Alfalfa genotype R2336 was vegetatively propagated in the greenhouse and used for genetic transformation. The CRISPR/Cas9 generated alfalfa mutants, wild type (control 1) and empty vector control (control 2) were grown under greenhouse conditions of 22°C/19°C day/night temperature, 16/8 h day/night photoperiod, 150 µmol∙m⁻²∙s⁻¹ light intensity and 70% to 80% relative humidity. All alfalfa plants used for this study were propagated by vegetative cuttings to avoid the outcrossing problem.

The MsSGR sequence was retrieved from the assembled alfalfa reference gene sequence DOBLAST (CADL) Genome Blast Server (https://www.alfalfatoolbox.org/doblast/?filterword=DOBLAST&function=function2) and http://blast.jcvi.org/Medicago-Blast/index.cgi based on the high sequence similarity between alfalfa and M. truncatula genomes. To avoid any potential genotypic SNPs that might affect genome editing through mismatch, the genomic DNA fragment was obtained by PCR amplification, cloned into pGEM-Teasy plasmids (TA-clone) (Promega, Madison, WI, USA) and subjected to sequence analysis. Four MsSGR-gRNAs were designed to target first, second, and third exon sites using the web-based tool CRISPR-P (http://cbi.hzau.edu.cn/cgi-bin/CRISPR) (Lei et al., 2014) (Figure 1A). The four 20-nts spacers were fused to gRNA scaffolds and clustered with tRNA in a tandem manner as described in the protocol given by Xie et al. (2015) using the Golden Gate assembly method (Engler et al., 2008). The pGTR plasmid, which contains a gRNA-tRNA fused fragment, was used as a template to synthesize polycistronic tRNA-gRNA (PTG) (Xie et al., 2015). To construct MsSGR-PTG, the gRNA scaffold fragment was amplified by PCR using a pair of specific primers (Bsal-MsSGR-gRNA1, 2, 3 and 4-F and MsSGR-gRNA1, 2, 3 and 4-R), whereas the tRNA fragment was amplified as a primer dimer of gRNA-F and tRNA-R and the fragments were fused as gRNA-tRNA by overlapping extension PCR using primers (Supplementary Table 3). The overlapping PCR products were separated and purified by the Spin Column PCR Product Purification Kit (Wizard SV Gel and PCR Clean-Up System) following manufacturer’s instructions (www.promega.com). Then the chain of multiplex tRNA-gRNA with four alfalfa spacers was inserted into three optimized vectors of pRGEB31 with AtU6-sgRNA, hSpCas9 backbone by digestion and ligation using Fok I (NEB) and BsaI enzyme step by step (Xie et al., 2015). We assembled three multiplex gRNA-CRISPR/Cas9 vectors: 1) AtU6-tRNA-gRNAs + 35S-Cas9, 2) AtU6-tRNA-gRNAs + 35S-Cas9 with modified gRNA scaffold sequence and 3) AtU6-tRNA-gRNAs + AtUBQ10-Cas9 (Figures 1B–D). CaMV35 (35S) was used to drive the expression of Cas9 in vector I; vector II was created by modifying the gRNA scaffold region with a point mutation according to the protocol given in Dang et al. (2015), whereas in vector version III, we replaced the 35S promoter by Arabidopsis ubiquitin promoter to drive Cas9 (Figures 1B–D). The subsequent multiplex gRNA-CRISPR/Cas9 binary vector modules were transferred into Agrobacterium tumefaciens and transgenic plants were obtained by Agrobacterium-mediated transformation method (Zhou et al., 2011). Briefly, fully developed trifoliate leaves were collected from 5- to 6-week-old alfalfa plants and sterilized; trifoliate leaf discs were infected with A. tumefaciens strain EHA 105 and co-cultivated for 24 to 36 h in the dark at 24 °C and then transferred onto selection media containing phosphinothricin. Resistant calli were produced and proliferated for five to six weeks. The resistant calli were then transferred onto a regeneration medium and cultured under light conditions of 150 µmol∙m⁻²∙s⁻¹ at 24°C with 16/8 h photoperiod. PCR verification of the transgenic plants was conducted with genomic DNA extracted from leaf tissues using bar gene primers (Supplementary Table 3).

To screen putative CRISPR mutants, genomic DNA was extracted from independent lines of alfalfa and the target region was amplified by PCR and then sequenced. Genomic DNA was extracted from young leaves of the regenerated lines of alfalfa using the CTAB method described by Rogers and Bendich (1988). Amplification of the target region was made by PCR using extracted genomic DNA as a template with specific primers designed from the border of the target site (Supplementary Table 3). The PCR products were treated with 1 µl of Antarctic Phosphatase and 0.25 µl Exonuclease I enzyme and incubated at 37°C for 1 h, PCR products were cleaned up by eliminating dNTPs residues. Then, 5 µl of PCR products were used for sequence analysis to verify CRISPR/Cas9-induced mutations in MsSGR transgenic lines. Furthermore, to identify and characterize the nature of mutation type induced by multiplex gRNA-CRISPR/Cas9 system at target sites (gRNA1, gRNA2, gRNA3 and gRNA4), the amplified fragments of mutated regions were sub-cloned into pGEM-T Easy plasmids. To precisely identify the Indels at the target site, 20 colonies were randomly selected and grown in LB medium for plasmid DNA isolation. The isolated single-colony DNA was subjected to Sanger sequencing for mutation events (Indels) analysis to determine homozygosity, heterozygosity and/or chimeric mutant. Reads were analyzed by aligning with the reference sequence using the SeqMan Pro 15.0.1 (DNASTAR software for life scientists) (https://www.dnastar.com/quote-request/).

To decide whether the MsSGR mutants identified by genotyping screening could show a true stay-green phenotype, leaves collected from the mutants were exposed to dark-induced senescence treatment assay (Zhou et al., 2011). Typically the phenotype of a sgr mutant can easily be identified by its chlorophyll retention capacity using detached-leaf dark treatment or whole-plant shade treatment (Ren et al., 2007; Zhou et al., 2011; Xu et al., 2018). Replications of four to five leaves of MsSGR mutants and controls were harvested from the middle part of each plant and incubated at room temperature in darkness. The phenotype of leaf appearance (senescence) was evaluated every five days and photos were taken. At the 10th day of the incubation period, overall phenotypic efficiency was determined. This phenotypic screening procedure helps to confirm the effectiveness of the multiplex gRNA system in detail. Based on phenotypic grouping, MsSGR mutants were categorized into three classes: strong, mild and weak phenotype after 10 days of detached-leaf incubation in darkness.

As single gRNA-CRISPR/Cas9 has been successfully used in genome editing of many plant species, we originally constructed binary vectors using a single gRNA-CRISPR/Cas9 system for alfalfa genome editing (Supplementary Figures 1A–E). Alfalfa stay-green (MsSGR), phytoene desaturase (MsPDS) and other genes were used for target-specific mutagenesis. A total of 1,531 independent transgenic alfalfa lines were generated and examined for mutations (Supplementary Table 1). Unexpectedly, out of such a large number of transformants, no mutant phenotype was observed because there was no complete knockout of the four copies of the target genes. The overall genotypic mutagenesis efficiency was 2.5% with only one or two copies of the target gene mutated (Figure 2A; Supplementary Table 1). The complexity of alfalfa genome might be a reason for such a low efficiency of genome editing involving single gRNA-CRISPR/Cas9 approach. Furthermore, the outcrossing nature of this species made it impossible to obtain complete knockout in the next generation by selfing. Therefore, we had to change the genome editing strategy in alfalfa by testing different multiplex gRNA-CRISPR/Cas9 vectors.

Multiplex gRNA-CRISPR/Cas9 vectors were constructed to mutagenize all four allelic copies simultaneously. Four MsSGR-gRNAs were designed to target the first, second, and third exons of MsSGR (Figure 1A). Three vectors were constructed and tested (Figures 1B–D). In vector I (Figure 1B), Cas9 was driven by the CaMV35S promoter; vector II (Figure 1C) was obtained by modifying the gRNA scaffold region according to Dang et al. (2015); in vector III (Figure 1D), the CaMV35S promoter (used in vector I&II) was replaced by the Arabidopsis ubiquitin promoter (AtUBQ10) to drive Cas9.

A total of 492 transformants were obtained with the new multiplex vectors targeting the MsSGR gene. Genotypic analysis was performed by sequencing PCR products obtained by amplifying the target sites. Transgenic lines with mutations at the target sites were considered as candidate mutants of MsSGR and the ratio of mutagenized lines versus total examined lines across the populations was used as genotypic efficiency (mutagenesis efficiency). Genotyping analysis showed that the mutagenesis efficiencies were 31% (65 of 210), 23% (29 of 128) and 49% (71 of 144) for vector versions I, II and III, respectively (Figure 2A; Supplementary Table 2). Notably, the mutagenesis efficiency of vector III was higher than that of versions I and II (Figure 2A). Furthermore, the mutagenesis frequencies at four target sites differ across tested vectors (Supplementary Figure 2). For instance, vector version I showed 3.2%, 2.2%, 15.5% and 9.8% mutagenesis efficiencies for MsSGR-gRNA1, -gRNA2, -gRNA4 and -gRNA3, respectively (Supplementary Figure 2). Similarly, vector version II, the least efficient vector, showed 2.3%, 1.5%, 9.6% and 7.2% mutagenesis efficiencies for MsSGR-gRNA1, -gRNA2, -gRNA4 and -gRNA3, respectively (Supplementary Figure 2). On the other hand, the most effective vector, version III, showed 35%, 16%, 22% and 12% for MsSGR-gRNA1, -gRNA2, -gRNA4 and -gRNA3, respectively (Supplementary Figure 2).

The multiplex system was also tested for other genes, the efficiency of vector III was consistently higher than vectors I and II for every gene tested (Supplementary Table 2). The genotypic efficiency of vector III was in the range of 44% to 75% (Supplementary Table 2), which is 18 to 30 folds higher than the overall efficiency of the single gRNA-CRISPR/Cas9 system.

To identify and characterize mutation events induced in MsSGR by the multiplex gRNA-CRISPR/Cas9 system at the target sites (gRNA1, gRNA2, gRNA3 and gRNA4), each fragment of the mutated region was amplified and sub-cloned into pGEM-Teasy plasmid vector to determine the mutation events by Sanger sequencing. Twenty single colonies were randomly selected for each independent line to characterize the mutation events at the targeted sites (Figures 3 and 4; Supplementary Figures 3A and 4A). The assessment of mutation events (deletion, insertion and substitution) at target regions also helps to determine the homozygosity/heterozygosity or chimeric nature of the mutant in depth. Ten MsSGR mutants, namely MsSGR-3 and MsSGR-16 for vector version I, MsSGR-21 and MsSGR-23 for vector version II, and MsSGR-33, MsSGR-36, MsSGR-53, MsSGR-68, MsSGR-83, and MsSGR-92 for vector version III, were subjected to this analysis (Figure 3). Different levels of nucleotide deletions (1–59 nts), insertions and substitutions (1–4 nts) at all target sites across the tested lines in alfalfa were detected (Figure 3; Supplementary Figure 5). Interestingly, one-nucleotide deletions, insertions and substitutions were the most frequent mutation events, with 37%, 26% and 20% occurrence, respectively, representing a total of 83% of overall mutation events (Supplementary Figure 5). Nucleotides “T” and “C” were most frequently involved in the insertion and/or substitution events, followed by “A” and “G.” In addition, large fragment deletions with 59 nts were the third most frequent mutation events, with 16% frequency between gRNA2 and gRNA4 in MsSGR-23, -33, -53, and -92 mutants (Supplementary Figure 5). Two- to 28 nts deletions were also detected across designed gRNAs with about 8% to 16% occurrence (Supplementary Figure 5). Notably, the occurrence of four induced allelic mutation copies was used to identify the homozygosity/heterozygosity of the line (Figures 3 and 4A; Supplementary Figures 3A and 4A). For instance, the homozygosity of MsSGR-21 was the combination mutation events of occurred at gRNA1, gRNA2 and gRNA3 (Figure 4). This result revealed that sequences of all the clones have at least one or more mutation event at the target site(s), which made the mutant 100% tetra-allelic homozygous (Figures 3 and 4A). Furthermore, the results suggested the capability and effectiveness of at least two sgRNAs in multiplex CRISPR/Cas9 modules to knockout all allelic copies in tetraploid alfalfa genome.

To determine whether the mutants display true stable stay-green phenotype, the mutants were subjected to dark-induced senescence assay. Three classes of phenotype (strong, mild and weak) were observed after detached leaves were incubated in darkness for 10 days (Figures 2B–E and 4B–D; Supplementary Figures 3B–D and 4A–D). Strong phenotype refers to the mutants with consistently retained greenish appearance throughout the incubation period (Figures 2E, F). Mild phenotype refers to the mutants with stay-green up to the fifth day that gradually changed to a mixture of light green and yellow (mosaic) by the 10th day of the incubation period (Figures 2E, F). Weak phenotype refers to those mutants that had light green color at the fifth day of dark incubation but became completely yellowish by the 10th day of the incubation period (Figures 2E, F). Interestingly, Vector III had the highest phenotypic efficiency, with 50%, 18% and 32% for strong, mild and weak phenotypes, respectively (Figures 2E, F; Supplementary Table 2). Vector I and II had a high proportion of weak phenotype (70% to 88%), with relatively low mild and strong phenotypic efficiency of 12% to 17% (Figures 2E, F). As illustrated in Figures 2B–D, MsSGR-18 and MsSGR-21 were the strong phenotype mutants generated by vector versions I and II, respectively (Figures 2B, C), whereas vector version III produced several strong mutants including MsSGR-53, MsSGR-68, MsSGR-70, MsSGR-83, MsSGR-32, MsSGR-33, MsSGR-34, MsSGR-35, and MsSGR-36 (Figure 2D).

Phenotype analysis was also carried out for other genes, vector III was consistently the most efficient one, with phenotypic efficiency in the range of 37% to 68% (Supplementary Table 2).

Similar to detached leaves dark treatment assay, when the whole plants of the strong mutants were kept under dry shade conditions (Figure 5), they consistently maintained the greenness after four weeks of shade treatment (Figures 5D, F–M). In contrast, the whole plants of wild type and empty vector became yellowish (bleached) after four weeks of shade treatment (Figures 5C, E). Taken together, phenotyping and genotyping analyses consistently revealed that the multiplex system is very effective in generating complete knockout mutants and version III is the most efficient vector for alfalfa genome editing.

We further compared the effectiveness of MsSGR mutants generated by the multiplex gRNA-CRISPR/Cas9 system with previously developed MsSGR-RNAi transgenic lines (Zhou et al., 2011) using detached-leaf dark treatment assay. The multiplex gRNA-CRISPR/Cas9-generated mutants (MsSGR-21 and MsSGR-53) had excellent chlorophyll retention with a consistently deep green appearance throughout the dark incubation period, whereas the MsSGR-RNAi lines showed less strong phenotype with light greenish phenotype after 10 days of dark treatment (Supplementary Figure 6).