Genetics and physiology of cell wall polysaccharides in the model C4 grass, Setaria viridis spp

Setaria viridis has emerged as a model species for the larger C4 grasses. Here the cellulose synthase (CesA) superfamily has been defined, with an emphasis on the amounts and distribution of (1,3;1,4)-β-glucan, a cell wall polysaccharide that is characteristic of the grasses and is of considerable value for human health. Orthologous relationship of the CesA and Poales-specific cellulose synthase-like (Csl) genes among Setaria italica (Si), Sorghum bicolor (Sb), Oryza sativa (Os), Brachypodium distachyon (Bradi) and Hordeum vulgare (Hv) were compared using bioinformatics analysis. Transcription profiling of Csl gene families, which are involved in (1,3;1,4)-β-glucan synthesis, was performed using real-time quantitative PCR (Q-PCR). The amount of (1,3;1,4)-β-glucan was measured using a modified Megazyme assay. The fine structures of the (1,3;1,4)-β-glucan, as denoted by the ratio of cellotriosyl to cellotetraosyl residues (DP3:DP4 ratio) was assessed by chromatography (HPLC and HPAEC-PAD). The distribution and deposition of the MLG was examined using the specific antibody BG-1 and captured using fluorescence and transmission electron microscopy (TEM). The cellulose synthase gene superfamily contains 13 CesA and 35 Csl genes in Setaria. Transcript profiling of CslF, CslH and CslJ gene families across a vegetative tissue series indicated that SvCslF6 transcripts were the most abundant relative to all other Csl transcripts. The amounts of (1,3;1,4)-β-glucan in Setaria vegetative tissues ranged from 0.2% to 2.9% w/w with much smaller amounts in developing grain (0.003% to 0.013% w/w). In general, the amount of (1,3;1,4)-β-glucan was greater in younger than in older tissues. The DP3:DP4 ratios varied between tissue types and across developmental stages, and ranged from 2.4 to 3.0:1. The DP3:DP4 ratios in developing grain ranged from 2.5 to 2.8:1. Micrographs revealing the distribution of (1,3;1,4)-β-glucan in walls of different cell types and the data were consistent with the quantitative (1,3;1,4)-β-glucan assays. The characteristics of the cellulose synthase gene superfamily and the accumulation and distribution of (1,3;1,4)-β-glucans in Setaria are similar to those in other C4 grasses, including sorghum. This suggests that Setaria is a suitable model plant for cell wall polysaccharide biology in C4 grasses.


Background
Setaria viridis spp viridis (L.) Beauv.SETVI, variously known as wild millet, green foxtail, green millet or green bristlegrass, is the wild ancestor of foxtail millet (Setaria viridis spp italica SETIT), a widely grown staple grain crop that is prevalent in regions of China, Korea, Japan and India [1]. Both types of millet are found in the Panicoideae subfamily of the order Poales, in the bristle clade of the tribe Paniciae [2]. This "bristle grass" clade includes the economically important C 4 food crops maize, sorghum, sugarcane and other types of millet, together with species specifically grown as biofuel feedstocks such as switchgrass (Panicum virgatum) and Miscanthus. Setaria viridis spp, collectively referred to as Setaria here, is a self-compatible diploid with a small genome of around 515 Mb but, consistent with its status as one of the most prevalent weeds on the planet [3], it is also small in stature, has a very rapid life cycle of 6-9 weeks and is capable of producing more than 10,000 seeds per plant [4]. Once an Agrobacterium-mediated transformation system became established [4,5] it became clear that Setaria would make an excellent model for the much larger, generally polyploid and therefore genetically more complex and intractable C 4 grasses, and it has been rapidly adopted in this role [6]. In the last few years reference genome sequences of Setaria spp. have been released [7,8] and are accessible from public databases exemplified by Phytozome [9]. Large collections of Setaria accessions have been gathered from geographically diverse and ecologically distinct regions of the world. These have facilitated association mapping, allele mining and transcriptomic analysis of traits related to abiotic stress tolerance [10,11], C 4 evolution and photosynthesis [12][13][14], domestication events [15,16] and biomass production [17]. This explosion of Setaria-related resources recently prompted the establishment of the foxtail millet Marker Database (FmMDb) [18]. Foxtail millet is globally the second-most consumed variety of millet behind pearl millet (Pennisetum glaucum), which has the major share of the market at 40 % [19]. Millets in general are seen as key crops in many developing countries, where they are cultivated on marginal agricultural land in areas of low rainfall. In these regions, more common cereal crops are not able to grow and millets in their many forms, which include over 100 wild and cultivated species, provide the majority of the energy and protein needs for millions of people, particularly those in sub-Saharan Africa and parts of Asia [20,21]. In developed countries there is also a resurgent interest in less common grains such as millets because, unlike wheat and related cereals, they contain no gluten. The nutritional composition of generic millets has been defined as being high in starch and protein, although there are relatively low levels of the essential amino acid lysine, and they also contain significant amounts of dietary fibre, calcium and polyphenols [19,22]. Of particular interest here is the content of non-starchy polysaccharides in green foxtail, but there are only a few reports where these have been examined. In the grain of some cereals there are appreciable levels of polysaccharides, or dietary fibres, which are particularly valuable to human health. These are mainly located in the bran and the endosperm tissues where they are major components of cell walls, which, upon consumption in human diets, undergo hydrolysis and fermentation in the lower digestive tract. Fermentation products include short chain fatty acids, which protect against intestinal disorders and/or colorectal diseases [23,24]. In cereals, the two key polysaccharides in walls of cereal grains are arabinoxylan and (1,3;1,4)-β-glucan [25,26].
The composition of plant cell walls and the polysaccharides embedded therein has a fundamental effect on human health, but also on the economics and efficient transition of plant biomass to biofuel. Current bioethanol production systems involve the harvesting and conversion to fermentable sugars of lignocellulosic biomasses, either in the form of residues arising from agricultural practices or from purpose-grown crops [27,28]. The C 4 grasses maize, sorghum, switchgrass and Mischanthus feature prominently in the suite of dedicated bioethanol crops due to attributes such as high yields, growth on marginal lands and drought tolerance [27,[29][30][31]. In general C 4 grasses consistently produce higher yields of biomass compared with C 3 species such as rice, wheat and barley, which are primarily grown as food crops [32]. The raw material that is harvested from the residues of biomass C 4 crops used for lignocellulosic biofuel production is largely comprised of plant cell walls. There is considerable variation in the composition of the walls in different C 4 plants and within the individual tissues of these plants [33]. Cellulose is generally the most abundant component of the cell wall in vegetative tissues; it consists of a linear polysaccharide comprised of (1,4)-linked β-glucosyl residues that are readily fermentable once they have been liberated from the polysaccharide. Also present are heteroxylans, which contain (1,4)-linked β-xylosyl residues, together with a range of substituents that are distributed along the xylan backbone, and variable amounts of less abundant polysaccharides that include mannans, pectins and xyloglucans. Relative to hexose sugars, pentose sugars released from these polysaccharides are fermented more slowly. (1,3;1,4)-β-Glucans are present in varying amounts in C 4 plants, and whilst they are also a linear polysaccharide containing (1,4)-linked β-glucosyl residues, the inherent asymmetry provided by the insertion of (1,3)-linked βglucosyl residues renders the molecule more soluble than cellulose [34,35]. Given its more soluble nature and the relative ease with which hydrolytic enzymes can convert it to its component monosaccharides, (1,3;1,4)-β-glucans constitute an ideal source of extractable and fermentable glucose [36]. Although the levels of (1,3;1,4)-β-glucans are generally low in most biomass sources, it has been shown that considerable natural variation exists and that increased levels of the polysaccharide can be engineered through standard genetic manipulation procedures [37,38]. Deconstructing the cell wall to access the monosaccharides, glucose in particular, can be expensive and complex because of the recalcitrant nature of cellulosic biomass [39][40][41]. The presence of lignin contributes to this recalcitrance, because lignin forms an interlinked network around the polysaccharide components. The accessibility of cell wall polysaccharides to hydrolytic enzymes is an important economic consideration in the conversion of biomass to bioethanol. The hydrolysis of different polysaccharides often requires pre-treatment of the biomass and at least three different types of enzymes, and the cost of these steps may be a large proportion of the overall budget of the ethanol production process [42].
The general cell wall composition of S. viridis has been analysed and compared with the other C 4 crops maize, sorghum and switchgrass [17]. The major components of the cell walls such as cellulose, lignin and neutral sugars have been reported. However, neutral sugars were used as a collective measurement of 'hemicelluloses' and specific levels of (1,3;1,4)-β-glucans were not quantitated [17]. Here, the phylogeny of gene families comprising the cellulose synthase superfamily in Setaria is defined, with a focus on the genetics and transcription of cellulose synthase-like genes that encode the synthases responsible for the production of (1,3;1,4)-β-glucan. The amount, structure and distribution of (1,3;1,4)-β-glucan in the tissues of S. viridis is reported, enabling this species to be assessed in the context of the value of its grain to human health and its utility as a model for other C 4 grasses in a biofuel context.

Results
The cellulose synthase superfamily in Setaria Cellulose synthase genes (CesA) The orthologous relationships of the CesA genes between Setaria italica, barley (Hordeum vulgare), sorghum (Sorghum bicolor), Brachypodium distachyon and rice (Oryza sativa) are represented on a Bayesian phylogenetic tree (Additional file 1: Figure S1). The analysis indicates that there are 13 cellulose synthase (CesA) genes in total in S. italica, which revises the previous estimate of eight reported by Petti et al. [17]. The genes are spread across five of the nine chromosomes (Additional file 2: Table S1) and they have been numbered according to their barley orthologues [43]. There are single representatives of CesAs 1, 3, 4, 5, 8 and 10 genes with no corresponding orthologues of the barley CesA7 or CesA9 genes [44]. There are four and three genes that are very closely related to HvCesA2 and HvCesA6, respectively. SiCesA2-1 is on chromosome 4, whereas SiCesA2-2 and SiCesA2-3 are located on chromosome 2 and SiCesA2-4 is on chromosome 9. Each of the three CesA6 paralogues are distributed on different chromosomes; SiCesA6-1 on chromosome 5, SiCesA6-2 on chromosome 4 and SiCesA6-3 on chromosome 3.

Cellulose synthase-like genes (Csl)
In S. italica a total of 35 cellulose synthase-like (Csl) genes have been identified, a sub-set of which are displayed on the Bayesian phylogenetic tree in Additional file 1: Figure S1. There are eleven CslAs and six CslCs (data not shown), which represent the more basal clades of this superfamily, together with five CslD, two CslE, seven CslF, two CslH and two CslJ genes (Additional file 1: Figure S1). Of the cellulose synthase-like gene families restricted to the Poaceae, Setaria is unlike barley, rice or sorghum with respect to CslHs. It has two members of this family whilst barley and Brachypodium have one, and rice and sorghum both have three (Fig. 1). Setaria also carries two distinct CslJ genes whereas all other grasses analysed to date have either none or just a single CslJ gene (Fig. 1). As expected, there are no representatives in the dicot-specific CslB and CslG families.
The CslF family, with only seven members, is smaller than the barley family, which has eleven members [45,46]. Like barley there are no equivalent Setaria orthologues to the CslF1 and CslF2 genes of rice [45] and there are single genes corresponding to CslF4, 6, 7, 8 and 9 (Fig. 1). Genes matching barley CslF10, 11, 12 and 13 [46] are also absent, but there appears to have been a recent duplication of the CslF3 gene in Setaria with two copies arranged as a tandem repeat on chromosome 2 (Additional file 2: Table S1). This is on the end of a gene cluster that also includes SiCslF4, 8 and 9.
All of the SiCslD genes are found on different chromosomes, and the two SiCslH genes are also separated onto chromosomes 7 (SiCslH1) and 1 (SiCslH2). However, the small sets of SiCslE and SiCslJ genes possibly arise from recent duplication events because they are present as adjacent pairs on chromosomes 2 and 3, respectively (Additional file 2: Table S1).

Transcript profiling of cellulose synthase-like genes
The genome sequence of the model plant Setaria is publicly available and is largely accounted for by reads from S. italica [8]. This was used as a reference in order to identify putative cellulose synthase-like sequences in the closely related S. viridis [8]. Primers were designed to the predicted 3' untranslated region of the S. viridis genes of the CslF, H and J clades (Additional file 2: Table S2) and used to examine the transcript levels by real-time quantitative PCR (Q-PCR) across a S. viridis tissues series (Fig. 2).
Within the CslF family SvCslF6 transcripts were present at relatively high levels in all tissues examined. Transcript levels were highest in RNA from stem internode 4, at around one million copies (Additional file 1: Figure S2). The transcripts of SvCslF8 and SvCslF9 were the next most abundant in tissues that have not reached full maturity, such as leaves and younger stem internodes (Additional file 1: Figure S2). However, in the more mature root and stem tissues, SvCslF4 transcripts were present at levels above those of SvCslF8 and 9 (Additional file 1: Figure S2). SvCslF6 and SvCslF4 transcripts were relatively high in the grain development series, where they started at high levels at 2 DAP but dropped rapidly to very low levels by 6 DAP (Fig. 3, Additional file 1: Figure S2).
Close examination of the transcript levels of the two SvCslH genes showed that SvCslH1 transcripts were significantly higher in leaf and spike tissues, which were still growing at the time of harvest. In contrast, SvCslH2 transcripts were high in older stem internodes and in mature roots (Additional file 1: Figure S3). This suggests there may be a negative correlation between the abundance of the two transcripts (Additional file 1: Figure S4). A positive correlation was observed between the transcript levels of SvCslF4 and SvCslH2 across vegetative tissues   Figure S4). Transcript levels of both SvCslH1 and 2, and CslJ2 were negligible during grain development (Additional file 1: Figures S3 and S5). The QPCR data indicated that SvCslJ2 transcripts were present at significantly high levels in the mature root and the older inflorescence, second in magnitude only to SvCslF6 (Additional file 1: Figure S5).
Abundance and fine structure of (1,3;1,4)-β-glucan in vegetative tissues The amount of (1,3;1,4)-β-glucan and the ratio of cellotriosyl to cellotetraosyl units (DP3:DP4 ratio) in the vegetative tissues of S. viridis were quantitated at different stages of plant development from vegetative-leaf to reproductive-floral stages (Additional file 1: Figure S11). . T-test compared differences within the same group. Tissues with the same letter indicate non-significant differences (P > 0.05); different letters indicate significant differences (P < 0.05) The amount of (1,3;1,4)-β-glucan changed during plant development and generally decreased as tissues matured (Fig. 4a). For example, the amount in the flag leaf decreased from 1.9 % (w/w) at the vegetative-leaf stage to 0.2 % (w/w) at anthesis (Fig. 4a). Similarly, the amount in the primary (seminal) root decreased from 0.9 % (w/w) at the early reproductive-floral stage to 0.5 % (w/w) at anthesis (Fig. 4a), whilst in the inflorescence, the amount decreased from 1.3 % (w/w) to 0.7 % (w/w) over the same time frame (Fig. 4a).

Distribution of (1,3;1,4)-β-glucan in the stem
The S. viridis stem has a typical monocot anatomy, with a sclerenchyma cylinder of vascular bundles embedded in chloroplast-containing mesophyll tissue [47], and this was evident in toluidine blue-stained sections (Fig. 7a, c). In fluorescence micrographs, there was stronger (1,3;1,4)-βglucan labelling in the sclerenchyma fibre cells than in the surrounding mesophyll cells of the stem rind area (Fig. 7b,  d). In the inner pith area, the labelling indicated an even distribution of the polysaccharide in all cell walls, including in the two rings of vascular bundles and in the ground tissue cells (Fig. 7b). TEM micrographs showed that there was more (1,3;1,4)-β-glucan in the walls of ground tissue cells compared with those of the vascular bundles (Fig. 7, Additional file 1: Figure S7). Values are means ± standard deviation measured using the Megazyme assay (n = 3) and HPAEC-PAD. d Grain with different superscript letters is significantly different (P < 0.05) GenStat 15 th Ed. SP2 Distribution of (1,3;1,4)-β-glucan in the root The anatomy of the main root of S. viridis was represented in a toluidine blue-stained section (Fig. 8a). Fluorescence micrographs of immunolabelled roots indicated that (1,3;1,4)-β-glucans were distributed evenly across all cell types except in the walls of endodermal cells (Fig. 8b). However, the more sensitive TEM and immunogold labelling procedures revealed the presence of some (1,3;1,4)-β-glucans in the walls of the endodermal cells (Fig. 8g, Additional file 1: Figure S8).
Distribution of (1,3;1,4)-β-glucan in Setaria grain The phenotype of the S. viridis grain during development was observed from 2 DAP to 24 DAP (Fig. 3). Three different stages, namely young (4 DAP), intermediate (10 DAP) and mature (24 DAP), were sectioned to map the distribution of (1,3;1,4)-β-glucan (Fig. 9). Fluorescence micrographs of Alexa Fluor® 488 and Calcoflour-White MR2 labelling represented the anatomical structure of cell walls in the grain in transverse sections (Fig. 9a-f ), which was consistent with previous reports of S. viridis anatomy by Winton and Winton [48] and of Setaria lutescens by Rost [49]. A higher magnification micrograph of the outer layer of the young grain displayed a developing pericarp consisting of different layers, namely the cuticle, cross cells, tube cells, testa and nucellus (Additional file 1: Figure S9A). The pericarp was fully developed in the intermediate grain (Additional file 1: Figure S9B). The pericarp surrounds the endosperm, which was comprised of aleurone cells and the starchy endosperm, and the scutellum of the embryo (Additional file 1: Figure S9B). Higher magnification micrographs  Fig. 9d-f) showed the typical embryonic structure of Setaria grain. The scutellum forms a cuplike structure that surrounds the axis of the coleoptile and coleorhiza [49,50]. Fluorescence micrographs indicated that (1,3;1,4)-β-glucans were most abundant in the pericarp and embryo of the younger grain (Fig. 9g, j) particularly in the coleoptile and coleorhiza. Lower levels of (1,3;1,4)-βglucans were present in the intermediate embryo (Fig. 9h, k) and (1,3;1,4)-β-glucans were almost completely absent in the pericarp and the mature embryo of the oldest grain (Fig. 9i, l). However, the use of TEM confirmed the presence of some (1,3;1,4)-β-glucans in cell walls where they were not clearly detected by fluorescence immunolabelling, such as in the embryo of the mature grain (Fig. 10a) or in the scutellum of the young and intermediate grain (Additional file 1: Figure S10). Labelling of (1,3;1,4)-β-glucan in the aleurone walls of the young (Fig. 10b), intermediate (Additional file 1: Figure S10) and mature grain (Fig. 10c) was also observed. In line with the fluorescence results, (1,3;1,4)-β-glucan in the walls of the pericarp was detected throughout development using TEM (Fig. 10d-f, Additional file 1: Figure S10).

Discussion
Genes in the Setaria cellulose synthase superfamily There are at least 48 genes in the cellulose synthase superfamily in Setaria ( Fig. 1 and Additional file 1: Figure S1), which is consistent with most other land plants. Of these, 13 are CesAs, five more than the eight previously reported by Petti et al. [17] and similar to numbers in other cereals in the Poales, where barley and rice have nine [43,44,51], The CesA genes are distributed over five of the nine chromosomes and do not appear to cluster (Additional file 2: Table S1, Additional file 1: Figure S1). In Setaria, CesA4 and CesA8 resolve as the ancestral lineages, while the phylogeny suggests that CesA5 was the next group to split, followed by a large diverse clade that contains genes corresponding to CesA1 and CesA3, CesA2 (4 genes) and CesA6 (3 genes). The CesA10 gene is grass specific and belongs to a highly divergent group that has lost the QXXRW catalytic motif [52]. It is not clear why there are so many closely related CesA genes in Setaria and why the CesA2 and CesA6 genes, associated with synthesis of cellulose in primary cell walls [43], have duplicated more than once in recent evolutionary history. The larger CesA gene family will make the identification of encoded proteins that participate in the formation of the classical heteromeric terminal rosette complexes, and the definition of their levels of redundancy, particularly difficult to unravel. The proposed participation of two sets of three different CesA genes in barley is based on co-transcription in tissues where primary or secondary wall synthesis predominates [43]. No such correlation could be detected in our transcript profiles from many tissues of Setaria (Additional file 1: Figure S2).
Setaria lacks a few members of the CslF family including orthologues of the most recently identified members in barley, namely CslF10, 11, 12 and 13, and the CslF1 and CslF2 genes of rice [45,46]. Not withstanding the 'missing' genes, the seven CslFs are organised across the genome akin to the family in other grasses with a syntenic main cluster on chromosome 2 ([52], Additional file 2: Table S1) and the remaining members are scattered on individual chromosomes. The exception are the two closely related genes SiCslF3-1 and SiCslF3-2, found next to each other on chromosome 2, at one end of the CslF Setaria has two genes in each of the CslH and the CslJ families, while sorghum and rice have three CslH genes, sorghum has a single CslJ (Fig. 1) and barley has a single member of both families [52,54]. The two SiCslH genes are separated on chromosomes 1 and 7 (Additional file 2: Table S1). This is the first report of a cereal that carries two CslJs where the second one derives from a gene duplication event because the two copies share 87.3 % identity at the amino acid level.

The SvCslF6 gene is highly transcribed
The CslF6 transcript is dominant in the majority of tissues of most C 3 and C 4 cereals and the corresponding CSLF6 protein is likely to be the major synthase driving the deposition of (1,3;1,4)-β-glucan [37,38,45,[55][56][57]. Setaria is not an exception to this, since the CslF6 transcript is found at the highest levels of all the CslF, H and J genes in vegetative tissues at all stages of development examined here. For example, in the elongating fourth stem internode SvCslF6 transcripts peaked at around one million copies per microliter and in tissues where expansion has ceased, such as the mature root (Additional file 1: Figure S2) approximately 100,000 copies of SvCslF6 per microliter were detected. The temporal and spatial occurrence of (1,3;1,4)β-glucan observed is similar to those described for other C 4 grasses, including maize and sorghum [58][59][60][61]. In general the amount of (1,3;1,4)-β-glucan in Setaria decreases during maturity in roots, leaves and the inflorescence and levels vary significantly according to the age of the tissues (Fig. 4a).
The physiological role of (1,3;1,4)-β-glucan in plants is debated in the literature [59,[62][63][64][65]. The decrease in the amount of (1,3;1,4)-β-glucan in leaves, roots and stem sections as they age indicates that it may be more important in developing tissues than mature ones; in the roots it decreased from 0.9 % w/w at inflorescence emergence to 0.5 % (w/w) at anthesis (Fig. 4a) possibly linked to tiller production, as grasses stop developing their primary (seminal) roots when mature tillers become established [66]. Similar trends have been reported for other C 4 grasses. In maize, sorghum and barley (1,3;1,4)-β-glucan is synthesised in seedlings and coleoptiles, leaves and stems when elongation starts and ceases when elongation stops [60,61].
The properties of (1,3;1,4)-β-glucans may be linked to their fine structure, as determined by the DP3:DP4 ratio [67,68]. A lower ratio is generally associated with higher solubility of the polysaccharide and this property is of interest in a human health context [69] and may affect ease of extraction for bioethanol production from lignocellulosic feedstocks [37,70]. There is evidence that individual CSLF enzymes synthesize (1,3;1,4)-β-glucans with different fine structures [36,71]. Subtle but significant differences in the (1,3;1,4)-β-glucan DP3:DP4 ratios in a range of different Setaria tissues were measured. Where (1,3;1,4)-β-glucan may be structurally important, such as in the actively growing parts of the plant (primary roots and stem internode 4, the ratio is relatively high. However, in tissues where cell growth slows or ceases and the remobilisation of carbon reserves is likely to occur, the ratio and solubility of (1,3;1,4)-glucans might be expected to be lower. This was observed in some tissues where remobilisation might be occurring, for example in mature stem internodes, but the correlation was not consistent.
The activity of CslH genes could significantly influence the amount and nature of (1,3;1,4)-β-glucan and thus the overall cell wall properties. Transcript levels of the barley HvCslH1 and the sorghum SbCslH3 genes were reported to be higher in mature tissues, such as the leaf tip, relative to young tissues [54,61]. This trend was also observed for the Setaria CslH2 transcript. The peak transcripts of the CslH gene have already been linked to the presence of (1,3;1,4)-β-glucan with a higher DP3:DP4 ratio, and thus lower solubility, suggesting that these synthases might be producing polysaccharides to strengthen or reinforce walls of cells that are undergoing secondary thickening. However, in this species transcript levels of SvCslH1 are higher in a completely different set of tissues to SvCslH2, such that the two transcripts show a pronounced negative correlation, whilst SvCslH2 transcripts are one of only a few that show  Figure S4). Thus, the influence of different CslH genes on (1,3;1,4)β-glucan properties is not clear. It has been observed previously that CslF transcript levels do not appear to be regulated in a coordinated pattern [45], in contrast to the CesAs where transcript co-expression is observed [43]. This may indicate the absence of a complex containing more than one CSLF, CSLH or CSLJ protein from the same family. In Setaria only the transcript patterns of SvCslF4 and SvCslH2 are highly correlated. SvCslF4 could be a relatively important member of the CslF family in Setaria, given that transcript levels of this gene were second only to SvCslF6 in a number of the more mature tissues, such as early stem internodes and mature roots (Additional file 1: Figure S2).
A specific role for CslJ genes in (1,3;1,4)-β-glucan synthesis has not yet been demonstrated unequivocally. The presence of CslJ genes across the Poales is patchy and these genes are absent from a number of important cereals, including rice [72]. Setaria is the only cereal so far examined that has more than a single CslJ gene. Transcripts of SvCslJ2 were higher than SvCslJ1 in all tissues tested and peaked in the mature root and the almost fully emerged inflorescence (Additional file 1: Figure S5). Our unpublished data (RA Ermawar, RA Burton and NJ Shirley) indicate that the sorghum CslJ is also found at high levels in root tissue suggesting that some cereal roots might require a particular cell wall structure conferred by the action of CslJ.
Fluorescent immunolabelling in the very early stages of Setaria grain development, just a few days after pollination, indicated that the (1,3;1,4)-β-glucan is predominantly located in the outer layers of the caryopsis in the maternally-derived pericarp tissues (Fig. 9g) and is present in specific areas of the embryo, the coleoptile and the coleorhiza (Fig. 9j). It persists in the pericarp in the intermediate stages of grain development but the amount decreases in the embryo and is negligible in walls of the starchy endosperm, or aleurone cells of mature grain (Fig. 9k). By the time the grain is mature almost no (1,3;1,4)-β-glucan can be observed with sensitive antibodybased immunocytochemical methods, consistent with the results of the biochemical analyses.
In developing sorghum grain, (1,3;1,4)-β-glucan is deposited more in the outer maternal layers, such as the pericarp and testa, than in the endosperm tissues [61]. Sorghum grain very often displays dual morphology in the starchy endosperm where a floury centre is surrounded by a more glassy outer layer [76]. Although this morphological feature is not evident in Setaria grain the accumulation of (1,3;1,4)-β-glucan in both C 4 grains is very similar. In contrast, the appearance of (1,3;1,4)-βglucan in developing barley grain, a C 3 species, is detected as early at 3 or 4 DAP in the maternal tissues, in the walls of starchy endosperm cells from 6 DAP onward [77] and continues to increase as the grain matures.
Here, an examination of the transcript levels of the CslF, H and J genes across a Setaria developing grain series reveals that minimal amounts of these transcripts can be detected, except for SvCslF4 and SvCslF6 in the very early stages at 2 to 4 DAP (Fig. 3, Additional file 1: Figure S2). This is similar to the CslF6 transcript in wheat, which peaked before 10 DAP and continuously decreased throughout maturity [56]. In all other cereals significant amounts of CslF6 or CslH transcripts are observed, where for barley [78], rice [79] Brachypodium distachyon [80] and other C 4 grain such as sorghum [61], the transcript peaked in the mid stages and much larger amounts of (1,3;1,4)-β-glucan are present. Also absent from the Setaria grain transcriptome is CslF9. In barley, the HvCslF9 transcripts peak during endosperm cellularisation in early grain development, whilst maximal amounts of HvCslF6 are found later at 18-20 DAP. Hence Setaria is somewhat unusual amongst the Poales in terms of its transcript profile in developing grain and until data from other millet species such as Pennisetum (pearl), Panicum (proso) and Eleusine (finger) are available it will not be clear whether Setaria is an outlier in this Order.

Conclusion
The genes in the cellulose synthase superfamily have been defined in Setaria and the analyses reveal subtle differences in numbers and chromosomal distributions when compared with other cereal genomes. Transcript analyses of Poales-specific genes in the CslF, CslH and CslJ families indicate that the CslF6 transcript dominates and that the abundance and distribution of (1,3;1,4)-βglucans in Setaria vegetative tissues are very similar to those described for the other C 4 grasses such as maize and sorghum [59][60][61]. These features indicate that Setaria it is a suitable model for these and other grasses in the context of manipulating this valuable polysaccharide [37,70]. However, Setaria grain contains only a very small amount of (1,3;1,4)-β-glucan, and the CslF6 transcript, and so may be less useful in a human health context than other cereal grains [26,69].

Bioinformatic analysis of Csl genes in Setaria
The CesA and Csl sequences of Setaria italica, sorghum (Sorghum bicolor), barley (Hordeum vulgare), rice (Oryza sativa) and Brachypodium (Brachypodium distachyon) were obtained from the public databases Phytozome 9.0 [9], NCBI GenBank [81], RGAP [82] and GRAMENE [83]. The hmmalign program in the HMMER package [84] was used to assign the retrieved full length protein sequences to the cellulose synthase pfam HMM (PF03552). The full length protein alignment was back-translated to codons using pal2nal [85] and residue assignments with a posterior probability of <0.6 were manually removed. The Bayesian Markov Chain Monte Carlo (MCMC) package BEAST v1.8.0 [86] was used to reconstruct the phylogenetic tree of Setaria, sorghum, barley and rice Csl and CesA genes. The final codon alignment was partitioned into the three separate codon positions, and each partition was unlinked, that is, substitution model parameters, rate heterogeneity model and base frequency were free to vary across partitions. Two analyses, replicated once, were reconstructed; one with a relaxed-clock (log-normal distribution of nucleotide rate variation), and one with a strict clock prior. The GTR + IG substitution model, as selected by jModelTest, and a Yule tree prior were used. Convergence was monitored in TRACER v1.5 [87] by assessing the Effective Sample Size (ESS) and likelihood of the estimated parameters. Each analysis was run for at least 75,000,000 states, or until stationarity, logging every 1000 states. CesA, CslD, and CslE clades on the tree were collapsed using nw_utils [88].

Plant material
The S. viridis A10 accession was grown in a greenhouse under a day/night temperature regime of 28°C/15°C (The Plant Accelerator, Waite Campus, University of Adelaide, South Australia). Plants were grown in a soil mix composed of coco peat (75 %) and sand (25 %) supplemented with 2.5 g L −1 agricultural lime and Osmo-cote® Exact® Mini (Scotts Australia Pty Ltd, NSW, Australia), 1.875 g L −1 iron sulphate and calcium nitrate, 1 g L −1 hydrated lime, 0.75 g L −1 dolomite lime, gypsum, superphosphate and MicroPlus (Langley Australia Pty Ltd, WA, Australia) and 0.125 g L −1 iron chelate. Soil pH ranged from 6-6.5. S. viridis plant development was defined based on growth stages of perennial forage grasses [89]. Tissues such as root, leaves, stem internodes and inflorescence were collected at various stages from vegetative-leaf (Additional file 1: Figure S11A) through to the reproductive-floral development stage (Additional file 1: Figure S11B). The flag leaf was harvested at three different stages: vegetative-leaf, inflorescence emergence (IE) and at anthesis. The 2 nd to 4th leaves were harvested from tillers at IE, while the 5 th and 6 th leaves were harvested from the plant at anthesis. Roots were harvested at IE and at anthesis. Stem internodes were harvested at anthesis. Whole grains were collected throughout development and at the ripening stages, from 2 to 24 DAP (Fig. 3). Glumes were removed manually using tweezers (Dumont, no.3, Kirwan, Queensland). Grain images (Fig. 3) were taken using the Stemi 2000 Stereomicroscope (Zeiss, Germany).
Total RNA isolation and cDNA synthesis S. viridis vegetative tissues and developing grain were snap frozen in liquid nitrogen immediately after harvest. Total RNA extraction and cDNA synthesis methods followed Burton et al. [38].

Q-PCR of the S. viridis CslF, CslH and CslJ genes
Real-time quantitative PCR (Q-PCR) of the S. viridis genes was performed as described by Burton et al. [45]. Primer design for the Cellulose synthase-like genes in S. viridis (SvCsl) and the control genes was based on S. italica sequences [Phytozome 9.0, 9] and tested on a cDNA sample generated from RNA from various tissues. The normalised expression values for the SvCsl genes were subjected to a correlation analysis and groups of genes sharing Pearson correlation coefficients (R 2 ) greater than or equal to 0.90 were ascribed as 'coexpressed'. The PCR primers and PCR product sizes in base pairs together with optimal acquisition temperatures are listed in Additional file 2: Table S2.
(1,3;1,4)-β-Glucan quantitative assay Samples of each S. viridis vegetative tissue were harvested from five individual plants. Vegetative tissues were freeze-dried, while mature grain samples were oven-dried at 37°C for two days. All samples were ground using a Retsch mill (Type MM 300) for 1 min at a frequency of 30 beats/second. The ground samples were weighed (15-20 mg) and assayed in four technical replicates. The samples were pre-treated with a series of 70-100 % ethanol washes before measurement using a small scale version of the Megazyme assay [90] according to Ermawar et al. [61]. The amount of glucose released from the (1,3;1,4)-β-glucan in the vegetative samples was quantified based on the colorimetric assay using GOPOD, Megazyme International Ireland Ltd., according to Ermawar et al. [61]. Due to the low level of (1,3;1,4)-β-glucan in the grain, the amount of glucose released from the lichenase digestion was quantified using high pH anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD) on a Dionex ICS-5000 column against a glucose standard.
Statistical analysis of (1,3;1,4)-β-glucan quantitative and structural data GenStat 15 th Ed. SP2 was used for statistical analysis. All data were initially analysed by one-way ANOVA using a 5 % least significance difference. Data was analysed with a t-test to identify values with a significant difference (P < 0.05).
Tissue fixation, embedding and sectioning of vegetative tissues and grain S. viridis vegetative tissue and grain samples were fixed in paraformaldehyde/glutaraldehyde and embedded in LR-white as described in Burton et al. [38]. Sections for toluidine blue staining, fluorescence and TEM microscopy were prepared following Wilson et al. [77]. Individual vegetative tissue and grain samples in resin blocks were transversely sectioned at 1 μm on a Reichert-Jung Ultracut microtome for toluidine blue staining and fluorescence microscopy. Sections for TEM were cut transversely at a thickness of 80-100 nm.

Fluorescence and TEM immunocytochemistry
Detection of (1,3;1,4)-β-glucan in S. viridis vegetative tissues and grain was carried out using fluorescence and TEM immunocytochemistry following Burton et al. [38] and Wilson et al. [77]. These methods rely on the use of the BG1 monoclonal antibody that is specific for (1,3;1,4)-β-glucans [92]. In the present study, dilutions of 1:50 and 1:500 of the anti-mouse primary antibody BG-1 (Biosupplies Australia, Parkville, Victoria Australia) were used for the fluorescence and TEM, immunocytochemistry. A 1:100 dilution of Alexa Fluor® 488 goat antimouse IgG (H + L) secondary antibody was used for the fluorescence detection, and a 1:30 dilution of an Aurion goat IgG/IgM anti-mouse secondary antibody conjugated to either 10 or 25 nm gold was used for the TEM. Fluorescent sections were counterstained with 0.001 % Calcofluor White MR2® (Sigma-Aldrich, St.Louis, MO). Three types of negative control of fluorescence micrographs from various tissues of Setaria were applied. The cell walls show no labelling when they are treated with no antibody, the primary antibody only or secondary antibody only. Fluorescence images were taken with an Axio Imager M2 Microscope (Zeiss, Germany). TEM images were taken as described in Wilson et al. [77]. Control transmission electron micrographs of H. vulgare leaf sections labelled with the antibody BG-1 was used as positive control, while the sections labelled with the secondary antibody only was used as negative control (Additional file 1: Figure S12).