Update on Mechanisms of Plant Cell Wall Biosynthesis: How Plants Make Cellulose and Other (1→4)-β-d-Glycans

CELLULOSE SYNTHESIS

In flowering plants, cellulose is a para-crystalline array of about two to three dozen (1→4)-β-d-glucan chains. Microfibrils of 36 glucan chains have a theoretical diameter of 3.8 nm, but x-ray scattering and NMR spectroscopy indicate that some microfibril diameters could be as small as 2.4 nm, or about two dozen chains (Kennedy et al., 2007). The microfibrils are synthesized at the plasma membrane by terminal complexes of six-membered “particle rosettes” that produce a single microfibril (Giddings et al., 1980; Mueller and Brown, 1980). Thus, each of the six components of the particle rosette is expected to synthesize four to six of the glucan chains, and 24 to 36 chains are then assembled into a functional microfibril (Doblin et al., 2002). In freeze fracture, the particle rosettes, found only on the P-face of the membrane, are about 25 nm in diameter, but this size represents only the membrane-spanning and short exterior domains (Fig. 2A). Hidden in surface views of rosette structures in the plasma membrane, the much larger catalytic domains of the cellulose synthases are estimated to be 50 nm wide and extend 35 nm into the cytoplasm (Bowling and Brown, 2008), a feature that has escaped consideration in many published models of the rosette structure (Fig. 2, B and C).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Particle rosette structures associated with cellulose synthesis in angiosperms. A, Freeze-etch images of the P-face of the plasma membrane showing clusters of rosettes associated with the developing of secondary wall spiral thickenings of a Lepidium tracheary element (from Herth, 1985). The inset shows the 6-fold symmetry of a single particle rosette from a Zinnia tracheary element developing in vitro (from C. Haigler, unpublished data, as seen in Delmer, 1999). A substructure can be observed in each of the particles. In these freeze-etch images, only the membrane-spanning domains and extracellular loops of the CesA proteins can be observed. B, Cytoplasmic structure (circled) underlying the rosettes in plasma membrane footprints (from Bowling and Brown, 2008). These structures always are at the terminus of a microfibril (arrow). Bar = 200 nm. C, A Markham rotational analysis of one of these shows the reinforcement of hexagonal shape with 60° rotational steps. All other angles of rotation cancel to circular (Bowling and Brown, 2008).

Cellulose synthase is an ancient enzyme (Nobles et al., 2001), and cellulose synthase genes in green algae are homologous to those of flowering plants (Roberts et al., 2002). The deduced amino acid sequences of CesAs share regions of similarity with the bacterial CesA proteins, namely the four catalytic motifs containing the D, DxD, D, Q/RxxRW that are highly conserved among those that synthesize several kinds of (1→4)-β-d-glycans (Saxena et al., 1995). The higher plant CesA genes are predicted to encode polypeptides of about 110 kD, each with a large, cytoplasmic N-terminal region containing zinc-finger (ZnF) domains, and eight membrane spans sandwiching the four U motifs of the catalytic domain (Fig. 1B; Delmer, 1999).

Further evidence for the functions of plant CesA genes in cellulose synthesis came from Arabidopsis mutants of three of the CesA genes involved in primary wall synthesis: the temperature-sensitive radial swelling mutant rsw1 (AtCesA1; Arioli et al., 1998), the dwarf-hypocotyl procuste mutant prc1 (AtCesA6; Fagard et al., 2000), and a stunted root phenotype with altered jasmonate and ethylene signaling (cev1) and ectopic lignification (eli1) mutant alleles (AtCesA3; Ellis and Turner, 2001; Caño-Delgado et al., 2003). Despite coexpression in the same cells and an expectation of redundancy, cellulose synthesis is impaired in each mutant. The same was observed with the irregular xylem mutants irx1, irx3, and irx5, which display a phenotype of collapsed mature xylem cells as a result of lowered cellulose content during secondary cell wall deposition (Taylor et al., 2000, 2003). A widely accepted hypothesis is that the AtCesA1, AtCesA3, and AtCesA6 proteins assemble to function in primary wall cellulose synthesis, while the AtCesA4, AtCesA7, and AtCesA8 proteins assemble to make secondary wall cellulose (Fig. 1A), with each member of the trio performing a nonredundant function in the complex (Taylor, 2008). Lack of one CesA prevents incorporation of the other two into the plasma membrane (Gardiner et al., 2003). However, at least some of the subunits are potentially interchangeable, as inferred by the dominant-negative inhibition of growth and primary wall thickness caused by constitutive expression of a mutated fra5 (irx3 allele) transgene (Zhong et al., 2003) and by the semi-dominant-negative phenotype observed in the heterozygous AtCesA3 mutant (Daras et al., 2009). AtCesA1 is essential for cellulose synthesis (Beeckman et al., 2002), whereas knockouts of AtCesA3 (Ellis and Turner, 2001; Caño-Delgado et al., 2003) and AtCes6 (Fagard et al., 2000) result in partially impaired synthesis but not in total inhibition. Desprez et al. (2007) indicated that the AtCesA2 and AtCesA5 proteins have partially redundant functions with AtCesA6.

A direct association of three distinct CesA polypeptides was demonstrated in vitro and by colocalization in vivo by Taylor et al. (2003). Domain-swap experiments with wild-type and mutant AtCesA1 and AtCesA3 proteins in their respective mutants resulted in dominant-positive and dominant-negative effects, consistent with both catalytic and C-terminal domains being important for function (Wang et al., 2006). Direct interactions of three distinct CesA polypeptides in vivo were shown by bimolecular fluorescence complementation (Desprez et al., 2007). Although some complementary pairs gave stronger fluorescence than others, both homodimers and heterodimers of AtCesA1, AtCesA3, and AtCesA6 are inferred. Wang et al. (2008) used pull-down experiments similar to those of Taylor et al. (2003) to show that these three primary wall CesA proteins interact. Furthermore, they showed that Triton-soluble microsomal preparations subjected to native PAGE gave an 840-kD complex and that null mutants, but not missense mutations, gave smaller 420-kD complexes (Wang et al., 2008). Atanassov et al. (2009) affinity trapped a ladder of complexes of CesA oligomers to about 700 to 730 kD. Consistent with the observations of Wang et al. (2008), only smaller oligomeric complexes of two of the CesAs are detected when the third is missing (Atanassov et al., 2009). Such an association of CesAs was indicated independently in yeast two-hybrid studies (Timmers et al., 2009).

DOES SYNTHESIS OF EACH (1→4)-β-d-GLUCAN CHAIN REQUIRE ONE OR TWO CATALYTIC POLYPEPTIDES?

After over four decades of study, the biochemical mechanism by which cellulose is made remains a mystery (Delmer, 1999; Saxena and Brown, 2005; Somerville, 2006; Guerriero et al., 2010), with only a few reports of cellulose synthesis in vitro with isolated membranes (Kudlicka and Brown, 1997; Lai-Kee-Him et al., 2002). For both cellulose and the related (1→4)-β-d-glycan, chitin, synthesis proceeds by the attachment of glucosyl residues to the nonreducing terminus of the acceptor glucan chain (Koyama et al., 1997; Imai et al., 2003). The simplest hypothesis is that each CesA polypeptide synthesizes a single glucan chain. In the Delmer (1999) model, the eight membrane spans form a channel through which a single glucan chain is extruded (Fig. 3A). This mode of synthesis comes with a very big steric problem for synthesis. To make a (1→4)-β-d linkage means that each glucosyl residue is turned 180° with respect to each neighbor. Thus, the O-4 position of nonreducing terminal sugar of the acceptor chain is displaced several angstroms upon addition of each successive unit (Fig. 3B). For the next glycosyl transfer to occur, the site of catalysis must move several angstroms within the protein, the acceptor chain must swivel 180°, or the catalytic or acid-base amino acids must toggle between two forms to account for the displacement. To overcome this steric problem, several models have proposed that two sites or modes of glycosyl transfer reside within the catalytic complex, so that disaccharide units are added iteratively (Carpita et al., 1996; Koyama et al., 1997; Carpita and Vergara, 1998; Buckeridge et al., 1999, 2001; Saxena et al., 2001) or that two polypeptides associate to form two opposing catalytic sites (Buckeridge et al., 2001; Vergara and Carpita, 2001). In either model, glycobiosyl units, or any even-numbered oligomeric units, are added to the nonreducing end to ensure that the (1→4)-β-d linkages are strictly preserved without inversion of substrate, active site, or terminus of the growing chain (Fig. 3C).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Models for cellulose synthase and the steric problem of making a (1→4)-β-d-glycosyl linkage. A, The first model of conformation for a single CesA protein subunit was proposed by Delmer (1999). Each CesA subunit must interact with other such subunits to form the synthase complex. The ZnFs, plant-specific conserved region (P-CR), and class-specific region (CSR) are potential interaction sites (figure modified from Delmer, 1999). B, The steric problem of synthesis is illustrated in top view and end view. Addition of a single glycosyl residue in a (1→4)-β-d linkage without rotation of the end of the chain or the active site of the synthase would result in movement of the O-4 several angstroms. C, The conceptual solution to the steric problem is a catalytic dimer of simultaneous glycosyl transfer to form a cellobiosyl residue to the O-4 position of the terminal glucosyl residue of the chain. Synthesis of an even number of units always maintains the acceptor position as the O-4 position as the chain is extruded. In the catalytic dimer model, if one of the sites is damaged, then the point of attachment becomes the O-3 position, which would maintain the point of attachment as the O-3, hence producing callose thereafter (after Buckeridge et al., 1999, 2001).

Despite the rationale for a two-site model of catalysis, biochemical evidence from other types of polysaccharide synthases indicate that a single polypeptide is sufficient. Hyaluronan (HA) is an unbranched polysaccharide composed of repeating units of (1→3)-β-d-GlcNAc and (1→4)-β-d-GlcUA (DeAngelis and Weigel, 1994). HA synthases exist in three distinct classes, with class I containing integral membrane proteins to transport the HA across the membrane (Fig. 4). Bacterial and mammalian HA synthases have been shown to contain both transferase activities in a single polypeptide (DeAngelis and Weigel, 1994; Yoshida et al., 2000; Williams et al., 2006). Such a finding argues that the synthesis of a single HA polymer requires only a single polypeptide.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Models of HA synthases. A, Class I synthases. B, Class II synthases (after Weigel and DeAngelis, 2007).

The crystal structure of a nonprocessive family 2 glycosyltransferase (GTs) sharing sequence similarity with a portion of the catalytic domain of a CesA, the Bacillus subtilis SpsA synthase, provided the first conformation of the active site and the role of the aspartyl residues in the positioning of the uridinyl group of a UDP-sugar (Fig. 5A; Charnock and Davies, 1999; http://www.pdb.org/pdb/explore/explore.do?structureId=1QGS). Charnock and colleagues (2001) argued that only a single site for a nucleotide-sugar substrate is accommodated within a single polypeptide of SpsA.

• Download figure
• Open in new tab
• Download powerpoint

Figure 5.

Crystal structures of type 2 glycosyl transferases. A, The SpsA synthase crystallizes as a monomer with a single binding site for UDP (Charnock and Davies, 1999; http://www.pdb.org/pdb/explore/explore.do?structureId=1QGS). B, The B. fragilis SpsA homologous polypeptide crystallizes as a dimer (http://www.pdb.org/pdb/explore/explore.do?structureId=3BCV), each with a single binding domain for UDP. C, Each member of the crystal dimer E. coli chondroitin polymerase has two UDP-GlcA- or UDP-binding domains (http://www.pdb.org/pdb/explore/explore.do?structureId=2Z86).

A CATALYTIC DIMER HYPOTHESIS

From the studies of the class I HA synthases and the SpsA crystal structure, it is inferred that a single polypeptide alone has all the features needed for synthesis. However, these features still do not address mechanistically the fundamental steric problem of synthesis of a repeating (1→4)-β-d-glucosyl linkage of the cellulose glucan chains. In fact, two recent studies demonstrate unequivocally that at least some HA synthases and homologs of SpsA synthase form dimers. An identical structure to the SpsA synthase is the 3BCV polypeptide from Bacteroides fragilis, which is also predicted to contain a single substrate-binding site. In contrast to SpsA synthase, the 3BCV protein crystallizes as a dimer, and each monomer possesses a bound UDP (Fig. 5B; http://www.pdb.org/pdb/explore/explore.do?structureId=3BCV). The dimerization occurs through the C-terminal regions, which appear to be flexible and, for this reason, were deleted from the crystal structure of the SpsA synthase (Charnock and Davies, 1999). This, to our knowledge, is the first direct evidence by crystal structure of a homodimer formed by GT2 proteins, but even more complicated structures are also found. For example, an Escherichia coli strain K4 chondroitin polymerase contains two “Rossmann fold-like” domains within a single polypeptide, each binding a UDP-GlcA or UDP, and it also crystallizes as a dimer, giving a total of four nucleotide-binding domains (Fig. 5C; http://www.pdb.org/pdb/explore/explore.do?structureId=2Z86). Although the fit is not exceptionally good, the CesA catalytic domain threads through the E. coli chondroitin polymerase best of all for known crystal structures of glycosyl transferases involving nucleotide sugars (D. Kihara, personal communication), and it is consistent with the suggestion by Brown and Saxena (2000) and Saxena et al. (2001) of a conformation within which the catalytic domain of a single CesA would allow the synthesis of cellobiose units of the chain within a single polypeptide.

The class II synthases contain two different types of GT2 modules but not the membrane-spanning domains (Fig. 4). One type of HA synthase possesses two repeats of the UDP-Glc and acceptor-binding domains, so the synthesis of the characteristic disaccharide of HA by a single synthase is rationalized (Jing and DeAngelis, 2000). However, a direct interaction of two synthases is inferred for HA synthesis to explain the finding that host cells harboring constructs in which each site is independently disrupted are still able to make HA (Jing and DeAngelis, 2000; Weigel and DeAngelis, 2007). Because the class I HA synthases both have activities on a single peptide does not preclude the possibility that the formation of homodimers of single isoforms of HA synthase is necessary for function, which, like the chrondroitin polymerase, would give four nucleotide sugar-binding sites per dimer.

Solving the steric problem aside, one must also ask if a channel of eight membrane spans proposed for the cellulose synthase is of sufficient size to extrude a (1→4)-β-d-glucan chain. The question about sufficient channel size was raised also with respect to the HA synthases by Weigel and DeAngelis (2007), who suggested that certain phospholipids required for activity possibly integrate with the membrane spans to widen the channel for extrusion. However, whether lipid interactions with a small number of domains would be significant is still in question. Callose synthases are about twice the size of CesAs and contain 16 membrane spans (i.e. double those of a CesA; Hong et al., 2001). Plasma membrane hexose and maltose transporters of prokaryotes and eukaryotes are homologous (Maiden et al., 1987), and virtually all of them contain a minimum of 11 to as many as 18 transmembrane spans per functional unit (Reifenberger et al., 1995; Pao et al., 1998; Sherson et al., 2000; Klepek et al., 2010).

The sensitivity of detergent-solubilized CesA complexes to dithiothreitol suggested to Atanassov et al. (2009) that disulfide bonds are involved in the coupling into larger complexes. Other features of the protein outside the region of catalysis, such as the ZnFs, which show high similarity to RING-finger domains that bind zinc in a “cross-brace” manner (Freemont, 2000), might function in the organization of these into the larger rosette structure. Kurek et al. (2002) proposed that CesAs are coupled through the ZnF domain in a redox-dependent manner, constituting the first step in the clustering of CesAs into rosettes. Moreover, the discovery that a thioredoxin-like protein associates with the CesA ZnF domain in a yeast two-hybrid screen led to the suggestion that reduction of the domains by oxidoreductases returns the CesAs to monomeric forms, which are directed to the ubiquitin-dependent turnover pathway (Kurek et al., 2002). The experimental herbicide CGA 325′615 blocks crystallization of the β-d-glucan chains into cellulose microfibrils, phenocopying the rsw1 swollen root tip (Peng et al., 2001). This phenotype can be abrogated completely in the presence of hydrogen peroxide, suggesting that the inhibitor blocks rosette assembly by enzymatic oxidation (Kurek et al., 2002).

If ZnF domains of two CesAs couple as part of the recruitment into rosette particles, the question remaining is how all the others interact to form a complete complex. Although not discussed specifically, the study by Kurek et al. (2002) presented data that full-length CesA proteins formed tetramers and even higher ordered pairings, while the ZnF domains were limited to coupling of a single pair. Timmers et al. (2009) showed that heterodimer interactions indicated by yeast two-hybrid analysis do not require the ZnF domains. Taken together, these data provide evidence that domains other than the ZnFs of the CesA participate in coupling reactions if two to three dozen CesAs or more are aggregated to form a rosette complex.

Comparison of CesA sequences suggests potential heterodimeric interaction domains within the catalytic domain. The initial scarcity of CesA protein sequences and the apparent variability within the so-called “hypervariable region” led to the assumption that this region was probably not essential in catalysis (Pear et al., 1996). However, it is now understood that these regions are well conserved across grass and dicot species with a distinct subclade structure. Potential protein-protein interactions through subdomains of this region containing conserved Cys residues, clusters of consecutive basic Lys and Arg residues, and clusters of acidic Asp and Glu residues form the basis of a class-specific region (Fig. 1B; Vergara and Carpita, 2001).

To test the catalytic dimer hypothesis, we expressed fusion proteins containing only the catalytic domain of Arabidopsis and maize (Zea mays) CesAs with affinity tags and observed that dimers and higher order aggregates collapse reversibly to monomeric forms by thiol-reducing agents (C. Rayon, A. Olek, L. Paul, and S. Ghosh, unpublished data). Because the ZnF was absent in these constructs, dimerization must occur through thiol-sensitive sequences in the catalytic domain. Such an interaction of CesAs to form homodimers or heterodimers solves the three basic problems of the single polypeptide-single polymer conundrum: (1) the steric problem is solved by coordinate synthesis and attachment of cellobiose units instead of monomers, preserving the integrity of the O-4 site of attachment at the nonreducing terminus of the chain; (2) a channel of 16 membrane-spanning domains is consistent with sugar transport and callose extrusion; and (3) the interaction produces two ZnF domains for recruitment of the catalytic dimer into rosette particles (Fig. 6). An exciting prospect is that conservation of space would be maintained if CesAs turn out to have a structure like the chondroitin polymerase dimers and function like class II HA synthases, because a CesA catalytic dimer with four nucleotide-binding domains would be capable of generating two (1→4)-β-d-glucan chains instead of just one. Further experiments are needed to establish preferred heterodimer interactions, the stoichiometry of UDP-Glc binding, and the role of the ZnF in recruitment of the catalytic dimers into larger complexes.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

A catalytic dimer hypothesis for cellulose synthase. A, A catalytic dimer model of two CesAs to form a complex that synthesizes a single (1→4)-β-d-glucan chain. Homodimerization or heterodimerization of CesAs gives mirrored active sites that generate cellobiosyl units, which are then attached to the nonreducing end of the extruded glucan chain. Dimerization also results in a channel composed of 16 membrane-spanning domains, equivalent to that of callose synthase and consistent with eukaryotic monosaccharide transporters. B, Dimerization results in two ZnF domains that are now able to couple two neighbors instead of just one. C, Six such complexes interact to constitute one particle of the six-particle rosette.

THE BIOLOGICAL SYNTHESIS OF CELLULOSE

Cellulose synthase has a half-life of less than 30 min, remarkably short for a membrane protein (Jacob-Wilk et al., 2006). Assembly of rosettes occurs in the Golgi stacks, and they must be continually secreted to the plasma membrane to maintain cellulose synthesis (Haigler and Brown, 1986). Additional proteins are suspected to be necessary for the formation of primers of polymer synthesis, metabolic channeling of substrates, crystallization of the chains, and termination of chains (Doblin et al., 2002; Somerville, 2006; Guerriero et al., 2010). Furthermore, a proteomics survey of plasma membrane proteins shows that certain CesAs are phosphorylated at several locations within the catalytic and N-terminal domains (Nühse et al., 2004). Modification of some potential phosphorylation sites with amino acids that either prevent (Ala) or mimic (Glu) phosphorylation has multiple effects that reduce either synthesis rates or interactions with the microtubule cytoskeleton independently (Chen et al., 2010).

In affinity-labeling experiments with [³²P]UDP-Glc, an 84-kD polypeptide was found to be associated with a plasma membrane fraction containing the highest activity of callose synthase (Delmer et al., 1991) and subsequently was identified as Suc synthase (SuSy). Confirmation of plasma membrane association was made immunocytochemically (Amor et al., 1995), and Delmer and Amor (1995) proposed that the association of SuSy represented a UDP-Glc delivery mechanism to cellulose synthase. β-Glucan microfibrils are synthesized from Suc and UDP on immobilized tobacco (Nicotiana tabacum) plasma membrane sheets (Hirai et al., 1998). More recently, SuSy was immunologically associated with CesA proteins in the rosette structures (Fujii et al., 2010), strengthening the idea of an association of SuSy directly with cellulose synthases for metabolic channeling of Glc through a localized pool of UDP-Glc. However, a quadruple mutant that eliminates all detectable SuSy in vegetative tissue does not impair cellulose synthesis (Barratt et al., 2009). Overexpression of SuSy in developing vascular tissue of transgenic poplar yields small but significant increases in cellulose content (Coleman et al., 2009). Taken together, SuSy does not appear to be required for cellulose synthesis but may enhance rates by concentrating substrate at the site of synthesis.

Peng et al. (2002) provided evidence that sitosterol-cellodextrins synthesized from sitosterol-β-glycoside serve as primers of glucan chain initiation, with the KORRIGAN glucanohydrolase trimming the sitosterol from the growing chain. DeBolt and colleagues (2009) question this role, as they found that double mutants of two major sterol-β-glucoside synthases result in severe defects in cuticle formation but not in cellulose synthesis. However, the sterol-glucosides are substantially reduced in the double mutant but not entirely eliminated, leaving open the question.

Outside the Golgi stacks, a membrane compartment also containing KORRIGAN (Robert et al., 2005) might represent a dynamic factory associated with both the microtubule network and the plasma membrane that aligns and directs the cellulose synthase complex and coordinate that function with the deposition of the many polysaccharides directed to it from packaged Golgi vesicles. Bimolecular fluorescence complementation techniques, as shown for CesA interactions (Desprez et al., 2007), can give important clues to selected participants in protein-protein interactions within a complex in vivo. Fluorescence tagging has also allowed visualization of the movement of cellulose synthase complexes at the plasma membrane (Paredez et al., 2006; Wightman and Turner, 2008; Gutierrez et al., 2009). These studies also established the dynamics of the relationships with the cortical microtubule network in real time. As reviewed by Baskin (2001) and Szymanski and Cosgrove (2009), these studies bring resolution to ideas on the alignment of cortical microtubules and cellulose microfibrils generated long ago through observations with inhibitors (Green, 1962) and in the electron microscope (Ledbetter and Porter, 1963). Improvements in imaging tools are still needed that permit visualization of the delivery of Golgi-derived vesicles to the sites of cellulose synthesis, and much progress has already been made along these lines (Held et al., 2008; Konopka and Bednarek, 2008; Crowell et al., 2009).

THE SYNTHESIS OF NONCELLULOSIC POLYSACCHARIDES WITH (1→4)-β-d-GLYCAN BACKBONES

Because of the same conserved domains of nucleotide-sugar binding and catalysis as those encoding CesAs, subfamilies of Csl genes were predicted to encode the synthases of noncellulosic polymers with (1→4)-β-d-glycan backbones, primarily (gluco)mannans and galacto(gluco)mannans, xyloglucans, glucuronoarabinoxylans (GAX), and the grass-specific (1→3),(1→4)-β-d-glucans (Delmer, 1999; Richmond and Somerville, 2000; Hazen et al., 2002). For the most part, this has turned out to be true, but GAXs are a clear exception. There is still an incomplete knowledge of most of the gene products and interactions among them to make specific β-d-glycans, even among families where at least one member has a confirmed glycosyl transferase activity. There is also a growing disconnection between the classic studies on the synthesis of these polysaccharides in vitro and the discovery of genes encoding the machinery that warrants a revisit.

CslF AND CslH: MIXED-LINKAGE (1→3),(1→4)-β-d-GLUCAN SYNTHASE IS THE TOPOLOGICAL EQUIVALENT OF CELLULOSE SYNTHASE

The mixed-linkage (1→3),(1→4)-β-d-glucan is made in the grasses (Poales; Carpita, 1996; Buckeridge et al., 2004), certain lichens (Wood et al., 1994), and Equisetum (Fry et al., 2008; Sørensen et al., 2008), but differences in the distribution of their cellodextrin oligomers indicate that they probably arose by convergent evolution of synthases. For the grasses, this glucan is not a random mixture of (1→3)-β-d- and (1→4)-β-d-glucosyl linkages but is composed primarily of cellotriosyl and cellotetraosyl units in a ratio of about 2.5:1 connected by single (1→3)-β-d linkages (Wood et al., 1994). Upon cleavage with a Trichoderma cellulase, smaller amounts of higher cellodextrin series are observed, with the odd-numbered cellodextrin 2-fold higher in abundance than the next even-numbered unit in the series. The synthesis of the (1→3),(1→4)-β-d-glucan has been demonstrated in vitro with isolated, intact Golgi membranes and UDP-Glc (Gibeaut and Carpita, 1993; Buckeridge et al., 1999, 2001; Urbanowicz et al., 2004). Whereas micromolar concentrations of UDP-[¹⁴C]Glc result in much shorter oligomers and polymers enriched in cellotetraosyl units rather than cellotriosyl units (Buckeridge et al., 1999), much larger polysaccharides enriched in cellotriosyl units are observed with higher concentrations of substrate (Buckeridge et al., 1999, 2001). The ratios of cellotriosyl-cellotetraosyl and cellopentosyl-cellohexaosyl are increased proportionally with substrate concentrations higher than 250 μm (Buckeridge et al., 1999), indicating that the mechanism of synthesis of the odd-numbered cellodextrin unit is fundamentally different from synthesis of the even-numbered units. Proteolysis protection assays show further that the active site of catalysis is on the outward-facing Golgi membrane (Urbanowicz et al., 2004). Golgi membranes treated with proteinase K specifically lost their ability to make the odd-numbered cellodextrin units, whereas the synthesis of the cellotetraosyl and higher order even-numbered units was unaffected. Again, loss of the ability to make cellotriosyl units is correlated with significant loss in size of the (1→3),(1→4)-β-d-glucan product (Urbanowicz et al., 2004). We proposed a similar catalytic dimer model wherein even-numbered units are synthesized by core cellulose synthase-like proteins and the odd-numbered units arise by an additional GT that has yet to be identified (Buckeridge et al., 2001, 2004).

Limited proteolysis and detergent reconstitution experiments with the mixed-linkage (1→3),(1→4)-β-d-glucan of grass species provides kinetic evidence for three sites of glycosyl transfer within the catalytic domain: two from the cellulose synthase-like core domain and a third, separable activity (Urbanowicz et al., 2004). (1→3),(1→4)-β-d-Glucan is the topological equivalent of cellulose synthase at the Golgi membrane. Limited proteolysis or detergent treatment causes loss of the ability to make the diagnostic odd-numbered cellotriose units for synthesis without affecting the ability to generate the even-numbered cellotetraosyl unit (Urbanowicz et al., 2004).

As the topologic equivalent of cellulose synthase at the Golgi membrane, the (1→3),(1→4)-β-d-glucan synthase shares another feature with cellulose synthase. When its resident membranes are damaged, cellulose synthase (Delmer, 1977) and the (1→3),(1→4)-β-d-glucan synthase (Buckeridge et al., 2001) “default” to synthesis of the (1→3)-β-d-glucan, callose, possibly by disruption of the complete active site to a single glycosyl transferase activity (Buckeridge et al., 2001; Urbanowicz et al., 2004). Loss of the cellobiosyl-generating system to a single site of glycosyl transfer would thereafter make only callose (Buckeridge et al., 2001), whose synthesis does not require turning the catalytic site or acceptor 180° (Fig. 3B). While it can be argued that membrane disruption activates callose synthase in vitro in plasma membrane preparations of all angiosperms (Nishimura et al., 2003), only Golgi membranes from grasses, the only angiosperms that make the (1→3),(1→4)-β-d-glucan, make callose when damaged (Gibeaut and Carpita, 1993; Buckeridge et al., 1999). The most direct evidence for the default synthesis of callose from a damaged cellulose synthase comes from the experiments of Blanton et al. (2000), who showed that isolated membranes of a cellulose synthase mutant of the cellular slime mold Dictyostelium discoideum also lost the ability to make callose in vitro. Whereas some cellulose is made in vitro with wild-type membrane preparations, callose linkages predominate. Because cellulose synthase is a single gene in Dictyostelium, loss of the ability to make (1→3)-β-d-glucan in vitro as well as (1→4)-β-d-glucan in membranes from the mutant strongly suggests that the single polypeptide is responsible for both activities.

Two groups of Csl genes, CslF and CslH, which are found only in grasses (Hazen et al., 2002), have been shown to catalyze (1→3),(1→4)-β-d-glucan biosynthesis. Heterologous expression of a rice CslF in Arabidopsis, a species that does not make (1→3),(1→4)-β-d-glucan, results in small amounts of the β-d-glucan in the cell walls (Burton et al., 2006). However, considerably greater amounts of the (1→3),(1→4)-β-d-glucan result when a CslH is coexpressed with CslF, suggesting a synergistic role for both CslH and CslF in the synthesis of the polysaccharide and that a catalytic heterodimer enhances the activity (Doblin et al., 2009). If an accessory glycosyl transferase is necessary to make the odd-numbered cellodextrin unit, then Arabidopsis must produce a related isoform. This finding of concerted action by two distinct group members highlights the possibility that synthases of other cross-linking glycans might be encoded by Csl genes of different groups.

CslA: MANNAN AND GLUCOMANNAN BIOSYNTHESIS

One of the first cell wall polysaccharides to be synthesized in vitro was glucomannan. An early conclusion that GDP-Glc is the substrate for (1→4)-β-d linkages of cellulose and that UDP-Glc is the substrate for (1→3)-β-d-glucans (Chambers and Elbein, 1970) had already been disproven, yet GDP-Glc is still listed erroneously as the substrate for cellulose synthesis on most wall charts of biochemical pathways. Kinetic evidence obtained with cotton fiber cells cultured in vitro showed unequivocally that UDP-Glc is the substrate for cellulose synthesis (Carpita and Delmer, 1981). Addition of both GDP-Glc and GDP-Man to membrane preparations resulted in marked stimulation of incorporation into a glucomannan product (Elbein and Hassid, 1966; Piro et al., 1993).

The knowledge of GDP-nucleotide sugars as substrates was instrumental in the discovery that a CslA gene encodes a mannan synthase by expression profiling of guar (Cyamopsis tetragonolobus) seed development, a species that accumulates large amounts of galactomannan as a cell wall storage carbohydrate (Dhugga et al., 2004). Liepman et al. (2005, 2007) confirmed that at least four members of the CslA group function in mannan and/or glucomannan synthesis. However, mixed substrates of GDP-Glc and GDP-Man in the heterologous expression system (Liepman et al., 2007) do not give the marked enhancement of glucomannan synthesis long ago observed in vitro (Elbein and Hassid, 1966). While the CesA genes appear orthologous across several species, the Csl genes are not (Penning et al., 2009). In fact, the CslA group is resolved into three subgroups that either are Arabidopsis dominated, grass dominated, or mixed. The CslA members defined as mannan synthase genes (Liepman et al., 2007) fall into both the Arabidopsis-dominated and the mixed subgroups (Penning et al., 2009). The functions of these other subgroup members of the grasses need to be defined.

CslC: XYLOGLUCAN BIOSYNTHESIS

Xyloglucans were among the first complex cell wall polysaccharides whose synthesis was demonstrated in vitro. Early studies showed that labeled sugars from UDP-Glc and UDP-Xyl are incorporated into several polysaccharides using microsomal membranes and were later refined by isolation of Golgi membranes (Ray et al., 1969; Ray, 1980). Small amounts of xyloglucan-like oligomers with the characteristic α-d-Xyl-(1→6)-d-glucosyl unit, isoprimeverose, are made with small amounts of UDP-Glc and UDP-Xyl, but Gordon and Maclachlan (1989) found that when concentrations of each nucleotide-sugar are increased to millimolar levels, large polymers containing the characteristic heptasaccharide XXXG (for nomenclature, see Table I) unit structure are synthesized. The tetraglucosyl unit of the xyloglucan backbone and the precisely repeated three xylosyl units added to make the XXXG structure are consistent with an even-numbered cellobiose unit synthase reaction for the glucan backbone. Even in the structural variant of solanaceous xyloglucan, where just two xylosyl units are added, a tetraglucosyl unit backbone is preserved by the replacement of the third xylosyl group with an acetate (Sims et al., 1996). An apparent exception is the ability of certain tree legumes, such as jatobá (Hymenaea courbaril), to make XXXXG units in addition to XXXG (Buckeridge et al., 2000). Curiously, partial digestion of this polymer with a Trichoderma cellulase, which cleaves only at unbranched positions, yields octomer, nonamer, and decamer backbone oligomers whose ratios predict that the polymer consists of 4-5-5-4 frameworks separated by variable amounts of XXXG, rather than a random distribution of XXXXG and XXXG units (Tiné et al., 2006). This is an intriguing result for two reasons: (1) the 4-5-5-4 framework of these types of xyloglucans preserves the even-numbered unit symmetry of the backbone; and (2) to make such a framework, as many as 18 glucosyl residues might be contained within the complex in order to be “read” properly to preserve the unit structure, making the complex much larger than expected.

The topology of xyloglucan synthesis at the Golgi membrane is still uncertain. Several lines of evidence suggest that synthesis relies on transporters for UDP-Glc for backbone synthesis of (1→4)-β-d-glucans in vitro (Orellana, 2005), and UDP-GlcA is transported to provide UDP-Xyl for transfer within the lumen (Hayashi et al., 1988). Lerouxel et al. (2006) proposed that synthesis of the backbone is the topological equivalent of cellulose synthase, but the addition of all subtending sugars of Xyl, Gal, and Fuc are within the lumen of the Golgi.

Based on heterologous expression in Pichia, Cocuron et al. (2007) provide evidence that CslC genes encode the synthases of the xyloglucan backbone. Although Pichia is unable to make UDP-Xyl, coexpression of the xylosyl transferase with CslC is sufficient to induce the extension of (1→4)-β-d-glucan chains, indicating that a close interaction of these proteins might stabilize the synthase to allow extension of the backbone. A complication to unequivocal annotation of function of this subfamily is the finding of a CslC at the plasma membrane instead of the Golgi membrane (Dwivany et al., 2009). Three members of the GT34 family are established as the xylosyl transferases involved in xyloglucan synthesis, and these Golgi-resident proteins are predicted to face the lumen (Cavalier et al., 2008; Zabotina et al., 2008).

Xyloglucan is decorated in various ways in a species-specific manner (Hoffman et al., 2005; Peña et al., 2008). Most are primarily galactosylated, with a characteristic α-l-Fuc-(1→2)-β-d-Gal-(1→2)-α-d-Xyl trisaccharide extension (Bauer et al., 1973), but others, such as those of solanaceous species, have a truncated unit structure substituted with α-l-Ara-(1→2)-α-d-Xyl extensions instead of Gal (Sims et al., 1996), and the Asteridae and Oleales species have mixtures of these two forms of xyloglucan substitution (Hoffman et al., 2005). Furthermore, the xyloglucans synthesized in the Golgi are modified in ways that make them structurally different from those that are assembled onto the cellulose microfibrils in the wall (Obel et al., 2009).

The purification of a xyloglucan-specific fucosyl transferase led to the discovery of a GT37 gene encoding it (Perrin et al., 1999). While the synthase complex must involve a close interaction between the glucan synthases and the xylosyl transferases that decorate it (Cavalier et al., 2008), the association of the galactosyl and fucosyl transferases might be more transient. Transferases extracted from the membrane are able to add Gal from UDP-Gal (Madson et al., 2003) and Fuc from GDP-Fuc (Perrin et al., 1999; Vanzin et al., 2002) to exogenous xyloglucan in vitro.

XYLAN BIOSYNTHESIS

The synthesis of grass (1→4)-β-d-xylans from UDP-Xyl with microsomal membranes was first demonstrated by Bailey and Hassid (1966). Cooperative action of two nucleotide-sugar substrates, in this instance UDP-Xyl and UDP-GlcA, resulted in the synthesis of (1→4)-β-d-xylans with subtending GlcA units (Waldron and Brett, 1983; Baydoun et al., 1989). Similar studies have shown that membrane preparations from grasses and mixtures of nucleotide sugars made GAXs, in part employing a nascent C-4 epimerase that interconverts UDP-Xyl and UDP-Ara (Porchia and Scheller, 2000; Kuroyama and Tsumuraya, 2001; Porchia et al., 2002; Zeng et al., 2008).

Apart from a hint that the AtCslD5 may be involved (Bernal et al., 2007) in the synthesis of (1→4)-β-d-xylans, it has yet to be directly demonstrated that any Csl gene plays a role. In fact, informatics approaches yield non-Csl genes as more likely candidates for encoding the machinery for xylan synthesis (Mitchell et al., 2007), and several mutants with deficiencies in normal xylan synthesis, such as parvus (Lao et al., 2003; Lee et al., 2007), irx8 (Brown et al., 2005), irx7/fra8 (Zhong et al., 2005; Brown et al., 2007), irx9 and irx14 (Brown et al., 2007), and irx10 and irx10-L (Brown et al., 2009), do not include a member of the Csl gene family.

(1→4)-β-d-Glucuronoxylan (GX) synthesis appears to involve a complex initiation sequence. Peña et al. (2007) discovered that collapsed xylem mutants deficient in xylan, irx8 and fra8, were essentially devoid of a complex tetrasaccharide, β-d-Xyl-(1→3)-α-l-Rha-(1→2)-α-d-GalA-(1→4)-d-Xyl, located at the reducing end of the xylan polymer, whereas an irx9 mutant also severely deficient in xylan contained an overabundance of the tetrasaccharide. The presence or absence of this tetrasaccharide greatly affected the size distribution of the xylans. In irx8 and fra8, a broader distribution is observed, with some polymers longer than observed in the wild type. Xylans of irx9 have short chains, with nearly all of them containing the tetrasaccharide (Peña et al., 2007). These results suggested a model whereby short chains of (1→4)-β-d-xylan are primed by the tetrasaccharide, and these are spliced, cleaving the primer, to make the long polysaccharides (York and O’Neill, 2008).

The IRX8 (GAUT12) and PARVUS (GATL1) genes encode members of the GT8 group C, and IRX7/FRA8 genes encode members of the GT47 group E, that synthesize the primer tetrasaccharide, whereas IRX9 and IRX14 encode members of GT43 that are likely to encode the synthases of the (1→4)-β-d-xylan oligomeric backbones that are stitched together by a yet unidentified glycosyl transferase (Brown et al., 2007; Peña et al., 2007). GT8 family members encode retaining-type transferases, and the GT47 members encode inverting-type transferases with respect to the anomeric linkages formed compared with the anomeric linkage in the nucleotide sugar, so GT8 members make α-d linkages, whereas the GT47 members make β-d or α-l linkages. While an α-d-GalA is found in the tetrasaccharide, (1→2)-α-d-GlcA (4-O-Me-GlcA) side groups are also attached at precise intervals along the xylan chain (Nishitani and Nevins, 1991) and a general blockwise synthesis of six consecutive branched xylosyl residues in grass xylans is observed (Carpita and Whittern, 1986). IRX10 and IRX10-L appear to encode xylosyl transferases also from GT47, but double mutants of these genes have greatly reduced GlcA substitutions along the shorter chains (Brown et al., 2009). It will be interesting to determine if the addition of these GlcA side groups plays a role as an attachment or recognition point where short xylan chains are grafted together to make a long chain, a suggestion foretold by the original work on the cooperative action of UDP-GlcA and UDP-Xyl in the in vitro synthesis of glucuronoxylan (Waldron and Brett, 1983; Baydoun et al., 1989).

York and O’Neill (2008) take a broader perspective of xylan synthesis and have suggested that reducing end addition should not be ruled out. In their model, a type of alternating “pendulum” mechanism is proposed for the introduction of the (1→4)-β-d-xylan linkages, a mechanism analogous to the dimer synthesis described here for (1→4)-β-d linkages in general. We are just learning the identities of the GlcA and Araf transferases of the GX and GAX polymers and the distinctions between the GX devoid of Ara that is abundant in the secondary xylem and the GAX with its rich Ara substitution that is the major polymer of the primary cell walls of grasses (Scheller and Ulvskov, 2010). Recently, proteomic approaches of isolated GAX synthase complexes from wheat membranes demonstrated a close interaction of GT43 and GT47 family members with a GT75 UDP-Ara mutase, an enzyme that interconverts the UDP-arabinopyranose and UDP-arabinofuranose conformations required for incorporation of the latter into the polysaccharide (Zeng et al., 2010). These kinds of studies represent the path forward to identify all the components of synthase complexes whose activities are preserved in vitro.

CONCLUDING REMARKS

While the plant cell wall community is making steady progress in defining biochemical functions of backbone synthases and the glycosyl transferases that decorate them, the biochemical details of these large, coordinated complexes are still unknown. Fading from memory are a great many foundation studies of polysaccharide synthesis in vitro that give insights to the differences between an isolated activity and the function of a complex as a whole.

The major differences between the behaviors of protein complexes in vitro and in vivo are compounded by the topology of the synthases and associated glycosyl transferases at the plasma membrane and Golgi membranes, across which both pH gradients and electrical potentials are generated. Early studies with excised cotton fibers cultured in vitro showed that resealing of the plasma membrane was essential to reconstitute cellulose synthesis, and evidence provided that regeneration of the electrical potential rather than the pH gradient led to that was critical to the preservation of synthesis (Carpita and Delmer, 1980). This work was extended to show that artificial potentials stimulate (1→3)-β-d-glucan synthesis in vitro (Bacic and Delmer, 1981). In Gluconacetobacter, rates of cellulose synthesis could be modulated directly through adjustment of the electrical potential in living cells under conditions that did not impair metabolism (Delmer et al., 1982). Such gradients exist across other compartments, and, in contrast to cellulose synthesis at the plasma membrane, it is maintenance of a pH gradient that prolongs the synthesis of the mixed-linkage (1→3),(1→4)-β-d-glucan in vitro in isolated maize Golgi membranes (Gibeaut and Carpita, 1993). The physiological and biochemical bases for the effect of these potentials and gradients on polysaccharide synthesis are still not understood. They do serve to illustrate that, beyond the biochemical mechanism of synthesis, technologies that preserve not only the protein complexes but also their cellular context need to be developed to truly understand the synthesis of macromolecules across membrane surfaces.