Biochemical and Genetic Analysis Identify CSLD3 as a beta-1,4-glucan Synthase that Functions during Plant Cell Wall Synthesis

One-sentence summary : A combination of genetic rescue and biochemical reconstitution experiments demonstrate that the Arabidopsis thaliana CSLD3 cell wall synthase is a beta-1,4-glucan synthase. The author(s) responsible for distribution of materials integral to the findings presented in this article in accordance with the policy described in the Instructions for Authors (www.plantcell.org) is: Nielsen ABSTRACT In plants, changes in cell size and shape during development fundamentally depend on the ability to synthesize and modify cell wall polysaccharides. The main classes of cell wall polysaccharides produced by terrestrial plants are cellulose, hemicelluloses, and pectins. Members of the Cellulose Synthase (CESA) and Cellulose Synthase-Like (CSL) families encode glycosyltransferases that synthesize the β-1,4-linked glycan backbones of cellulose and most hemicellulosic polysaccharides that comprise plant cell walls. Cellulose microfibrils are the major load bearing component in plant cell walls and are assembled from individual β-1,4-glucan polymers synthesized by CESA proteins that are organized into multimeric complexes, called cellulose synthase complexes (CSCs), in the plant plasma membrane. During distinct modes of polarized cell wall deposition, such as in the tip growth that occurs during the formation of root hairs and pollen tubes, or de novo formation of cell plates during plant cytokinesis, newly synthesized cell wall polysaccharides are deposited in a restricted region of the cell. These processes require the activity of members of the cellulose synthase-like D subfamily. However, while these CSLD polysaccharide synthases are essential, the nature of the polysaccharides they synthesize has remained elusive. Here, we use a combination of genetic rescue experiments with CSLD-CESA chimeric proteins, in vitro biochemical reconstitution, and supporting computational modeling and simulation, to demonstrate that Arabidopsis thaliana CSLD3 is a UDP-glucose-dependent β-1,4-glucan synthase that forms protein complexes displaying similar ultrastructural features to those formed by the cellulose synthase, CESA6.


INTRODUCTION
Cellulose is one of the most abundant organic polymers in nature and is the principal component of the plant cell wall, providing most of its tensile strength (Baskin, 2005;Cosgrove, 2005). Cellulose microfibrils contain multiple β-1,4-glucan chains that associate via intermolecular hydrogen bonds, and are synthesized by large, membrane-localized complexes called "rosette complexes" (Baskin, 2005;Cosgrove, 2005). In Arabidopsis thaliana, CESA proteins interact to form rosette subunits, and six of these subunits then assemble into multimeric rosette complexes, often referred to as cellulose synthase complexes (CSCs) (Kimura et al., 1999). CSCs contain several CESA subunits, each thought to be capable of synthesizing β-1,4-glucan polysaccharides. Arabidopsis contains ten CESA genes of which the proteins encoded by at least three, CESA1, CESA3, and CESA6 or CESA2/5/9, are required for cellulose synthesis during primary cell wall formation (Arioli et al., 1998;Fagard et al., 2000;Scheible and Pauly, 2004). While earlier models suggested each rosette subunit may contain six CESA proteins, recent studies combining ultrastructural analysis and computer-modelling of plant CESAs using bacterial cellulose synthase structures have proposed rosette subunits may contain as few as three CESAs each Vandavasi et al., 2016). Furthermore, in vitro reconstitution of cellulose synthase activity was observed in proteoliposomes containing only Populus tremula x tremuloides CESA8 (PttCESA8) or Physcomitrella patens CESA8 (PpCESA8) (Purushotham et al., 2016;Cho et al., 2017), highlighting that our understanding of the composition of functional cellulose synthase rosette complexes remains incomplete.
CSLD proteins were initially proposed to synthesize cellulose in tip-growing pollen tubes, consistent with the high degrees of sequence similarity and overall domain organization CSLDs share with CESAs (Doblin et al., 2001). Confirming an important role in tip-restricted cell expansion, Arabidopsis CSLD2 and CSLD3 are required for proper root hair growth, and csld1 and csld4 mutants are male sterile, presumably due to defects in pollen tube growth (Favery et al., 2001;Wang et al., 2001;Bernal et al., 2008). The functional roles of CSLD enzymes are not restricted solely to cells undergoing tip-restricted expansion, with CSLD2, CSLD3, and CSLD5 all participating in construction of newly-forming cell walls during plant cytokinesis (Gu et al., 2016). Furthermore, CSLD5 also displays cell cycle-specific accumulation in dividing cells (Yoshikawa et al., 2013;Gu et al., 2016;Yang et al., 2016). Cellulose polysaccharide epitopes have been observed in tip-growing root hairs and pollen tubes (Park et al., 2011;Chebli et al., 2012), and in newly-forming cell plates (Miart et al., 2014), although the predominant cell wall polysaccharide in pollen and cell plates is likely the β-1,3-glucan callose (Meikle et al., 1991;Samuels et al., 1995;Ferguson et al., 1998;Chen and Kim, 2009;Drakakaki, 2015). In Arabidopsis, CSLD5 insertional mutants accumulated less xylan in stems and had reduced pectin (Bernal et al., 2007), and in a weak mutant allele of CSLD3, rhd7, the organization of xyloglucan and cellulose was altered in root hairs (Galway et al., 2011). Microsomal membranes isolated from tobacco (Nicotiana benthamiana) heterologously expressing A. thaliana CSLD2, CSLD3, and CSLD5 proteins were shown to contain elevated mannan synthase activity, specifically utilizing GDP-mannose (GDP-Man) as an activated nucleotide-sugar donor Yin et al., 2011), and examination of cell wall epitopes in newly-forming cell plates in shoot apical meristems in csld5 mutants displayed altered β-1,4-mannan accumulation (Yang et al., 2016). Alternatively, a functional YFP-CSLD3 fusion protein localized to apical plasma membranes in the tips of growing Arabidopsis root hairs, and genetic chimeras, where the CSLD3 catalytic domain (residues 340-921) was replaced with the corresponding CESA6 catalytic domain, rescued root hair defects in csld3 mutant plants, supporting a UDP-glucose (UDP-Glc)-dependent cellulose synthase-like activity for CSLD3 (Park et al., 2011). 4 Here we show using a combination of genetic and biochemical analysis, combined with in vivo localization of fluorescently-tagged fusion proteins, that a Citrine-CESA6 chimeric fusion protein containing the catalytic domain of CSLD3, integrates into plasma membrane-localized CSCs and is able to fully rescue both the hypocotyl elongation and cellulose accumulation defects in the prc1-1 (CESA6 null) mutant. In addition, we show that proteoliposomes containing purified CESA6 and CSLD3 utilize UDP-Glc but not GDP-Man, and accumulate β-1,4-glucan when supplied with UDP-Glc, while CSLA9 instead only utilized GDP-Man. These results are further supported by computational modeling and simulation of substrate binding for CESA6, CSLD3, and CSLA9 enzymes. Finally, both CESA6 and CSLD3 proteins could be purified as higher-order complexes which form ~10-12 nm particles with apparent three-fold symmetry when examined by electron microscopy.

A genetic chimera with a CSLD3 domain restores cellulose synthase functions in cesa6 mutants
CESA and CSLD proteins share overall membrane topology and maintain high degrees of sequence identity, especially in the central domain where critical catalytic residues are absolutely conserved ( Figure 1A; (Morgan et al., 2013;Sethaphong et al., 2013;Slabaugh et al., 2014)). We previously used this structural similarity to demonstrate that a fluorescently-tagged chimeric fusion protein in which the CSLD3 catalytic region was replaced with the corresponding CESA6 catalytic domain was able to quantitatively rescue kjk-2 (csld3 null) root hair defects (Park et al., 2011). While these results indicated that a chimeric CSLD3 fusion could restore root hair growth, it remained unclear whether CSLD catalytic domains could replace CESA sequences. To address this, we generated stably-transformed Arabidopsis lines expressing a fluorescently-tagged Citrine-CESA6 chimera containing a CSLD3 catalytic region (Citrine-CESA6:D3CD) under control of the endogenous CESA6 promoter ( Figure 1A).
The CESA6 chimera quantitatively rescued both dark-grown hypocotyl elongation defects ( Figure 1B, D) and root elongation defects observed in cesa6 (prc1-1) mutant plants grown either in the dark ( Figure   1B, D) or in light ( Figure 1C, E), indicating CSLD3 catalytic domain sequences could functionally replace CESA6 catalytic domain sequences in the CESA6 primary cell wall cellulose synthase protein.
Earlier genetic studies indicated that at least three distinct CESA proteins are required for cellulose synthesis during primary cell wall formation, and therefore CSCs involved in primary cell wall cellulose synthesis are thought to assemble with both essential cellulose synthases, CESA1 and CESA3, and at least one of either CESA6 or CESA2/5/9 (Arioli et al., 1998;Fagard et al., 2000;Scheible and Pauly, 2004 Figure   3D; magenta), indicating that both fluorescently-tagged CESA6 and CESA6:D3CD proteins occur in multimeric CSCs with other CESA proteins.
While these results supported the integration of Citrine-CESA6:D3CD chimeras into CSCs and rescue of prc1-1 mutant phenotypes, they did not directly address whether the Citrine-CESA6:D3CD chimera was catalytically active, or simply allowed for assembly of complexes with non-functional subunits. To address this directly, we mutated a conserved TED motif responsible for formation of β-6 1,4-glucosidic bonds in the extending glucan polymers (Morgan et al., 2013;Morgan et al., 2016), replacing both aspartic acid and glutamic acid with alanine residues. Neither stably-transformed Citrine-CESA6-TAA, nor Citrine-CESA6:D3CD-TAA plants were able to fully rescue dark-grown hypocotyl and root elongation defects ( Figure 4A, B, and Supplemental Figure 2). Interestingly, while Citrine-CESA6:D3CD-TAA expressing plants were indistinguishable from prc1-1 mutants, we did observe a small, statistically significant increase in dark-grown hypocotyl length in seedlings expressing the Citrine-CESA6-TAA protein (~15%; Figure 4A, B). Quantification of crystalline cellulose in these seedlings using the Updegraff method confirmed that expression of both YFP-CESA6 and Citrine-CESA6:D3CD restored crystalline cellulose content to wild-type levels ( Figure 4C), while cellulose content of Citrine-CESA6:D3CD-TAA seedlings was indistinguishable from that of the prc1-1 mutant background. Interestingly, consistent with earlier phenotypic analysis ( Figure 4A, B), we also observed a small, statistically-relevant increase in crystalline cellulose content in the Citrine-CESA6-TAA expressing seedlings ( Figure 4C). Taken together, these results strongly support the interpretation that the quantitative rescue of prc1-1 mutant phenotypes and chemotypes observed in seedlings expressing YFP-CESA6 and Citrine-CESA6:D3CD chimeras requires the catalytic function of a b-1,4-glucan synthase.
While the introduction of point mutations in both Citrine-CESA6-TAA and Citrine-CESA6:D3CD-TAA mutants are unlikely to affect overall folding of these proteins, we wanted to address whether these mutant proteins also associated into CSCs in these transgenic seedlings, and whether these "catalytically dead" constructs displayed any differences in either CSC trafficking or dynamics in these plants. As with their wild-type counterparts, both Citrine-CESA6-TAA ( Figure 5A Figure 3E-F). However, when we examined CSC motility events, we observed that particle tracks containing Citrine-CESA6-TAA were markedly shorter than those containing YFP-7 CESA6 (Figure 5A-C; green lines in 5C right panel), and CSC particles containing Citrine-CESA6:D3CD-TAA were virtually immobile ( Figure 5D-F; red lines in 5F right panel). Speeds calculated for CSC motility events for particles containing TAA point mutations in their catalytic domains were also significantly slower than their catalytically active counterparts, with Citrine-CESA6-TAA particle speed reduced to 207.2 +/-63.7 nm/min ( Figure 5G; green bars; 6 cells, 6 seedlings, 283 particles), and Citrine-CESA6:D3CD-TAA to 142.3 +/-87.5 nm/min ( Figure 5G; red bars; 6 cells, 6 seedlings, 253 particles). All together, these results confirm that "catalytically-dead" Citrine-CESA6-TAA and Citrine-CESA6:D3CD-TAA subunits successfully integrate into CSCs and are delivered to plasma membranes in rates largely indistinguishable from their catalytically-active counterparts. However, CSCs containing these TAA mutant subunits show distinctly slower particle speeds and dramatically shorter particle trajectories, consistent with the impaired rescue of prc1-1 phenotypes and cellulose deposition observed earlier ( Figure 4).

Purification and reconstitution of catalytically-active CSLD3, CESA6, and CSLA9 enzymes into proteoliposomes
While the in vivo rescue of cell wall defects in the CESA6 mutant prc1-1 is consistent with a UDP-Glc dependent β-1,4-glucan synthase activity for CSLD proteins, earlier studies of isolated N.
benthamiana microsomal membranes over-expressing CSLD proteins identified increased GDP-Mandependent β-1,4-mannan synthase activities Yin et al., 2011). To directly test whether CSLD proteins utilize UDP-Glc to synthesize β-1,4-glucan, or GDP-Man to synthesize β-1,4-mannan, we generated His-tagged versions of CSLD3, as well as CESA6 and CSLA9 which were used as a positive and negative controls for UDP-Glc and GDP-Man dependent polysaccharide synthase activities. These cell wall synthases were expressed in Saccharomyces cerevisiae under control of a galactose-inducible promoter ( Figure 6A). While S. cerevisiae do not contain β-1,4-glucans, they do contain significant amounts of β-1,3glucan and chitin (b-1,4-linked GlcNAc) polysaccharides in their cell walls and b-1,4-glucosidic linkages often connect these β-1,3-glucan and chitin polymers (Lesage and Bussey, 2006). These, or similar, endogenous yeast transglycosylase activities might therefore complicate assessment of potential b-1,4-glucan synthase activities of the heterologously expressed CESA6 and CSLD3 proteins. To more specifically assess the enzymatic activities of these plant cell wall synthases, microsomal membranes were isolated from yeast expressing CESA6, CSLD3, and CSLA9, and treated with a panel of nondenaturing detergents and lipid analogs to determine their ability to efficiently solubilize these proteins 8 (Supplemental Figure 4A-C). Interestingly, CSLD3 and CSLA9 proteins were unable to be efficiently solubilized in the presence of the non-denaturing detergent Triton X-100 (Supplemental Figure 4A, C), while about 50% of CESA6 was recovered in the soluble fraction under these detergent conditions. Similarly, while CSLA9 and about 50% of CESA6 were solubilized in the presence of lysophosphatidylcholine (LPC), CSLD3 remained in the insoluble fraction in these conditions. These three integral membrane polysaccharide synthases were most efficiently solubilized in the presence of a pair of lipid analogues, lysoFos-Choline Ether-14 (LFCE-14), or lauryl dimethyl amine-N-oxide (LDAO). As >90% of both CSLD3 and CSLA9, and around 50% of CESA6 were solubilized in the presence of LFCE-14, this detergent was used for further purification steps. Detergent-solubilized His-CSLD3, His-CESA6, and His-CSLA9 protein fractions were enriched by affinity-purification on Niagarose columns (Supplemental Figure 5A-C). Purified His-CSLD3, His-CESA6, and His-CSLA9 were mixed with S. cerevisiae total lipid extracts, and reconstituted proteoliposomal fractions containing purified cell wall polysaccharide synthases were isolated using sucrose density gradient ultracentrifugation (Supplemental Figure 5D). The presence of His-tagged CSLD3, CESA6, and CSLA9 proteins in these proteoliposomes was determined by immunoblotting with anti-His antibodies ( Figure   6B). Reactions catalyzed by GTs, such as members of the CESA/CSL superfamily, are bi-substrate reactions, and the nucleotide product (UDP, GDP) can be measured to quantify GT activity. Substrate specificity and catalytic activity of these proteins were assessed in the presence of 1 mM Mn 2+ and 1 mM Mg 2+ , and either UDP-Glc ( Figure 6C) or GDP-Man ( Figure 6D). When reconstituted into proteoliposomes and provided UDP-Glc, saturable UDP-forming activities were observed for both CSLD3 and CESA6 proteins, but not for CSLA9, which utilizes GDP-Man ( Figure 6C-D). The apparent Km values of CSLD3 and CESA6 for UDP-Glc were 65 µM and 73 µM, respectively, consistent with values recently reported for reconstituted PttCESA8 (~30 µM; (Purushotham et al., 2016)), and significantly lower than values determined for the bacterial cellulose synthase, RsBCSA (~500 µM; (Omadjela et al., 2013)). When reconstituted proteoliposomes were provided with GDP-Man, saturable GDP-forming activities (Km value of 17 µM) were only observed in proteoliposomal fractions containing the β-1,4-mannan synthase, CSLA9, but not for either CSLD3 or CESA6 ( Figure 6D). Both UDPforming activities for CESA6 and CSLD3 proteins and GDP-forming activities for CSLA9 were timeand concentration-dependent (Supplemental Figure 6).
To determine the nature of the polysaccharides produced in these reactions, proteoliposomes containing purified CESA6, CSLD3, CSLA9, and Ni 2+ -agarose eluted proteins from an empty vector control were incubated with UDP-Glc, Mg 2+ and Mn 2+ , and a UDP-[ 3 H]-Glc as a tracer. Time-dependent accumulation of [ 3 H]-Glc containing reaction products were observed for both CESA6 and CSLD3 upon 9 sedimentation and subsequent purification of insoluble reaction products by paper chromatography ( Figure 6E), but not for CSLA9 or the empty vector control proteoliposomes ( Figure 6F). To determine whether in vitro-synthesized material represented β-1,4 glucan, these reaction products were incubated with glucanases specifically degrading β-1,3-linked or β-1,4-linked glucans. Consistent with the formation of cellulose, both CESA6 and CSLD3 reaction products were selectively degraded only by a β-1,4-specific glucanase, and were largely resistant to treatment with a β-1,3-specific glucanase ( Figure 6F). Although fibrillar structures, structurally similar to cellulose microfibrils, were observed in recent reconstitution experiments with plant CESAs (Purushotham et al., 2016;Cho et al., 2017), we were unable to detect similar fibrils in reconstituted proteoliposome fractions with either CESA6 or CSLD3. Taken together, these results strongly support the conclusion that CSLD3 utilizes UDP-Glc as a substrate and synthesizes β-1,4-glucan and not β-1,4-mannan.
To further understand how UDP-Glc and GDP-Man substrates might be selectively bound within CSLD, CESA, and CSLA catalytic domains, we generated 3D atomistic models of the cytosolic catalytic  (Guex et al., 2009;Benkert et al., 2011;Bertoni et al., 2017;Bienert et al., 2017;Waterhouse et al., 2018). CSLD3, CESA6, and CSLA9 models all showed substantial structural conservation when aligned with the RsBCSA crystal structure, with the exception of two plant-specific PCR and CSR domains that displayed randomly-disordered structures (Supplemental Figure 7A). However, these domains, which are suspected to participate in rosette complex formation (Sethaphong et al., 2013), are too far away from the enzyme activity sites to substantially influence substrate binding in the highly conserved structural cores of these catalytic domains.
To identify additional amino acid residues that might participate in binding of nucleotide sugar substrates, we initially focused on amino acids whose 3D positions were within seven angstroms of the bound UDP-Glc in the crystal structure of BCSA, and compared this with the 3D structural models 10 (Supplemental Figure -7A; blue spheres). These amino acid sequences from RsBCSA, CESA6, CSLD3, and CSLA9 were aligned with the other Arabidopsis CESA, CSLD, and CSLA family members (Supplemental Figure -7B). In addition to the conserved GT2 catalytic motifs, additional amino acids within the CESA and CSLD binding pockets were highly conserved with one another and not with CSLA proteins, possibly reflecting differing nucleotide sugar-binding specificities. When UDP-Glc binding was modeled with RsBCSA, the UDP-Glc consistently adopted a binding mode in a conformation similar to that observed in the RsBCSA crystal structure (Average RMSD 0.8 Å; Supplemental Figure Figure 7E; bold red amino acids). Taken together, these results indicate that nucleotide-sugar binding pockets are more highly-conserved between CESA and CSLD proteins than CSLA proteins. Additionally, the presence of several conserved aromatic residues in CSLA binding pockets may be associated with GDP-Man nucleotide-sugar selection in these enzymes.

Detergent solubilized CSLD3 and CESA6 proteins assemble into higher-order complexes
In plants, β-1,4-glucans synthesized by plant CESA proteins coalesce into microfibrils of 18-24 individual glucan polymers held together via hydrogen bonding (Cosgrove, 2018). This efficient microfibril bundling is thought to occur due to the assembly of higher order CSCs, comprised of at least three CESAs organized into subcomplexes that then further associate into six-lobed rosettes (Polko and Kieber, 2019). The ability to detect UDP-Glc dependent formation of b-1,4-glucan polysaccharides in reconstituted proteoliposomes containing CSLD3 and CESA6 ( Figure 6) suggested that these detergent-solubilized protein complexes maintain appropriate conformation during their isolation and purification. We were interested therefore in whether the detergent-solubilized His-CESA6, His-CSLD3, and His-CSLA9 proteins might also be found in higher-order complexes. Detergent-solubilized complexes were therefore separated by size-exclusion chromatography, and resulting fractions were 11 analyzed by SDS-PAGE and immunoblotting with anti-His antibodies ( Figure 7A). Both His-CSLD3 and His-CESA6 eluted as high-molecular weight complexes with estimated sizes significantly larger than the ~700 kDa thyroglobulin complexes observed in molecular weight standards typically used for analysis of globular protein complexes ( Figure 7B). While His-CSLA9 complexes also migrated as higher-order complexes, these were markedly smaller, with estimated molecular size of ~150-200 kDa when compared to globular protein molecular weight standards.
Based on their large apparent molecular sizes, we attempted to visualize these purified, detergentsolubilized CSLD3 and CESA6 fractions by electron microscopy. When detergent-solubilized membrane fractions from untransformed yeast, and yeast expressing either His-CSLD3 or His-CESA6 were affinity purified, placed on EM grids, negatively stained, and imaged at 50,000x magnification, clear enrichment of ~10-12nm particles in the CSLD3 and CESA6 fractions was observed ( Figure 7C).
When examined at higher magnification these particles often displayed structural detail reminiscent of a "wagon-wheel," with a bright central mass often surrounded by three pairs of smaller circular structures oriented at roughly 120-degree intervals around the central mass ( Figure 7B and 7C, insets).

DISCUSSION
Plant cells are surrounded by a load-bearing cell wall comprised of cellulose, hemicelluloses, pectins, and a variety of cell wall proteins (Cosgrove, 2005). Cellulose synthases, or CESA proteins, are contained within a larger superfamily of Cellulose Synthase-Like, or CSL proteins (Richmond and Somerville, 2000). While a number of different biosynthetic activities have been proposed for members of the CSLD family of glycan synthases (Doblin et al., 2001;Bernal et al., 2008;Park et al., 2011;Verhertbruggen et al., 2011;Yin et al., 2011), here we present multiple lines of evidence that support classification of members of the CSLD family as β-1,4-glucan synthases. Using a genetic rescue approach, we demonstrated that a CESA6 protein chimera in which the CESA6 catalytic domain is replaced with CSLD3 catalytic sequences (CESA6:D3CD) fully rescues cesa6 mutant alleles ( Figure   2; Figure 4C). We had previously reported that a CSLD3 protein chimera containing a CESA6 catalytic domain could rescue csld3 null mutant phenotypes consistent with CSLD3 being a β-1,4-glucan synthase (Park et al., 2011). However, it remained unclear whether CSLD proteins assembled into higher order protein complexes, or if β-1,4-glucan polymers generated by CSLD enzymes could produce microfibrillar cellulose. YFP-CESA6 has been observed to organize into punctate structures similar to CSCs (Paredez et al., 2006;Gutierrez et al., 2009), and loss of CESA6 activity in the prc1 null mutant results in plants that produce less crystalline cellulose (Fagard et al., 2000;MacKinnon et al., 2006). Quantitative genetic rescue of prc1 mutant phenotypes by Citrine-CESA6:D3CD chimeras ( Figure 1) and restoration of crystalline cellulose content back to wild-type levels ( Figure 4C) strongly supports the conclusion that the CSLD3 catalytic domain synthesizes β-1,4-glucan. These CESA6:D3CD chimeras integrate into CSCs (Figure 2) that also contain GFP-CESA3 (Figure 3), migrate along cortical microtubule tracks with similar speeds to YFP-CESA6 containing CSCs ( Figure   2), and must be catalytically active in order to fully rescue cesa6 mutant phenotypes (Figures 4, 5).
The fact that Citrine-CESA6:D3CD chimeras co-localized and appear to be integrated into CSCs and were delivered to plasma membranes at similar rates as YFP-CESA6 (Supplemental Figure 3), indicates that the amino-terminal and transmembrane domains of CESA6 are likely more important for assembly of these proteins into primary cell wall CSCs and for their subsequent subcellular targeting.
While it is somewhat surprising that wholesale replacement of the ~70 kDa CESA6 cytosolic catalytic domain with CSLD3 sequences does not affect this assembly, this may perhaps be mitigated because the CESA6 position in primary cell wall CSCs may alternatively be occupied by CESA2 and CESA5 (Desprez et al., 2007). The presence of plant-conserved (PCR) and class-specific (CSR) regions has been proposed to play important roles in the assembly of plant cellulose synthases into primary and secondary cell wall specific CSCs (Vergara and Carpita, 2001;Scavuzzo-Duggan et al., 2018). At least in the case of integration of the Citrine-CESA6:D3CD into primary CSCs, the absence of CESA6-specfic PCR and CSR domains does not appear to interfere with their assembly into CSCs, although the presence of similar PCR and CSR sequences in CSLDs may perhaps indicate the general ability of CSLD proteins to also assemble into higher-order protein complexes. Similarly, the absence of the cytosolic CESA6 catalytic domain does not appear to negatively affect movement of these CSCs along While rescue of hypocotyl growth defects and accumulation of crystalline cellulose required a catalytically active cytosolic domain in either YFP-CESA6 or Citrine-CESA6:D3CD proteins (Figure 4), replacement of glutamate and aspartate residues in the catalytic TED motif with alanines (TAA) did not appear to affect assembly of CSCs or their delivery to the plasma membrane. However, motility of Citrine-CESA6-TAA and Citrine-CESA6:D3CD-TAA containing CSCs along microtubule tracks were significantly affected, with the speed of Citrine-CESA6-TAA containing CSCs reduced by roughly onethird, and Citrine-CESA6:D3CD-TAA CSCs by slightly more ( Figure 5). These reduced speeds are 13 consistent with the proposal that microtubule-associated motility is primarily driven by biosynthetic activity and the elongation and assembly of β-1,4-glucan polymers into cellulose microfibrils (Paredez et al., 2006), with a loss of catalytic activity in the CESA6 positions resulting in an associated reduction of ~1/3 of the particle velocity. Interestingly, in addition to reduced particle velocities, we also noted a reduction in the overall length of linear tracks in CSCs containing TAA mutant subunits, perhaps indicating that "pausing" caused by catalytically-inactive CESA6 may increase the chance of disengagement with cortical microtubules and/or endocytosis from the plasma membrane.
Based on genetic reconstitution experiments, as well as biochemical reconstitution of CSLD3 activity in vitro, the catalytic domain of CSLD3 appears to prefer UDP-Glc. These studies further support the synthesis of β-1,4-glucan polymers that can integrate into crystalline cellulose. Both of these activities are consistent with a β-1,4-glucan synthase activity for CSLD3, which we directly confirmed A UDP-Glc substrate specificity for CSLD3 would appear to be inconsistent with recent descriptions of GDP-Man dependent synthesis of β-1,4-mannan by CSLD proteins Yin 14 et al., 2011). However, it should be noted that mannan synthesis in these studies was assessed in plant membranes overexpressing CSLD proteins that contained endogenous β-1,4-mannan activity, and that the levels of β-mannan synthesis reported were significantly lower than those observed for GDP-Man dependent mannan synthesis described for CSLA proteins (Liepman et al., 2005;Goubet et al., 2009).
However, consistent with earlier in vitro reconstitution studies with purified hybrid aspen PttCESA8, or moss (Physcomitrella patens) PpCESA8 (Purushotham et al., 2016;Cho et al., 2017), we observed accumulation of β-1,4-glucan in membranes containing only CESA6 proteins, suggesting that at least in vitro activities may not require assembly of hetero-oligomers. It should be noted however, that the specific activity we report in reconstituted CESA6 proteoliposomes is significantly lower than those reported in vivo (Reiss et al., 1984).
In Arabidopsis, CSLD cell wall synthases are required for cell wall deposition during tip growth in root hairs and pollen tubes as well as for de novo cell wall deposition in newly-forming cell plates in dividing cells (Favery et al., 2001;Wang et al., 2001;Bernal et al., 2008;Gu et al., 2016;Yang et al., 2016). While cellulose-specific stains have recently been used to show the presence of cellulose-like polysaccharides in growing root hairs and in newly-forming cell plates during cytokinesis (Park et al., 2011;Miart et al., 2014), callose is likely the major cell wall component in these walls. At least during the early stages of deposition, characteristic arrays of cellulose microfibrils are not regularly observed in root hairs, pollen tubes, and forming cell plates (Meikle et al., 1991;Samuels et al., 1995;Ferguson et al., 1998;Chen and Kim, 2009;Drakakaki, 2015). Therefore, a major question is what kinds of β-1,4-glucan polymers do CSLD proteins synthesize? Based on size exclusion chromatography we observed that detergent-solubilized protein complexes of both CSLD3 and CESA6 were organized into higher order complexes of similar molecular size ( Figure 7A), and ultrastructural analysis of these complexes by transmission electron microscopy revealed that both CSLD3 and CESA6 were present in particles of ~10-12 nm diameter with an apparent three-fold symmetry ( Figure 7C). Both the overall size and three-fold symmetry are remarkably similar to proposed trimeric structures described for individual lobes of rosette terminal complexes from P. patens  and trimeric 15 complexes observed for recombinantly expressed and purified CESA1 catalytic domains .
Why might land plants have evolved two distinct families of β-1,4-glucan synthases that can assemble into higher-order complexes? One of the distinguishing features of cellulose deposition in plant lineages is the organization of cellulose into paracrystalline microfibrils, whose deposition is associated with the orientation of an underlying array of cortical microtubules (Ledbetter and Porter, 1963). Indeed, the delivery (Crowell et al., 2009;Gutierrez et al., 2009), and migration upon cortical microtubules (Paredez et al., 2006;Desprez et al., 2007) appear to be intimately regulated by association of plasma membrane CSCs with a number of microtubule-associated proteins (Bringmann et al., 2012;Li et al., 2012;Lei et al., 2013;Liu et al., 2016). Earlier analysis of the CESA and CSLD gene families revealed that the CSLD genes display more diversity in their intron/exon organization, perhaps indicating that these are the older evolutionary group (Richmond and Somerville, 2001;Hazen et al., 2002;Little et al., 2018). Consistent with this, CSLD gene families are generally larger and more diverse in green algae such as Coleochaete orbicularis, and bryophytes including the moss P. patens, and the vascular seedless plant Selaginella moellendorffii (Harholt et al., 2012;Mikkelsen et al., 2014).
CSLDs are essential for the protonemal tip-growth that occurs during vegetative growth in P. patens (Roberts and Bushoven, 2007). CSLD-dependent cell wall deposition appears to be essential in the apical plasma membranes of tip-growing cells (Doblin et al., 2001;Favery et al., 2001;Wang et al., 2001;Park et al., 2011) and during cell plate deposition in dividing cells (Gu et al., 2016;Yang et al., 2016): two cellular contexts in which plasma membrane-associated cortical microtubules are generally absent (Emons and Ketelaar, 2012). One possibility is that CSLD complexes represent ancestral CSCs that synthesize cellulose microfibrils, but in a manner not as tightly associated with cortical microtubule organization. Evidence for a distinct, randomly distributed fibrillar cell wall element has been described in the apical domain of tip-growing root hairs (Newcomb and Bonnett, 1965;Akkerman et al., 2012;Emons and Ketelaar, 2012). Alternatively, while CESA-containing CSCs assemble into larger six-lobed rosette-like complexes, the CSLD-containing complexes we have observed (Fig 7C), may not. A major unresolved question is whether these CSLD3-containing particles assemble into similar rosette complex configurations in vivo, as with CESA-containing complexes, and whether the β-1,4-glucan polymers synthesized by these oligomeric CSLD complexes assemble into cellulose microfibrils that are similar or distinct in nature to those generated by CSCs containing CESA proteins.

Plant material and growth conditions
Arabidopsis thaliana lines used in this study were derived from Col-0 ecotype. The pCESA6::eYFP-CESA6 expressing line was kindly provided by Chris Somerville, UC-Berkeley (Paredez et al., 2006). Seeds were sterilized with 10% Clorox bleach solution, rinsed five times with distilled water, then stored at 4°C for 2 days before being plated on growth medium comprised of 0.25X Murashige and Skoog Basal Medium,1% sucrose, and 0.6% phytagel. Plates were placed vertically in a growth chamber at 21°C and grown under long-day conditions (16 hours light (200 uE/m 2 s)/8 hours dark photoperiod). For dark-grown conditions, plates were wrapped in aluminum foil. Three-day-old dark-grown seedlings were used for microscopy analysis. Five-day-old seedlings were used for morphology analysis. For propagation of mature plants, 14-day old seedlings were transferred to soil and grown in environmental chambers at 21°C under long-day conditions (16 hours light/8 hours dark photoperiod).

Yeast expression plasmid construction and growth conditions
S. cerevisiae (Strain INVSc1, Thermo Fisher, Cat#: C81000) was used for protein expression.
Untransformed yeast was cultured in YPD medium. Positive colonies containing pYES2/NT C plasmids (Thermo Fisher, Cat#: V825220), expressing N-terminal His-tagged CESA6, CSLD3, or CSLA9, were selected and cultured overnight at 30°C and 180 rpm in SC-Ura + Glucose medium composed of 1.9 g/L SC-Ura (uracil drop-out) powder, 1.7 g/L yeast nitrogen base without amino acids and ammonium sulfate, 5 g/L ammonium sulfate, and 20 g/L glucose. Yeast cells were harvested, rinsed in sterile water, and used to inoculate 200 ml of SC-Ura + Raffinose medium with the same nitrogenous base composition containing 20 g/L raffinose to an OD600 equal to 0.03. Cultures were grown for 14 to 16 h at 30°C and 180 rpm until the OD600 reached 2.0. Protein expression was induced by addition of 800 ml of SC-Ura + Galactose medium containing 20 g/L galactose, and cells were incubated for an additional 6 h at 30°C and 180 rpm. Yeast cells were harvested, weighed, flash frozen in liquid nitrogen, and stored at -80°C.

Plant expression plasmid construction and plant transformation
The CESA6 promoter was amplified (2251 bp upstream of ATG, primers shown in Supplemental Figure   8) and cloned into a pCAMBIA1301 vector upstream of a Citrine fluorescent protein coding sequence 17 (Griesbeck et al., 2001), replacing the 35S promoter to generate pCambia1301:pCESA6. To construct the CESA6:D3CD chimeric protein coding sequence, the following three fragments were assembled: a CESA6 N-terminal domain fragment corresponding to CESA6 amino acids 1-321, a CSLD3 catalytic domain fragment corresponding to CSLD3 amino acids 340-921, and a CESA6 C-terminal fragment corresponding to CESA6 amino acids 861-1084. These three DNA fragments were ligated together and integrated into the pCambia1301:pCESA6 vector using the Gibson assembly method (Gibson et al., 2009). Point mutations for CESA6 (E463A, D464A) and the CESA6:D3CD chimera (corresponding to CSLD3 E508A and D509A) coding sequences were generated using PCR (shown in Supplemental Figure 8). Resulting N-and C-terminal fragments of the CESA6 and CESA6:D3CD chimera sequences were ligated together and integrated into the pCambia1301:pCESA6 vector by Gibson assembly.
Plasmids were transformed into Agrobacterium tumefaciens strain GV3101 and then transferred to Arabidopsis using the standard floral dip method (Clough and Bent, 1998).

Hypocotyl and root length morphology analysis
Images of 5-day-old seedlings were recorded using an Epson Perfection 4990 Photo scanner. The lengths of hypocotyl and root regions were measured using Fiji-ImageJ (Schindelin et al., 2012). All transformed lines were grown side by side on the same plate, and at least 15 individuals were measured per line. Three independent biological replicates were performed for each line.

Cellulose content
Ten-day-old dark-grown seedlings (with seed coats attached) were collected and rinsed five times with distilled water to remove sucrose and residual MS medium. Samples were frozen in liquid nitrogen, ground to a fine powder, suspended in 80% (v/v) ethanol, filtered through a 45 µm nylon mesh (Industrial Netting, Cat#: WN0045), and then washed with 80% (v/v) ethanol followed by 100% ethanol.
Cell wall residue was resuspended in a solution of chloroform: methanol (1:1) and shaken slowly for 2 h at room temperature. Cell wall residue was collected by filtration through 45 µm nylon mesh, and washed extensively with acetone, yielding AIR (alcohol insoluble residue) for Updegraff analysis to determine cellulose content (Updegraff, 1969). Briefly, 3 ml of acetic/nitric reagent was added to 2 mg of AIR and boiled in a water bath for 30 minutes. Insoluble crystalline cellulose residue was collected by sedimentation in a Sorvall ST 16R centrifuge with TX-400 swinging bucket rotor at 5000 rpm for 5 minutes at room temperature., rinsed with 5 ml distilled water, resuspended in 1 ml of 67% sulfuric acid, 18 and incubated for 1 h at room temperature. 5 mg of pure cellulose (Sigma, Cat#: C0806) was dried at 105°C for 6 h and dissolved in 1 ml of 67% sulfuric acid, and then 50 ml distilled water was added to generate 100 µg/ml cellulose-sulfuric acid solution stock. The cellulose content was quantified by the anthrone assay with a standard curve containing 0, 4, 10, 20, 30 µg/ml cellulose-sulfuric acid solution diluted from the 100 µg/ml cellulose-sulfuric acid solution stock.

Fluorescent imaging, FRAP analysis, and co-localization analysis
Images were acquired using a Leica confocal laser scanning microscope SP8 using a 100X oil lens (Type F Immersion oil, NA=1.518) and processed with the Leica Application Suite X (LAS X) Life Science Microscope software. YFP and citrine fluorescence were excited at 514 nm and visualized from 519 nm to 650 nm. CSC particle movements were collected at 10 second intervals. Raw images were enlarged from 256*256 pixel images to 512*512 pixel images with Adobe Photoshop and imported into Fiji-ImageJ (Schindelin et al., 2012) to generate time projections using the Stacks function. CSC tracks were recorded using the segmented lines tool and analyzed by the Kymograph function in ImageJ. Particle velocities were calculated based on the distances measured in the kymograph over time. Photobleaching experiments (shown in Supplemental Figure 3) were performed by excitation of an ROI (red box) using a 405 nm laser, followed by collection of images at 10 second intervals. The red boxed region was cropped post-collection, using Adobe Photoshop, and CSC particle numbers were analyzed using the Spot Counter function in ImageJ (Box size, 3; Noise Tolerance, 30). For colocalization analysis, YFP/citrine and GFP fluorescence were excited at 488 nm and visualized from 524 nm to 650 nm, and 494 nm to 520 nm, respectively.
The supernatant was discarded and the total membrane pellet was gently resuspended in 5 ml 19 resuspension buffer (50 mM Hepes pH=7.4, 300 mM NaCl, 1 mM MgCl2, 1 mM MnCl2, 5 mM cellobiose, 50 mM Pefabloc SC plus, 1 mM PMSF, 2% LFCE-14 (Anatrace, Cat#: L414)) and incubated at 4°C for 30 minutes with gentle end-over-end shaking. Resuspended membranes were then spun at 100,000 x g in a Fiberlite F65L-6 X 13.5 rotor at 4°C for 1 h, and the supernatant was carefully collected and incubated at room temperature with Ni-NTA slurry (Thermo Fisher, Cat#: 88221) for 1 h. The slurry was transferred to a disposable chromatography column (BioRad, Cat#: 7321010) and washed with 5 ml wash buffer (50 mM Hepes pH=7.4, 300 mM NaCl, 1 mM MgCl2, 1 mM MnCl2, 5 mM cellobiose, 0.05% LFCE-14, 30 mM imidazole). Protein fractions (2.5 ml) were eluted from the Ni-NTA column with a 10 ml linear gradient of 30-250 mM imidazole in wash buffer. Protein fractions containing the His-tagged cell wall synthase enzymes were concentrated into wash buffer lacking imidazole using an Amicon Ultra 15 ultracel 100k centrifugal filter units (Sigma, Cat#: UFC910024) at 4,000 x g for 10 minutes. For reconstitution of purified cell wall synthases into proteoliposomes, 10 mg of yeast lipids (Avanti, Cat#: 190000P) were dissolved in 1 ml chloroform in a glass test tube, and then evaporated with nitrogen and dried in a vacuum chamber at room temperature for 1 h. The resulting yeast lipid film was resuspended in reconstruction buffer (50 mM Hepes pH=7.4, 300 mM NaCl, 1 mM MgCl2, 1 mM MnCl2, 5 mM cellobiose, 6% LFCE-14), and mixed with vigorous vortexing. Purified, detergent-solubilized proteins were mixed with 300 µl of the solubilized lipid fraction in a protein-to-lipid molar ratio of 1: 4000 and incubated for 1 h at 4°C with gentle end-over-end shaking. Meanwhile, 0.2 g SM2 Adsorbent beads (Bio-Rad, Cat#: 1528920) were washed with 1 ml bead buffer (50 mM Hepes pH=7.4, 300 mM NaCl, 1 mM MgCl2, 1 mM MnCl2, 5 mM cellobiose) for 1 h at 4°C with gentle end-over-end shaking. The 300 µl protein-lipid mixture was diluted with 600 µl bead buffer and incubated with 0.2 g of pre-washed SM2 Adsorbent beads for 1 h. An additional 900 µl of bead buffer was added to the protein-lipid mixture and the resulting 1800 µl protein-lipid mixture was transferred to a new tube containing 0.2 g of pre-washed SM2 Adsorbent beads and incubated for 1 h with gentle end-over-end rotation. This step was repeated twice more (for a total of four SM2 Adsorbent bead extractions) to completely extract the detergent.

Size exclusion chromatography and immunoblot analysis
3 mg of purified, detergent-solubilized His-CSLD3, His-CESA6, and His-CSLA9 proteins in 5 ml wash buffer (50 mM Hepes pH=7.4, 300 mM NaCl, 1 mM MgCl2, 1 mM MnCl2, 5 mM cellobiose, 0.05% LFCE-14), were injected onto a HiPrep Sephacryl S-300 HR column equilibrated with wash buffer and run at a continuous flow rate of 0.5 ml/minute. 2 ml fractions were collected starting at 30 ml, concentrated using Ultra 15 ultracel 100k centrifugal filter units, and every second fraction was analyzed by SDS-PAGE followed by either Coomassie Blue staining or immunoblotting. For immunoblotting, proteins were transferred to nitrocellulose membrane (GE Healthcare, Cat#: 10600003) by wet transfer at 15 V overnight. Membranes were blocked with 5% skim milk in PBST containing 0.05% Tween-20.

Computational Model Building
The catalytic domain of RsBCSA (PDB ID: 5EIY (Morgan et al., 2016), residue indices 13 to 740 in UniProt sequence Q3J125) was extracted for our model. Structures for CSLD3,CESA6, and CSLA9 were constructed via homology modeling with PDB ID: 5EIY using the SWISS-MODEL webserver (Guex et al., 2009;Benkert et al., 2011;Bertoni et al., 2017;Bienert et al., 2017;Waterhouse et al., 2018). Protonation of titratable residues was performed at pH 7.2 using the H++ webserver (Anandakrishnan et al., 2012). Simulation-ready models were prepared using the Amber 16 software suite (David A. Case, 2016). The models were solvated in explicit TIP3P water molecules (Jorgensen et al., 1983) in a periodic box such that there were at least 10 Å of solvent on each side, and the protein was parameterized using the Amber ff14sb force field (Maier et al., 2015). The divalent cations in each enzyme's active site (Mg 2+ in BCSA and Mn 2+ in the plant enzymes) were parameterized using the Li/Merz parameters for TIP3P (Li et al., 2015), the product sugar chains used the GLYCAM06j force field (Kirschner et al., 2008), and the UDP-Glc and GDP-Man substrate parameters were taken from (Pavla Petrová, 1999). After solvation, chlorine ions were added to neutralize the overall system charge.
Product sugar chains were added to each model based on homology to the position of the cellulose chain in the PDB ID: 5EIY crystal structure. The mannan chain was used in the CSLA9 structure by inversion of the chirality at the second carbon. Amino acid sequences from Arabidopsis CESA, CSLD, and CSLA family members were aligned by MUSCLE (Edgar, 2004) using MEGA 7 (Kumar et al., 2016.

Simulations and Energy Measurements
The initial coordinates for the UDP-Glc substrate in RsBCSA were taken directly from the PDB ID: 5EIY crystal structure. Each model was prepared for simulation in Amber 16 with 2500 steps of minimization followed by heating from 100 K to 300 K over 10,000 2-fs steps in the NTV ensemble. Production 22 simulations then followed in the NTP ensemble with a 2-fs timestep. The cutoff distance for non-bonded interactions was 8 Å and the temperature was maintained at 300 K using the Andersen thermostat scheme with a randomization interval of 100 steps (Andersen, 1980). Covalent bonds to hydrogen atoms were restrained using the SHAKE algorithm (Jean-Paul Ryckaert, 1976). Five parallel production simulations were performed for each combination of enzyme and substrate. First, 2-ns simulations were performed for BCSA, and the lowest-energy conformations of each substrate were fitted into the homology models as initial coordinates for longer, 6-ns simulations, of which the first nanosecond was discarded from each for equilibration. Substrate binding energies were measured using the MMPBSA.py software (Miller et al., 2012) packaged in Amber 16. The solvent was stripped out from the models using the CPPTRAJ (Roe and Cheatham, 2013) and ParmEd software (https://github.com/ParmEd/ParmEd) packaged in Amber 16. The product chain in each model was included in the "protein" portion of the binding energy calculations (as opposed to the "ligand" portion), in order to isolate the binding energies of the nucleotide sugars alone.