Targeted mutation of barley (1,3;1,4)- b -glucan synthases reveals complex relationships between the storage and cell wall polysaccharide content

Barley ( Hordeum vulgare L) grain is comparatively rich in (1,3;1,4)- b -glucan, a source of fermentable dietary ﬁbre that protects against various human health conditions. However, low grain (1,3;1,4)- b -glucan content is preferred for brewing and distilling. We took a reverse genetics approach, using CRISPR/Cas9 to generate mutations in members of the Cellulose synthase-like ( Csl ) gene superfamily that encode known ( HvCslF6 and HvCslH1 ) and putative ( HvCslF3 and HvCslF9 ) (1,3;1,4)- b -glucan synthases. Resultant mutations ranged from single amino acid (aa) substitutions to frameshift mutations causing premature stop codons, and led to speciﬁc differences in grain morphology, composition and (1,3;1,4)- b -glucan content. (1,3;1,4)- b -Glucan was absent in the grain of cslf6 knockout lines, whereas cslf9 knockout lines had similar (1,3;1,4)- b -glucan content to wild-type (WT). However, cslf9 mutants showed changes in the abundance of other cell-wall-re-lated monosaccharides compared with WT. Thousand grain weight (TGW), grain length, width and surface area were altered in cslf6 knockouts, and to a lesser extent TGW in cslf9 knockouts . cslf3 and cslh1 mutants had no effect on grain (1,3;1,4)- b -glucan content. Our data indicate that multiple members of the CslF / H family fulﬁl important functions during grain development but, with the exception of HvCslF6, do not impact the abundance of (1,3;1,4)- b -glucan in mature grain.


INTRODUCTION
Cell walls are a distinctive structural feature of plant cells. They are composed of polysaccharide-rich layers that determine cell size and shape, and impact plant growth and development. The natural diversity of inter-and intra-species cell wall composition has an impact on the processing properties of plant-derived materials, both for industrial applications and human nutrition. For example, non-cellulosic polysaccharides are a key source of soluble dietary fibre that influence the nutritional quality and digestibility of plantbased foods (Doblin et al., 2010;Burton and Fincher, 2012).
Although HvCslF6 transcripts are the most abundant of all putative (1,3;1,4)-b-glucan synthase genes in developing grain and many vegetative tissues, HvCslF9 is expressed during early grain development and in root tips (Burton et al., 2008). HvCslF3 is also highly expressed in root tips and coleoptiles. Due to the presence of (1,3;1,4)-b-glucan in the tissues where they are transcribed, both HvCslF3 and HvCslF9 are considered potentially capable of synthesising (1,3;1,4)-b-glucan (Burton et al., 2008;Aditya et al., 2015). In addition, the expression of HvCslH1 in leaf tissue supports previous findings that this gene contributes to synthesis of (1,3;1,4)-b-glucan in this tissue (Burton et al., 2008(Burton et al., , 2011Doblin et al., 2009). Despite distinct expression profiles, in planta functions of the HvCslF3, HvCslF9 and HvCslH1 genes have yet to be confirmed in barley by lossof-function mutants. This contrasts with the barley CslF6 gene, where several cslf6 alleles containing amino acid substitutions (bgla, bglb and bglc) have been described that exhibit reduced levels of (1,3;1,4)-b-glucan, susceptibility to chilling and alterations in grain morphology (Taketa et al., 2012). Loss-of-function alleles for members of the HvCslF/H families would provide an opportunity to assess their role in (1,3;1,4)-b-glucan accumulation as well as plant growth and development.
Here we utilised a reverse genetics approach to provide insight into the contribution of each of four members of the HvCslF/H subfamilies to barley grain morphology and composition. We used CRISPR/Cas9-based gene-editing technology to generate site-specific double-strand breaks leading to the introduction of targeted mutations via the non-homologous end-joining (NHEJ) repair mechanism. We analysed plant and grain morphology/development, grain monosaccharide composition, and the structure, distribution and content of (1,3;1,4)-b-glucan in the grain to provide insight into the function of each of the four genes. Our findings suggest that each of the genes contributes to the phenotype of barley grain, thereby providing opportunities to modify composition and morphology via HvCslF/H genes other than HvCslF6.

CRISPR/Cas9-induced mutations in members of the HvCslF/H gene family
We catalogued the mutations in each of the target genes over three generations (T 0 , T 1 and T 2 ). In the T 0 plants all CRISPR/Cas9-induced mutations were in the heterozygous state. The frequency of mutations ranged from 35.3% for HvCslF3, to 9.3% for HvCslF6, 22% for HvCslF9 and 4.7% for HvCslH1 (Table S1). In putative T 0 mutants, 83% of the plants carried short InDels (ranging from 1 bp deletions to 2 bp insertions; Table 1; Figure S1). An exception to this trend was identified for HvCslF9 where a 39-bp deletion was detected in the heterozygous state in a single T 0 plant.
Homozygous lines were detected in the T 1 (14 out of 58 for HvCslF3, 8 out of 169 for HvCslF9 and 2 out of 58 for HvCslH1). These lines contained InDels of varying sizes, some of which had the potential to change the protein sequence. Screening for the presence of Cas9 by polymerase chain reaction (PCR) indicated a high prevalence of transgene retention in HvCslF6 and HvCslH1 genotypes (92% and 97%, respectively) compared with HvCslF3 (55%) and HvCslF9 (56%) in T 1 plants (Table S2). A range of T 1 alleles were selected to maximise the possibilities of identifying potential phenotypic and allelic changes in the subsequent generation. A subset of 10 T 2 alleles (Table 1; Figure 1; cslf3, cslf6, cslf9 and cslh1) was used to characterise the effect of these mutations on grain morphology and composition.

Mutations in HvCslF/H genes influence grain morphology
Using grain from a subset of 10 T 2 lines of cslf3 (2), cslf6 (2), cslf9 (3) and cslh1 (3), we assessed the effect of each mutation on thousand grain weight (TGW), grain area, grain length and grain width, and compared these with wild-type (WT) grain of cv. Golden Promise. Cas9-free T 2 lines were identified for cslf3, cslf9 and cslh1; however, multiple Cas9 insertions were suspected in cslf6-2 lines based on a preliminary segregation test for the selectable marker ( Figure S2). Grain from the cslf6-2 mutants containing a premature stop codon (Table 1) in the first exon exhibited a significantly lower TGW (P = 0.002, Tukey's test) compared with WT ( Figure 2b). In addition, the grain was significantly longer (P = 0.006, Tukey's test) compared with WT. In heterozygous mutants (cslf6-2/+), TGW was affected to a lesser extent, due to higher phenotypic variation for this trait. Grains were significantly longer (P = 0.002, Tukey's test) than WT, whereas grain width was not significantly different, possibly leading to larger grains in terms of overall grain area. Significant differences in grain from cslf6 homozygous and heterozygous plants compared with WT indicate there may be some dosage dependency of HvCslF6 in sporophytic (maternal) tissues, similar to that reported previously for substitution alleles (Taketa et al., 2008).
Based on BLAST searches, the sgRNA sequences for HvCslF3 and HvCslH1 have potential off-targets that could result in mutations being introduced into additional genes (Table S3). To assess if these off-target mutations were present in the lines characterised here, we amplified the putative off-target amplicons in the same 10 lines (four cslf3 lines and six cslh1 lines) used for phenotypic characterisation of the grain (Table S4). For HvCslF3, the sgRNA could potentially bind to HORVU2Hr1G042350 (encoding a putative Cellulose synthase-like D2 based on gene annotation), and for HvCslH1 the potential off-target site was in HORVU2Hr1G074940 (encoding a putative Cellulose synthase-like B4); however, no mutations were observed in these genes. There were no putative off-targets for the The mutation location was described with respect to the cDNA start. PAM represents proto adjacent spacer motif, aa represents amino acid. Mutation locations are provided as nucleotides from the start codon followed by 'ins' for an insertion or 'del' for a deletion.
© 2020 The Authors. sgRNAs designed to HvCslF6 and HvCslF9 because any potential matches identified contained mismatches within the PAM ( Figure S6). Therefore, we conclude that for all the lines characterised that the phenotypes observed are due to mutations in the targeted genes.

Impact of the induced mutations on plant development
The same lines used for phenotypic characterisation of grain development were analysed for changes in whole plant growth under glasshouse conditions. Due to the small number of seed available for some lines it was not feasible to carry out a full replicated study of gross plant morphology during development. However, several phenotypes were striking, and therefore worth describing here.
We observed a delay in development and a reduction in plant height in both cslf6-2 (homozygous) and cslf6-2/+ (heterozygous) mutants compared with WT (cv. Golden Promise). cslf6-2 mutants had fewer tillers than WT after 10 weeks (Figure 2a). Five weeks later, spikes were developing in the cv. Golden Promise control, whereas the cslf6-2 mutant line was still in the vegetative phase. At maturity, only two to three spikes set seed in the cslf6-2 mutants, limiting further grain phenotypic studies ( Figure S3). The phenotypic appearance of cslf6-2 homozygous grain was vastly different to other csl mutants (~50% reduction in TGW; Figures 2 and 3c). No major whole plant phenotypic changes were detected in cslf9 mutants compared with WT ( Figure 2a). A similar plant phenotype to WT was observed in cslf3 and cslh1 gene-edited lines (Figures S4a and S5a).
In thin sections of mature WT grain, (1,3;1,4)-b-glucan was detected in the starchy endosperm and cell walls of aleurone layers. In contrast, and as expected based on the (1,3;1,4)-b-glucan assay, labelling was completely absent in the endosperm of the cslf6-2 (homozygous) grain. We observed weak fluorescence in the cell walls of sub-aleurone cells and outer endosperm cells (Figures 3c and 4b), but this may be partially due to autofluorescence of phenolic acids ( Figure S8). cslf6-2/+ (heterozygous) grain showed intermediate levels of (1,3;1,4)-b-glucan labelling across the aleurone and endosperm cell walls compared with WT grain (Figures 3c and 4). These results are consistent with the quantification of (1,3;1,4)-b-glucan in these lines (Figure 3a). Differences in grain shape and structure between cslf6 mutants and WT were apparent based on calcofluor counterstaining (Figure 3c,d). Grain sections were also stained with iodine to detect starch granules revealing that both homozygous and heterozygous mutants showed more even, compact staining than WT in mature endosperm tissues. Additionally, a potential difference in polysaccharide distribution was observed in the sub-aleurone and young endosperm tissue of the cslf6-2 (homozygous) mutant as revealed by intense staining with Toluidine blue (TB; Figure 3c).
Developing grains (15 DPA) from cslf9 mutants (all homozygous) exhibited a similar (1,3;1,4)-b-glucan labelling pattern (i.e. distribution) compared with WT grain (Figures 3d and 4). However, this was accompanied by a reduction in calcofluor staining intensity in the cslf9-2 mutant, the outer layers of the cslf9-3 mutant (both of which contain a premature stop codon) and to a lesser extent in the cslf9-1 mutant (which contains a 39-bp deletion; Figure 3d). Further characterisation of cslf9 alleles by linkage analysis (Table S5; Method S3) showed a 24.11 AE 1.56% reduction in cellulose content compared with WT grain, which was also verified by a crystalline cellulose biochemical assay (18.27 AE 0.62% reduction). Moreover, cslf9-2 and cslf9-3 knockout mutants appear to have a different distribution of starch and cell wall polysaccharide staining in transverse grain sections compared with WT grain. A stronger TB stain in the outer endosperm cells of mutants was observed, whereas in WT grain a darker blue colour was seen around the central grain area, indicating a potential difference in polysaccharide distribution (Figure 3d).
Mutations in HvCslF6 and HvCslF9 alter grain polysaccharide composition The differences in staining intensity indicated that polysaccharide distribution, amount and/or cell structure may be altered between cslf6 and cslf9 mutant grain and WT. To assess the monosaccharide composition of these mutant lines, mild acid hydrolysis was used on mature grain flour. Standard conditions were followed such that only non-cellulosic polysaccharides (but not cellulose) would be hydrolysed. Consistent with previous studies, the most abundant non-cellulosic monosaccharides detected in WT nondestarched mature grain flour samples were glucose, followed by xylose, arabinose and galactose (Table 2). These represent the building blocks of the three most abundant non-cellulosic polysaccharides reported in barley grain; starch, (1,3;1,4)-b-glucan and arabinoxylan . Significant differences in non-cellulosic monosaccharide composition were identified between WT and mutants, particularly in glucose levels between WT (54.04 AE 0.93% w/w) and cslf6-2 (36.85 AE 0.89% w/w), cslf9-2 (43.37 AE 1.77% w/w) and cslf9-3 (40.18 AE 2.13% w/w) alleles ( Table 2). The reduced levels of glucose in cslf6-2 mirrored the reduction in (1,3;1,4)-b-glucan content (Figure 3a). The reduction of glucose observed in cslf9-2 and cslf9-3, both of which carry a premature stop codon, was not detected in cslf9-1 that carries the 39-bp deletion. Moreover, no significant difference was detected in (1,3;1,4)-b-glucan content (% w/w) in any of the cslf9 mutant lines compared with WT ( Figure 3a), suggesting that the reduction in glucose may reflect a change in starch abundance (Table S5). Small but significant increases in xylose and arabinose content were identified in cslf6-2 and cslf9-3 lines compared with WT (Table 2). Due to the altered non-cellulosic monosaccharide composition of grain from cslf9 alleles compared with WT, total starch content was analysed in these lines (Table S5). A significant reduction in total starch was observed across all cslf9 alleles (50.90 AE 0.61% w/w cslf9-1; 45.62 AE 0.19% w/w cslf9-2; and 40.24 AE 0.45% w/w cslf9-3) compared with WT grain (59.32 AE 0.97% w/w, P < 0.001, Tukey's test; Table S5).
Taken together, these findings suggest that several of the CRISPR-induced mutations in HvCslF6 and HvCslF9 appear to have an effect on the abundance of cellular carbon stores in the barley grain.

DISCUSSION
Multiple genes have been reported to play a role in (1,3;1,4)-b-glucan biosynthesis. These include genes from the CslF family, of which CslF6 has a prominent role . Gain of function assays in tobacco and Arabidopsis indicate that CslF2, CslF4, CslH1 and CslJ may also contribute to (1,3;1,4)-b-glucan biosynthesis (for review, see Amos and Mohnen, 2019), while QTL and GWAS studies in barley have hypothesised roles for other CslF family members (Han et al., 1995;Molina-Cano et al., 2007;Houston et al., 2014). Despite this, loss-of-function mutants have only confirmed the function of CslF6 in barley and rice (Taketa et al., 2012;Vega-Sanchez et al., 2012;Cu et al., 2016), while functional evidence supporting a role for the other genes has been lacking. In this study a collection of mutants was successfully generated for several members of the HvCslF/H subfamilies of GT2 enzymes, and a subset of these lines was characterised at both the phenotypic and molecular levels. This pool of gene-edited mutants represents a resource for the functional assessment of these and other cell-wall-related genes.
Despite the differences in alleles generated and genetic background used, a similar phenotype to the naturally occurring barley betaglucanless (bgl) mutant (Taketa et al., 2012) was observed in our cslf6-2 knockouts; a reduction in plant height, vigour and spike development. Such extreme phenotypes are to be expected given that HvCslF6 is expressed in a wide range of tissues and developmental stages (Burton et al., 2008). Moreover, HvCslF6 is the main gene responsible for the synthesis of (1,3;1,4)-b-glucan, which has an integral role in the primary cell wall of barley. When DP3:DP4 ratios were quantified, no differences were observed in cslf6 homozygous and heterozygous mutant grain compared with WT. Burton et al. (2011) described a reduced DP3:DP4 ratio in grain of HvCslF6 over-expressing lines, affecting (1,3;1,4)-b-glucan solubility. Although a higher DP3:DP4 ratio was noticed in cslf6-2/+ heterozygous mutants, this was not significantly different from controls (P = 0.285, Tukey's test). Immunohistochemical detection of (1,3;1,4)-b-glucan confirmed the absence of this polysaccharide in mature grain of cslf6-2 homozygous mutants, whereas a weak labelling was detected across aleurone and endosperm tissues in heterozygous lines. It is known that an inverse relationship exists between (1,3;1,4)-b-glucan and starch content in grains of several grass species, ). The precise mechanism/s that regulate this relationship remain unknown. Sections of cslf6 mutant grain (homozygous and heterozygous) analysed for starch distribution suggest a stronger and more compact staining in sub-aleurone layers of knockout and heterozygous lines, potentially due to lower levels of (1,3;1,4)-b-glucan. In addition to confirming the utility of the gene-editing system, the cslf6-2 lines described here provide an important 'null' background for future experiments. These lines can be used to either investigate the molecular, spatial and temporal details of HvCslF6 or other gene function in grain (1,3;1,4)-b-glucan synthesis. Previous studies of wheat addition lines carrying HvCslF6 revealed a 60% increase in grain (1,3;1,4)-b-glucan content although, when compared with barley controls, this only reached 20% of the (1,3;1,4)-b-glucan level in the grain (Cseh et al., 2013). This indicates that in a heterologous system, HvCSLF6 alone is insufficient to replicate the complete (1,3;1,4)-b-glucan biosynthetic activity of barley, and other activities must be present that support (1,3;1,4)-b-glucan synthesis. Further analysis of cslf6-2 knockout lines could potentially reveal regulatory proteins affecting this gene. It is known that HvCesA1, HvCesA2 and HvCesA6, members of the Cellulose synthase gene subfamily within the GT2 family, are co-expressed with HvCslF6 Wilson et al., 2012). It currently remains unclear whether their function or regulation is compromised through lack of functional HvCSLF6 in cslf6-2 lines. The cslf6-2 knockout alleles also provide an opportunity to create double mutants, for example with cslh1 lines carrying frameshift mutations, to assess the contribution of multiple members of the Csl gene family to cell wall composition in leaf, grain and other tissues.
Unlike HvCslF6, previous studies have not reported lossof-function phenotypes for the HvCslF9 gene. HvCslF9 expression in developing grain is generally low but shows a prominent peak at about 8 DPA (cv. Sloop; Burton et al., 2008), although this varies depending on the cultivar (Garcia-Gimenez et al., 2019). In cv. Sloop, this peak coincides with initiation of (1,3;1,4)-b-glucan synthesis, which by 10 DPA is uniformly distributed throughout the cellularised endosperm , and accumulation continues until late grain development (Wong et al., 2015). One possibility is that HvCslF9 contributes to (1,3;1,4)-b-glucan synthesis in a temporally and/or spatially restricted manner. Our data confirm that HvCslF9 has a role in barley grain development. Knockout alleles exhibited pleiotropic changes in grain morphology, monosaccharide, polysaccharide and starch composition, but only minor changes in grain (1,3;1,4)-b-glucan content. This was confirmed in four independent knockout mutants, two lines carrying a 5-bp deletion (cslf9-2) and another two carrying a 1-bp insertion (cslf9-3), all leading to premature stop codons. Similar results were obtained for cslf9-1 mutants carrying a 39-bp deletion in the first exon.
© 2020 The Authors. could also have an impact on malt (1,3;1,4)-b-glucan modification (Stuart et al., 1988;Slakeski and Fincher, 1992), and has been reported in QTL studies that quantified malt (1,3;1,4)-b-glucan content (Han et al., 1995Ullrich et al., 1997;Sz} ucs et al., 2009). Evidence presented in the current study suggests that HvCslF9 is unlikely to influence mature grain (1,3;1,4)-b-glucan content in the Golden Promise cultivar and, hence, HvCslF9 may not be the gene underlying this 1H QTL. Alternatively, HvCslF9 may contribute to (1,3;1,4)-b-glucan biosynthesis only in selected genotypes, at an earlier stage of grain development and/or in a defined grain tissue compartment not investigated here. Despite there being no significant difference in the amount of grain (1,3;1,4)-b-glucan present compared with WT, the cslf9 mutants exhibited a lower TGW compared with WT grain. In the case of the two lines containing premature stop codons in HvCslF9, this reduction in TGW was coupled with a decrease in total starch, cellulose and arabinoxylan compared with WT. Subtle spatial differences in staining intensity were identified in the same lines using iodine (starch), TB (general cell walls) and calcofluor white [general cell walls including cellulose, callose and (1,3;1,4)-b-glucan] stains. These changes suggest that HvCslF9 may contribute to polysaccharide synthesis in specific regions of the grain, and/or specific time points during development, and this warrants further investigation via molecular and biochemical assays. A recent publication using transient heterologous expression in Nicotiana benthamiana leaves indicated that the HvCslF3 and HvCslF10 genes may be involved in the synthesis of a novel linear (1,4)-b-xyloglucan that consists of (1,4)-b-linked glucose and xylose residues (Little et al., 2019). The same study investigated activity of HvCslF9, but failed to identify any clear biosynthetic activity in the heterologous system. Combined with the results described by Little et al. (2018Little et al. ( , 2019, our data agree with the hypothesis that some members of the CslF/H gene families could either influence or be involved in the synthesis of polysaccharides other than (1,3;1,4)-b-glucan.
The final two genes examined were HvCslF3 and HvCslH1, which are expressed at very low levels during grain development (Burton et al., 2008;Doblin et al., 2009). Therefore, as expected, frameshift mutations affecting HvCslF3 and HvCslH1 did not significantly alter either grain monosaccharide abundance or (1,3;1,4)-b-glucan content (Tables 2 and S4). The highest mRNA abundance is found for HvCslF3 in roots (Aditya et al., 2015) and for HvCslH1 in leaves (Doblin et al., 2009). However, while a clear role for CslH1 in (1,3;1,4)-b-glucan synthesis was demonstrated by Doblin et al. (2009) using heterologous expression in tobacco leaves, when Burton et al. (2011) overexpressed HvCslF3 in barley grain there was no detectable change in (1,3;1,4)-b-glucan content. Loss-of-function mutations in either gene impacted mature grain morphology, suggesting they are likely to have a function during plant development, but changes in mature grain composition were relatively small and statistically insignificant. Further phenotypic characterisation of additional plant tissues and developmental stages will be required to determine the precise contribution of these genes to cell wall composition; whether they have an organ-or tissue-specific role in (1,3;1,4)-b-glucan biosynthesis, are impacted by compensatory mechanisms (P erez et al., 2019), or synthesise other polysaccharides such as glucoxylan (Little et al., 2019).
Taken together, our collection of gene-edited alleles for HvCslF/H with knockout and frameshift mutations represents a valuable genetic resource for studying barley cell walls, and demonstrates the effectiveness and potential of CRISPR/Cas9 genome-editing technology for the analysis of gene function. Importantly, our results show that members of the CslF/H family other than HvCslF6 have an in planta function in growth and development, providing greater opportunities to understand and optimise cell wall composition for specific downstream applications.

Plant material and Agrobacterium-mediated transformation
Barley cv. Golden Promise was used for embryo transformation via Agrobacterium tumefaciens following Bartlett et al. (2008) at the Functional Genomics (FUNGEN) facility, The James Hutton Institute (UK). Plants were grown in glasshouse conditions (16-h light/8-h dark photoperiod) until maturity in T 0 , T 1 and T 2 generations. At harvest, spikes were collected and manually threshed to obtain seed for subsequent generations. sgRNA design and construct assembly sgRNA sequences were manually designed using the criteria stated on 'Addgene CRISPR Guide' website (Addgene, 2018) to the 5' end of the first exon for all four genes. This was to maximise the possibility of obtaining lines containing mutations that caused a knockout of gene function induced by CRISPR/Cas9. To prevent off-target mutations, a BLAST search of sgRNA plus PAM sequences was performed on Barlex (Colmsee et al., 2015) to determine sequence specificity to HvCslF3, HvCslF6, HvCslF9 and HvCslH1, avoiding sgRNAs with potential matches to non-target genes and conserved protein motifs ( Figure S4). Final sgRNA sequences were selected based on their target location and BLAST scores (Tables S3 and S6). Each sgRNA was cloned into the pC95-gRNA entry vector downstream of the rice U6 small nuclear RNA (snRNA) (OsU6p) by Gibson Assembly. sgRNA for HvCslF3, HvCslF6, HvCslF9 and HvCslH1 were inserted into the pBract214m-HvCas9-HSPT expression vector, which contains a barley codon optimised Cas9, and independently transformed into cv. Golden Promise (Method S1; Figure S7).

Screening of CRISPR/Cas9-induced mutations
Mutations were identified by a nested PCR method combined with InDel Detection by Amplicon Analysis (Yang et al., 2015). Genomic DNA was isolated from a leaf disc (2 mm diameter) taken from individual barley seedlings using the Phire Plant Direct PCR Kit (Thermo Fisher Scientific, Waltham, USA). An aliquot of 0.5 µl crude plant extract was used as a template for cas9 and InDel PCR detection in a reaction containing: 10 µl 2 9 Phire Plant PCR Buffer, 1 µl of each external forward and reverse primer at 10 mM (Table S8), 0.4 µl Phire Hot Start II DNA polymerase and 7.1 µl sdH 2 O in a total volume of 20 µl. External PCR amplicons, about 1.5 kb for each targeted gene, served as a DNA template (1 µl aliquot) for the nested (internal) PCR containing: 2.5 µl 10 9 Hot Start Taq buffer, 2.5 µl dNTPs, 0.12 µl Hot Start Taq DNA polymerase (Qiagen, Hilden, Germany) and 1 µl each primer: external forward and reverse at 1 and 10 mM, respectively, and internal forward (6-FAM labelled) at 10 mM (Table S8) in a total volume of 25 µl. The internal forward (6-FAM) primer contains a 21-nucleotide tail complementary to the external forward primer. PCR conditions are described in Table S7. Fluorescent-labelled PCR amplicons were processed using an ABI3730 DNA Analyzer (Applied Biosystems, Foster City, USA). The resulting chromatograms were visualised and analysed using GeneMapper â (Applied Biosystems, v4.1), which allowed InDel identification based on size differences compared with cv. Golden Promise control. Mutations were confirmed by Sanger sequencing after reaction clean up and preparation as described in Houston et al. (2012). Resulting sequences were manually trimmed, aligned and analysed with Geneious V.9 (Kearse et al., 2012) to identify sequence variations. Potential off-target effects were checked by PCR using primers described in Table S8, amplified using Hot Start Taq as described in Table S7. Only putative off-targets with zero mismatches in the PAM, i.e. HvCslF3 and HvCslH1, were included as this motif is vital for Cas9 to recognise the sequence. DNA sequencing was carried out as described above.

Phenotypic assessment of grain characteristics
Mature spikes were collected and hand threshed. A subset of~30 bulked grains per genotype was used for phenotypic studies. All phenotypic data were collected from T 2 plants (T 3 grain). The number of grain, average grain area, length and width were measured with Marvin Seed Analyzer (GTA Sensorik GmbH, 2013).
The grains were weighed to combine with the grain number estimate and derive TGW.

Imaging
Imaging of immunolabelled grain sections was performed on a Zeiss LSM 710 Confocal Laser Scanning Microscope (Carl Zeiss Microscopy GmbH, Jena, Germany) using a PL APO 20 9/1.0 water dipping objective (Zeiss). Calcofluor was excited using 405 nm light from a blue diode laser, and emission was collected between 420 and 460 nm. Alexa Fluor 488 dye (BG1) was excited using 488 nm laser light from an argon ion laser, and emission was collected between 504 and 543 nm. Grain autofluorescence was detected by excitation using 594-633 nm laser light from a red diode laser, and emission was collected between 631 and 730 nm. Cultivar Golden Promise with 5.00% w/w AE 0.03 of grain (1,3;1,4)-b-glucan content, was used as a positive control for (1,3;1,4)-b-glucan detection. Negative controls showed no fluorescent labelling when neither the primary antibody nor secondary were used ( Figure S8). Brightfield images of grain sections stained with Lugol's iodine solution and TB O were collected using a Zeiss Axioskop 2 Plus microscope (Carl Zeiss Microscopy GmbH, Jena, Germany) equipped with a 5 9 lens (Zeiss). Subsequent to collection, images were processed with Zen Software v6.0, utilising global adjustment tools only.

Non-cellulosic monosaccharide analysis
Non-cellulosic monosaccharide analysis was carried out as described in Hassan et al. (2017) using reversed-phase high-performance liquid chromatography separation coupled to diode array detection (Agilent Technologies). Dehulled grain was ground, and resulting flour samples (20 mg) were prepared according to Pettolino et al. (2012). Acid hydrolysis of nondestarched alcohol insoluble residue was undertaken by adding 1 M sulphuric acid, as described previously (Burton et al., 2011), to the insoluble material (~15 mg). Between 3 and 12 replicates were analysed for each genotype. The moisture content of the ground flour was determined as outlined in AOAC 925.10 (AOAC International, 2005

AUTHOR CONTRIBUTIONS
KH, MRT, RAB and RW conceived this work. KH designed and generated the sgRNA constructs. JS carried out barley transformations. Abdellah Barakate adapted the genotyping method for mutation detection. GG-G and KH performed the genotypic characterisation of the CRISPR lines. GG-G carried out the grain phenotyping, (1,3;1,4)-b-glucan quantification, immunolabelling experiments and data analyses. PS processed the CRISPR lines/samples for subsequent assays. SFK and RAB carried out the grain monosaccharide and starch analyses. MSD and PH performed the linkage and cellulose analyses. The manuscript was drafted by GG-G, KH and MRT, and reviewed by Abdellah Barakate, Antony Bacic, RAB, MSD, RW and GBF. All the authors read and approved the manuscript.

CONFLICT OF INTEREST
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

DATA AVAILABILITY STATEMENT
All relevant data can be found within the manuscript and its supporting materials.

SUPPORTING INFORMATION
Additional Supporting Information may be found in the online version of this article. Figure S1. Frequency of mutations in T 0 , T 1 and T 2 generations across the target genes. Figure S2. Determination of cas9 copy number by hygromycin segregation test. Figure S3. Images of whole plant phenotype in the cslf6-2 homozygous mutant. Figure S4. (a) Plant phenotype of cslf3 mutants. (b) Assessment of grain characteristics. Figure S5. (a) Plant phenotype of cslh1 mutants. (b) Assessment of grain characteristics. Figure S6. Representation of sgRNA + PAM sequences in the context of BLAST hits. Figure S7. Vector map of pBract214m-HvCas9-HSPT-sgRNA (destination vector). Figure S8. Negative controls used for grain immunolabelling of (1,3;1,4)-b-glucan. Methods S1. sgRNA construct assembly.