Structural snapshot of a glycoside hydrolase family 8 endo-β-1,4-glucanase capturing the state after cleavage of the scissile bond

The structure of an endo-β-1,4-glucanase from the cellulose-producing bacterium Enterobacter sp. CJF-002 bound to cellooligosaccharide was determined.


Introduction
Worldwide, cellulose (-1,4-glucan) production mainly relies on plants, but some bacteria can also produce a type of cellulose named bacterial cellulose (BC; Somerville, 2006;Ross et al., 1991). Cellulose-producing bacteria secrete BC as an exopolysaccharide, forming a hierarchical assembly outside the cell. Unlike the lignocellulosic matrix derived from plant cell walls, BC displays a naturally high purity and high crystallinity (Chundawat et al., 2011;Czaja et al., 2004). BC also has high mechanical strength (Yamanaka et al., 1989), biodegradability (Torgbo & Sukyai, 2020) and biocompatibility (Helenius et al., 2006). These unique characteristics make BC an attractive environmentally friendly resource for the development and manufacture of biomedical or industrial materials such as wound dressings (Czaja et al., 2006), artificial blood vessels (Lee & Park, 2017), electronic paper displays (Nogi & Yano, 2008;Shah & Brown, 2005) and speaker diaphragms (Nishi et al., 1990). Despite numerous efforts to improve BC production (Lee et al., 2014), such as optimization of the composition of culture media, bioreactors and bacterial strains, bacteria do not satisfy the demand for BC. Understanding the mechanism of BC biosynthesis will therefore help to produce BC more efficiently, and has become a hot topic in BC research.
Cellulose-producing bacteria biosynthesize BC using a multi-protein complex (terminal complex; TC) localized on the outer and inner membranes. Komagataeibacter xylinus (formerly called Acetobacter xylinum) serves as a model bacterium for BC biosynthesis. In K. xylinus, the TC is composed of BC synthase subunits A, B, C and D (BcsA, BcsB, BcsC and BcsD, or alternatively BcsAB, BcsC and BcsD; Kawano et al., 2002;Wong et al., 1990). Analyzing the individual structures of the subunits clarified their roles in BC biosynthesis (Acheson et al., 2019;Nojima et al., 2017;Fujiwara et al., 2013;Morgan et al., 2013Morgan et al., , 2016Hu et al., 2010). The production of high-quality BC depends on both the TC and some accessory factors such as cleaner proteins. The cmcax gene, which is located upstream of the BC synthase operon in K. xylinus, encodes CMCax, an endo--1,4-glucanase (EC 3.2.1.4) that hydrolyzes carboxymethyl cellulose (CMC) and cellooligosaccharides (Yasutake et al., 2006;Standal et al., 1994). Disruption of the cmcax gene reduced the amount of BC produced (Nakai et al., 2013), while overexpression of CMCax increased it (Kawano et al., 2008). Furthermore, adding a CMCax antibody to the culture medium inhibited the formation of BC fibers (Koo et al., 1998). Considering the hydrolytic activity of CMCax, endo--1,4-glucanase could act as a cleaner protein to produce high-quality BC by trimming disordered cellulose. Interestingly, BcsZ, an endo--1,4glucanase encoded in the BC synthase operon, also exists in most bacterial lineages (Rö mling & Galperin, 2015). Recent studies showed that Enterobacter sp. CJF-002, isolated from rocks in oil reservoirs, efficiently produces BC and is a promising bacterium for producing large amounts of BC (Sunagawa et al., 2012). In Enterobacter sp. CJF-002, a single operon encodes BcsA, BcsB, BcsZ and BcsC. Although CMCax and BcsZ have low sequence similarity (19% identity), the residues located in the catalytic site are highly conserved ( Supplementary Fig. S1). BcsZ from Enterobacter sp. CJF-002 (EbBcsZ; GenBank accession No. BAM44856) and cellulolytic Escherichia coli (EcBcsZ) also hydrolyze cellooligosaccharides and CMC (Pang et al., 2019;Sunagawa et al., 2012;Mazur & Zimmer, 2011), indicating that BcsZ and CMCax are functionally homologous. Therefore, a biological function of EcBcsZ could be to promote the formation of cellulose fibers by degrading -1,4-glucan chains (Mazur & Zimmer, 2011). Similar to endo--1,4-xylanases (EC 3.2.1.8), licheninases (EC 3.2.1.73), chitosanases (EC 3.2.1.132) and reducing-end xylose-releasing exo-oligoxylanases (EC 3.2.1.156), CMCax and BcsZ belong to glycoside hydrolase family 8 (GH8; CAZy; http://www.cazy.org). The catalytic reaction of GH8 enzymes is a single-displacement mechanism accompanied by an inversion of the anomeric configuration of the sugar unit in subsite À1, which changes the conformation of pyranose from -4 C 1 to -4 C 1 (Koshland, 1953;Cremer & Pople, 1975; Fig. 1, Supplementary Fig. S1). The general acid catalyst donates a proton to the C1 atom to release the hemiacetal sugar. The general base catalyst activates a water molecule for a nucleophilic attack on the anomeric center (C1) of the substrate. Displacement of the activated water molecule and liberation of the leaving group of the product occur on the C1 atom of a distorted oxocarbenium ion intermediate, forming an -configuration at the C1 atom. To understand the reaction mechanism of GH8 enzymes, structures of GH8 enzymes and their acid catalyst mutants bound to oligosaccharides have been analyzed to date. In the structure of the E55Q mutant of EcBcsZ bound to cellopentaose (CPT) [EcBcsZ(E55Q) CPT ; PDB entry 3qxq], the glucosyl unit in subsite À1 (G À1 ) presented an -4 C 1 conformation, showing that EcBcsZ (E55Q) CPT is the product-binding form (Mazur & Zimmer, 2011). In contrast, the structure of the E95Q mutant of CelA from Clostridium thermocellum bound to CPT [CtCelA (E95Q) CPT ; PDB entry 1kwf] showed that G À1 presented a 2,5 B conformation and thus CtCelA(E95Q) CPT was an intermediate in the transition state (Gué rin et al., 2002). Interestingly, the structure of wild-type endo--1,4-xylanase from Teredinibacter turnerae bound to xylotriose (XTO) (TtGH8 XTO ; PDB entry 6g0b) confirmed the existence of a 2,5 B conformation of the xylosyl unit in subsite À1 (X À1 ; Fowler et al., 2018). Based General reaction mechanism of inverting glycoside hydrolases. The upper part shows the proposed conformational changes of the sugar in subsite À1 (G À1 ). -4 C 1 , 2,5 B and -4 C 1 are the conformations of G À1 in the substrate, oxocarbenium ion intermediate (transition state) and product, respectively. 2 S O and 5 S 1 are the conformations of the other intermediates. The lower part shows the proposed catalytic process of GH8 enzymes. The reaction follows the single-displacement mechanism through the oxocarbenium ion intermediate. The numbers in the left part represent the order of the displacement reaction, i.e. protonation of the glycosidic bond occurs followed by nucleophilic attack of an activated water molecule. G À1 is shown as a six-membered sugar. The substrate, transition intermediate and product are shown in the left, middle and right panels, respectively. Before and after forming the oxocarbenium intermediate, G À1 forms 2 S O and 5 S 1 conformations, respectively. on these structures, a computational simulation predicted that in the reaction process of GH8 enzymes the G À1 conformational change sequence was -4 C 1 ! 2 S O ! 2,5 B! 5 S 1 !-4 C 1 (Ardè vol & Rovira, 2015;Petersen et al., 2009;Fig. 1). However, the coordinates of the 2 S O and 5 S 1 intermediates remain unclear since no structural data are available to prove the existence of such intermediates of GH8 enzymes.
In this study, we assessed the hydrolytic activity of EbBcsZ on BC produced by K. xylinus. Based on the idea that BcsZ hydrolyzes cellulose through an inverting mechanism, we mutated the base catalyst Asp242 to alanine (D242A) and determined the structure of this mutant in complex with a cellooligosaccharide [EbBcsZ(D242A) CPT ]. The EbBcsZ (D242A) CPT structure showed a novel snapshot of a reaction stage immediately after scissile-bond cleavage. It displayed a distorted 5 S 1 conformation of G À1 . Moreover, the structure of wild-type EbBcsZ in complex with glycerol (EbBcsZ GOL ) supported the existence of the 5 S 1 conformation in the reaction process of BcsZ. We finally compared our structures with those of other GH8 endo--1,4-glucanases, extended the previously described reaction mechanism of GH8 enzymes and proposed a -1,4-glucan-trimming mechanism for EbBcsZ.

Cloning, expression and purification
Except for the N-terminal signal peptide (Met1-Ala20), we amplified the gene fragment encoding EbBcsZ (amino acids Ala21-Gln367) by PCR using the genome of Enterobacter sp. CJF-002 as a template using the primers 5 0 -GGAATTCCA TATGGCCTGCACATGGCCTGC-3 0 and 5 0 -CCGCTCGAG TTACTGTGAACTTGCGCATGCCTG-3 0 . We used NdeI and XhoI to digest the amplified gene fragments and the protein expression vector pET-28b (Novagen) with a hexahistidine tag and thrombin cleavage sequence at the N-terminus, and then ligated them. We transformed the expression plasmids into E. coli strain BL21(DE3) by electroporation. We grew the cells in LB medium supplemented with 25 mg ml À1 kanamycin at 37 C until the OD 600 reached 0.5. We then induced overexpression of recombinant EbBcsZ by adding 100 mM isopropyl -d-1-thiogalactopyranoside to the culture medium and cultured the cells overnight at 25 C. After resuspending the harvested cells in buffer A [50 mM HEPES-KOH buffer pH 8.0 containing 500 mM potassium chloride and 10%(v/v) glycerol], we lyzed the cells by sonication. We centrifuged the cell lysate at 40 000g for 15 min at 10 C and then loaded the supernatant onto a nickel-affinity chromatography column (5 ml HisTrap HP; Cytiva) equilibrated with buffer A. We then eluted the bound EbBcsZ with buffer B [50 mM HEPES-KOH buffer pH 8.0 containing 500 mM potassium chloride, 10%(v/v) glycerol and 500 mM imidazole] in a linear gradient of imidazole from 50 to 500 mM. We collected the eluate containing EbBcsZ and then removed the N-terminal hexahistidine tag using thrombin (Fujifilm Wako) overnight at 4 C. We then again loaded EbBcsZ onto a nickel-affinity chromatography column and collected the eluate containing the protein, which had lost its affinity for the column. Finally, we purified EbBcsZ on a Superdex 200 16/600 column (Cytiva) equilibrated with buffer C (50 mM HEPES-KOH buffer pH 8.0 containing 150 mM potassium chloride). We introduced the D242A mutation using the designed primers by the QuikChange technique and confirmed it by sequencing. We prepared the EbBcsZ(D242A) mutant as described above for wild-type EbBcsZ. We concentrated the purified EbBcsZ and EbBcsZ(D242A) to a final concentration of 9 mg ml À1 by ultrafiltration using Vivaspin 6 concentators with a 10 kDa cutoff membrane (Cytiva).

Crystallization and X-ray diffraction data collection
We performed initial crystallization screening for both proteins using sparse-matrix crystallization kits from Qiagen, Hilden, Germany. We mixed 0.5 ml protein solution with an equal volume of reservoir solution and used the sitting-drop vapor-diffusion method. We obtained EbBcsZ and EbBcsZ(D242A) crystals using the same reservoir solution consisting of 0.2 M disodium tartrate, 20%(w/v) PEG 3350. Crystals grew to dimensions of 0.3 Â 0.2 Â 0.2 mm within three days at 20 C. We transferred the EbBcsZ(D242A) crystals into soaking solution [0.2 M disodium tartrate, 20%(v/v) PEG 3350, 20%(v/v) glycerol, 2 mM CPT] and incubated them for 20 h at 20 C. We soaked the EbBcsZ crystals in reservoir solution supplemented with 20%(v/v) glycerol and then flash-cooled both crystals under a stream of liquid nitrogen. We collected diffraction data on beamlines NW-12A at Photon Factory, Tsukuba, Japan and BL41XU at SPring-8, Hyogo, Japan and processed the data using XDS (Kabsch, 2010). Table 1 summarizes the data statistics.

Structure determination and refinement
To determine the structure of EbBcsZ in complex with glycerol (EbBcsZ GOL ), we used the molecular-replacement (MR) method with AutoMR from the Phenix package (Liebschner et al., 2019) using the EcBcsZ apo structure (PDB entry 3qxf; Mazur & Zimmer, 2011) as the search model. We calculated the rotation and translation functions using data in the resolution range 45.0-3.0 Å . Similarly, we determined the EbBcsZ(D242A) CPT structure by the MR method with AutoMR using the EbBcsZ GOL structure as the search model. We calculated rotation and translation functions using data in the resolution range 45.0-2.5 Å . We performed several rounds of refinement using phenix.refine from the Phenix suite, alternating with manual fitting and rebuilding based on 2F o À F c and F o À F c electron-density maps using Coot (Emsley et al., 2010). We confirmed the conformation of the -1,4-glucan chains based on the omit map. Table 1 summarizes the final refinement statistics and geometry defined by MolProbity (Chen et al., 2010). We prepared the structural figures using PyMOL (http://www.pymol.org).

research papers 2.4. BC hydrolysis assay
To perform the BC hydrolysis assay, we cultured K. xylinus to obtain BC based on the method reported previously (Czaja et al., 2004). BC is porous and incorporates water molecules, forming a gelatinous structure. To reduce the variation in the produced BC, we collected BC from a cultured medium as follows. We placed K. xylinus cells into 5 ml Schramm-Hestrin (HS) medium (Schramm & Hestrin, 1954) and initially grew them at 30 C with stirring at 170 rev min À1 for four days. We then inoculated 0.1 ml cultured medium into 3 ml fresh HS medium and grew the cells at 30 C with stirring at 120 rev min À1 for three days. The dry mass of BC was referenced each time to control the amount of produced BC. We collected the gelatinous BC membrane and gently washed it with 3 ml buffer D (50 mM sodium citrate buffer pH 5.0 containing 200 mM sodium chloride). To measure BC hydrolysis, the solution containing the purified EbBcsZ and EbBcsZ(D242A) was changed to buffer D. We performed repetitive concentration and dilution of purified EbBcsZ and EbBcsZ(D242A) by ultrafiltration using Vivaspin 6 concentators with a 10 kDa cutoff membrane, and finally adjusted the concentration of the samples to 10 mg ml À1 . We dissolved bovine serum albumin (BSA; Sigma-Aldrich) and cellulase ONOZUKA RS (Yakult Pharmaceutical Industry) in buffer D to a concentration of 10 mg ml À1 as negative and positive controls, respectively. We transferred the prepared gelatinous membrane of BC into 3 ml of each enzyme solution and incubated the mixtures at 30 C with stirring at 170 rev min À1 for 20 h. Since the turbidity of the BC solution increases as BC is hydrolyzed, we observed the turbidity of each enzyme solution to estimate the BC hydrolysis. Finally, we measured the absorbance of each sample at 600 nm. We performed four technical replicates for each condition.
To estimate the degree of BC hydrolysis, we measured the absorbance at 600 nm representing the turbidity of BCcontaining solutions in the presence or absence of EbBcsZ. In the positive control, BC hydrolysis allowed the cells embedded in gelatinous BC to disperse into the solution, increasing the turbidity (Fig. 2, left). Incubating the BCcontaining solution with wild-type EbBcsZ partly degraded  i jI i ðhklÞ À hIðhklÞij= P hkl P i I i ðhklÞ, where i is the number of observations of a given reflection and hI(hkl)i is the average intensity of the i observations. ‡ R work = P hkl jF obs j À jF calc j = P hkl jF obs j. § R free was calculated using a randomly selected 5% of reflections that were excluded from refinement. gelatinous BC (Fig. 2a). The turbidity of this solution was 40% of that of the positive control (Fig. 2b), indicating that EbBcsZ has a moderate hydrolytic activity on BC. In contrast, the base catalyst mutant EbBcsZ(D242A) barely changed the shape of gelatinous BC (Fig. 2a, right), and the turbidity of the BC solution incubated with EbBcsZ(D242A) was comparable to that of the negative control (Fig. 2b, right).

The structure of EbBcsZ in complex with cellooligosaccharide
Similar to the homologous enzyme EcBcsZ (Mazur & Zimmer, 2011), we prepared EbBcsZ without the N-terminal signal peptide (Met1-Ala20) and with a mutation of the base catalyst (D242A) to elucidate the reaction mechanism of EbBcsZ. We determined the structure of EbBcsZ(D242A) CPT at 1.30 Å resolution by soaking EbBcsZ(D242A) crystals in a solution containing CPT. In this EbBcsZ(D242A) CPT structure, the residues Cys22-Trp359 were visible, while the N-terminal Ala21 and the eight C-terminal residues (Gly360-Gln367) could not be built due to poor electron density. The overall structure of EbBcsZ(D242A) CPT formed a classical (/) 6 barrel composed of six inner and lateral antiparallel helices. Although EbBcsZ is a monomeric enzyme, the asymmetric unit contained four EbBcsZ(D242A) CPT monomers. These four monomers were essentially identical, with a root-mean-square deviation (r.m.s.d.) value of less than 0.41 Å for 335 C atoms. We selected monomer A for structural comparison and description of the active site. Structural comparison using PDBeFold (Krissinel & Henrick, 2004) showed that EbBcsZ(D242A) CPT was similar to the other GH8 endo--1,4-glucanases EcBcsZ(E55Q) CPT (PDB entry 3qxq), CtCelA(E95Q) CPT (PDB entry 1kwf) and CMCax (PDB entry 1wzz), with r.m.s.d.s of 0.65 Å for 338 C atoms, 2.09 Å for 270 C atoms and 2.02 Å for 278 C atoms, respectively. Although we soaked the crystals in a CPT-containing solution, we did not obtain full-length CPT in the EbBcsZ(D242A) CPT structure, unlike in the EcBcsZ(E55Q) CPT and CtCelA (E95Q) CPT structures. In each monomer of the asymmetric unit, two cellooligosaccharides occupied the plus and minus subsites (both sides of the cleavage position) of the active site (Fig. 3a, Supplementary Fig. S3), indicating an intermediate state that differs from those of reported GH8 endo--1,4glucanase structures. In all four monomers, cellobiose binds to subsites +1 to +2 similarly, whereas the cellooligosaccharide binds to the minus subsites differently. Cellotetraose bound to subsites À1 to À3 and extended to an additional subsite À4 in three monomers (monomers A-C), while cellotriose bound to subsites À2 to À4 in the remaining monomer (monomer D) (Fig. 3a). We next compared the EbBcsZ (D242A) CPT structure with those of other GH8 endo--1,4-glucanases. The cellobiose orientation in subsites +1 to +2 of EbBcsZ(D242A) CPT was similar to that of CtCelA(E95Q) CPT (Gué rin et al.,  2002). G +1 and G +2 formed stacking interactions with the aromatic residues in the same manner. However, there were differences in the interaction between the hydroxy group of G +1 and polar residues in the structures of both EbBcsZ(D242A) CPT and CtCelA (E55Q) CPT (Fig. 3b). Note that in the EbBcsZ(D242A) CPT   The structure of the active site in EbBcsZ(D242A) CPT . (a) Electron densities of the -1,4-glucan chains bound to the four monomers of the asymmetric unit. The electron densities are the omit map contoured at 3.0. (b) Interactions between the -1,4-glucan chains and the residues in the active site. We selected monomer A as the representative active site. The water molecules and the residues interacting with the -1,4-glucan chains are represented as red spheres and light blue lines, respectively. The hydrogen-bond network is represented as dotted lines. In (a) and (b), the -1,4glucan chains are shown as sticks and O, N and C atoms are shown in red, blue and dark cyan, respectively. (c) The conformation of G À1 (dark cyan) and five other glucosyl units in subsites G À2 , G À3 , G À4 , G +1 and G +2 (light gray) of EbBcsZ(D242A) CPT . (d) Superposed views of glucosyl units. Left: superposition of G À1 (dark cyan) and G À2 (light gray) in EbBcsZ(D242A) CPT . Right: superposition of G À1 in EbBcsZ(D242A) CPT (dark cyan) and G À1 in CtCelA(E95Q) CPT (yellow). Arg245 side chain presented two different conformations (Arg245 conf1 and Arg245 conf2 ). The Arg245 conf1 side chain interacts with the 4-OH of G +1 . Additionally, the side chain of the general acid catalyst Glu54, corresponding to E95Q in CtCelA and E55Q in EcBcsZ, interacted with the 3-OH and 4-OH of G +1 . Our structure also displayed an interaction network associated with Glu54. Glu54 interacts with Arg245 and Tyr330 through a salt bridge and a hydrogen bond, respectively. Except for G À1 , four glucosyl units interacted with the minus subsites in the same manner as those in EcBcsZ(E55Q) CPT and CtCelA(E95Q) CPT (Mazur & Zimmer, 2011;Gué rin et al., 2002). In EbBcsZ(D242A) CPT , the Asp115 side chains formed a bidentate interaction with G À1 , and the Glu54 side chain interacted with the 2-OH of G À1 . Furthermore, we observed an interaction between G À1 and G +1 in the EbBcsZ(D242A) CPT structure. The 6-OH and sugar-ring O (O5) atoms of G À1 interacted with the 3-OH and 4-OH of G +1 , respectively, through hydrogen bonds.

The glucosyl unit conformation in subsite À1
In the EbBcsZ(D242A) CPT structure, G À1 displayed a distorted 5 S 1 conformation, whereas the glucosyl units in the other subsites (G À2 , G À3 , G À4 , G +1 and G +2 ) presented a stable -4 C 1 chair conformation (Fig. 3c, Supplementary Table   S1). Moreover, considering the position of the 1-OH of G À1 , G À1 has the -configuration at the C1 atom. Comparing the C5 position of G À1 with those of the other glucosyl units clearly shows this difference (Fig. 3d, left). G À1 displayed a 2,5 B conformation in the CtCelA(E95Q) CPT structure, and comparing the C1 positions of G À1 in EbBcsZ(D242A) CPT and CtCelA(E95Q) CPT also shows the difference in the conformation of G À1 (Fig. 3d, right). To confirm whether the alanine mutation of the general base catalyst Asp242 causes structural changes, we determined the structure of EbBcsZ in complex with glycerol (EbBcsZ GOL ). We obtained EbBcsZ crystals in the same crystal form as those of EbBcsZ(D242A) CPT . Residues Cys22-Trp359 of each EbBcsZ GOL monomer were built in the asymmetric unit. EbBcsZ GOL and EbBcsZ (D242A) CPT had very similar structures, with an r.m.s.d. of 0.17 Å for 335 C atoms, demonstrating that the D242A mutation scarcely changed the overall EbBcsZ structure. In the EbBcsZ GOL structure, subsites À1, À2 and +1 were occupied by glycerol molecules used as a cryoprotectant for the diffraction experiment ( Supplementary Fig. S4a). These glycerol molecules can be considered to be a mimic of the bound glucosyl unit of -1,4-glucan. In each EbBcsZ monomer the glycerol molecule in subsite À1 adopted a remarkably similar orientation. Superposing EbBcsZ GOL on EbBcsZ(D242A) CPT showed that three hydroxy groups of the glycerol and two water molecules were close to the O5 atom and 3-OH, 6-OH, 1-OH and 2-OH of G À1 , respectively ( Supplementary Fig. S4b, Supplementary Table S2), supporting the plausibility of the formation of a distorted 5 S 1 conformation of G À1 during the hydrolytic reaction process. Besides, the existence of the 5 S 1 conformation is consistent with the predicted result of a quantum mechanics/molecular mechanics (QM/ MM) dynamic simulation on the reaction process of GH8 enzymes (Ardè vol & Rovira, 2015;Petersen et al., 2009).

The secondary binding site of the b-1,4-glucan chain
Among the GH8 endo--1,4glucanases, the structure of EbBcsZ (D242A) CPT is the first to show that -1,4-glucan binds to the surface away from the active site (Fig. 4a), which is called the secondary binding site (Cuyvers et al., 2012). Two of the four EbBcsZ(D242A) CPT monomers (monomers A and C) bind to cellotriose through a tryptophan-rich cleft (Fig. 4a). However, electron density for the cellooligosaccharide did not appear in the corresponding region in monomers The structure of the secondary binding site in EbBcsZ(D242A) CPT . (a) Close-up view of the secondary binding site. The cellotriose and the residues interacting with it are represented as sticks and lines, respectively. The hydrogen bonds between cellotriose and the residues of EbBcsZ (D242A) CPT are represented as dotted lines. (b) Superposition of the residues interacting with cellotriose in EbBcsZ and the other endo--1,4-glucanases. The residues of EbBcsZ(D242A) CPT (dark cyan), EcBcsZ(E55Q) CPT (PDB entry 3qxq, orange), CMCax (PDB entry 1wzz, pink), Cel10 (PDB entry 5gy3, blue), BcsZ (PDB entry 4q2b, gray) and hydrolase from Vibrio fischeri (PDB entry 5cd2, yellow) are labeled and represented as lines. The -hairpin region of EbBcsZ (D242A) CPT and the corresponding regions in the other endo--1,4-glucanases are represented as ribbons.
B and D, which can be considered to be an effect of crystal packing ( Supplementary Fig. S5). This marked difference implies that the physical and chemical environment easily affects the binding of -1,4-glucan at the secondary binding site. Cellotriose bound to monomers A and C in the same manner. The glucosyl units at the central part and the reducing end formed a T-shaped stacking with Trp97 andstacking with Trp105, respectively. Stacking interactions with aromatic residues are common in secondary binding sites and seem to be essential for oligosaccharide binding (Cockburn et al., 2016;Cockburn & Svensson, 2013;Cuyvers et al., 2011). The 2-OH of the reducing-end sugar and the 6-OH of the central sugar interacted with the side chains of Arg40 and Trp78, respectively. The 4-OH of the non-reducing-end sugar interacted with the main-chain carbonyl of Asp81 through a hydrogen bond. These contacts are probably conserved among the other GH8 endo--1,4-glucanases, except for the stacking with Trp105 in a -hairpin motif (Fig. 4b).

Discussion
BC, which is generated by cellulose-producing bacteria, is a biomaterial with remarkable characteristics that are superior to those of the cellulose fiber extracted from plants. It has potential applications in various fields. In BC-producing bacteria, the endo--1,4-glucanase serves as a cleaner protein that degrades disordered cellulose fibers, efficiently producing high-quality BC. Our work is focused on investigating the structural and functional properties of the endo--1,4-glucanase BcsZ in order to help to improve BC production.
A BC degradation assay showed that EbBcsZ has hydrolytic activity, although it is weaker than that of cellulase ONOZUKA RS, a multi-component cellulase. This weak activity may arise from limited degradation of the disordered region in gelatinous BC, which is in agreement with previous studies showing that EbBcsZ can degrade cellooligosaccharides and CMC to noncrystalline cellulose (Koo et al., 1998). Considering that the endo--1,4-glucanase EbBcsZ needs to access the -1,4-glucan chain to cleave internal bonds, secondary binding sites probably assist in the binding of the target substrate -1,4-glucan chain. Moreover, immunostaining of CMCax (Standal et al., 1994) and digestion of BcsZ from Salmonella typhimurium by a protease (Ahmad et al., 2016) demonstrated that these endo--1,4-glucanases are located close to the outer membrane or periplasmic space, allowing them to degrade -1,4-glucan chains before crystallization (Ahmad et al., 2016;Hu et al., 2010;Standal et al., 1994). A phylogenetic relationship shows that the gene encoding endo--1,4-glucanase appears with the bcsC gene in the bcs gene cluster, with some exceptions (Rö mling & Galperin, 2015). This little-studied BcsC subunit of TC localizes at the outer membrane and periplasmic space, and contributes to the export of BC to the extracellular matrix. Thus, endo--1,4-glucanase may be associated with BcsC to access noncrystalline -1,4-glucan chains efficiently.
Several studies have reported structures of wild-type GH8 endo--1,4-glucanases and general acid catalyst mutants. These studies considered that the apo-form structures of these GH8 endo--1,4-glucanases represented the 'resting state'. Here, we select EcBcsZ apo as representing the resting state (Fig. 5a, i) since it is most similar to EbBcsZ in both sequence and structure. In the structures of CtCelA(E95Q) CPT (Gué rin et al., 2002) and TtGH8 XTO (Fowler et al., 2018), a single oligosaccharide was incorporated into the subsites, and G À1 and X À1 showed a distorted 2,5 B conformation. These complex structures represent the 'reactive state' that readily forms the oxocarbenium intermediate (Fig. 5a, ii). In the EcBcsZ (E55Q) CPT structure cellopentaose was only located in the minus subsites, meaning that EcBcsZ(E55Q) CPT represents the 'post-hydrolysis state' that occurs after release of the leaving group (Fig. 5a, iv). We employed a base-catalyst mutant EbBcsZ(D242A) and determined its cellooligosaccharide-  Interactions between the bound -1,4-glucan chains and the polar residues of GH8 endo--1,4-glucanases in each reaction state. (a) The structures of (i) the resting state EcBcsZ apo , (ii) the reactive state CtCelA(E95Q) CPT , (iii) the post-cleavage state EbBcsZ(D242A) CPT and (iv) the post-hydrolysis state EcBcsZ(E55Q) CPT . The cellooligosaccharides and the residues are represented as sticks and lines, respectively. (b) Schematic diagram of the motion of the GH8 endo--1,4-glucanase residues during the reaction process. The glutamate and arginine of interest are represented as orange and blue sticks, respectively. When the sugar units interact with the glutamate and/or arginine, the solid circles are in the same color as the glutamate and/or arginine. bound structure, unlike previous studies. The EbBcsZ (D242A) CPT structure provides a novel snapshot of the reaction state, in which two -1,4-glucan chains lying in subsites À1 to À4 and +1 to +2 are separated. The distance between the C1 atom of G À1 and the 4-OH of G +1 is approximately 3 Å , which is longer than the typical length of a C-O single bond (1.43 Å ). Thus, EbBcsZ(D242A) CPT represents the 'postcleavage state', with G À1 in a distorted 5 S 1 conformation, immediately after the scissile-bond cleavage (Fig. 5a, iii).
The residues Glu54 and Arg245 in EbBcsZ(D242A) CPT differed in orientation compared with the corresponding residues in EcBcsZ(E55Q) CPT (E55Q and Arg246) and CtCelA(E95Q) CPT (E95Q and Arg281). Similar to the reactive state [CtCelA(E95Q) CPT ] and the post-hydrolysis state [EcBcsZ(E55Q) CPT ], the side chain of the acid catalyst Glu54 was located between G +1 and G À1 in the post-cleavage state [EbBcsZ(D242A) CPT ] (Fig. 5a, ii-iv). However, comparison between EbBcsZ(D242A) CPT and EcBcsZ apo indicated that when the substrate binds, the Glu54 side chain would simultaneously flip and interact with Tyr330 and Arg245 to avoid a clash with the 3-OH of G À1 (Figs. 3b and 5a, i). Thus, upon substrate binding, the acid catalyst side chain flips and keeps this conformation, while the cellooligosaccharide occupies subsite À1. In the resting state (EcBcsZ apo ), the N 1 atom of Arg246 forms a salt bridge to the O 2 atom of Asp116 and was distant from the binding subsites (Fig. 5a, i). In the reactive state [CtCelA(E95Q) CPT ], the corresponding Arg281 was distant from Asp152 and shifted to interact with G À1 (Fig. 5a, ii). In the post-cleavage state [EbBcsZ(D242A) CPT ], Arg245 presented two conformations: Arg245 conf2 was the same as that of the corresponding Arg281 of CtCelA in the reactive state, whereas Arg245 conf1 shifted to interact with G +1 (Figs. 3b and 5a, iii). In the post-hydrolysis state [EcBcsZ(E55Q) CPT ], Arg246 returned to interact with Asp116 and was distant from subsite +1 (Fig. 5a, iv). Therefore, this arginine residue has an important role in tethering the cleaved -1,4-glucan chains. Based on these findings, we summarized a series of conformational changes in the catalytic process of GH8 endo--1,4glucanase to illustrate the whole catalytic mechanism (Fig. 5b). The flexibility of the arginine residue allows the post-cleavage state to be captured. The superimposed structures of EbBcsZ(D242A) CPT and EbBcsZ GOL clearly showed that the side chain of Asp242 does not sterically hinder the conformational changes of G À1 and Arg245 (Supplementary Fig. S6). Considering the reaction mechanism of an inverting glycoside hydrolase (Fig. 1), the acid catalyst mutation prevents the formation of the transition intermediate, and the 5 S 1 conformation is invisible. The base catalyst activates a water molecule for a nucleophilic attack on the C1 atom of G À1 and does not directly attack G À1 . Even though the base catalyst is mutated, the reaction may proceed by the very weak nucleophilicity of water. Indeed, previous studies demonstrated that mutating the base catalyst drastically reduced, but did not completely remove, the activity of GH8 enzymes (Fowler et al., 2018;De Vos et al., 2006;Collins et al., 2005). The structural comparison of G À1 in the post-cleavage state 5 S 1 [EbBcsZ (D242A) CPT ] with the substrate state -4 C 1 [the other glucosyl unit in EbBcsZ(D242A) CPT ] and the transition state 2,5 B [CtCelA(E95Q) CPT ] clearly showed the difference in planarity of G À1 (Supplementary Fig. S3c). The boat 2,5 B conformer changes to a skew-boat 5 S 1 conformer, indicating that the hydrolysis reaction of GH8 enzymes precedes via an unstable conformation to meet the requirements of an antiperiplanar lone-pair hypothesis (Nerinckx et al., 2005).
In the structure of EbBcsZ(D242A) CPT , we obtained electron density for six glucosyl units in monomers A-C as shown in Fig. 3(a), although we utilized cellopentaose. Considering that no electron density appeared at subsite À1 in monomer D and the electron density for G À1 in monomers A-C clearly showed the 5 S 1 conformation, the electron density might be derived from a mixture of the post-cleavage state of cellopentaose in subsites À3 to +2 and cleaved cellotriose in subsites À4 to À2, but not noncleaved cellopentaose in subsites À4 to +1. Combining the binding of oligosaccharides in the structure of EbBcsZ(D242A) CPT with a previous thinlayer chromatography (TLC) analysis (Sunagawa et al., 2012), we proposed the binding patterns of oligosaccharides in the trimming mechanism of BcsZ (Fig. 6). The structure shows that EbBcsZ has six subsites, À3 to +2 and additionally À4, for binding the -1,4-glucan chain. The TLC analysis demonstrated  that EbBcsZ degrades cellopentaose into cellobiose and cellotriose but does not degrade cellotetraose. Thereby, five cellopentaose glucosyl units should lie in subsites À3 to +2 and not in subsites À4 to +1 or in the four subsites À2 to +2 (Fig. 6,  G4 and G5). The TLC analysis also showed that EbBcsZ degrades cellohexaose into two cellotrioses or into cellobiose and cellotetraose. Cellohexaose needs to have five glucosyl units in subsites À3 to +2 and a free glucosyl unit to produce two cellotrioses. The six glucosyl units of cellohexaose need to be in subsites À4 to +2 (Fig. 6, G6) to obtain cellobiose and cellotetraose. Consequently, EbBcsZ hydrolyzes cellopentaose or longer cellooligosaccharides when the -1,4glucan chain occupies at least subsites À3 to +2 (Fig. 6,  bottom). Moreover, considering that glycerol molecules preferentially bind to subsites À2 to +1 in the EbBcsZ GOL structure ( Supplementary Fig. S3c), the sugar units in subsites À3 and +2 might have a relatively lower affinity than those in subsites À2 to +1. Therefore, subsites À3 and +2 assist in binding the -1,4-glucan chain.

Conclusion
This work demonstrated that the GH8 endo--1,4-glucanase EbBcsZ partly degrades gelatinous BC, implying that it plays a potential role in the quality control of BCs. Unlike previous studies, we obtained the EbBcsZ(D242A) CPT structure using a base-catalyst mutant, revealing a 5 S 1 intermediate as the 'postcleavage state'. By comparing GH8 endo--1,4-glucanase snapshots, we provided insights into the motion of the residues coupled to the conformational change of the glucosyl unit in subsite À1 during the reaction process. Moreover, based on structural analysis of EbBcsZ(D242A) CPT , we proposed a -1,4-glucan-trimming mechanism for EbBcsZ to enhance the understanding of the biosynthesis of high-quality cellulose by bacteria, and our findings open up a route to a deep understanding of the enzymology of glycoside hydrolases.