Efficient genetic transformation and CRISPR/Cas9‐mediated genome editing in Lemna aequinoctialis

Summary The fast growth, ease of metabolic labelling and potential for feedstock and biofuels production make duckweeds not only an attractive model system for understanding plant biology, but also a potential future crop. However, current duckweed research is constrained by the lack of efficient genetic manipulation tools. Here, we report a case study on genome editing in a duckweed species, Lemna aequinoctialis, using a fast and efficient transformation and CRISPR/Cas9 tool. By optimizing currently available transformation protocols, we reduced the duration time of Agrobacterium‐mediated transformation to 5–6 weeks with a success rate of over 94%. Based on the optimized transformation protocol, we generated 15 (14.3% success rate) biallelic LaPDS mutants that showed albino phenotype using a CRISPR/Cas9 system. Investigations on CRISPR/Cas9‐mediated mutation spectrum among mutated L. aequinoctialis showed that most of mutations were short insertions and deletions. This study presents the first example of CRISPR/Cas9‐mediated genome editing in duckweeds, which will open new research avenues in using duckweeds for both basic and applied research.


Introduction
Duckweeds are small, fast-growing and morphologically simple flowering aquatic plants (Landolt, 1986;Ziegler et al., 2015). For decades, these features made duckweeds an attractive model for understanding fundamental biological processes of plants, for example, photoperiodic control of flowering (Cleland and Tanaka, 1979), cellular circadian clocks (Muranada and Oyama, 2016), as well as biosynthesis of auxin (Rapparini et al., 1999) and assimilation of sulphur (Datko et al., 1978). Recent studies showed that duckweeds have an efficient photosynthesis capacity, high starch accumulation rate and low lignin content, which make duckweeds excellent candidates for biofuels production (Xu et al., 2011(Xu et al., , 2012. In addition, duckweeds are capable of rapidly accumulating heavy metals, nutrients and pollutants, which can be used for bio-remediation of wastewater (Cheng and Stomp, 2009;Khellaf and Zerdaoui, 2010;Megateli et al., 2009;Stout et al., 2010;Xu et al., 2018). The extremely fast population doubling rate (3-5 days per asexual doubling under ideal conditions) also makes duckweeds a good candidate for the production of recombinant proteins including antigens, monoclonal antibodies and exogenous enzymes (Cox et al., 2006;Firsov et al., 2015Firsov et al., , 2018Khvatkov et al., 2018;Ko et al., 2011;Naik et al., 2012;Stomp, 2005;Sun et al., 2007), as well as dietary supplements for humans and animals because of its high nutritional value (Appenroth et al., 2017;Beukelaar et al., 2019;Bhanthumnavin and Mcgarry, 1971). Despite the potential for both basic research and industrial applications that established duckweeds as one of the most promising alternative future crops Lam et al., 2014), current research and development in duckweeds are constrained by the lack of efficient genetic manipulation tools that allow to knockout or knock-in a gene of interest in the duckweed genome within a short time.
While using the established transformation system, one can manipulate the transcript abundance of the gene of interest through RNA interference (RNAi) or over-expression, it can suffer from low efficiency and instability due to DNA methylation of the transgene and incomplete target gene silencing (Klose and Bird, 2006;Weinhold et al., 2013). Therefore, to understand the function of genes and traits in duckweeds, developing tools that can stably and precisely manipulate the gene of interest is critical. To this end, the recently developed CRIPSR/Cas9 system, which was adapted from a naturally occurring genome editing system in bacteria and has been widely used in both animals and plants, is ideal (Cao et al., 2016;Ma et al., 2016;Yang, 2017). However, to our knowledge, no study has reported using CRISPR/Cas9 in duckweeds. Due to the feature of asexual budding, which is different from many other plants on which CRISPR/Cas9 has been applied, whether and to what extent this system will work in duckweeds remains unknown.
Here, we describe the optimization of a transformation protocol and the application of CRISPR/Cas9-mediated targeted mutagenesis in L. aequinoctialis. We demonstrate that using Agrobacterium to transform the duckweed plants with a CRISPR/ Cas9 vector can efficiently generate biallelic mutant plants in 5-6 weeks with high success rate.

Results
A stable and efficient genetic transformation system for L. aequinoctialis To establish a stable and efficient genetic transformation system in L. aequinoctialis (Figure 1a-h), we optimized previous protocols in several steps. First, by using an available callus induction protocol (Cant o-Pastor et al., 2015), we screened more than 100 clones of duckweed that were collected in China (Table S1). Among them, seven had more than 90% callus induction rate, and L. aequinoctialis 6002, which showed highest callus induction rate was used for further investigations. Second, we compared the infection efficiency among three different Agrobacterium tumefaciens strains that were used in previous studies: EHA105, AGL1 and GV1301. Among them, both the strain EHA105 and AGL1 showed overall high infection efficiency ( Figure S2). We used EHA105 for establishing the transformation protocol in L. aequinoctialis as it is a strain widely used for plant transformation. To estimate the transformation rate, a GUS vector was used for the transformation. At 3 days after Agrobacterium inoculation, we found more than 94% (80 out of 85) of the calli expressed GUS (Figure 1e), suggesting the overall high infection rate. The transformed calli (survived in the selection medium) were then transferred to the regeneration medium (Figure 1f), and transgenic fronds were obtained after 4 weeks (Figure 1g). Measuring GUS signals in transgenic fronds at 70 days after the transformation suggests that all of the transformations are stable (Figure 1h). The whole transformation procedure (from callus infection to the establishment of transgenic fronds) took only 5-6 weeks.
The detected GUS signal could be due to T-DNAs from living A. tumefaciens. To further examine whether the detected GUS signal was resulted from T-DNAs from living A. tumefaciens, we first detected the A. tumefaciens curdlan gene via PCR. After 35 cycles of PCR, no gene signal found ( Figure S4). Second, to examined whether the T-DNAs were integrated into the duckweed genomic DNA, we performed thermal asymmetric interlaced PCR (TAIL-PCR). Using the combinations of T-DNA specific and degenerate primers, we examined four independently transformed lines ( Figure S3). Sequencing four amplified DNA fragments confirmed that in three lines T-DNAs were inserted into the region that showed high homology to S. polyrhiza genomic sequences (Table 1). In one line, the surrounding genomic sequence near the T-DNA insertion showed no homology to any other sequences in the NCBI database, likely because the T-DNA was inserted to a L. aequinoctialis species-specific region (Table 1). These results showed that the observed GUS expression resulted from T-DNAs that were integrated into the L. aequinoctialis genomes.

CRSPR/Cas9 target selection and vector construction
Using the optimized transformation system, we examined the efficiency of CRSPR/Cas9-mediated genome editing in L. aequinoctialis. To this end, we focused on the duckweed phytoene desaturase gene (LaPDS), a common marker gene for testing genetic manipulation tools in plants (Fan et al., 2015;Shan et al., 2013). Based on the genomic sequence of LaPDS ( Figure S1), we selected three targets (T1-T3, Table 2) using the CRISPR-P 2.0 design tool and RNA Folding Form . While T1 and T3 are located in the first and second exon, respectively, T2 is located in the first intron (Figures 2a and S1). To increase the chance of obtaining a high sgRNA transcription, three promoter sequences from rice (a monocot species, same as duckweeds), OsU3, OsU6a and OsU6b, were used to drive the transcription of the sgRNA against T1, T2 and T3, respectively ( Figure 2b). The expression cassettes including the three targets were inserted into the binary vector pYLCRISPR/Cas9-MH using BsaI (Ma et al., 2015a). An overview of the constructed vectors is shown in Figure 2c. The CRISPR/Cas9 vector was transformed into the L. aequinoctialis calli using the optimized Agrobacterium transformation protocol.
Generating CRISPR/Cas9-mediated mutagenesis in L. aequinoctialis To examine the efficiency of the constructed CRISPR/Cas9 vector in L. aequinoctialis, we amplified and sequenced the LaPDS gene from the 105 independently regenerated plants. In total, while 25 plants contained only WT alleles, 80 plants contained at least one mutated allele (Figure 3), indicating a mutation success of 76.2%. Among these 80 mutated plants, 15 showed no WT allele, suggesting that LaPDS was completely mutated ( Figure 3). Interestingly, all of the 15 plants that contained no WT allele were biallelic mutants, in which two different mutated alleles were found. Similarly, all of the 65 plants that contained at least one WT alleles contained also at least two additional mutated alleles (chimeric mutants). Consistently, biallelic mutants showed an albino phenotype and significantly less chlorophyll a, chlorophyll b and total carotenoids, and the chimeric mutants showed slightly lower levels of pigments ( Figure 3 and Table S3).
Some of the biallelic mutants still showed a slightly green colour (Figure 3), which can be due to a small amount of cells that still contain WT allele or other unknown mechanisms. To further check the presence/absence of WT alleles, we designed WT allele-specific primers to amplify the genomic DNA of the two selected biallelic mutants. No PCR product can be found, which suggests that the slightly green colour of the biallelic mutants is likely not due to the presence of wide-type allele containing cells ( Figure S6).

Spectrum of CRISPR/Cas9-mediated mutations in L. aequinoctialis
Analysing the mutation spectrum of the 105 transgenic lines showed that 76.2% of them had at least one mutation at the site T3, while 25.7% of them had at least one mutation at the site T2. No plants showed any mutation at the site T1 (Table 3). At the site T3, while 66 plants contained chimeric mutations, 14 had biallelic mutations (and showed albino phenotype). At the site T2, 26 plants contained chimeric mutations and one contained biallelic mutations (and showed an albino phenotype).
Most of the mutations at the sites T2 and T3 are short insertions and deletions (Figures 4 and S5), with 1-bp insertion as the most frequent mutation type. In addition, large insertions and deletions (more than 10 bps) were found. Interestingly, base substitutions (C?G, C?A, G?T, G?C, G?A and A?G) followed by insertions and deletions were also found (Figures 4 and S5). Furthermore, we observed that the plants of line #19, #82, #94 and #96 contained large deletions (>50 bps) due to combinational effects of the two target sites (Figures 4 and S5). Together, these results demonstrate that CRISPR/Cas9 can be used to efficiently create different types of mutations in L. aequinoctialis.

Discussion
Here, we report an efficient genetic transformation system and for the first time demonstrated that CRISPR/Cas9 can be an effective tool for genome editing a duckweed species. These tools will significantly accelerate duckweed research, both for understanding basic plant biology and for industrial applications. An efficient genetic transformation protocol is essential for high-throughput genetic manipulations in plants. Although stable genetic transformation methods have been used in different duckweed species, the efficiency varied among species (Table S2). Here, by optimizing the transformation protocol, we achieved a 94% transformation success rate, which is higher than previous reports on duckweed transformations. For example, in L. minor, a different species of the same genus, the efficiency obtained was approximately 10% (Chhabra et al., 2011), 40% (Yang et al., 2018a) and 59% (Cant o-Pastor et al., 2015). Furthermore, the whole transformation only took 5-6 weeks, which is similar to the fastest protocol that was previously reported in L. minor (Cant o-Pastor et al., 2015), and the genetic transformation was stable in the regenerated plants for at least 70 days (~30 generations).
In comparison to previous Agrobacterium-mediated transformation protocols in L. minor (Cant o-Pastor et al., 2015;Chhabra et al., 2011;Firsov et al., 2015Firsov et al., , 2018Ko et al., 2011;Sun et al., 2007;Yamamoto et al., 2001;Yang et al., 2013Yang et al., , 2017Yang et al., , 2018a, the increased efficiency of our protocol is likely resulted from adopting sonication coupled with vacuum infiltration infection method. In several previous studies, such as banana (Subramanyam et al., 2011), chickpea (Indurker et al., 2010 and cowpea (Bakshi et al., 2011), it has been shown that the combination of vacuum infiltration and sonication can greatly improve the efficiency of Agrobacterium infection. Although it remains to be tested, we believe that this infection method can also be used to improve the transformation efficiency in other duckweed species.
A major limitation of our transformation protocol is that the success rate might be highly dependent on the choice of the genotypes, which can vary in their callus induction rate (Table S1). A similar effect was also found in L. minor (Yang et al., 2018a), indicates that this might be a general pattern in species of the genus Lemna. To resolve this issue, two possible approaches can be applied. First, in addition to the conventional callus-based transformation, previous studies showed that a direct transformation of fronds (Ko et al., 2011;Yang et al., 2018a) or clusters (the specific structure for Wolffia) (Heenatigala et al., 2018;Khvatkov et al., , 2018 can work in duckweeds. Therefore, for the genotypes in which callus induction rate is low, direct frond transformation, which worked for different tested genotypes of L. minor (Yang et al., 2018a), can be used. The disadvantage of direct frond transformation is that the whole process requires~12 weeks, which is more than two times longer than the callus-based transformation protocol. Second, an optimized callus induction protocol can be used. Recently, a study showed that optimizing the callus induction medium can significantly increase the rate of callus induction in S. polyrhiza (Wang, 2016). Our preliminary results also showed that using the optimized medium reported from Wang (2016) can result in 100% callus induction among all 16 tested S. polyrhiza genotypes. It seems plausible that the optimization of the callus induction medium can be used to increase the callus induction efficiency also in other Lemna species.
The efficient and stable genetic transformation system in L. aequinoctialis enabled us to establish the CRISPR/Cas9-based genome editing tool in this species. As a proof-of-concept, we targeted three sites of the LaPDS gene in L. aequinoctialis. Among the 105 regenerated plants from the transformed calli, 80 of them contained mutated alleles, of which 15 were biallelic mutants (have no WT alleles), 65 were chimeric mutants (at least one WT alleles and two additional mutated alleles). This represents an average of 14.3% biallelic mutant generation rate. Because under normal conditions, L. aequinoctialis plants reproduce through asexual budding, the biallelic mutants will be completely maintained in the plant genome over the generations, thus the plants and their offspring can be directly used for analysing gene functions or further bioengineering.
The relatively high frequency of chimeric mutants could arise due to two non-exclusive reasons. First, it could be due to different independent CRISPR/Cas9-mediated mutagenesis among different cells. Since CRISPR/Cas9 is integrated into the plant genome, the mutagenesis machinery should be continuously active at different stages of plant regeneration and development. When the mutagenesis took place in the first embryogenic cell, all cells in the regenerated plant should contain the same allele at the targeted locus, resulting in heterozygous or biallelic mutants. However, when the mutagenesis took places in cells after the first embryogenic cell division, the regenerated plants may contain cells with independently mutated alleles (chimeric). Second, the observed chimeric mutants may also arise  if some of the regenerated L. aequinoctialis plants are polyploid. In this scenario, even the CRISPR/Cas9-mediated mutagenesis took place in the first embryogenic cell, it may still result in more than two alleles at the targeted locus. Future studies that directly measure the ploidy level (e.g. using flow cytometry) of the regenerated plants are required to distinguish these two possible reasons.
Phytoene desaturase (PDS) encodes one of the key enzymes in the carotenoid biosynthesis pathway (Qin et al., 2007). We found that L. aequinoctialis plants in which PDS were completely disrupted showed albino phenotype (Figures 3) which is consistent with a previous report in Arabidopsis (Qin et al., 2007). Loss-of-function of PDS in L. aequinoctialis likely have disrupted the biosynthesis of carotenoid, which indirectly destroyed chlorophyll, as well as many other genes involved in carotenoid, chlorophyll and GA biosynthesis pathways (Qin et al., 2007). As a consequence, these biallelic mutants showed albino phenotype.
Analysing the CRIPSR/Cas9-mediated mutagenesis showed that the gene editing efficiency varied substantially among target sites. For example, no mutations were found at the site T1. Because different promoters were used to drive different sgRNAs (Ma et al., 2015a), the capacity in driving the expression of sgRNA might have resulted in the observed differences. In addition, the sgRNA efficiency of T1 site was lower in comparison to the other two sites (Table 2), which might also have contributed to the low mutation efficiency at this site. These factors should be considered for future applications of the CRIPSR/Cas9 system in duckweeds.
A major concern of CRIPSR/Cas9-mediated genome editing is the off-target effects. Although several online tools have been developed to reduce the off-targets effects (Bae et al., 2014;Liu et al., 2017), it requires complete genomic information, which is only available for S. polyrhiza (Hoang et al., 2018;Michael et al., 2017;Wang et al., 2014). Except sequencing the complete genome of the mutant plants, which is still relatively expensive due to the large genome size (>600 Mb), it remains unclear how many off-targets were generated by the CRISPR/Cas9 in our mutant plants. To further evaluate the off-target effects and optimize the CRISPR/Cas9 system in duckweeds, S. polyrhiza, which has a relatively small genome size (158 Mb), high-quality reference genome and low mutation rates (Michael et al., 2017;Xu et al., 2019), would be an ideal system.
Together, we present the first CRISPR/Cas9 study in a duckweed species, which is one of the important future crops. We demonstrated that CRISPR/Cas9 coupled with an optimized transformation protocol can efficiently generate biallelic mutant lines in duckweed within 5-6 weeks, which is significantly shorter than the currently available protocols in model plants such as Arabidopsis and rice (Ma et al., 2015a;Shan et al., 2014;Zhang et al., 2014). Together with the rapidly accumulating genomic information in duckweeds, we anticipate that the established gene manipulation tools will boost duckweed researches and pave the way to fully utilize the remarkable duckweed plants for both industrial applications and researches on the mechanisms of plant-environment interactions.

Plant material and callus induction
A duckweed stock collection stock was established containing more than 100 accessions. The duckweed species was identified by atpF-atpH noncoding spacer (Wang et al., 2010) and registered in the Duckweed Stock Cooperative (http://www.ruduc kweed.org/database.html). Plants were cultivated for 2-3 weeks in Schenk and Hildebrandt medium (SH) with 1% (w/v) sucrose at pH 5.6 (Schenk and Hildebrandt, 1972 -D), 0.45 lM thidiazuron (TDZ) and 7.9 g/L bacteriological agar at pH 5.9. The cultures were kept in the dark at 24°C. Embryogenic calli were obtained after 3-4 weeks of culture. The developed calli were propagated on the fresh induction medium for another 4-5 weeks before they were used in transformation assays.

Agrobacterium preparation and genetic transformation
We compared the infection efficiency of Agrobacterium tumefaciens using EHA105, AGL1 and GV1301. Single colonies of A. tumefaciens were transferred to liquid LB medium containing 50 mg/L kanamycin and 25 mg/L rifampicin. The cultures were grown at 28°C with shaking (200 rpm) until the OD 600 reached 0.8-1.0. Freshly prepared 100 lM acetosyringone (3 0 ,5 0dimethoxy-4 0 -hydroxyacto phenone, ACROS Organics, Morris Plains, NJ, USA) was added to the cultures, and the mixture was continuously shaken for another 2 h. Cells were then pelleted by centrifugation at 24009g for 15 min and resuspended in the induction medium (same as mentioned above). The density (OD 600 ) of the resuspended Agrobacterium was adjusted to approximately 0.4. The GUS expression vector pANIC6B (Mann et al., 2012) was introduced into the Agrobacterium strain EHA105 and used for genetic transformation. Embryogenic calli of L. aequinoctialis were immersed in the Agrobacterium suspension in 350-mL tissue culture bottles. The bottles were placed in a vacuum chamber and vacuum was applied for 10 min. Then, the bottles were placed in a bath-type sonicator (Branson ultrasonic cleaner CPX2800, Branson Ultrasonics Corp., Danbury, CT, USA) and then subjected to ultrasound at a frequency of 40 kHz for 5 min at 17°C. After sonication, the bottles were vacuum infiltrated again for 10 min. After releasing the vacuum, the callus pieces and Agrobacteria were incubated for 30 min with gentle shaking. Excess bacteria were removed after the incubation by transferring the infected callus pieces onto filter papers wetted with induction medium and placed in empty petri dishes in the dark at 25°C for cocultivation.

Selection and regeneration of transgenic plants
Three days after co-cultivation, the infected calli were transferred onto regeneration medium: Gamborg's B5 basal medium supplemented with 4.65 lM kinetin, 2.57 lM indole-3-acetic acid (IAA), 1% (w/v) sucrose, 9.48 lM hygromycin (PhytoTechnology Laboratories, Shawnee Mission, KS, USA), 600 lM timentin and solidified with 7.9 g/L agar. After 1 week on the agar in dark, the calli were cultivated under a 16 h photoperiod of approximately 50 lmol/m 2 /s of white light for 4 weeks. The regenerated fronds were transferred onto a conservation medium: SH medium supplemented with 0.6% (w/v) sucrose, 300 mM timentin and solidified with 7.9 g/L agar. Regenerated fronds were proliferated on liquid SH medium.

Cloning of the LaPDS gene
The DNA fragment of LaPDS was amplified with gene-specific primers (Table S4), which were designed based on the gene sequence obtained from the L. aequinoctialis transcriptome sequences (Yu et al., 2017). PCR reactions were carried out with Pfu DNA polymerase (Takara, China) in a total volume of 50 lL contained 19 buffer, 0.2 mM (each) dNTPs, 0.4 lM primers, 5 U Pfu DNA polymerase and 50 ng of template DNA. PCR cycling parameters were: denaturation at 95°C for 3 min; 35 cycles of 95°C for 45 s, 56°C for 30 s and 72°C for 120 s, followed by a final extension of 72°C for 10 min. PCR products were cloned into pMD18-T Vector (Takara, China) and sequenced.

CRISPR/Cas9 target site selection
Target sequences were designed within the LaPDS gene using the online tools CRISPR-P 2.0 (http://cbi.hzau.edu.cn/crispr/) . To further selected target sequence, secondary structure analysis of target-sgRNA sequences was carried out with the program RNA Folding form (http://mfold.rna.albany. edu/?q=mfold/ RNA-Folding-Form2.3) (Zuker, 2003). Target sequences were further selected to avoid those paired to the sgRNA with more than six continuous bp. On the other hand, target site near the 5 0 end of LaPDS gene was the preferred target to increase the gene targeting effect. The efficiency score of sgRNA was predicted by CRISPR efficiency predictor (Housden et al., 2015).

Assemble Cas9/sgRNA construct and plant transformation
A CRIPSR/Cas9 construct carrying three sgRNA cassettes was generated using the binary pYLCRIPSR/Cas9 multiplex genome targeting vector system provided by Yao-Guang Liu of South China Agricultural University (Ma et al., 2015a). Three plasmids with sgRNA cassettes driven by OsU3, OsU6a and OsU6b, respectively, were assembled according to the golden gate cloning protocol.
The CRISPR/Cas9 constructs were introduced into A. tumefaciens strain EHA105 by electroporation as previous described (Ma et al., 2015a). Transformation of duckweed and stable genetic transformation validation was performed as described above.

Validation of stable transformations
We validated stable genetic transformations by analysing the integration of T-DNA into the L. aequinoctialis genome. To identify the insertion sites of the T-DNA, thermal asymmetric interlaced PCR (TAIL-PCR) was performed as previously described (Liu and Chen, 2007;Wang et al., 2011). To further examine the presence/absence of living Agrobacteria in the regenerated tissue, the A. tumefaciens strain EHA105 curdlan gene was amplified by PCR with the gene-specific primers for 35 cycles, EHA105-crds-F and EHA105-crds-R (Table S4). The PCR conditions were same as described above.
Genomic DNA extraction and detection of CRISPR/Cas9mediated mutations The genomic DNA was extracted from both transgenic and wildtype plants following the CTAB method (Porebski et al., 1997).  For each plant,~100 mg fresh frond was ground in liquid nitrogen and 400 lL of pre-heated CTAB buffer was added. After incubating at 65°C for 30 min, 200 lL of chloroform was added and the resulting mixture was kept at room temperature for 10 min. After centrifugation at 16 000 g for 5 min, the supernatant was transferred to a new tube, mixed with 300 lL of isopropanol and incubated at 4°C for 30 min. Then, genomic DNA was precipitated by centrifuge at 16 000 g for 10 min, the supernatant was removed and the DNA pellet was washed with 0.5 mL of 70% ethanol. The genomic DNA pellet was dissolved in 100 lL of H 2 O, and the concentration was determined using a spectrophotometer (NanoDrop 2000, Thermo Fisher Scientific, Inc., Pittsburgh, PA, USA). The extracted genomic DNA was then used as a template to amplify the endogenous PDS fragment by PCR. PCR was performed using the specific primers, LaPDS-F1 and LaPDS-R1 (Table S4), which covered the region of target site 1, 2 and 3. The PCR products were sequenced directly using the specific primers with Sanger-sequencing approach. Biallelic and chimeric mutations that produced superimposed sequence chromatograms from direct sequencing were decoded using the Degenerate Sequence Decoding method (Ma et al., 2015b). Biallelic mutants showed two distinct allelic mutations, but no wild-type allele, while chimeric mutations had more than two distinct allelic mutations and an additional wild-type allele. PCR products from all biallelic mutants and some chimeric mutants were cloned into the pMD18-T Simple vector (Takara, China), and ten to twenty clones for each sample were sequenced using the same method mentioned above. DNAMAN (version 9; Lynnon Biosoft, Inc., San Ramon, CA) was used for sequence alignment analysis. Based on the deletion information obtained from the two biallelic mutant #34 and #78, wt LaPDS allele-specific primers were designed(34-F and 34-R; 78-F, 78-R) (Table S4) to examine the presence of wt cells in the transgene lines.

Chlorophyll and total carotenoids quantification
Chlorophyll a, chlorophyll b and total carotenoids were extracted in ethanol and quantified with spectrophotometry (GeneQuant 1300, GE Healthcare, San Diego, CA, USA) using three biological replicates (Lichtenthaler and Wellburn, 1983).

Statistical analysis
Data were presented as the mean AE standard deviation of the mean of triplicate samples. Significant differences between means of WT and transgenic lines were determined using oneway analysis of variance followed by Duncan's multiple-range tests, using the SPSS statistical package (version 16.0; SPSS Inc., Chicago, IL, USA) at a significance level of P < 0.05.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Figure S1 Genomic sequence of the LaPDS. Figure S2 Comparison of transformation efficiency among three different Agrobacterium strains based on GUS staining. Figure S3 Analysis of T-DNA insertions. Figure S4 Examine living presence/absence of Agrobacterium in the regenerated tissue. Figure S5 Different types of mutations detected in the transgenic duckweed after CRISPR/Cas9-mediated gene editing. The plant numbers are shown in black. Figure S6 Examine presence/absence of WT allele containing cells in two biallelic transformants.