAAAC-3′ for CBF3; RD29A_RT-F, 5′-CTTGTCGACGAGAAGCAAAGAA-3′ and RD29A_RT-R, 5′-TCTTGATGGAGAATTCGTGTCC-3′ for RD29A; COR15A_RT-F, 5′-ACTCAGTTCGTCGTCGTTTCTC-3′ and COR15A_RT-R, 5′-TCTCACCATCTGCTAATGCCTC-3′ for COR15A; COR47_qRT-F, 5′-TGTCATCGAAAAGCTTCACCGA-3′ and COR47_qRT-R, 5′-ACCGGGATGGTAGTGGAAACTG-3′ for COR47; KIN1_qRT-F, 5′-ATGCCTTCCAAGCCGGTCAGAC-3′ and KIN1_qRT-R, 5′-CCGGTCTTGTCCTTCACGAAGT-3′ for KIN1; Clathrin-F, 5′-CTGACTGGCCCTGCTT-3′ and Clathrin-R, 5′-ATACGCGCTGAGTTCCC-3′ for Clathrin as the internal control).

For protein blot analysis for CBF123 protein detection, 14-day-old seedlings of wild type and cbf123LUC-2 on MS agar plates were cold-treated at 0°C for 6 h, the seedlings were frozen-ground and suspended in protein sample buffer [130 mM Tris-Cl (pH 8.0), 4.6% (w/v) SDS, 0.02% Bromophenol blue, 2% DTT, 20% (w/v) Glycerol]. Samples were boiled for 3 min and centrifuged at 13,200 rpm for 10 min at 4°C. The supernatant was run on the sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) gel (Nguyen et al., 2016). The resultant gel was blotted on to the polyvinylidene fluoride (PVDF) membrane and the existence of CBF proteins were detected with rabbit anti-CBF123 antibody and anti-rabbit goat anti-rabbit IgG-HRP conjugate (1:10000, Abcam, United Kingdom). Anti-CBF123 antibody was raised with full length protein of CBF2. Further detail on the method of anti-CBF123 generation is provided in Supplementary Figure S3.

Statistical analyses of the results from the experiments of three accessions were performed using two-way analysis of variance with Microsoft Excel program (2010 version), and significant differences between each index among accessions and among T2 ratios of the same accession were determined using Student’s T test.

To generate a triple mutant of CBF1, CBF2, and CBF3 genes using the CRISPR/Cas system, two possible sgRNA target sites were selected by aligning CBF1 and CBF2, and CBF2 and CBF3 with a multiple sequence alignment program (Clustal W)² (Figures 1A,B). The two sgRNA target sequences were named sgRNA12, which targets CBF1 and CBF2, and sgRNA23, which targets CBF2 and CBF3 (Figures 1A–C). There were no single sgRNA site that could target all three CBF genes in the coding sequence (CDS) region at the same time. By using these two sgRNAs, the CRISPR/Cas system can target the three CBFs in the four sites (Figure 1C). Since the sgRNAs will be used in three different accessions, Col-0, C24RDLUC and Ler, the CBF regions of these accessions were sequenced. We found that the DNA sequences for sgRNA12 and sgRNA23 were identical in all three accessions (data not shown). Then, the possible off-target sites of sgRNA12 and sgRNA23 were analyzed by the online program CRISPR-P³ (Lei et al., 2014). The analysis revealed that the off-target possibility of both sgRNA12 and sgRNA23 was extremely low (scores of 0.0–5.7 for off-target sites vs. 100 for CBF1, CBF2, and CBF3 genes), suggesting that each would very specifically target CBF1/CBF2 and CBF2/CBF3, respectively (Table 1).

In order to target CBF1, CBF2, and CBF3 genes, DNA oligomers for sgRNA12 and sgRNA23 were inserted into psgR-Cas9-At vector containing the Cas9 sequence with a chimeric RNA backbone for the sgRNA (Feng et al., 2013; Mao et al., 2013; Liu et al., 2015). The whole cassette with Cas9 was then subsequently subcloned into pCAMBIA1300 to generate the CBF targeting CRISPR/Cas binary vector construct. The construct was used to transform three Arabidopsis accessions through Agrobacterium mediated plant transformation. The seeds from the floral-dipped plants (T1 seeds) were harvested and plated on hygromycin plates for the positive transformant selection. The presence of transgenes including sgRNA12, sgRNA23, and Cas9 in the positive T1 seedlings was then tested by PCR with the primers used for construct component confirmation. While generally high (75–100%), the detection ratios of each transgene in each accession varied, implying various occurrences of the whole transgene insertion in each accession (Table 2). The ratios of plants with a complete set of three transgenes were higher than 60% in all accessions with Columbia-0 being the highest (80%) (Table 2). These transgenic plants with a whole transgene set were further analyzed for CRISPR/Cas efficiency on the CBF genes.

We then asked in T1 generation if our sgRNAs were functional in CRISPR/Cas-mediated genome editing and in what ratio the CRISPR/Cas-caused CBF123 deletions would occur in each accession. Our sgRNA design intended to target four sites (two sites per sgRNA) in CBF1, CBF2, and CBF3 genes (Figure 1C) to maximize the gene editing efficiency. Among the deletions that would be produced by our CRISPR/Cas system, the deletions for the cbf123 triple mutation should occur between the sgRNA12 site on CBF1 and sgRNA12 or sgRNA23 site on CBF2 (Figure 1C). In order to detect these deletions of the CBF gene region, we performed PCR using a primer set of one primer aligning to the 5′ UTR region of CBF1 and the other aligning at the 3’ UTR region of CBF2. Ca. 1070 bp and/or 1005 bp PCR products were expected if the deletion for cbf123 mutation took place (Figure 2A). Our initial PCR produced only several faint bands. It is highly likely because of the low numbers of cells with cbf123 deletion mutation in these T1 plants. It should be noted that T1 plants would contain heterogeneous cells with various CRISPR/Cas effects. Therefore, nested PCR was conducted with the 1000 times diluted initial PCR products. Nested PCR would also ensure that the faint bands are not non-specific and that the bands are also sufficiently intensified to be visible. As expected, in the nested PCR results, even the samples without initial visible bands displayed clear bands (Figure 2B). Among the T1 plants with all transgenes, the large deletion ratios for cbf123 triple mutations varied between 45.71 and 84.85% among the accessions with Ler being the lowest and C24RDLUC the highest (Table 3). These results suggest that the CRISPR/Cas system has accession-dependent efficiency in the generation of CBF123 deletion.

To determine whether the deletion occurred in the sgRNA cut sites, the nested PCR products from T1 plants were sequenced. Although the sizes of the PCR bands varied considerably, most bands showed a size of ca. 800 base pairs, which was in agreement with the assumption that the CBF123 deletion mutations would occur at the sgRNA cut sites present on CBF1 and CBF2 (Figure 2). CRISPR/Cas action can result in mutations with a few nucleotide additions or deletions at the sgRNA target site due to the error-prone non-homologous end joining pathway (Belhaj et al., 2013; Hsu et al., 2014; Lozano-Juste and Cutler, 2014; Liu et al., 2015). Because of this, T1 plants become genetically mosaic with cells containing different mutations at the sgRNA target sites. Thus, the PCR products similar in size could be heterogeneous and the sequencing of the PCR products could produce multiple peaks at the nucleotide after the sgRNA cut sites. Our sequencing results showed that the deletion started at the cut site of sgRNA or in close proximity to it, implying a specific double strand break on the target sites (Figures 3A,B). Our results also reveal that addition or deletion of nucleotides at the sgRNA target also occurred in our CBF123 deletion mutations (Figures 3A,B). Occasionally, we observed the sequencing results from some unexpected size products (Figures 3A,C). These could be due to the prolonged DNA repair which might lead to extra base pair deletion after the sgRNA cut site (Gorbunova and Levy, 1999).

The T2 seeds from the T1 plants were harvested and used for the analysis of genetic inheritance of the CBF123 deletion and the cbf123 triple mutant screening. We examined T2 progenies from all T1 plants with a complete set of sgRNAs and Cas9 transgenes; T2 progenies from 16 T1 Col-0 plants, 33 T1 C24RDLUC plants, and 35 T1 Ler plants were analyzed (Table 3 and Supplementary Tables S1–S3). In most cases, we analyzed 24 T2 seedlings per T1 line. We found that CBF123 gene deletions could be observed in T2 seedlings not only from the T1 lines with the CBF123 deletion, but also from the T1 lines without the deletions. Also, there were some cases when the CBF123 gene deletion was not inherited to T2 seedlings from the T1 plants that contained the CBF123 deletions (Supplementary Tables S1–S3). This indicates that the CRISPR/Cas9-induced mutations in germ cells do not always occur even when the CBF123 deletion in somatic cells (leaf cells) exists at T1. Interestingly, the transmission ratios of CBF123 deletion from T1 to T2 generation were different among the Arabidopsis accessions. Although not significantly different, the ratios of the CBF123 deletion at the T2 generations in Col-0 was the highest (33.56%) followed by C24RDLUC (19.80%). Ler showed the lowest CBF123 gene deletion ratio (3.74%) (Figure 4A). We also noticed the different accession-dependent inheritance ratios of the CBF123 deletion among the T2 progenies from T1 lines with and without the CBF123 deletions (Figure 4B). In Col-0, the CBF123 deletions were detected at very similar ratios in the T2 progenies, regardless of the presence of CBF123 deletions at the T1 generation (32.20% in T2 from CBF123-deleted T1 vs. 35.71% in T2 from CBF123 not-deleted T1) (Figure 4B). By contrast, C24RDLUC showed a higher transmission ratio of the CBF123 deletions in T2 lines from T1 lines with the deletion rather than in those from T1 lines without the deletions (29.0% vs. 6.40%) (Figure 4B). Though similar in the CBF123 deletion ratios in either T2 lines of Ler (5.77% vs. 2.08%), Ler showed very low CBF123 deletion ratios in T2 (Figure 4B).

In order to isolate the homozygote mutants, we analyzed homozygosity of the candidate cbf123 triple mutants in each accession by PCR with three pairs of primers that were designed to detect the combination of CBF123 deletions (Figure 5A). Through the intensive PCR screening, we identified several homozygote mutants for the CBF123 deletion in each accession (Figure 5B).

In the case of the C24RDLUC background cbf123 mutant, we found two different homozygous mutants. cbf123#1 in C24RDLUC (hereafter cbf123LUC-1) had a 6,551 bp deletion between the cut sites of sgRNA 12 and sgRNA 23 in CBF2 while cbf123#2 in C24RDLUC (hereafter cbf123LUC-2) contained two deletions; one deletion was a 3,633 bp deletion between the cut sites of CBF1 sgRNA 12 and CBF3 sgRNA23 with an 11 bp insertion, and the other was a 65 bp deletion between the cut sites of sgRNA12 and sgRNA23 of the CBF2 region (Figures 5C,D and Supplementary Figure S2). Ler background cbf123 mutant had a 6,551 bp large deletion between the cut sites of CBF1 sgRNA 12 and CBF2 sgRNA 23 and Col-0 background cbf123 mutant showed a 6485 bp large deletion between the cut sites of CBF1 sgRNA 12 and CBF2 sgRNA 12 (Figures 5C,D and Supplementary Figure S2). In addition, no mutation was found in the potential off-target region of a closely related CBF paralog, the CBF4 gene (data not shown).

Among these cbf123 triple mutants, we decided to further characterize the cbf123LUC-2 because this mutant’s background line, C24RDLUC, contains a stress-inducible RD29A promoter-driven luciferase, which will be beneficial in monitoring the CBF target gene expression in vivo. In addition, the Cas9 transgene was segregated out of cbf123LUC-2, but not in other cbf123 triple mutants in other accession backgrounds. RNA transcripts and protein blot analysis confirmed the lack of the CBF proteins in cbf123LUC-2, indicating that cbf123LUC-2 is a null mutant (Figures 6A,B). Luminescence imaging after cold treatment showed clear reductions of the RD29A-LUC luminescence in cbf123LUC-2 in comparison to C24RDLUC (Figures 6C,D). This RD29A-LUC expression patterns were well correlated with those of the endogenous RD29A expression in cbf123LUC-2 (Figure 6E). Additionally, expressions of other CBF123 target genes were also reduced in cbf123LUC-2 (Figure 6E). In growth and development, cbf123LUC-2 appeared slightly smaller than C24RDLUC (Figures 7A–C) in early development stages. At the fully grown stage, there appeared to be no big differences in growth and development between C24RDLUC and cbf123LUC-2 (Figure 7D).