Ecient Editing of CSLD2 Orthologue by CRISPR/Cas9 Affects Cell Morphogenesis of Root Hair in Spinach

CRISPR/Cas9 is a valuable tool and has been extensively employed to perform gene editing in plants. However, CRISPR/Cas9 has not been successfully used in spinach, an important leafy vegetable crop. Here, we precisely edited Spo23361 and Spo10340, two cellulose synthase-like D (CSLD) genes involved in root hair formation of spinach hairy roots, using CRISPR/Cas9 system. Four mutation types (i.e., replacement, insertion, deletion, and combined mutations) were observed, among which the deletion accounted for the vast majority (about 64.1%). Mutation rate differed largely among different targets. Seven homozygous/bi-allelic and eight heterozygous/chimeric mutated lines of Spo23361 were obtained from 15 independent transgenic hairy root lines. All of the seven homozygous/bi-allelic lines displayed bulking and short root hairs, which exhibited the characteristics of Arabidopsis csld2 mutants. Thirteen heterozygous/chimeric mutated lines, but no homozygous/bi-allelic lines, of Spo10340 were obtained from 15 independent transgenic hairy root lines, all of which showed similar phenotype of root hair with normal hairy roots. The transcriptomic analysis further revealed that multiple gene expressions for cell wall modulation and membrane tracking were disturbed, which might result in the inhibition of root hair growth in Spo23361 mutants. Our results indicate that Agrobacterium rhizogenes-mediated transformation using CRISPR/Cas9 is a simple and ecient genome editing tool in spinach. It lays a solid foundation for large-scale genome editing in spinach in future. genes edited by CRISPR/Cas9 system in hairy roots of spinach. root hairs were observed in homozygous/biallelic mutated lines, and the transcriptome analysis provide the signal and metabolic mechanisms in the CRISPR/Cas9- edited hairy roots of spinach. To our knowledge, this is the rst report for the CRISPR/Cas9 system applied in spinach.


Introduction
Spinach is an important nutritious green leafy vegetable, which is rich in carotenoids, folate, vitamin C, as well as calcium and other irons (Xu et al. 2017). We have published the draft genome sequence of a Chinese inbred spinach cultivar Sp75 (Spinacia oleracea, 2n=12) (Xu et al. 2017) and provided a primary A. rhizogenes-mediated transformation system of S. oleracea ). We also have cultured new spinach varieties with higher stress (e.g., heat, salinity, and diseases) tolerance using traditional and molecular breeding strategies (Ge et al. 2017). Besides, the speci c heat-responsive signaling and metabolic mechanisms in heat-tolerant spinach variety Sp75 and heat-sensitive Sp73 were reported using proteomics and phosphoproteomics methods (Zhao et al. 2018; Li et al. 2019). Although numbers of genes/proteins have been proposed to involved in transcription, calcium signaling, ROS homeostasis, endomembrane tra cking, and cross-membrane transport, their functions were not proved by molecular genetics, due to lack of a highly e cient gene editing system (Zhao et al. 2018; Li et al. 2019). Therefore, establishment of a gene editing system is critical for the investigation of molecular mechanisms in spinach development and stress response.
The clustered regularly interspaced short palindromic repeat (CRISPR)-associated protein (Cas9) system was rstly discovered as a component of the immune system from the bacterium Streptococcus thermophilus (Barrangou and Marra ni 2014). The DNA fragments of invading virus were caught by the host cell and integrated into the CRISPR loci. Then, the host cell used the transcribed RNA to guide the nuclease digesting the target site of virus transcript when the pathogenic organism infected again (Barrangou and Marra ni 2014). The simpli ed type II CRISPR/Cas9 system has been widely used in higher organisms. In this system, sgRNA could form a complex with Cas9 nuclease and guide the complex to target DNA sequence which can pair with the former 20 bases of sgRNA and was followed by an "NGG" sequence for the binding and cleavage of Cas9 nuclease (Jinek et al. 2012; Ma and Liu 2016;Dong et al. 2017; Wang et al. 2018). Homologous recombination (HR) and non-homologous endjoining (NHEJ) were promoted in the higher organisms by the site-speci c double-strand breaks generated Cas9 nuclease which results in targeted mutagenesis.
In recent years, the type II CRISPR-Cas9 system through Agrobacterium tumefaciens transformation has become a powerful tool for investigating plant gene function and crop molecular breeding, but lacking of a mature transformation platform for some plants limited its applications (Gao 2021; Zhu et al. 2020). Fortunately, Agrobacterium rhizogenes-mediated transformation system is relatively easier to succeed in plants, because it induces the production of hairy roots rather than aims at germline transmission. It would be a satisfactory system after the regeneration from hairy roots was gured out in the coming future (Georgiev et al. 2012).
So far, CRISPR-Cas9 system has been successfully applied in hairy roots from many plant species for various purposes. For example, knocking out the committed diterpene synthase gene SmCPS1 by CRISPR/Cas9 in hairy roots from a traditional Chinese medicinal herb Danshen (Salvia miltiorrhiza) can completely inhibit the synthesis of various active ingredient tanshinones, such as cryptotanshinone, tanshinone IIA and tanshinone I (Bai et al. 2018). Besides, the CRISPR/Cas9 system-edited rosmarinic acid synthase gene SmRAS, functioned in the water-soluble phenolic acid biosynthetic pathway, obviously reduced the contents of various phenolic acids (i.e., rosmarinic acid and lithospermic acid B) in the hairy roots from S. miltiorrhiza (Zhou et al. 2018). In addition, CRISPR-Cas9 mediated-knockout of St16DOX gene caused the complete abolition of the accumulation of steroidal glycoalkaloids (SGAs) (i.e., alpha-solanine and alpha-chaconine) that showed the toxicity against various organism in potato hairy roots (Nakayasu et al. 2018). Besides of elucidating distinct reaction within a biosynthetic pathway, hairy root was considered as a transformative tool for interrogating gene function of root through CRISPR-mediated gene mutation, because type-speci c expression of gene was similar in transformed hairy root by A. rhizogenes as compared with primary roots of A. tumefaciens transformants (Ron et al. 2014). In Brassica carinata, deletions in FASCICLIN-LIKE ARABINOGALACTAN PROTEIN 1 (BcFLA1) gene in the hairy roots resulted in shorter root hairs under inorganic phosphate de cient conditions, which proved a model role of hairy root system in rapidly assess gene function again (Kirchner et al. 2017).
To establish CRISPR/Cas9 system in spinach, we chose the established A. rhizogenes-mediated hairy root platform as a model. Interestingly, the hairy roots of spinach have abundant root hairs, which implied that the phenotype of root hairs could be taken as a better indicator for optimizing the CRISPR/Cas9-based gene editing system. Root hairs have important role in uptake of water and nutrients, and also decide the surface area between plants and soil microbes (Libault et al. 2010). Root hairs are tip-growing and tubular outgrowths of specialized root epidermal cells, which needs rapid cell wall assembly in the tips. The de cient and/or disrupted cell wall components lead to root hairs deformed and even ruptured . A number of genes involved in the cell wall modulation during hair tip growth. Cellulose synthase-like D (CSLD) gene belongs to a subfamily of cellulose synthase-like (CSL) gene family, which functions in the cellulose synthesis in cell wall of root hairs (Bernal et al. 2008). It has been reported that, among six AtCSLDs, AtCSLD2 and AtCSLD3 were required for normal root hair growth in Arabidopsis (Bernal et al. 2008;Yin et al. 2011;Qi et al. 2013). AtCSLD3 was involved in the biosynthesis of β-glucan-containing polysaccharides in cell wall for root hair elongation (Favery et al. 2001), while more AtCSLD2 was required during late stage of hair development than AtCSLD3 (Bernal et al. 2008). The csld2 mutant exhibited abnormal root hair and many root hairs bulging and rupturing late during root development process (Bernal et al. 2008).
In the present study, SpCSLD2 (Spo10340) and SpCSLD3 (Spo23361) were chosen as the target genes edited by CRISPR/Cas9 system in hairy roots of spinach. The abnormal root hairs were observed in homozygous/biallelic mutated lines, and the transcriptome analysis provide the signal and metabolic mechanisms in the CRISPR/Cas9-edited hairy roots of spinach. To our knowledge, this is the rst report for the CRISPR/Cas9 system applied in spinach.

Materials And Methods
Plant materials, growth and culture conditions Chinese inbred spinach cultivar Sp75 was used in this study (Xu et al. 2017). The seeds were surface-satirized with 75% ethyl alcohol and 5% sodium hypochlorite ). All the sterile seeds were germinated on solid 1/2 MS basal medium containing 30g/L sucrose and 10 g/L agar, and cultured under 26 ℃ and 16 h light/ 8 h dark. The explants from leaves were used to induce hairy roots. The different tissues and organs from two-month-old seedlings were used for qRT-PCR analyses. Two-month-old hairy roots from normal wild type (WT) and Spo23361 mutant (line 3) were collected for transcriptomic analysis.

Bioinformation analysis of CSLD protein sequence
The ve SoCSLDs gene sequences were achieved by BLAST using the AtCSLD3 gene as query sequence in the website (http://www.spinachbase.org/). Multiple alignment analysis of the full-length protein sequences was performed by Clustal X program. The phylogenetic tree was constructed with MEGA 3.1 program using the Neighbor-joining (NJ) method with 1,000 bootstrap replicates. The pI and molecular weight (MW) of Sp23361 and Sp10340 were predicted in the website (https://web.expasy.org/compute_pi/). The conserved domains of Spo23361 and Spo10340 were online analyzed by the Conserved Domain Database (CDD) of NCBI (https://www.ncbi.nlm.nih.gov/cdd).

Quantitative real-time PCR
Total RNAs were extracted from stem, leaf, ower, stipe, root and hairy root using Trizol reagent (Invitrogen Corporation). RT-PCR was carried out using 500 ng of total RNA with PrimeScript™ RT reagent Kit with gDNA Eraser (TaKaRa) following the manufacturer's protocol. Gene expression was assayed using an Rroche LightCycler480 real-time PCR System. The expression of actin gene (Spo23599) was used as an internal control. The primers used to amplify transcripts were listed in Table S8.

Target sites design and vector construction
The speci c sequences of Spo23361 and Spo10340 were ampli ed by two pairs of primers in Table S8, respectively. Three target sites were selected for each gene according to Ma et al. (2016) (Fig. S1A,1B). Brie y, the 20 bp sequence before the NGG region should have appropriate GC content (≥30%), avoid four consecutive T nucleotides, and exclude the target candidates that form the hairpins of longer than 6 bp with sgRNA sequence.
The tRNA-gRNA cassette strategies were used (Xie et al. 2015). Cas9 was driven by the CaMV 35S promoter. Target sequences separated by tRNA and followed by one gRNA were driven by A. thaliana U3b promoter. Bar gene driven by CaMV 35S promoter was used as a selective marker (Fig.  S1c).We constructed a modi ed pYLCRISPR/Cas9 35s-B (GenBank accession number: AI133729.1) vector. The segment of two Bsa I cutting sites and CCDB in the vector was replaced with the sequence containing a AtU3b snRNA promoter, two new Bsa I cutting sites, and a sgRNA sequence. AtU3b snRNA promoter was ampli ed from pYLsgRNA-AtU3b (GenBank accession number: KR029097.1) (Fig. S1d). The tRNA-gRNA cassette was linked to the constructed vector according the methods of Ma et al. (2016) and Xie et al. (2015). The fragments containing tRNA and/or sgRNA sequence were ampli ed from pGTR vector (Beijing Genomics Institute) using the primers listed in Table S8. The mixed fragments and the modi ed pYLCRISPR/Cas9 35s-B vector, were digested by Bsa I and linked by T4 ligase in one step (Ma et al. 2016;Xie et al. 2015). The CRISPR/Cas9 vector for Spo23361 and Spo10340 containing three target sequences, and tRNA-gRNA expression cassette was transferred into A. rhizogenes strain LBA9402 for subsequent experiment.
Induction and identi cation of transgenic hairy roots of spinach A. rhizogenes strain LBA9402 carrying the CRISPR/Cas9 vector for Spo23361 and Spo10340 was used for the gene transfer experiment. The Agrobacterium strain was cultured by shaking in liquid LB containing 50 mg/L kanamycin and 20 mg/L rifampicin at 28 ℃ and under 200 rpm. The OD 600 of 0.6-0.8 was the proper concentration for inoculation. The bacterial liquids were collected and re-suspended to OD 600 of 0.2 using liquid MS medium containing 100 µM acetosyringone. The leaf discs (~ 0.5 cm × 0.5 cm) taken from the 30-day-old aseptic seedlings were submerged in A. rhizogenes suspension for 10 min, and then co-cultured for two days in darkness. The discs were sub-cultured in a selection medium (SH base medium containing 1 mg/L phosphinothricin and 300 mg/L timentin). Hairy roots (2 cm in length) were cut from the discs and sub-cultured in the selection medium for further identi cation.
The genomic DNA of the putatively transformed plants was extracted using CTAB method. In order to identify the co-transformed hairy root lines, PCR analyses were performed to detect the presence of bar, Cas9 and rol B genes using primers in Table S8.

Mutation type analysis
Fragments containing the target sites were ampli ed using the primer in Table S8. The PCR products were sequenced to detect mutagenesis. The PCR products with mixed peaks and/or sequence changes were puri ed with DNA Puri cation Kit and ligated in the pMD19-T easy vector. The ligated products were transformed into Escherichia coli strain Top10, and ve positive clones of each transgenic line were sequenced to analyze their speci c mutation types.

Phenotype observation
The phenotype of root hairs was observed under a microscope OLYMPUS SZX16. The length and density of root hairs were measured from three lines of mutants and normal hairy roots, respectively (n = 8).

Off-target analysis
The potential off-target sites were predicted according the method of (Li et al. 2017). Spinach genome was downloaded from website (http://www.spinachbase.org/). The putative off-target locus were blasted among the whole genome of spinach. The sites containing no more than 3-bp mismatches in the 20-bp target sequence were taken as potential off-target sites (Table S2). We ampli ed the sequence containing the potential off-target sites by PCR using gene-speci c primers (Table S8). The PCR products were sequenced to verify the occurrence of mutation.

Transcriptome sequencing of root hairs
Total RNA was extracted from hairy root of normal wild type (control) and Spo23361 mutants using the mirVana miRNA Isolation Kit (Ambion). The samples with RNA Integrity Number (RIN) ≥ 7 were subjected to the subsequent analysis. The cDNA library was constructed using TruSeq Stranded mRNA LTSample Prep Kit (Illumina, San Diego, CA, USA), and the libraries were sequenced on the Illumina sequencing platform (HiSeqTM 2500 or Illumina HiSeq X Ten) according to the method of Xu et al. (2017). 125bp/150bp paired-end reads were generated. The transcriptome de novo assembly was carried out using Trinity (http://trinityrnaseq.sourceforge.net). The clean reads were mapped to spinach reference genome using hisat2 .
Fragments Per Kilobase of transcript per Million (FPKM) of each gene and read counts value of each transcript (protein_coding) were calculated using bowtie2 and eXpress, respectively. The analysis of differential expression gene (DEG) among samples was performed using the DESeq (2012) R package. The p value < 0.05 and Fold Change >2 (or < 0.5) were applied to evaluate the difference signi cance. Hierarchical cluster of DEGs was performed to explore transcripts expression pattern. GO and KEGG pathway enrichment of DEGs were performed using R based on the hypergeometric distribution.
Heatmap was plotted using an OmicShare tool (http://www.omicshare.com/tools/). The KEGG pathway, as well as GO biological processes and cellular component enrichment analyses were conducted using the Metascape analysis (http://metascape.org/). Terms with a p-value < 0.01, a minimum count of 3, and an enrichment factor (the ratio between the observed counts and the counts expected by chance) >1.5 were collected and grouped into clusters on the bases of their membership similarities.
Protein-protein interaction between DEG encoding proteins were predicted by the online tool STRING 10 (http://string-db.org/) with the minimum required interaction score set as > 0.4. All other parameters were set as defaults, and all active prediction methods were used. The network was visualized using Cytoscape 3.7.1 software (http://www.cytoscape.org/). All the identi cation (ID) input for the interaction network analyses were protein homologs in Arabidopsis thaliana, which were found by sequence BLASTing in the TAIR database. Proteins (nodes) were shown as bubbles whose color gradient from light blue to dark green according to the node degree distribution. The related interactions (edges) between proteins were shown as gray lines, whose width represented the strength of interaction score. The function categories were represented by different colors of bubbles in network and columns.

Results
Spo23361 and Spo10340 are strongly expressed in roots and hairy roots Five spinach genes, Spo23361, Spo10340, Spo16352, Spo12722 and Spo16366, were predicted from the spinach (S. oleracea) genome based on the similarity to the Arabidopsis AtCSLD3 gene (Favery et al. 2001). Phylogenetic analysis of the spinach and Arabidopsis CSLDs indicated that the ve spinach genes clustered with different Arabidopsis CSLD genes respectively. Among them, Spo23361 and Spo10340 were clustered with AtCSLD2 and AtCSLD3 together (Fig. 1a). The open reading frames of Spo23361 and Spo10340 have 3,444 and 3,459 nucleotides, respectively. They encode two putative proteins containing 1,447 and 1,152 amino acid resides with similar molecular mass (128,557 Da and 130,011 Da) and pI (6.9 and 7.06), respectively. Pairwise comparison analysis showed that Spo23361 and Spo10340 exhibit 80% and 81% identity to both AtCSLD2 and AtCSLD3 (Fig. 1b), respectively. Conserved domain analysis showed Spo23361 and Spo10340 proteins have the conserved DDDQ/RxxRW motif (Richmond and Somerville 2000), indicating they belong to the CSL family (Fig. 1b).
The expression of ve SoCLSD genes were examined in stems, roots, leaves, petioles, and owers from two-month-old seedlings of spinach, as well as in hairy roots induced from leaves by A. rhizogenes strain LBA9402 using qRT-PCR analysis. Spo23361 and Spo10340 expressed in all these organs with similar patterns, and both of them showed obviously higher expressions in roots and hairy roots, respectively (Fig. 1c). Especially, the expression of Spo23361 was signi cantly higher in both roots and root hairs than that in other organs/tissues (Fig. 1c). In addition, the others (Spo16352, Spo12722, and Spo16366) were also highly expressed in roots.
All these implies that Spo23361 and Spo10340 would be good candidates for investigating the function of AtCSLD2 and AtCSLD3 homologues in root hair development of spinach using CRISPR/Cas9 gene editing system.

CRISPR/Cas9 vector construction and co-transformation e ciency in hairy roots
Partial sequences of Spo10340 and Spo23361 were ampli ed from cDNA of roots. Three targets were designed for each gene. The tRNA-gRNA cassette strategies were used to construct the CRISPR/Cas9 vector (Fig. S1) (Xie et al. 2015).
T-DNA of Ti plasmid harboring CRISPR/Cas9, tRNA-gRNA and bar expression cassette were transferred into leaf explants of spinach using A. rhizogenes strain LBA9402. Hairy roots appeared after 10-20 days of infection, which looked very similar with normal roots of spinach (Fig. 2a, b). Numerous independent transgenic hairy roots were produced for each construct (22 root lines for Spo10340, and 16 for Spo23361) (Fig. 2c, d). The PCR detection implied that rolB, Cas9, and bar genes were successfully transformed into the spinach, and about 80% hairy root lines (30 from 38 lines) were co-transferred into Ri T-DNA and Ti T-DNA (Fig. 2c, d).
Genotyping and phenotyping of edited hairy root lines of Spo26631 Total 15 hairy root lines of Spo26631 edited by CRISPR/Cas9 system were obtained (Fig. 3 and Table S1), among of which seven lines (i.e., line 2, line 3, line 4, line 8, line 12, line 13, and line 14) showed obvious phenotypes of short root hair (Fig. 3a, b). Besides, the abnormal middle and/or top of root hairs from these lines looked like bulging balls (Fig. 3a, b). Sequence analysis showed that all of seven lines were bi-allelic/homozygous mutated lines with edited sequence in the same or different target sites in two sister chromatids ( Fig. 3e and Table S1). The root hair density of seven lines was not changed compared with that in normal hairy roots (Fig. 3f). However, the lengths of root hairs of seven lines were signi cantly reduced to the one twentieth of normal length (Fig. 3g). Another eight co-transferred lines were also sequenced. Among them, ve heterozygous mutated lines (i.e., line 5, line 6, line 7, line 9, and line 11) contained one mutation type in each line, two chimeric mutated lines (line 1 and line 10) had three or two mutation types in each line, and the line 8 had a wide-type genotype in three target sites (Table S1). All of the heterozygous or chimeric mutated lines showed no difference in the length and density of root hairs when compared with normal hairy roots ( Fig. 3c-g).

Genotyping and phenotyping of edited hairy root lines of Spo10340
Twenty-two hairy root lines were obtained for the gene Spo10340, among of which 15 lines were found co-transferred with rol B and Cas9 genes (Fig. 2d). No lines showed obviously stunted root hairs. The 15 lines were submitted to target sequence analysis. Among them, 13 lines (line 1 to line 13) were edited in more than one target sites, the remaining 2 lines had no change at the target sites. However, no bi-allelic and homozygous mutants were recovered. All of them were either chimeric or heterozygous lines (Fig. 4a-d and Table S1). The density and the length of root hairs were similar with those of normal hairy roots (Fig. 4e, f).

Mutation variety and frequency of CRISPR/Cas9-edited hairy roots
Fifteen co-transferred hairy root lines of Spo23361 and Spo10340 were performed PCR analyses, and their PCR amplicons with the target sites were sequenced. Among them, fourteen lines of Spo23361 and thirteen lines of Spo10340 showed mixed peaks, indicating that these lines were edited by CRISPR/Cas9 at the target sites ( Table 1). The PCR amplicons from the 27 hairy roots were further con rmed by TA cloning and sequencing (Table  S1). The editing e ciency varies widely among the three targets of each gene of Spo23361 and Spo10340 (Table 1). The third target site of Spo23361 and the rst target site of Spo10340 showed the highest mutation rate of 86.7%. Four mutation types were observed, including replacement, insertion, deletion and combined mutations (i.e., more than one mutation type in one target site). For Spo23361, the insertion, deletion, and combined mutations occurred at the third target site, the replacement and deletion mutations occurred at the rst target site, and only replacement mutation occurred at the second target site (Table 1). Interestingly, almost all the target sites except the rst and second target of Spo23361 showed that the editing mode of deletion accounted for the majority. For Spo10340, four mutation types appeared at the rst target site, while two mutation types (deletion and combined mutation) appeared at the second and third target site (Table 1). i: insertion, d: deletion, r: replacement, c: combined mutation (more than one mutation type in one allele).
To detect the off-target events in CRISPR/Cas9 edited hairy roots of spinach, potential off-target loci following protospacer adjacent motif (PAM) sequences were predicted using the program written by Li et al (2017). They were highly homologous and not more than three mismatches of each target sequence in the hairy roots of Spo23361 and Spo10340 (Table S2). Only one off-target locus was predicted in the rst and second target sequence of Spo23361, respectively. However, ve off-target locus were found in the third target sequence of Spo23361 (Table S2). For Spo10340, there were one and two off-target sequences in the rst and second target sequence, respectively. To examine whether these predictions happened in our results, 15 hairy root lines of Spo23361 and Spo10340 were submitted for sequencing, respectively, and no site mutation was detected (Table  S2).

Transcriptomic analysis of CRISPR/Cas9 edited hairy roots
To evaluate the differentially-expressed genes (DEGs), the transcriptomics analyses of Spo23361 mutants and normal hairy roots were performed by using RNA-seq based on Illumina platform. A total of 25,495 unique genes were identi ed and quanti ed, 305 of which were more than 2.0-fold change (p < 0.05) in abundance and considered as DEGs in hairy roots of Spo23361 mutant when compared with normal hairy roots. The hierarchical clustering indicated that 305 DEGs were clustered into two main groups (Figure 5a and Table S3). Metascape analysis showed that 13 GO biological processes of the corresponding genes were enriched (Figure 5b, c and Table S4). Cluster I showed a signi cantly-reduced pattern of 169 DEGs in Spo23361 mutants. The 57 reduced DEGs in Cluster I were enriched in 13 GO biological processes including biosynthesis of salicylic acid, phenylpropanoid and avonoid, response to salicylic acid, iron ion, bacterium, and hypoxia, metal ion and intercellular transports, defense response, protein folding, as well as cell death (Figure 5b, c and Table S4). Besides, 136 DEGs in Cluster II were induced in Spo23361 mutants. Among them, 32 genes were enriched in response to various stresses (e.g., salicylic acid, iron ion, hypoxia, and bacterium), ion transport, cell death, as well as biosynthesis of avonoid and phenylpropanoid (Figure 5b, c and Table S4). Interestingly, DEGs were highly enriched in several clusters, including the metal ion transport (blue node), salicylic acid biosynthesis (yellow node), hypoxia (green node), avonoid biosynthesis (purple node), and host interaction and intercellular transport (orange and dark green nodes) (Figure 5b and Table S4). In addition, these DEGs were enriched in several cellular components including cell wall, cell periphery, extracellular region, plasmodesma, plasma membrane, vacuole, and whole membrane ( Figure 5d and Table S5), which indicated the cell wall modulation and membrane tra cking were seriously disturbed in the Spo23361 mutants.
To predict the relationship of DEGs in mutant roots, the Arabidopsis homologs of spinach proteins encoding by DEGs were found by sequence BLASTing in TAIR database (Table S6), and then were subjected to the web-tool SRING 10 for predicting protein-protein interaction (PPI) networks. The network showed 116 nodes (represent proteins encoding by DEGs) with 172 node degrees (represent PPIs) in mutants (Figure 5e and Table S7). The node colors with gradient from light blue to dark green indicated the node degree distributions (Figure 5e and Table S7). In this network, the dark green nodes represented cinnamyl alcohol dehydrogenase 3, calmodulin-binding protein 60-C, receptor-like protein kinase, and 70 kDa heat shock protein have more than 10 node degrees for each, implying their complicated interaction with other proteins. Interestingly, the nodes representing proteins encoding by DEGs for cell wall metabolism, protein folding and processing, membrane and transport, as well as signaling have more node degree than other nodes, suggesting that these signal and metabolic processes were obviously affected in mutants (Figure 5f, g, and Table S7). The transcriptomic results indicate that the gene expression patterns were signi cantly disturbed in Spo23361 mutants and might affect the root hair phenotype.

Discussion
Spinach Spo23361 was predicted to be the homologous gene of AtCSLD2 In Arabidopsis, CSLD family has six members encoding glycosyltransferases that are responsible for the synthesis of the glycan backbones of cellulose and most polysaccharides (Richmond and Somerville 2000). In our results, the analyses of phylogenetic tree and homology alignment indicated that both Spo23361 and Spo10340 from ve SpCLSDs were the homologs of AtCSLD3 and AtCLSD2 (Fig. 1a). Besides, we found that Spo23361 and Spo10340 were highly expressed in roots and hairy roots of spinach (Fig. 1c), which were similar with the highest expression of AtCSLD3 and AtCLSD2 in Arabidopsis roots (Bernal et al. 2008). Interestingly, the two poplar genes, PdCSLD5 and PdCSLD6, also have higher similarity with AtCSLD2 and AtCSLD3, which were grouped into one branch in the phylogenetic tree (Qi et al. 2013). It has been found that overexpression of PdCSLD5 and PdCSLD6 can rescue the root hair defect phenotype of atcsld3 mutant, suggesting that PdCSLD5 and PdCSLD6 are the homologous genes of AtCSLD3 (Qi et al. 2013).
In our study, all the bi-allelic or homozygous mutants of Spo23361 gene generated by CRISPR/Cas9 system exhibited short and/or bulking root hairs. However, the root hair numbers were not changed compared with the wide type (Fig. 3). Importantly, the phenotype of these lines was stable during the four-month subculturing process. This indicates that Spo23361 has similar function of AtCSLD2 rather than AtCSLD3. However, in our study, all these CRISPR/Cas9-edited lines of Spo10340 were either chimeric or heterozygous, leading to no stable root hair phenotype observed (Fig. 4). Even so, we still tried our best to repeat this experiment for obtaining bi-allelic or homozygous mutated lines for Spo10340 genes. Unfortunately, there was still no lines showing the similar phenotype with atcsld3 mutant, although 17 co-transferred lines were induced (data not shown). Therefore, we suspect that the homozygous or bi-allelic mutation of Spo10340 resulted in thorough growth inhibition of hairy roots, which is similar with the phenotype of root hair loss in atcsld3 mutants (Wang et al. 2001). However, the homolog relationship of Spo10340 and AtCSLD3 still needs to be further investigated.
The tRNA-target-gRNA cassette expression system was an e cient tool for multiply targeted genome modi cation in spinach In this study, we tested the use of plant-optimized Streptococcus pyogenes CRISPR-associated protein 9 (SpCas9) and AtU3b promoter to selectively modify the spinach genome. Our results revealed that 27 from 30 hairy root lines were edited successfully, indicating that the SpCas9 and AtU3b promoter worked e ciently in spinach (Table 1). All the six target sites of Spo23361 and Spo10340 were edited to diverse mutation types, including replacement, insertion, combined mutation, and deletion, and the highest mutation frequency reached to 86.7% (Table 1). The high editing rate was resulted from the tRNA-gRNA cassette expression system and multiple target sequences for each gene, since tRNA could not only serve as enhancer for AtU3b but also as a "divider" for the sgRNAs.
The mutation frequency of different target was ranged from 6.7-86.7% in our results, which was probably affected by various factors. It was reported that GC percentage of overall targets (Gagnon et al. 2014) and the 6-PAM-proximal nucleotides (Ren et al. 2014) are positively associated with editing e ciency in zebra sh (Danio rerio) and Drosophila, respectively. In the present study, the GC contents of the six target sites were range from 40-60%, which was reported to be favored for e cient on-target cleavage . For Spo23361, the GC content of the 1st, 2nd, and 3rd target sequence was 45%, 60% and 40%, while the GC content of the 1st, 2nd, and 3rd target sequence of Spo10340was 45%, 45% and 50%, respectively. However, the highest mutation rate was occurred in the 3rd target site of Spo23361 and the 1st target sequence of Spo10340, which have the lowest GC content both in the target sequences (40% and 45%) and in the 6-PAM-proximal nucleotides (2/6 and 3/6). This is not consistent with the previous notion Transcriptome analysis implied the cell wall remodulation in spcsld mutant CSLD is essential for the synthesis of polymers, which is critical for the fast-growing primary cell wall at the tip of root hairs (Wang et al. 2001). The root hair tip-localized CSLD3 was predicted to regulate the root hair elongation and cell integrity. In atcsld3 mutant, the cell wall changed at the site of root hair tip, leading to root hairs elongation was severely inhibited (Wang et al. 2001). Our transcriptomics results revealed large number of genes were involved in cell wall remodulation and membrane tra cking in the Spo23361 mutant. In the mutant, the expression levels of some expansins and xyloglucan endotransglucosylase/hydrolases (XTH) were increased, which could enhance the dissociation of polysaccharide complex and adjustment of xyloglucans, respectively (Cosgrove 2005), therefore resulting in cell wall relaxation and extension. Moreover, the increase of plasma membrane H + -ATPase could facilitate the extracellular acidi cation activating expansin activity (Cosgrove 2005). Besides, the increased endo-β-1,3glucanase and β-1,3-1,4-glucanase may induce the hydrolysis of the glucan polymer cross-links as well (Thomas et al. 2000;Li et al. 2020). In addition, the decrease of arabinogalactan proteins has potential to reduce the cross-link of pectin and arabinoxylan impairing the wall thickening (Tan et al. 2013;Showalter 2001). All these changes implied that the cell wall loosening was enhanced, while its strength was reduced in mutant. On the other hand, the decease of galacturonosyltransferases (GATL) could affect the pectin synthesis, and the increase of pectin acetylesterase has capacity to induce the cleavage of the acetylester bond from pectin (Kong et

Conclusion And Prospects
The present study indicates that the CRISPR/Cas9 via tRNA-sgRNA strategy is a highly e cient tool for genome editing in spinach. Homozygous and bi-allelic mutations were induced at different loci by CRISPR/Cas9 in hairy roots of spinach, which provides a powerful tool and model to study the gene function in spinach roots. Importantly, the CRISPR/Cas9 system has potential to be utilized for investigating the development and stressresponsive mechanism in spinach and breeding novel spinach germplasm with useful agronomic traits.    of normal hairy roots, bi-allelic/homozygous mutated and heterozygous/chimeric mutated hairy root lines. Root hair length and density were measured at the region 1.5 to 2 mm from the root tip. Root hair density is shown as root hair number per mm 2 . 24 roots of three lines (8 roots for each line) were measured for each mutation type, values were mean ± SD. Asterisk (**) indicates that difference is signi cant at p < 0.01. the inserted bases were in green. (e) Root hair density and (f) root hair length analysis of normal hairy roots and heterozygous/chimeric mutated hairy root lines. Root hair length and density were measured at the region 1.5 to 2 mm from the root tip. Root hair density is shown as root hair number per mm 2 . 24 roots of three lines (8 roots for each line) were measured for each mutation type, values were mean ± SD.

Figure 5
Cluster analysis and enrichment of differential expression genes, and the predicted interaction networks of their encoding proteins in Spo23361. (a) Hierarchical clustering analysis of differential expression genes. Two columns represent two replications of fold change (Log2 transformed) of DEGs in Spo23361 mutant (R1, R2). The rows represent individual genes. The detailed information is listed in Table S3. The increased or decreased genes are indicated in red or blue, respectively. (b) Network of enriched terms of GO biological processes and KEGG pathways across the DEG lists of the two clusters. The network is visualized using Cytoscape software, where each node represents an enriched term. The term genes next to the nodes are shown in Table S4. The terms with a similarity > 0.3 are connected by edges. The same color nodes represent a subset of enriched terms, where the terms with the best p-values were highlighted with bold words, and is according to the GO biological process or KEGG pathway shown under the network. The color of the term genes (i.e., blue and pink) are corresponded to the Cluster I and Cluster II, respectively.