Hydroxycinnamic acid-modiﬁed xylan side chains and their cross-linking products in rice cell walls are reduced in the Xylosyl arabinosyl substitution of xylan 1 mutant

The intricate architecture of cell walls and the complex cross-linking of their components hinders some industrial and agricultural applications of plant biomass. Xylan is a key structural element of grass cell walls, closely interacting with other cell wall components such as cellulose and lignin. The main branching points of grass xylan, 3-linked L -arabinosyl substitutions, can be modiﬁed by ferulic acid (a hydroxycinnamic acid), which cross-links xylan to other xylan chains and lignin. XAX1 (Xylosyl arabinosyl substitution of xylan 1), a rice ( Oryza sativa ) member of the glycosyltransferase family GT61, has been described to add xylosyl residues to arabinosyl substitutions modiﬁed by ferulic acid. In this study, we characterize hydroxycinnamic acid-decorated arabinosyl substitutions present on rice xylan and their cross-linking, in order to decipher the role of XAX1 in xylan synthesis. Our results show a general reduction of hydroxycinnamic acid-modiﬁed 3-linked arabinosyl substitutions in xax1 mutant rice regardless of their modiﬁcation with a xylosyl residue. Moreover, structures resembling the direct cross-link between xylan and lignin (ferulated arabinosyl substitutions bound to lignin monomers and dimers), together with diferulates known to cross-link xylan, are strongly reduced in xax1 . Interestingly, apart from feruloyl and p -coumaroyl modiﬁcations on arabinose, putative caffeoyl and oxalyl modiﬁcations were characterized, which were also reduced in xax1 . Our results suggest an alternative function of XAX1 in the transfer of hydroxycinnamic acid-modiﬁed arabinosyl substitutions to xylan, rather than xylosyl transfer to arabinosyl substitutions. Ultimately, XAX1 plays a fundamental role in cross-linking, providing a potential target for the improvement of use of grass biomass.


INTRODUCTION
Grass cell walls are an abundant and renewable source of energy-rich polymers for a great variety of industrial applications ranging from animal nutrition to the production of second-generation biofuels. A major constraint in the industrial use of grass cell walls lies in their recalcitrance to breakdown due to the complexity associated with the arrangement and interactions of cell wall components (Abramson et al., 2010;Bhatia et al., 2017;Pauly and Keegstra, 2008). The composition and cross-linking of cell wall components differs greatly from that in eudicot cell walls (Hatfield et al., 2016;Vogel, 2008).
In general, plant cell walls are composed of a scaffold of cellulose microfibrils embedded in a matrix containing a mixture of hemicelluloses, pectins, and structural proteins, and in some instances non-polysaccharide, nonprotein components such as lignin. In grasses, xylan is the main hemicellulosic polysaccharide in the cell wall. Xylan consists of a backbone of b-(1,4)-linked hydroxycinnamic acids like ferulic acid (FA) and pcoumaric acid (pCA) through an ester linkage (Hatfield et al., 2016). Specifically, FA has been described to crosslink xylan chains through oxidative coupling with other FA groups (Ishii, 1997). 5-5-, 8-O-4-, 8-5-, and 8-8-coupled ferulate dehydrodimers as well as dehydrotrimers and -tetramers have been identified (Bento-Silva et al., 2018;Burr and Fry, 2009;Ralph, 2010;Waterstraat and Bunzel, 2018). The diversity of coupling products highlights that these ferulate structures are synthesized in a combinatorial radical cross-coupling process similar to that which occurs during lignification (Ralph, 2010;Ralph et al., 2004). Lignification results from the polymerization of the monolignols p-coumaroyl, coniferyl, and sinapyl alcohol, yielding the p-hydroxyphenyl (H), guaiacyl (G), and syringyl (S) units in lignin (Freudenberg and Neish, 1968;Ralph et al., 2019;Vanholme et al., 2019). In the oxidative environment of the cell wall, peroxidase-catalyzed oxidation of ferulates and lignin units and monomers and their subsequent radical-radical cross-coupling is expected and this idea is supported by biomimetic co-polymerization of ferulates into synthetic lignins (Grabber et al., , 2000Quideau and Ralph, 1997;Ralph et al., 1992). Many of the resulting cross-coupling units were indeed identified in lignin Jacquet et al., 1995;Ralph et al., 1995). Because ferulates in monocot cell walls are mostly present as FA-modified a-L-arabinofuranosyl (5-Oferuloyl arabinose; FA-A) residues (1,3)-linked to the backbone (Ishii, 1997), it is generally accepted that these latter ferulates are involved in the lignin-polysaccharide connections (Grabber et al., 2002). This has been recently supported by the detection of the cross-coupling products of either coniferyl or sinapyl alcohol with FA-A in hydrolysis extracts of Miscanthus and maize (Zea mays) lignified cell walls . Cross-linking of lignin and xylan through FA is seen as a key element in enhancing the structural complexity and recalcitrance of the cell wall to breakdown (Buanafina, 2009;de Oliveira et al., 2015;Terrett and Dupree, 2018). In some instances, FA-A can adopt more complex forms when decorated with additional sugars such as b-(1,2)-linked xylopyranosyl (X), which leads, e.g., to the formation of X-(FA)-A (2-O-xylopyranosyl-(5-O-feruloyl) arabinose) or the a-(1,2)linked L-galactopyranosyl modification of X-(FA)-A (Allerdings et al., 2006;Saulnier et al., 1995;Schendel et al., 2015;Wende and Fry, 1997). However, the function of the additional sugar modifications is unknown. Quantitative analysis of side chains in Miscanthus xylan showed that X-A substitutions are rare with a frequency of 1-2% in leaves and considerably less in stem; similar substitution profiles were seen in rice (Oryza sativa) leaves (Tryfona et al., 2019). In contrast to FA, the combinatorial radical cross-coupling of pCA or pCA esters, resulting in ligninhemicellulose connections and pCA dimers, cross-linking polysaccharide chains have not been reported in planta (Ralph, 2010), although pCA can be involved in photoinduced formation of cyclodimers with FA Hartley, 1989, 1990).
Because FA modifications occur on A side chains of xylan, A side chains are fundamental for the cross-linking of cell wall components. Members of the CAZY family GT61 have been described as a-1,3-arabinosyltransferases adding A side chains to the xylan backbone (Anders et al., 2012;Zhong et al., 2018), while others were described as b-1,2-xylosyltransferases for xylan (Chiniquy et al., 2012;Voiniciuc et al., 2015;Zhong et al., 2018) or for N-glycans (Bencur et al., 2005). Xylan arabinosyltransferases TaXAT1, TaXAT2, OsXAT2, and OsXAT3 add a-(1,3)-A side chains to xylan in planta and this catalytic activity has been confirmed for OsXAT2 in in vitro assays and for its homolog in Sorghum (Anders et al., 2012;Shao et al., 2020;Zhong et al., 2018). Interestingly, analysis of water-extractable xylan from wheat (Triticum aestivum) flour showed that the reduction of a-(1,3)-A in TaXAT1 RNA interference lines is accompanied by a reduction of bound FA (Freeman et al., 2017). A significant reduction in FA and pCA released upon base hydrolysis, along with enhanced saccharification, was also reported in xax1. In contrast to other GT61 xylan b-1,2-xylosyltransferases, XAX1 has been inferred to act on A side chains rather than the xylan backbone forming b-(1,2)-xylopyranosyl-a-(1,3)-arabinofuranosyl (X-A) side chains (Chiniquy et al., 2012), but it is unclear how changes in a minor side chain substitution result in a large decrease of bound FA and pCA.
Here, we investigated the significance of XAX1 in hydroxycinnamic acid modification of xylan and its impact on cross-linking. We established polysaccharide analysis by carbohydrate gel electrophoresis (PACE) and applied liquid chromatography-mass spectrometry (LC-MS) as methods for analyzing hydroxycinnamic acid-modified xylan oligosaccharides of wild-type (WT) rice mature leaves. We then compared their abundance in WT to that in the xax1 mutant. Our analysis showed that not only the X-(FA)-A structure was reduced in xax1, but surprisingly also hydroxycinnamic acid-decorated side chains which were not modified by a xylosyl residue like FA-A and pCAmodified a-L-arabinofuranosyl (5-O-p-coumaroyl arabinose; pCA-A) residues (for reference, Figure 1 shows the main structures and abbreviations used in this manuscript of the most abundant hydroxycinnamoyl-modified A structures found in the LC-MS analysis). This finding could be explained by XAX1 acting as hydroxycinnamoyl a-1,3arabinosyltransferase rather than b-1,2-xylosyltransferase. The observed changes in xylan side chain abundance in xax1 lead to a plethora of downstream changes in plant cell wall cross-linking, namely a reduction of diverse hydroxycinnamate-hydroxycinnamate and monolignolhydroxycinnamate coupling products. Figure 1. Abbreviations used in this study with their corresponding chemical structures, confidently assigned by MS n fragmentation in the phenolic profiling. Linkages in ferulate dimers are tentatively shown as 5-5, but 8-8, 8-O-4, or 8-5 linkages are equally likely. Compounds classified as hydroxycinnamic acid-modified A are shown in red, xylan cross-linking moieties in green, and xylan-lignin cross-linking moieties in blue; see Table S2 for trivial names and further information. Boxed-in structures contain a xylose residue from the xylan backbone.

PACE as a method to detect hydroxycinnamic aciddecorated xylan side chains
To analyze hydroxycinnamic acid modification of xylan in an easy and robust way, we first established PACE as a method using WT rice. Xylan substitutions decorated with hydroxycinnamic acid can be analyzed using mild trifluoroacetic acid (TFA) hydrolysis (Bowman et al., 2011;Saulnier et al., 1995;Wende and Fry, 1997). Mild TFA treatment leads to the preferential cleavage of arabinofuranosyl linkages over cleavage of the xylose backbone, removing A side chains from the xylan backbone, while largely preserving the ester-linked hydroxycinnamic acid decorations on A side chains. Alcohol-insoluble residue (AIR) from mature rice leaves, the tissue in which changes in xax1 were originally reported (Chiniquy et al., 2012), was analyzed in a time course of 0-100 min of mild TFA hydrolysis. To identify structures harboring the alkali-sensitive ester-linked hydroxycinnamic acid groups on A, TFA hydrolysis was followed by NaOH treatment. This treatment also removes acetyl groups present in rice xylan (Gao et al., 2017); however, removal of acetyl groups on xylan oligosaccharides results only in minor changes to the band positioning in PACE (Busse-Wicher et al., 2014).
As expected, mild TFA hydrolysis leads to the immediate release of large amounts of arabinose, co-migrating with the xylose (X 1 ) standard in the PACE gel ( Figure 2). Following 60 min of TFA hydrolysis, a ladder of NaOH-resistant oligosaccharides (presumably hydrolysis products of the xylan backbone) are observed, which co-migrate with the xylooligosaccharide standard X 1 to X 6 . Most importantly, from 60 min of hydrolysis time onwards, at least three major bands were apparent, which did not co-migrate with the xylooligosaccharide standard and were sensitive to NaOH treatment (Figure 2), suggesting the detection of hydroxycinnamic acid-modified A side chains by PACE.

Identification of feruloyl and p-coumaroyl-modified A side chains
To characterize the NaOH-sensitive structures, TFA hydrolysates were subjected to solid phase extraction (SPE) using C18 as stationary phase and elution with different ethanol concentrations and the fractions were analyzed by PACE (Figure 3). While the xylooligosaccharide ladder obtained from TFA hydrolysis elutes with 10% ethanol (Figure S1), the NaOH-sensitive bands eluted at ethanol percentages of 15-18%, indicating greater hydrophobicity. Fractions harboring the hydroxycinnamic acid-modified structures were selected for derivatization with procainamide hydrochloride (217 Da) and analyzed by matrixassisted laser desorption ionization (MALDI)-MS. MALDI-MS analysis allowed the detection of masses corresponding to procainamide hydrochloride-modified p-coumaroyl pentose (m/z 514 Da), feruloyl pentose (m/z 544 Da), and feruloyl dipentose (m/z 676 Da), as well as small amounts of p-coumaroyl dipentose (m/z 646 Da) ( Figure 3). Monosaccharide analysis of the 100% ethanol SPE eluate showed the presence of arabinose and xylose (Table S1). These data together with the published data on rice xylan structure and the reported effect of TFA hydrolysis on xylan led to the assignment of p-coumaroyl pentose as pCA-A, feruloyl pentose as FA-A, p-coumaroyl dipentose as X-(pCA)-A (2-O-xylopyranosyl-(5-O-p-coumaroyl) arabinose), and feruloyl dipentose as X-(FA)-A.
While MALDI is not a quantitative technique per se, it provides reproducible ratios between the peak intensities, allowing the assessment of relative quantities (Lerouxel et al., 2002). By comparing the relative peak intensities of the MALDI spectra of the four assigned structures in a specific fraction (Figure 3(a-d)) with the relative band intensities of the NaOH-sensitive bands of the correlating fractions in the PACE gel (Figure 3(e)), the NaOH-sensitive structures in the PACE gel can be assigned. FA-A and pCA-A migrate as doublet, which is indicative of their similarity in size and properties with pCA-A running slightly quicker in the PACE analysis. This is consistent with pCA-A eluting Time (min) NaOH Figure 2. Identification of NaOH-sensitive xylan hydrolysis products by PACE. PACE analysis of mild TFA hydrolysis products from rice leaf AIR at successive time points of hydrolysis (0-100 min). The hydrolysates were tested for the presence of ester-linked groups by comparing hydrolysis profiles in the absence (À) or presence (+) of NaOH. NaOH-sensitive bands are marked with an arrow. S: xylooligosaccharide standard X 1 -X 6 . Note that the accumulation of NaOH-resistant bands below X 3 that do not align with the xylooligosaccharide standard are most likely small non-xylan oligosaccharides released during TFA hydrolysis.
© 2021 The Authors. slightly before FA-A in the ethanol elution. The further modified X-(FA)-A band runs significantly more slowly in the PACE gel, while the low-abundance X-(pCA)-A could not be robustly detected by PACE. To support further the assignment of hydroxycinammic acid structures by PACE, the 100% ethanol SPE eluate was digested with a GH3 bxylosidase from Chaetomium globosum (CgGH3), which has been found to cleave b-(1,2)-linked xylose side chains (Tryfona et al., 2019). Of the three assigned NaOH-sensitive structures X-(FA)-A, FA-A, and pCA-A, only X-(FA)-A was sensitive to hydrolysis with GH3 xylosidase ( Figure S1), confirming its assignment.
Not only X-(FA)-A, but also FA-A and pCA-A xylan side chains are reduced in xax1 Chiniquy et al. (2012) reported that bound FA and pCA were approximately halved in the plant cell wall of mature rice leaves in xax1. To analyze how the reduction of FA and pCA relates to modification of the specific xylan side chains, i.e., A and X-A, quantitative changes of hydroxycinnamic acid-modified structures in WT and xax1 mutant rice were analyzed. AIR of mature leaves was subjected to mild TFA hydrolysis and analyzed by PACE, as established above, and the signal intensity of relevant bands was quantified. Four different time points of hydrolysis (180, 240, 330, and 450 min) were used to ensure that the comparison of WT and xax1 is not compromised by different rates of release and degradation of the structures of interest, particularly given that xax1 is reported to show increased extractability of sugars (Chiniquy et al., 2012). Consistent with the findings of Chiniquy et al. showing the loss of X-A side chains, X-(FA)-A is strongly reduced in xax1 ( Figure 4). FA-A and pCA-A could not be completely separated using PACE to quantify the bands separately; however, quantification of both bands together revealed a combined reduction in xax1 compared to WT. This

Complexity of hydroxycinnamic acid structures in WT and xax1 mutant rice
To achieve better resolution of the different hydroxycinnamic acid-modified xylan side chains and to analyze the downstream effects of the change in xylan structure on the cross-linking of xylan and lignin in the plant cell wall, hydroxycinnamic acid structures obtained from mild TFA hydrolysis of WT and xax1 AIR were analyzed using ultrahigh-performance LC (UHPLC) hyphenated to a Fourier transform ion cyclotron resonance mass spectrometer (FTICR-MS; Figure 5(a)). This allowed the profiling of up to 6008 compounds (see Experimental Procedures). Some of the compounds were structurally assigned using accurate mass (m/z) determination and MS n spectral elucidation (see Experimental Procedures). This led to the highly confident structural characterization of 32 of the most abundant compounds ( Figure 1, Tables 1 and S2). Although MS n data provided strong evidence for the structures of the 32 compounds, their full authentication needs either purification for NMR or chemical synthesis of reference compounds. In addition, based on the structures of these 32 compounds, the phenolic profiles were searched for masses corresponding to isomers and alternative structures that were assumed to be present (see Experimental Procedures). This yielded a list of 60 features of which the chemical formulae matched the mass of those searched structures (Table S2). However, as MS n spectra were absent or of low quality, these 60 features could be only tentatively assigned. Below, the 32 confidently characterized compounds are referred to by a shorthand name and a bold number, whereas only a shorthand name is mentioned for each of the 60 tentative compounds.
To unravel the differences in the cell wall phenolic profiles of WT and the xax1 mutant, the compound abundances were subjected to piecewise regression modeling including the genotype (WT versus xax1 mutants) and hydrolysis time, and their interaction as factors as shown in Figure 5(b); significant abundance changes due to genotype were observed for 29.5% of the 6008 compounds (see Experimental Procedures). Remarkably, when considering the 92 (confident and tentatively) annotated compounds, which are mainly hydroxycinnamoyl arabinose derivatives, the abundances of 77 of them (83.7%) were significantly affected by the genotype, indicating that the main impact of xax1 is on hydroxycinnamoyl arabinose derivatives.
To pinpoint similarly behaving compounds, the regression-fitted mean abundances were subjected to hierarchical cluster analysis (HCA). This yielded six main clusters ( Figure 5(c) and Table S2): Cluster A and F were unaffected by genotype and comprise compounds with increasing (cluster A, 2/92 compounds) and decreasing (cluster F, 4/92 compounds) abundance with hydrolysis time. Clusters B, C, D, and E are all genotype-dependent: while cluster E (3/92 compounds) is the only cluster with an increase of abundance in the xax1 mutant, clusters B (11/92 compounds), C (19/92 compounds), and D (52/92 compounds) all show a reduction of compound abundance compared to WT, with cluster B and C showing additional dependency of compound abundance with hydrolysis time (increased in cluster B and decreased in cluster C). Finally, one of the 92 compounds was affected by genotype, yet did not belong to one of the main clusters (Table S2).

Hydroxycinnamic acid-decorated A xylan side chains are reduced in xax1
Consistent with the results from the PACE analysis, abundant compounds detected are hydroxycinnamic acidmodified A: pCA-A 8 and 9, FA-A 12 and 15, X-(FA)-A-X 10, X-(FA)-A 13 and 16, and FA-A-X 18. All of these compounds are reduced in xax1, confirming that hydroxycinnamic acid modifications on arabinosyl substitutions are affected, but importantly, independent of the presence of the additional side chain xylose residue ( Figure 5(b)). X-(pCA)-A, a structure detected by MALDI-MS, is not among the highly   (Table S2) and they are also reduced in xax1. The MS 2 spectrum recorded for one of them was indicative for the presence of two pentose moieties, supporting an X-(pCA)-A or pCA-A-X structure with the latter compound containing a xylose from the xylan backbone, which is expected to be sensitive to mild acid hydrolysis. Because a hydrolysis timedependent change in abundance was observed for only one of the isomers, one isomer is tentatively annotated as pCA-A-X and the other as X-(pCA)-A. A targeted search also revealed two isomers for each of the X-(pCA)-A-X/   Observed means are shown in black, fitted means obtained following piecewise linear regression are shown in red. Error bars, n = 10. Note that the retention time is chromatogram-specific rather than the median retention times as shown in Table S2. (c) Heatmap obtained via hierarchical cluster analysis of the fitted means using a distance matrix based on Pearson correlations. Main clusters A-F are indicated on the right side of the heatmap. Color in the heatmap represents the row-based z-score varying between À2 (blue) and 2 (yellow).
© pCA-A-X-X, in addition to X-(FA)-A-X 10 and two other X-(FA)-A-X/FA-A-X-X isomers (Table S2). Interestingly, a third hydroxycinnamic acid-modified A, i.e., 5-O-caffeoyl arabinose (CA-A 4), was detected and also reduced in abundance in xax1. The gas phase fragmentation of the anion of CA-A is shown in Figure S2. This finding indicates that A residues, although mainly decorated with FA and pCA, can also be decorated with CA. Similar to the FA and pCA derivatives, a targeted search revealed one CA-A isomer and X-(CA)-A-X/CA-A-X-X.
Moreover, one abundant compound, also reduced in xax1, was elucidated as 5-O-feruloyl arabinose oxalate   ester (FA-A-Ox 6). The gas phase fragmentation of the anion of FA-A-Ox is shown in Figure S3. Searching for other hydroxycinnamoyl arabinose oxalate esters based on the accurate mass, a putative caffeoyl derivative (CA-A-Ox) was detected, yet no p-coumaroyl arabinose oxalate ester could be pinpointed (Table S2). Taken together, we detected a reduction of diverse hydroxycinnamoyl arabinose derivatives, revealing interesting caffeoyl and oxalyl modifications on arabinosyl side chains, all of which were reduced in xax1.

Cross-linking of xylan via diferulate linkages to arabinoses is reduced in xax1
The phenolic profiling revealed additional highly abundant compounds in WT, which were characterized as radicalradical coupling products between hydroxycinnamic acids and/or hydroxycinnamic acid-modified A (FA-FA-A 20, FA-FA-A-X 21, and A-FA-FA-A 25 and 26; Table 1 and  Table S2, Figure 1). A targeted search allowed the detection of many more isomers of radical-radical coupling products between hydroxycinnamic acids and/or hydroxycinnamic acid-modified A (Table S2). In case of diferulates, five and four additional isomers of FA-FA-A 20 and FA-FA-A-X 21/A-FA-FA-A 25 and 26 could be traced, respectively. In contrast, only two and three putative dicoumarates and dicaffeates were observed, indicating that diferulate linkages are more abundant and more structurally diverse than either dicoumarates or dicaffeates in cross-linking of xylan. Except for A-FA-FA-A 26, all other ferulate dehydrodimers, i.e., FA-FA-A 20, FA-FA-A-X 21, and A-FA-FA-A 25, were reduced in abundance in xax1, suggesting that cross-linking of xylan chains is reduced in the mutant.

Cross-linking of lignin monomers via ferulic acid to arabinose is reduced in xax1
Some of the abundant compounds in WT that we could structurally elucidate (Table 1) were radical-radical crosscoupling products between FA-A and the lignin monomer coniferyl alcohol (G(8-O-4)FA-A 22-24 and G(8-5)FA-A 27).
A targeted search revealed four additional G-FA-A isomers and seven H-FA-A isomers (i.e., products arising from the coupling of p-coumaroyl alcohol with FA-A) that were much less abundant (Table S2) (Table S2). The latter two compounds, representing cross-coupling products of the lignin dimer G(8-O-4)G with FA-A, were too low abundant to obtain clear MS 2 fragmentation data, yet the presence of a pentose moiety could still be inferred from their MS 2 spectra. Similar to FA-A, CA-A seems to enter coupling reactions with coniferyl alcohol as well:  (Table S2). In addition, a radical-radical coupling product of tricin, which is both a flavone and a lignin monomer in monocots (Lan et al., 2015(Lan et al., , 2016del Rio et al., 2012), and feruloyl arabinose (i.e., T-FA-A 32) was present (Table 1). Tricin explicitly couples via its 4ʹ-O-position (Lan et al., 2015), suggesting that tricin is coupled to FA-  (Table S2). Cross-coupling products between tricin, coniferyl alcohol, and FA-A seemed also to be present based on the targeted detection of five isomers (Table S2). The comparison of the WT with xax1 revealed that clusters C and D, with decreased compound abundance in xax1, comprised feruloyl arabinose linked to either coniferyl alcohol or tricin, i.e., G(8-O-4)FA-A 22, 23, and 24, G(8-5)FA-A 27, and T-FA-A 32, all of which represent cross-coupling moieties between xylan units and lignin monomers (Table 1). Cluster D also included some lignin dimers, i.e., the dilignols H(8-O-4)T 29 and G(8-O-4)T 30 and 31 (Table S2). These latter two compounds in cluster D suggest that the xax1 mutation also affects tricin-comprising dilignols.
Taken together, the cell wall-based phenolic profiling revealed a large variety of radical-radical coupled feruloyl arabinose-derived structures, supporting the presence of feruloyl-involved linkages between hemicelluloses and lignin. Most of these structures are reduced in xax1, highlighting the complex downstream effects of the xax1 mutant. The results are consistent with the xax1 mutation leading to a reduced abundance of hydroxycinnamic acidmodified A side chains on xylan, and this leads to altered xylan cross-linking and reduced xylan coupling to lignin.

DISCUSSION
Chiniquy et al. reported that in the xax1 mutant of rice, X-A side chains of xylan are absent and the amount of bound hydroxycinnamic acids FA and pCA in the plant cell wall is reduced (Chiniquy et al., 2012). Their results suggested that XAX1 is a xylan side chain xylosyltransferase, and that absence of X-A side chains in xax1 leads to the reduction of cell wall-bound FA and pCA observed in xax1. Here, we studied the hydroxycinnamic acid-modified xylan side chains and their cross-linking products prepared from xax1 mutant cell walls. Unexpectedly, the mutation causes a general defect in hydroxycinnamic acid modifications of A side chains of xylan. An alternative and simpler explanation for the xax1 cell wall phenotype is that XAX1 is a glycosyltransferase that catalyzes the transfer of hydroxycinnamic acid-modified A to the xylan backbone.
The reduction in FA-A, pCA-A, and CA-A xylan side chains in xax1 is hard to explain with the role currently attributed to XAX1 as a b-1,2-xylosyltransferase. If XAX1 were a b-1,2-xylosyltransferase, either an increase or no change in abundance might be expected to occur in hydroxycinnamic acid-modified A side chains lacking the b-(1,2)-linked xylose ( Figure 6, models a and b). Our results in contrast showed a substantial decrease in abundance of these structures. Unexpectedly, none of the abundant cross-linking structures described in our study possess the xylose branch, even though uncross-linked FA-A and X-(FA)-A may be similarly abundant (Table 1, Figures 4 and  6). This result does not support the proposal that the xylose branch protects the hydroxycinnamic acid substitutions from esterases (Chiniquy et al., 2012). Finally, the quantity of released FA-A and X-(FA)-A was similarly reduced in the xax1 mutant (Table 1, Figure 4), indicating there is no specific reduction in the xylosylated branches. The proposition of XAX1 acting as a xylosyltransferase was supported by the detection of microsome xylosyltransferase activity upon transient expression in tobacco (Nicotiana benthamiana) leaves (Chiniquy et al., 2012). Low activity was detected by measuring incorporation of radioactivity from UDP-[ 14 C]-xylose into an endogenous acceptor. However, the endogenous acceptor in this assay for XAX1 activity is unclear, as 3-linked arabinosyl side chains, the postulated substrate of XAX1, have not been described in dicot leaves.
Apart from XAX1, all so far characterized GT61 Clade A family members have been described to add either 2-linked xylose or 3-linked arabinose to the xylan backbone itself (Anders et al., 2012;Zhong et al., 2018). The more general decrease in arabinose sugars decorated with hydroxycinnamic acids found in the xax1 mutant therefore could be explained by XAX1 acting as an a-1,3-arabinosyltransferase tolerating hydroxycinnamic acid modifications on an UDP-Larabinofuranose (UDP-Araf) substrate (Figure 6, model c). The reduction of both xylosylated and non-xylosylated hydroxycinnamoyl arabinose side chains in xax1 can be explained with this model. Indeed, all the highly abundant cross-linking structures described in our analysis did not contain the xylosyl modification. Moreover, these nonxylosylated hydroxycinnamoyl arabinose cross-linking structures were reduced in xax1. If XAX1 is an a-1,3-arabinosyltransferase, UDP-Araf-FA, UDP-Araf-CA, and UDP-Araf-pCA are the potential substrates. These type of sugar donors have not been identified in planta, however their presence was suggested based on the finding that several BAHD enzymes (Benzylalcohol O-acetyl transferase, Anthocyanin O-hydroxycinnamoyl transferase, anthranilate N-Hydroxycinnamoyl/benzoyltransferase, Deacetylvindoline 4-O-acetyltransferase) have been described to be involved in the incorporation of hydroxycinnamic acids into xylan (Piston et al., 2010, Bartley et al., 2013, Buanafina et al., 2016, de Souza et al., 2018, de Souza et al., 2019. BAHDs are thought to be cytosolic and therefore the acyl transfer is believed to occur in the cytosol and not in the Golgi lumen ( Figure 6; D' Auria, 2006;Mnich et al., 2020). UDP-Araf is synthesized by UDP-arabinopyranose mutase (UAM) in the cytosol as well (Konishi et al., 2007;Rautengarten et al., 2011) and could be ferulated or p-coumaroylated by BAHD enzymes and the hydroxycinnamic acid-modified UDP-sugar transferred into the Golgi (Buanafina, 2009;Chateigner-Boutin et al., 2016;Hatfield et al., 2016;Rennie and Scheller, 2014) by an unknown transporter or by the UDP-Araf Transporter (UAfT). Challenging experiments such as in vitro enzyme assays and identification of the predicted modified sugar nucleotides are needed to investigate whether XAX1 indeed utilizes UDP-Araf-FA, UDP-Araf-CA, and UDP-Araf-pCA substrates to modify xylan.
The fact that hydroxycinnamic acid-modified A and X-A side chains are not completely absent in our analysis of xax1 suggests redundancy between the GT61s of Clade A enzymes. Extensive phylogenetic analysis of GT61s in monocots revealed that XAX1 belongs to the Poalesspecific subgroup GT61-A7, which has only two other members in rice, Os06g27560 and Os01g02910 (Cenci et al., 2018), which could have a similar function to XAX1 in tolerating hydroxycinnamic acid on the UDP-Araf substrate. Interestingly, FA, pCA, and CA modifications of xylan appear not equally affected in the xax1 mutant, which might hint to a preference for specific hydroxycinnamic acid modifications on the donor substrate: pCA-A 8 and 9 (0.32-and 0.35-fold change) are much more reduced in xax1 than either CA-A 4 (0.67-fold change) or FA-A 12 and 15 (0.75-and 0.76-fold change; Table 1). However, in contrast to pCA-A, many derivates of FA-A were present. Therefore, when comparing the abundance changes of the FA-A derivate pool with those of pCA-A, an unequal effect of the xax1 mutation on at least the FA and pCA modifications of xylan cannot be unambiguously determined.
We did not detect additional more complex structures in the phenolic profiling, such as the a-(1,2)-linked Lgalactopyranosyl sugar of X-(FA)-A, which have been previously detected in leaf tissues from several grasses (Wende and Fry, 1997). This might be due to their lower abundance or due to their reduced hydrophobicity, leading to weaker interactions with the C18 stationary phase. However, the detection of minor amounts of galactose in the monosaccharide analysis of the SPE eluent fractions (Table S1) might suggest the presence of such structures. However, we detected a structure, which most likely shows an Ox modification of the A side chain of xylan in the plant cell wall. Ox modifications were suggested to exist in the plant cell wall based on polysaccharide acyltransferase activity detected in vitro (Dewhirst and Fry, 2018).
The changes in xylan side chain biosynthesis facilitated by XAX1 result in enhanced saccharification and extraction of xylan in xax1 hinting to altered interactions of xylan in the plant cell wall, while the cellulose content and the size of xylan molecules remained unchanged (Chiniquy et al., 2012). Our study shows that enhanced saccharification is due to lowered cross-linking through FA as summarized in Figure 6. Except for A-FA-FA-A 26, all diferulates detected and all molecules derived from a lignin monomer coniferyl alcohol cross-coupled to 5-O-feruloyl arabinose (G-FA-A) were less abundant in xax1. In addition, dicoumarates, dicaffeates, and all putative G-CA-A structures were reduced in xax1 (Table S2).
Xylans are linked via radical-radical cross-coupling of the ferulate moieties present on the xylan A side chains to lignin and other xylan chains. The cross-coupling versatility is illustrated by the detection of a wide variety of diferulate isomers, i.e., 8-O-4', 8-8', 8-5', 5-5', and 5-O-4', with the 5-5'-linkage being the most frequent (Bento-Silva et al., 2018;Ralph, 2010;Ralph et al., 2004;Waterstraat and Bunzel, 2018). Our findings further substantiated this through the detection of (i) different linkages, i.e., 8-O-4' and 8-5', in the G-FA-A cross-coupling products, (ii) lignin units other than G units (H units) attached to FA-A, (iii) crosscoupling products between ferulates and minor lignin monomers such as tricin, i.e., T-FA-A 32, and (iv) detection of dicaffeates and several G-CA-A isomers, representing dicaffeates as potential novel lignin-xylan bridging structures. CA is abundant in plants as an intermediate of lignin biosynthesis; to our knowledge it has not been described to modify xylan in rice, but it was identified in LC-MS analysis as one of the phenolic acid compounds bound to xylan in Kodo millet (Bijalwan et al., 2016).
In studies with synthetic lignins and monocot lignins (Ralph, 2010;Ralph et al., 1995Ralph et al., , 2019, monolignols preferentially couple with their 8-position to ferulate esters yielding mainly G(8-O-4)FA, G(8-5)FA, and G(8-8)FA moieties. The strong prevalence of the 8-O-4-linkages in the here characterized G-FA-A cross-coupled structures are consistent with the 'end-wise' polymerization process of lignification, which postulates that monolignols slowly diffusing into the lignification zone will preferentially couple with an existing polymer rather than with another monomer (Ralph  , 2004). As the hemicellulose network is laid down before lignification starts, the first monolignols that enter the cell wall will likely couple with the FA-A moieties that are linked to the hemicellulose network (Grabber et al., 2002). Most of these cross-couplings lead to the formation of the coniferyl alcohol coupled via its 8-position to the phenol function (4-O position; the two most abundant peaks associated with G-FA-A products in Figure 5 Apart from laccases and peroxidases, which have been characterized as key factors in the spatio-temporal regulation of lignin polymerization (Tobimatsu and Schuetz, 2019), the xylan structure with the presence or absence of FA-A, pCA-A, or CA-A side chains will direct lignification as well, the latter possibly regulated by the activity of XAX1 and its homologs.
Our data show that the GT61 XAX1 is a key player in the synthesis of xylan hydroxycinnamoyl arabinose moieties. We therefore propose an alternative role for the XAX1 glycosyltransferase in that it directly transfers FA-A, pCA-A, and, interestingly, also CA-A side chains to xylan. As a result of this activity, the levels of various hydroxycinnamate-hydroxycinnamate and monolignol-hydroxycinnamate coupling products are depleted in xax1. Furthermore, the characterized coupling products between lignin, FA, and A suggest that XAX1-dependent FA-A moieties on xylan initiate lignin polymerization. All of this emphasizes the importance of XAX1 in establishing both xylan-xylan and lignin-xylan cross-links.

Plant material and growth conditions
The rice GT61 (O. sativa L. ssp. japonica) transfer DNA (T-DNA) insertion mutant seeds for XAX1 were generously provided by Dr. Pamela Ronald and Dr. Henrik Scheller (Lawrence Berkeley National laboratory, US). Rice GT61 T-DNA insertion mutants were genotyped using the primer sequences described in (Chiniquy et al., 2012). Segregating WT was used as control in all experiments.
Rice dehulled seeds were surface-sterilized by shaking in a 3% bleach solution at 1200 rpm for 10 min. Seeds were rinsed four times with autoclaved milliQ water. The seeds were germinated at 30°C on a humidified filter paper and then transferred to Silica Sand. Once the plants reached the tillering developmental stage, they were transferred to a 1:1 Silica Sand/soil mixture. Plants were grown at 28°C (day)/23°C (night), 60% humidity, 60% ambient CO 2 , and 12 h light (400 lmol/m 2 ) and kept under constant watering conditions, supplemented weekly with a 5009 diluted Poliverdol â solution and 0.1 g L À1 of Sequestrene Rapid â .

Alcohol-insoluble residue preparation
Tissue of mature rice leaves was harvested in 96% ethanol and incubated in a water bath at 70°C for 30 min. After cooling down, the tissues were ball-milled using a mixer mill MM400 at 20 rotations per second, three times for 5 min each. The homogenized tissue was centrifuged and the pellet was subjected to several washes to remove proteins, lipids, and other non-polysaccharide components. This procedure included washing the pellet with 100% ethanol, followed by overnight treatment with 2:3 (v/v) of methanol: chloroform and a sequential set of washes with 100, 65, 80, and 100% ethanol. The pellet obtained after these washes corresponded to the AIR, which was dried and stored at room temperature.

Mild acid hydrolysis
Mild acid hydrolysis was performed on AIR samples using 50 mM TFA, taking the appropriate measures to prevent TFA evaporation. This procedure was carried out at 100°C in a heat block, for different time intervals, as indicated in the respective experiments. A TFA hydrolysis time of 240 min was chosen for the analysis by MALDI-MS in order to reach maximal release of the hydroxycinnamic acid-modified structures. Samples were cooled down on ice and centrifuged and the supernatant was collected. The pellet was washed and the supernatant was pooled with the supernatant collected previously. Supernatant samples were dried in a vacuum concentrator overnight at room temperature.

Solid phase extraction with C18 Cartridges
Sep-Pak C18 Classic Cartridges were washed with 100% ethanol, milliQ water, and 100% acetonitrile to remove potential contaminants (Ishii, 1991). The cartridge was then equilibrated with 20 mM ammonium acetate buffer (adjusted to pH 7 with acetic acid). Lyophilized samples obtained from mild TFA hydrolysis were resuspended in 20 mM ammonium acetate buffer (pH 7) and loaded on the cartridge. The C18 Cartridges were washed with milliQ water and 10% ethanol and the sample was eluted with 5 ml of increasing percentages of ethanol of 15, 18, and 20%. Four fractions were collected for each ethanol elution. Full elution of the bound oligosaccharides was achieved with a single 100% ethanol wash. Fractions were dried in a vacuum concentrator at 55°C.

Enzyme hydrolysis
Solid phase extracts were hydrolyzed using xylosidase GH3 from C. globosum (Tryfona et al., 2019), a generous gift from Novozymes (CgGH3, NS39127). Enzyme was added at a final concentration of 2 lM and digestions were carried out overnight at room temperature in 0.1 M ammonium acetate buffer (pH 5.5) under constant shaking. Enzymes were inactivated at 100°C for 10 min and the sample was dried in a vacuum concentrator at 45°C.

Alkali treatment
Solid phase extracts were alkali-treated with 4 M NaOH for 1 h at room temperature to cleave ester linkages. Samples were subsequently neutralized to pH 7 with 1 M HCl and the samples were dried in a vacuum concentrator at 45°C. After SPE and washing the cartridges with milliQ water and 10% ethanol, the samples were eluted with 100% ethanol. Monosaccharide detection was performed by high-performance anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD) as described in .

Polysaccharide analysis by carbohydrate gel electrophoresis
Samples and xylooligosaccharide standard X 1 -X 6 were derivatized by reductive amination with 8-aminonaphthalene-1,3,6-trisulfonic acid (ANTS). The derivatization reaction was carried out with a mix containing 5 µl of ANTS, 5 µl of 0.2 M of 2-picoline-borane (2-PB), and 10 µl of a freshly prepared buffer containing 30 ll of acetic acid, 170 ll of milliQ water, and 200 ll of DMSO. Derivatization was carried out at 37°C overnight. Derivatized samples were lyophilized in a vacuum concentrator at 60°C and resuspended in 100 µl of 3 M urea to solubilize the labeled oligosaccharides. Samples were loaded on polyacrylamide gels and electrophoresed at 10°C at 200 V for 30 min, followed by 1000 V for 1 h and 40 min. A 0.1 M TRIS-borate (pH 8.2) solution was used as the running buffer. Labeled products on the PACE gel were detected using a G-Box CCD camera with a transilluminator with long-wave tubes emitting at 365 nm. Images were captured using GeneSnap software.
Quantification was carried out on three biological replicates of WT and xax1 leaf AIR using Genetools software. Aliquots of 20, 40, and 100 picomol of a mix containing ANTS-labeled xylose, xylobiose, and xylotriose were loaded on the gel. These standards were used to build a calibration curve in Genetools. This procedure allowed the software to determine the mol quantities of the oligosaccharides in PACE bands by linear regression. Significance of the differences detected was assessed at the 5% level using Student's t-test.

Mass spectrometry
SPE was carried out on mild acid hydrolysates as described above, with a C18 SPE Cartridge with 1 g of packing to allow for the purification of higher amounts of sample. Eluted ethanol fractions were dried in a vacuum concentrator at room temperature. Samples were resuspended in water and manually spotted on a MALDI plate with procainamide hydrochloride as described previously (Lavanant and Loutelier-Bourhis, 2012). Procainamide hydrochloride was used as a derivatizing comatrix to add mass to the products of the C18 fractions, which are expected to be considerably small and could have their masses obscured by matrix peaks in the MALDI spectrum. After air-drying, the sample spots were overlaid with 1 ll 2,5-DHB matrix (10 mg ml À1 in 50% aqueous methanol) and analyzed by MALDI-ToF/ToF-MS on an AB-Sciex 4700. The MS spectra were acquired with an average of 10 000 laser shots per spectrum in positive ion mode (mass range 250-1500 Da). The observed ions are protonated.

Liquid chromatography-mass spectrometry
Mild acid hydrolysis (50 mM TFA, 100°C) was performed on AIR samples from mature leaves of five biological replicates of WT and xax1 and for two technical replicates of each biological replicate. Three time points of mild acid hydrolysis (180, 240, and 330 min) were analyzed for each sample. The lyophilized samples were dissolved in 200 µl milliQ water, of which 10 µl was profiled by a reverse phase (Acquity UHPLC BEH C18 column, 2.1 9 150 mm, 1.7 µm; column temperature 80°C) Accela UHPLC system coupled to FTICR-MS (LTQ FT Ultra) via electrospray ionization (ESI) operated in negative ionization mode (spray voltage À3.5 kV, sheath gas 30 [arb], aux gas 10 [arb], capillary temperature 300°C). Other instrumental conditions were as previously described with minor modifications (Morreel et al., 2014). LC solvents were acidified with 0.1% formic acid. A reversed phase gradient was applied in which the proportion of acetonitrile increased from 1% (0 min) to 50% (30 min). LC-MS feature integration, feature grouping, and chromatogram alignment methods have been described previously (Morreel et al., 2014). In total, 41 542 m/z features were integrated that were further grouped into 6008 m/z feature groups. This is necessary as each compound is represented by multiple m/z features; hence, the number of m/z feature groups is an estimation of the number of profiled compounds. However, for many m/z feature groups, not all m/z features belonging to the same compound were included based on visual inspection of the chromatograms. Therefore, the number of m/z feature groups still represents an overestimation of the real number of profiled compounds. Nevertheless, the low number of data points across the peak of an m/z feature in an FT full MS spectrum hinders a more efficient deconvolution. Statistical analyses were performed on the most abundant m/z feature in each group, as this was the pseudo-molecular ion in most cases. The phenolic profiling results are described in terms of compounds rather than m/z feature groups.

MS-based structural elucidation
Structural elucidation of the negative ion MS n data was based on the accurately recorded m/z value of the compound (using either the base peak or the 13 C isotope peak dependent on which one served as precursor ion for MS n recording; see Table S2), previously published MS n data of phenylpropanoids, (neo)lignans and oligolignols (Morreel et al., 2014), and 5-O-feruloyl-L-arabinose (Quemener and Ralet, 2004), and knowledge of MS-based sugar cross-ring cleavages for various di-and oligosaccharides (Carroll et al., 1995;Dallinga and Heerma, 1991;March and Stadey, 2005;Mulroney et al., 1999;Quemener et al., 2006). A more in-depth description of the effect of the glycosidic bond on the cross-ring cleavages is given as Methods S1. Whenever possible, the aglycone moiety was verified via spectral matching to the MS n spectra of a standard compound. This set of structurally characterized compounds was then used to trace other isomers or structurally similar compounds in which the aglycone moiety was represented by another phenylpropanoid/monolignol using the R-based RDyn-Lib package (Desmet et al., 2021). For the latter approach, the presence of the putative compound was only based on the chemical formula that was computed from the accurately recorded m/z value.

LC-MS data mining
R version 3.2.3 (R Core Team, 2013) was used for all statistical analyses. For each m/z feature group, changes in the abundance of the selected m/z feature were modeled via piecewise linear regression (lm function). Missing data were replaced by an arbitrarily chosen threshold value (100). Technical replicates were averaged before modeling and all data were logarithmically (natural logarithm) transformed. Starting from a full model including the mild acid hydrolysis time, genotype, and the appropriate interaction terms, those terms showing the least significant Wald test P-value (P term > 0.05) and those that were not included in any higher-order interaction terms were iteratively removed. Model comparison was performed using the anova() function. Based on a P model < 0.001 criterion, 2480 m/z feature groups (i.e., 10092480/ © 2021 The Authors. The Plant Journal published by Society for Experimental Biology and John Wiley & Sons Ltd., The Plant Journal, (2021), doi: 10.1111/tpj.15620 6008 = 41.3% of the feature groups) showed a significantly changed abundance between samples that differed in the applied hydrolysis time and/or genotype. However, for two of these 2480 m/z feature groups, none of the model terms was significant. Of the remaining 2478 m/z feature groups, the abundances of 494 (8.2%), 707 (11.8%), and 1093 (18.2%) m/z features were due to the effect of the genotype (P genotype < 0.05), the hydrolysis time (P time < 0.05), and both the genotype and the hydrolysis time (P genotype < 0.05 and P time < 0.05). The abundances of only 184 (3.1%) m/z features were governed by a genotype 9 hydrolysis time interaction effect (P genotype 9 time < 0.05 and/or P genotype 9 knot < 0.05). Thus, 29.5% (8.2% + 18.2% + 3.1%) and 33.1% (11.8% + 18.2% + 3.1%) of the m/z feature groups were affected by genotype and hydrolysis time, respectively. HCA (hclust() function) was performed on the regressionderived fitted mean abundances for the presumed pseudomolecular ion from each of the m/z feature groups using the Pearson correlation matrix as distance matrix; a heatmap was returned with the heatmap.2() function (gplots package). HCA yielded six main clusters (A-F). HCA clustering results were much improved by including the regression-fitted mean values rather than the raw data-based mean abundances. Nevertheless, sometimes isomeric compounds were classified into two different HCA clusters because the curvature in the raw data-based feature abundance profile of only one of them reached the significance threshold during regression modeling and, hence, was retained in the model. For example, FA-A 12 and FA-A 15 are likely the Z and E isomers of 5-O-feruloyl arabinose, which are mutually converted due to UV light. Thus, although their abundances should be correlated, they are members of different HCA clusters, i.e., clusters D and C. Cluster D is comprised of compounds that have horizontal abundance profiles across the various TFA hydrolysis times but are higher in WT than in xax1. Cluster C is comprised of compounds that have abundances that decrease with hydrolysis time but are on average higher in WT than in xax1. The cluster D classification of FA-A 12 is evident from the red regression profile for this compound in Figure 5(b). However, the feature abundance profile of the raw data shows a decreasing trend with increasing hydrolysis time. Clearly, this decrease was not significant enough to classify FA-A 12 as a cluster C member. Figure S1. PACE analysis of NaOH-sensitive xylan hydrolysis products in SPE eluates. Figure S2. Gas phase fragmentation of the anion of caffeoyl arabinose (CA-A 4). Figure S3. Gas phase fragmentation of the anion of 5-O-feruloyl arabinose oxalate ester (FA-A-Ox 6). Table S1. Monosaccharide composition of WT AIR.