F5H is a cytochrome P450 monooxygenase enzyme and a member of the CYP84 family (Meyer et al., 1998). Putative barley F5Hs were shortlisted in two steps. First, we blast-searched the barley whole-genome Morex assembly and used multiple sequence alignment to identify proteins that had all the conserved structures and signature motifs in F5Hs as described previously (Meyer et al., 1996; Bak et al., 2011). Next, using phylogenetic analysis, candidate sequences from barley were compared with functionally characterised F5H genes in Arabidopsis (Meyer et al., 1998) and grasses including rice (Takeda et al., 2017), switchgrass (Wu et al., 2019), and sugarcane (Bewg et al., 2016). This led to identification of five putative F5H genes in barley, named HvF5H1-5 (HORVU1Hr1G047220, HORVU3Hr1G014930, HORVU2Hr1G109440, HORVU7Hr1G084500, and HORVU2Hr1G092360, respectively). The derived amino acid sequences of these genes are shown in Figure S1 .

A phylogenetic tree was constructed using the Maximum Likelihood method and a JTT matrix-based model that shows the relationships between these five putative barley F5H amino acid sequences and those from other C3 and C4 grasses ( Figure 1 ). Spike moss (Selaginella moellenodorffii) smF5H was brought into this analysis as an outlier (Weng et al., 2008). The phylogenetic tree shows that the majority of putative F5H proteins in grasses group closely together, forming a large cluster that encompasses all previously functionally characterised grass F5Hs ( Figure 1 red clade). This cluster includes protein sequences for barley HvF5H1, HvF5H2, and HvF5H3 genes, hypothetically supporting a direct role for these in lignin biosynthesis. The presence of three barley F5Hs in this clade represents an expansion of the gene family in barley compared to some other grasses that have only one F5H in the clade (e.g., Brachypodium and rice). By comparison, wheat has two or three TaF5H counterparts for each barley sequence representing the three ancestral genomes that compose modern hexaploid wheat. Collectively these data suggest duplications of the F5H gene in barley and wheat and possibly all Triticeae after their divergence from the Brachypodieae lineage. Predictably, SmF5H and AtFAH1 proteins appeared to be less closely related to the grass F5Hs and are located in isolated distant clades. We obtained similar phylogenetic results using the UPGMA (Unweighted Pair Group Method with Arithmetic mean) and Neighbour Joining methods.

The phylogenetic tree also encompasses five other small clades of amino acid sequences that carry F5H signature motifs but have unknown functions potentially distinct from lignin biosynthesis. The OsF5H2 (CYP84A6) gene was reported to be expressed in rice ovary 3 days after flowering (Takeda et al., 2017) and in mature leaves (Kim et al., 2006) but is not considered to be an essential gene in lignifying tissues (Takeda et al., 2017). Similarly, in Arabidopsis, duplication and neofunctionalisation of AtFAH1 has produced a paralogous gene, AtCYP84A4, of which the corresponding enzyme has a widely altered catalytic activity in α-pyrone metabolite biosynthesis (Weng et al., 2012).

The HvF5Hs show 60-65% amino acid sequence identity to AtFAH1 and 65-81% identity to orthologs in the Poaceae family. Out of the five identified F5H homologs in barley, HvF5H1 shared the highest amino acid sequence identity to the functionally characterised F5Hs from Arabidopsis (AtFAH1, 65%), rice (OsF5H1, 80.7%), switchgrass (PvF5H1, 80.8%), and sugarcane (SaH_AOR81843, 81%). Furthermore, barley RNAseq data (Mascher et al., 2017) revealed that the expression pattern of HvF5H1 is characteristically different from the other HvF5Hs ( Figure S2 ). HvF5H1 is the only HvF5H expressed to substantial levels with the highest transcript abundances found in lignifying tissues as well as leaves ( Figure S2 ). This is particularly important as it demonstrates HvF5H1 is highly active in lignocellulosic biomass but not in the grain or inflorescence tissues. HvF5H3 shows some expression only in embryo and roots, whereas HvF5H4 expression was found only in roots. Very limited expression was detected for HvF5H2 or HvF5H5 in any of the 16 tissues presented in the dataset. The expression data combined with phylogenetic analysis clearly suggest the HvF5H1 gene as the strongest candidate for an involvement in lignin biosynthesis.

To manipulate lignin composition and structure and investigate the influence of HvF5H1 on S unit biosynthesis, we used the RNAi approach to produce transgenic plants with suppressed HvF5H1 expression. A fragment of 669 bp ( Figure S3 ) was used to produce a hairpin inverted repeat sequence in the plasmid pBract207 under the control of the constitutive maize ubiquitin promoter. Subsequently, 23 T0 independent transgenic plants were produced following the transformation of the Golden Promise cultivar ( Figure S4A ). Southern blot analysis identified plants with a single T-DNA locus and qPCR was employed to assay plants for reduced HvF5H1 expression relative to wild-type plants ( Figure S4B ). The qPCR was carried out on samples of second internode taken at growth stage 32 on the Zadoks scale (Zadoks et al., 1974). Of 10 T0 plants that had a single insertion site of the T-DNA locus, nine also exhibited at least 60% reduction in HvF5H1 expression compared to the wild type and greater than 67% reduction compared to the empty vector (EV) plants; the best four lines showed approximately 95% or more reduction in HvF5H1 expression. From the pool of nine T0 plants, we selected seven lines to develop into the T1 generation. These lines were named B, I, K, M, T, W, Y. T1 generation plants were screened to identify HvF5H1-RNAi homozygote and azygote plants that had lost the transgene through segregation. Consequently, we selected three independent lines B, T, and W, to analyse further in the T2 generation. For this purpose, homozygous RNAi T2 plants along with control lines and wild type were grown in the greenhouse in a Randomised Complete Block design with five replications. Levels of HvF5H1 transcript reduction were evaluated to be 90% for line B, 88% for line W and 93% for line T ( Figure S5 ). Using Mäule staining, the second internodes of the T2 plants were investigated for a change in S lignin relative to control plants. Whereas all control plants (wild type, empty-vector controls, and azygote plants that had lost the RNAi transgene through segregation) showed bright red staining in lignified tissues, in the RNAi lines the vascular bundles and cortical fibre tissues exhibited more of a brown colour, indicating a reduction in the proportion of S lignin ( Figure S6 ).

To investigate the consequence of suppressed HvF5H1 expression on lignin content and composition/structure, extract-free cell wall residue (CWR) was prepared from the main tiller of fully-senesced T2 generation RNAi homozygote plants and related control plants. We removed leaf sheaths from the tillers as their cell wall structure and lignin content differs from culms. Klason lignin determinations on tiller CWR showed no significant differences in Klason lignin content between the HvF5H1-RNAi plants and the controls ( Figure 2A ).

Analytical thioacidolysis and 2D NMR (HSQC) were employed to study lignin composition of HvF5H1 RNAi lines in more detail. Thioacidolysis degradation releases H, G and S thioethylated monomers only from H, G and S lignin units linked by β-aryl ether (β–O–4) bonds to reflect relative abundance of these monomers (Lapierre et al., 1986). The thioacidolysis yield (in µmol/g Klason lignin) was reduced by up to 27% in the HvF5H1-RNAi lines indicating a reduction in the proportion of units linked by β-aryl ether bonds. The S/G ratio of thioacidolysis released monomers in HvF5H1-RNAi lines decreased by up to 85% from 1.62 in line B azygotes to 0.24 in the corresponding RNAi plants (P<0.01; Figure 2B ) reflecting a shift from 60% S units and 37% G units in lignin in azygote plants to 19% S units and 78% G units in lignin in RNAi plants ( Figure S7 ). Changes in RNAi lines W and T were similar but slightly less severe. These findings are in line with F5H’s role in hydroxylating G-lignin precursor units (coniferaldehyde or coniferyl alcohol) to produce S lignin monomers. Our analyses also revealed a significant reduction (P < 0.01) of 28% and 41% in the amount of ester-linked pCA released by mild alkaline hydrolysis in HvF5H1-RNAi lines T and B respectively ( Figure 2C ), although the 6% reduction in line W was not significant. Ester-linked FA was also significantly reduced in all RNAi lines ( Figure 2D ). Reductions in ester-linked pCA are consistent with the reduction of S lignin given that much of the cell wall pCA in grasses is ester-linked to S units (Ralph et al., 1994).

A 2D-NMR (HSQC) analysis of line B cell wall material further verified the lignin compositional changes in the HvF5H1-RNAi plants that had been demonstrated through analytical thioacidolysis ( Figures 3A, B ). The greatest alterations were observed in S and G units. The G units in the RNAi line increased proportionally to 86% compared to 44% in the control. On the other hand, suppression of HvF5H1 led to reduction of S unit content to 12% in the RNAi plant compared to 53% in the control plant. The H-unit content slightly decreased in the RNAi plant (2%) relative to control (3%). In agreement with thioacidolysis, NMR detected 12% of pCA in the RNAi plant versus 20% in the control and 4% tricin in the RNAi plant versus 3% in the control; such mobile end-units are, however, over-quantified by NMR (Mansfield et al., 2012).

The aliphatic subregion of the 2D-NMR spectra shows the signals that are derived from interunit linkages in the CWR lignin polymer ( Figures 3C, D ). In both RNAi and control plants, the dominant units are β-aryl ethers followed by substantially lower percentages of phenylcoumaran (β–5) and resinol (β–β). The proportion of β-aryl ether bonds, which are the main intermolecular linkages in native lignin, were reduced from 93% in control to 83% in the RNAi line, consistent with the thioacidolysis yield reduction which also indicated reduced β-aryl ether linkages. On the other hand, the share of phenylcoumaran linkages in the RNAi line increased to 15% compared to 4% in the control. There was little change in the proportion of resinol bonds. These data are all consistent with the higher proportion of G units in the HvF5H1-downregulated plants.

To study the impact of HvF5H1 suppression on intracellular phenolic metabolites, we sampled the first three internodes of plants from two HvF5H1-RNAi lines (lines B and W) to compare with wild-type and EV control plants. UHPLC-MS was employed to analyse phenolic metabolites extracted from the internodes. Initially, our analysis detected 7486 features with an abundance of more than 100 counts in at least one sample. Features were selected for further investigation if their abundance was significantly (P<0.01) different from the controls for both HvF5H1-RNAi transgenic lines. This led to the identification of 173 features (i.e., m/z traces) with at least three-fold change in both RNAi lines compared to control values. Of these, 95 features showed higher intensity in the HvF5H1 RNAi plants whereas 78 m/z traces exhibited lower intensity ( Table S2 ). Taking into account the accurate m/z, retention time, and MS/MS fragmentation, we characterised 47 metabolites. The 21 identified metabolites with higher intensity in the RNAi lines were mainly small oligomers composed of G units only, or in combination with units derived from ferulic acid or tricin. In contrast, the metabolites with reduced abundance were sinapyl alcohol, three conjugates with syringic acid, and 22 di- and tri-lignols containing at least one S unit.

We performed a transcript profiling in order to evaluate the effect of HvF5H1 downregulation on the expression of other genes in stems. For this purpose, the first three internodes were sampled from two HvF5H1-RNAi lines (lines B and W) and compared with wild-type and EV control. To identify differentially regulated genes, expression data was first log-normalised and the genes that were significantly (P<0.01) different between the HvF5H1-RNAi lines and both controls were identified. Then we shortlisted genes that were up- or down-regulated greater than three-fold. In the process, probes that had a significant (P<0.001) expression difference between the wild-type and empty vector controls were excluded due to lack of consistency. Subsequently, fourteen genes were identified as upregulated and four genes as downregulated ( Table S3 ) in the RNAi lines compared to controls. HvF5H1, the targeted gene, was downregulated 4.7-fold. Of the other HvF5H genes, HvF5H2 and HvF5H5 expression was not detected by transcript profiling and expression of both HvF5H3 and HvF5H4 remained unchanged. The top-downregulated gene (HORVU2Hr1G052170) belonged to the alpha/beta hydrolase superfamily and was downregulated about 7-fold.

The most differentially overexpressed genes in the RNAi lines were a DNA/RNA helicase which is involved in meiotic crossover formation (HORVU4Hr1G049830, 22-fold), and a hydroxyjasmonate sulfotransferase (HORVU2Hr1G117700, 15-fold). An NBS LRR disease resistance gene along with a MYB transcription factor were upregulated 12.4 and 6.6-fold, respectively, in the HvF5H1-RNAi lines.

Given the considerable modification in lignin structure of the HvF5H1-RNAi plants, we phenotyped them for a variety of agronomic traits to determine whether such dramatic changes to lignin might impact directly on crop productivity. Visually, the HvF5H1-RNAi lines did not exhibit noticeable phenotypic differences compared to the control plants ( Figure 4A ). Plants were phenotyped for shoot height, tiller number, straw biomass, spike number, and seed characteristics including seed length, width, and area and thousand-grain weight (TGW) ( Table S4 ). Employing ANOVA, none of these traits in the RNAi lines was significantly different from the ones in control plants (P<0.05).

Lignin modification could impact straw strength and potentially the plant’s resilience or lodging resistance, with indirect effects on productivity and yield. We therefore tested the mechanical properties of stems sampled from the RNAi lines at harvest time by quantifying the maximum stretching force straw can tolerate before it breaks, i.e. the ‘Flexural Strength Stress at Yield’. For this purpose, we sampled second and fourth internodes from the main tiller and employed an Instron Universal Testing System for precision phenotyping. Predictably, in each genotype significantly (P<0.05) less force was required to break the fourth internode compared to the corresponding second internode. However, we did not find a significant difference between the RNAi lines and the controls ( Figure 4C , Table S5 ) when either the second or fourth internodes were analysed, suggesting that the S/G ratio reduction did not impact the mechanical strength of barley straw.

To investigate the consequence of lignin structural changes on straw recalcitrance, we carried out saccharification assays after subjecting the straws to hot water pretreatment. The enzymatic digestion phase lasted 96 hours and samples were collected at 4, 24, 48 and 96 hours. However, there was not a significant improvement or deterioration in saccharification efficiency when each RNAi line was compared with its most relevant controls following Student’s t‐test (P < 0.05) ( Figure 5 ).

Putative HvF5H genes were initially retrieved from a public database (Mayer et al., 2012) and reconfirmed with the latest released version (Mascher et al., 2017). Barley amino acid sequences of HvF5H that showed 40% or more identity to the well-studied AtFAH1 (Meyer et al., 1996) were selected. Next, the best aligning protein sequences were blasted back into the barley database to pull out putative genes with lower identity to AtFAH1. Subsequently, all the selected protein sequences were aligned with those of functionally characterised F5H genes in rice (Takeda et al., 2017), switchgrass (Wu et al., 2019), sugarcane (Bewg et al., 2016), spike moss (Weng et al., 2008), Brachypodium (Sibout et al., 2017), maize (Andersen et al., 2008) and foxtail millet (Muthamilarasan et al., 2015). In the phylogenetic analysis we also used unpublished F5H sequences that stemmed from bioinformatics screenings. These genes were identified or revisited by blasting against the latest relevant genome database and selecting those containing all five signature motifs of F5Hs as described before (Chapple, 1998).

Phylogenetic analysis and construction of the tree were carried out using MEGA X software (Kumar et al., 2018). Amino acid sequences were aligned using ClustalW and a Neighbor-Joining tree constructed tested by 100 bootstrap replications. Evolutionary distances were estimated using the Poisson model based on the number of amino acid substitutions per site.

Barley genes transcriptomic profiling data from different plant tissues were extracted from the publicly available database (Mayer et al., 2012; Milne et al., 2021).

Barley cultivar Golden Promise was used to generate transgenic plants. All plants were grown in the greenhouse with light and temperature controls. High-pressure sodium lamps provided supplementary lights when necessary. Plants were grown in seven-inch pots filled with standard compost mix (150 L Levingtons M2 compost, 75 L perlite, 500 g Osmocote Plus Mini, 125 g Celcote Certis and 75 g Horti-Wet). Whole above-ground biomass of individual plants was harvested after senescence at the Z9.2 growth stage based on the Zadoks scale (Zadoks et al., 1974). Subsequently, biomass was air-dried at 35°C overnight. Following removing the leaves, second internodes from the main tillers were used to conduct cell wall characterization assays or saccharification. Second internodes were identified as described previously (Tottman, 1987).

A cDNA of HvF5H1 (HORVU1Hr1G047220) was used to amplify a 669 bp fragment using primers containing Gateway AttB sites. Subsequently, the amplicon was cloned into the entry vector using BP clonase II (Invitrogen) and then the fragments were transferred from the entry clone to the RNAi vector, pBract207 (John Innes Centre) using LR Clonase (Invitrogen). The RNAi cassette was transferred into Agrobacterium tumefaciens AGL1 containing the helper pSOUP plasmid. Barley transgenic lines were produced by the FunGen Group at The James Hutton Institute through the transformation of embryos of the Golden Promise cultivar (Harwood et al., 2009). The T0 transgenic lines were grown to produce seeds and perform preliminary screenings. The T1 generation was screened using a hygromycin root assay to establish plants’ zygosity and the southern blot assay to determine numbers of T‐DNA loci. Briefly, twenty seeds were germinated on agar plates with 0.5% phytogel and 100 µg/mL of hygromycin as described by Jacobsen et al. (2006). In subsequent generations, hygromycin assays and PCR was used on germinating progeny from parental plants to identify parents homozygous for the RNAi transgene and those that had lost it through segregation (azygotes).

RNA extraction was performed using TRI-Reagent (Sigma-Aldrich) as described by the manufacturer. The resulting RNA was further cleaned up using the Qiagen RNeasy Mini Kit RNA Cleanup Protocol (Cat No 74536). 500 ng of RNA was used to synthesise cDNA using qScript (Perfecta). The resulting products contain a supermix that has both OligoDT and random primers for better amplification. This was then amplified using the protocol outlined by the manufacturer. The expression analysis was conducted using the Applied Biosystems StepOne machine and related software. SnRK1 (SNF1-related kinase 1) was selected as the internal control gene and we used primers designed against the 3 different orthologs of the HvF5H1 gene. Primers and probes were designed (allocated) through the Roche Universal Probe Library version.

Freehand stem cross sections were obtained from the second internodes using a No. 11 Scalpel blade. Sections were fixed in 4% glutaraldehyde for 1 h at room temperature, rinsed with water, and immersed in 0.5% (w/v) potassium permanganate solution for 10 min at room temperature. Then they were rinsed with water several times and incubated in 10% hydrochloric acid for 5 min. Following rinsing twice with water, the sample was dipped in concentrated ammonium hydroxide under a fume hood for 2 min and mounted on the slide. A Nikon light microscope was employed to study stem sections. Images were taken using a microscope mounted Lecia DC500 camera.

Lignin contents were quantified following modification of a previously described method (Dence, 1992). In summary, whole tillers (excluding sheaths and leaves) of three biological repeats for each line were powdered to 0.5 mm particle sizes using a Retsch ZM-200 centrifugal rotor mill and dried at 35°C overnight. Followed by an extraction process using exhaustive hot water and hot ethanol via Soxhlet. Samples of 300 ± 5 mg were then hydrolysed in 5 mL of 72% H₂SO₄ for 2 h. Distilled water (12.5 mL) was added and the samples were kept at room temperature for 1 h before being transferred into 250 mL Schott bottles by washing with 150 mL dH₂O. The samples were incubated at 121°C for 1 h and filtered through a crucible to collect both the solid (acid-insoluble fraction) and the liquid filtrate (acid-soluble fraction). The lignin was collected by vacuum filtration on Whatman GF/C 55 mm filter paper and subsequently washed and dried at 110°C overnight. The lignin content was calculated as a percentage of the dry weight of the starting material. Lignin values were adjusted for ash content by burning the extracted lignin at 545°C in a furnace.

Lignin structure was evaluated by the simplified thioacidolysis procedure as described previously (Lapierre et al., 2021). Ester-linked p-coumaric and ferulic units were measured after mild alkaline hydrolysis as described previously (Sibout et al., 2016).

Cell walls were characterised without fractionation using two-dimensional (2D) solution-state NMR (Kim and Ralph, 2010; Mansfield et al., 2012). Straw (2-mm pieces) was pre-ground using a Mixer Mill MM400 (Retsch; 30/s vibrational frequency for 90–120 s). Samples were extracted three times with water, three times with 80% ethanol and once with acetone, then allowed to dry. The pre-ground extracted samples were ball-milled using a Fritsch Planetary micro mill Pulverisette 7 vibrating at 600 rpm with zirconium dioxide (ZrO₂) vessels containing ZrO₂ ball-bearings (10 mm × 10) with 5-min milling and 5-min cooling per milling cycle (cycle number depended on the amount of sample). The ball-milled samples were subjected to digestion (72 h × 2) to obtain ‘enzyme lignin’ (EL) by Cellulysin^® cellulase from Trichoderma viridae (Calbiochem), at 35°C in acetate buffer (pH 5.0). The ELs were dissolved into DMSO-d₆/pyridine-d₅ (4:1 v/v) and subjected to NMR on a Bruker Biospin AVANCE-III 700 MHz spectrometer equipped with a 5-mm QCI ¹H/³¹P/¹³C/¹⁵N cryoprobe with inverse geometry (proton coil closest to the sample). 2D-¹H–¹³C HSQC spectra were acquired using Bruker’s pulse program (hsqcetgpsip2.2). Bruker’s Topspin 3.2 (Mac) software was used to process spectra. The central DMSO peak was used as internal references (δ_C: 39.51, δ_H: 2.49 ppm).

Five plants from each line were grown for 61 days in a Randomised Block Design (RBD) in a greenhouse with supplementary lighting. The first three internodes of each plant were collected and ground in liquid nitrogen, subsequently 100 mg of the ground stem from each replication was used for transcriptome or metabolite analysis as described before (Daly et al., 2019).

Straw was sampled from the RNAi and control lines at harvest time, Zadoks stage 97 (Zadoks et al., 1974), and mechanical properties were analysed by quantifying the Flexural Strength Stress at Yield. Second and fourth internodes from the main tiller were fixed into in an Instron Universal Testing System (https://www.instron.com/) and the force needed to break the straw was measured.

A robotic platform (Labman Automation Ltd, UK) was employed for alkaline pre-treatment and saccharification as described before (Oakey et al., 2013). A modified version was used for hot water pre-treatment and subsequent saccharification. Briefly, 7.0 ± 0.5 mg extracted pulverised sample was pre-treated with hot water in the autoclave at 130°C for 30 min. The solid was washed 3× with deionised water. The samples were then incubated while shaking at 50°C for 96 h in the presence of an enzyme cocktail (4:1 ratio of Celluclast and Novozyme 188). The GOPOD assay kit (K-GLUC) (Megazyme, Ireland) was used to quantify the glucose released during saccharification.

Data were collected as described for each trait. Where required, a square root transformation of response was used to ensure model assumptions of normality and homogeneity of variance in residuals were met (Miller, 1997). T-test was employed when comparing two groups. To compare significance levels for three or more groups, ANOVA was employed. Where ANOVA showed significance at the 5% level, post-hoc Tukey’s test was used for comparisons. The Minitab Statistical Software version 18.1 or R (R Core Team, 2021) was used for statistical analysis. Model-adjusted means were used for the metabolomics/transcriptomic experiment. Metabolites/probes for which the combined mean was at least threefold and significantly (P<0.01) different from combined controls in both HvF5H1-RNAi lines were considered to be significantly different.