Crystal structure of a homotrimeric verrucomicrobial exo-β-1,4-mannosidase active in the hindgut of the wood-feeding termite Reticulitermes flavipes

Graphical abstract


Introduction
Wood-feeding termites are known to cause substantial damage to man-made wood-based constructions, buildings and trees, with an estimated economical loss of 15 to 40 billion USD per year (Govorushko, 2019). Notwithstanding their destructive potential, termites are attracting considerable interest for biomass processing due to their gut microbiomes that produce a rich variety of wood-degrading enzymes (Peterson and Scharf, 2016). For a wood-based diet, termites rely on the obligate symbiotic relationship with specialized microbes that inhabit the saliva and digestive tract.
The well studied and highly invasive Eastern Mediterranean termite Reticulitermes flavipes (Kollar) is considered to be the most harmful and destructive termite, and responsible for causing considerable damage and economical loss (Su and Scheffrahn, 1990). The efficient saccharification of wood by R. flavipes is enabled by symbionts that evolved to target specifically recalcitrant lignocellulose and hemicellulose constituents, and the usefulness of this termite for processing of lignocellulosic biomass has been reported (Karl and Scharf, 2015;Rajarapu et al., 2015;Rajarapu and Scharf, 2017). The hindgut lumen symbiota of R. flavipes has been characterized (Boucias et al., 2013), and the microbiome was found to include mainly bacteria (nine major phyla, 4761 species-level phylotypes), with a minor population of archaea (phyla Korarchaeota and Euryarchaeota).
One of the bacterial genomes identified in the R. flavipes hindgut is that of the gram-negative Opitutaceae bacterium strain TAV5 of the Verrucomicrobia phylum (Kotak et al., 2015). Verrucomicrobial members are widespread, but the relative abundance of verrucomicrobial communities differs (Nixon et al., 2019) (2% in the R. flavipes hindgut (Boucias et al., 2013). Despite their low relative abundance, these bacteria play an important role in carbon cycling, and their genomes carry complete sets of genes for degradation of a range of polysaccharides, including hemicelluloses (Nixon et al., 2019;Martinez-Garcia et al., 2012;Herlemann et al., 2013;Cardman et al., 2014). Opitutaceae bacterium strain TAV5 has a comparably large genome among known verrucomicrobial members, approximately 7.4 Mbp, including predicted enzyme activities for degradation of wood-based components (Kotak et al., 2015;Nixon et al., 2019). For the achievement of improved efficiency of biomass processing, considerable efforts have been devoted to identify microbial enzymes capable of degrading the recalcitrant fractions of crystalline cellulose in plant-based biomass, while less progress has been made to identify specialist enzymes for the conversion of the hemicellulose fraction, which depending on the source of wood can amount to about 40% of the dry weight (Fan et al., 1982). Hemicellulose components of plant cell walls include, e.g., xylan, glucan, xyloglucan, glucuronoxylan, galactoglucomannan, glucomannan, glucuronomannan, arabinoxylan, arabinogalactan, arabinoglucuronoxylan.

Op5Man5 sequence and biochemical properties
The gene coding for Op5Man5 (locus 10225; UniProt W0J1H8) has been annotated as a putative endo-β-mannanase (EC 3.2.1.78) that belongs to the GH5 family of glycoside hydrolases (www.cazy.org (Lombard et al., 2014). The primary structure comprises 594 amino acids, lacks a secretion signal, and corresponds to a theoretical isoelectric point of 5.6 and molecular weight of ~ 66 kDa. According to SDS-PAGE analysis, the molecular mass is ~ 68 kDa for the recombinant Op5Man5 produced in Escherichia coli, and 66824 g/mol according to electrospray ionization mass spectrometry (ESI-MS) ( Figure S1). The ESI-MS-derived molecular mass agrees well with the theoretical molecular weight for full-length Op5Man5.
Analysis of the Op5Man5 oligomeric state was performed using sizeexclusion chromatography (SEC) using a set of reference proteins with known molecular weights to calculate a standard curve ( Figure S2(A)). At pH values 5.5-6.5 ( Figure S2(B-D)), the SEC profile is monodisperse corresponding to the homotrimer assembly with a molecular weight of ~214 kDa according to the standard curve. At pH 7.5 ( Figure S2(E)), the SEC profile becomes bidisperse with the homotrimer species appearing in equilibrium with a species of twice its molecular weight (~434 kDa), presumably corresponding to a homohexamer. Increasing the pH further (pH 8.0-8.5) shows a gradual disappearance of the presumed hexamer with a concomitant increase of the homotrimer (Figure S2(F-G)). Thus, hexameric Op5Man5 appears to exist only within a relatively narrow pH range around 7.5. Below pH 5.5, the enzyme precipitates (data not shown), a process that starts already at pH 5.5, as indicated by the lower amount of Op5Man5 at pH 5.5 ( Figure S2(B)). Following the SEC analysis, we performed chemical cross-linking analysis, which confirmed a major homotrimeric species, and a minor hexameric assembly ( Figure S3).

Substrate screening and kinetic characterization
Of the aryl glycosides tested, only pNP-β-D-mannopyranoside (pNP-βMan) served as substrate for recombinant Op5Man5 (Table 1), which establishes the specificity for β-1,4-mannosidic bonds. The apparent K m and k cat values for pNP-βMan were 1.0 mM and 238.9 s − 1 , respectively, and the specificity constant (k cat /K m ) of 238.9 mM − 1 s − 1 ( Table 2). The pH optimum of activity for pNP-βMan as the substrate was 6.0 ( Fig. 1 (A)), at which the temperature (T) optimum was 60 • C ( Fig. 1(B)). showed that the activity was decreasing rapidly above 60 • C ( Fig. 1(C)). Based on the inactivation profile, pH 6.0 and 40 • C were chosen as the standard assay conditions. The catalytic dependency of EDTA-treated Op5Man5 on various metal ions was investigated, and no significant enhancement of activity was observed for the metal ions tested, however, several metal ions showed an inhibitory effect at 5 mM concentration, i.e., Fe 2+ , Mg 2+ and Zn 2+ (Fig. 2). For β-1,4-linked mannooligosaccharides (MOSs) with a degree of polymerization (DP) of 2 to 6 (M2, M3, M4, M5, M6) the specific activity increased with increasing DP with the lowest being 1.3 U mg − 1 for M2, and the highest value of 2.6 U mg − 1 observed for M6 (Table 1). For the polysaccharides, activity was only observed for ivory nut mannan (0.29 U mg − 1 ) and Konjac glucomannan (0.18 U mg − 1 ). The lower activity on glucomannan indicates that the presence of glucose in the backbone of this type of mannan negatively affects hydrolysis compared with a pure β-mannan backbone. Michaelis-Menten kinetics for MOSs of DP 3 to 6 (pH 6.0, T = 40 • C, t = 0-12 h) shows that the specific catalytic efficiency increases with DP, with the highest k cat /K m value observed for M6 ( Table 2). The catalytically active residues Glu156 and Glu281 were inferred from sequence comparisons with related GH5 enzymes. The single replacement E156Q showed no activity against MOSs, but weak activity against pNP-βMan. The single replacement E281Q and the double mutant E156Q/E281Q displayed no activity against either pNP-βMan or MOSs, which supports the assignment of Glu281 as the catalytic nucleophile.
The reaction products from hydrolysis of various MOSs were studied by TLC and HPLC ( Fig. 3 and Fig. 4). Op5Man5 hydrolyzed the β-1,4linkages in M3, M4, M5 and M6 to produce M1 and M2 as the major products. M1 was successively released from all MOSs, supporting an exo-type action. An increasing efficiency of conversion to M1 was observed with increasing substrate DP. Whereas M6 was almost completely converted to M1 after 72 h, hydrolysis of M2 and M3 was considerably slower.
It is common for glycoside hydrolases to be activated by an alcohol and to catalyze the production of alkyl glycosides, which is largely attributed to transglycosylation (Lundemo et al., 2013;Morrill et al., 2018), and there are several examples in the literature of transglycosylation by β-mannosidases and β-mannanases with alcohols as acceptors (Zhang et al., 2009;Morrill et al., 2018;Kurakake and Komaki, 2001;Rosengren et al., 2014). The acceptor specificity of Op5Man5 suggests that it is attractive for synthesis of oligosaccharides and alkyl glycosides, which are compounds that can be used in industrial applications for the production of food additives, animal feed, detergents, and personal care products. β-mannosidases reported to have transglycosylation activity are also indicated in Table S1.

Structure of Op5Man5
Crystals of Op5Man5 grew from a solution containing phosphorus citrate (pH 4.2), 30% (v/v) ethanol, 5% (w/v) polyethylene glycol 1000, and were subjected to soaking with 1 mM UO 2 (NO3) 2 . Op5Man5 crystallized in space group P2 1 2 1 2 1 with one homotrimer per asymmetric unit and diffracted to 2.2 Å resolution (Table S2). The rationale for adding the uranyl compound was to obtain experimental phases by isomorphous replacement, however, no significant signal was obtained. A large number of alternative heavy atoms were tested, but despite extensive screening, no crystals were sufficiently derivatized to allow heavy-atom substructure determination, which was likely due to the large protein content of the asymmetric unit (200 kDa), a relatively high solvent content (65%), and overall poor resolution of heavy-atom data sets.
We also performed structure prediction using the fold-recognition server Phyre2 (Kelley et al., 2015). The only high confidence hits were homotrimeric GH42 β-galactosidases (e.g., PDB codes 1KWG, 5E9A, 3TTS, 4UZS, 4OJY). Based on this information, extensive attempts were made to obtain preliminary phases by molecular replacement, but without success. All else having failed, an attempt was made to estimate  phases by single particle electron cryo-microscopy (cryo-EM). Images were recorded and processed for Op5Man5 at pH 7.5 to yield a 3.7 Å resolution Coulomb potential map that allowed a partial model to be built. The partial model was then used to phase the 2.2-Å X-ray diffraction data using molecular replacement to determine the crystal structure of Op5Man5. Although only SEC fractions at pH 7.5 ( Figure S2 (E)) containing the trimeric species were pooled for the cryo-EM experiment, the EM micrographs show clear evidence of larger particles presumably corresponding to a hexamer or heptamer ( Figure S5). Due to the low number of large particles, only the homotrimeric state was considered for image processing. Determination of the precise identity and oligomeric state of the larger assembly would require EM data presenting a significant increase of the number of large particles, which is currently not available.
As expected for an exo-acting β-mannosidase, the active site in Op5Man5 is located in a pocket with the two catalytic glutamate residues (Glu156 and Glu281) buried roughly 10 Å below the surface of domain A. All three active sites are accessed from the central void of the homotrimer. By manually placing mannooligosaccharides in the active site of the A-domain, possible binding sites for three mannose units are readily identified (Fig. 6(A)). The innermost binding site, subsite -1, includes Glu281, Trp317, Glu335, Ser105, Tyr242, and Asn155. Glu156 is positioned between subsites -1 and + 1. Trp103, Gly106 and possibly Asp196 are suitably positioned to interact with a mannose unit in subsite + 1, Trp247 can act as a platform roughly below the second glycosidic bond between subsites + 1 and + 2 in a docked mannotriose molecule. Gln198 is positioned to interact with a mannose unit in subsite + 2. The observed increase in k cat /K m from 2.2 mM -1 s − 1 for M3 to 18.9 mM -1 s − 1 for M6 (Table 2) suggests that there are binding sites beyond the subsite + 2 that contribute to binding and hydrolysis efficiency. Residues with longer side chains at the interface between domains A and B could possibly provide a framework for additional binding sites, e.g., Glu405, Arg407, Arg411 and Arg517.
Two loop regions in Op5Man5 (residues 103-108 and 320-339) cooperate to form a steric block that closes off the active site to form a pocket ( Fig. 6(B)). This type of steric block was first described for CmMan5A (Dias et al., 2004), and later also for the fungal GH5_7 enzyme RmMan5B from Rhizomucor miehei (PDB code 4LYQ (Zhou et al., 2014). In CmMan5A and RmMan5B, a continuous loop region forms the steric block (residues 379-410 in CmMan5A and residues 357-386 in RmMan5B), rather than two separate loops. The steric block has been assigned a structural role in converting a GH5-type endo-mannanase into an exo-acting mannosidase (Zhou et al., 2014;Dias et al., 2004). The trajectory of a mannan chain that occupies the blocked negative subsites in an endo-acting GH5 mannanase is exemplified by the GH5_8 β-mannanase TfMan from Thermobifida fusca in complex with mannotriose where the three mannose units occupy subsites -2, -3 and -4 (PDB code 3MAN (Hilge et al., 1998), which are disabled by the steric block in the exo-acting Op5Man5, CmMan5A and RmMan5B (Fig. 6(B)).
The B-domain is referred to as a trimerization domain in homotrimeric GH42 β-galactosidases. Analyzing the Op5Man5 structure using PISA gives a CSS (complexation significance score) of 1.0 for a homotrimer, which is expected for a natural oligomerization state. Pairwise monomer interface areas range between 1112 and 1154 Å 2 with a solvation free energy gain of − 8.7 to − 10.0 kcal/mol. There are no salt bridges formed between individual monomers, and trimer stabilization is partly provided through 17 hydrogen bonds between the B-domain and the A-domain of a neighboring monomer. Six regions in the trimerization domain provide trimer interactions: 399-402, 453-458, Two regions in Op5Man5 map have weak, unresolved electron density, and were therefore not modeled: a flexible region in domain A between β5 and β6, including the region expected to form α5 (residues 213-221); and the linker region that connects domains A and B (residues 368-381). Additionally, the last seven residues of the C-terminus appear disordered. The missing regions are located at the surface, on the side of the TIM barrel opposite to the active-site entrance, and neither is therefore expected to influence substrate binding or catalysis. However, it cannot be excluded that these flexible regions may play a role in formation of the high-order oligomer that is predicted to be a hexamer or heptamer.

Structural and functional similarity to other GH enzymes
Despite insignificant sequence identity (12-17%) to sequences of GH42 β-galactosidases, the A and B domains of the GH5 member Op5Man5 and have the same topology and overall structure as the homotrimeric GH42 β-galactosidases ( Figure S6 and Figure S7). There are several structures of GH42 β-galactosidases, and RaBGal from Rhanella sp. R3 (PDB code 5E9A (Fan et al., 2016) was chosen for the purpose of structural comparison. Root-mean-square deviation values for aligned Cα atoms for the individual domains of Op5Man5 and RaBGal are 2.2 Å for 269 (of 349) aligned Cα positions (16.0% sequence identity) for domain A; and 2.6 Å for 165 (of 204) aligned Cα positions (16.4% sequence identity) for domain B. While the overall structure of the Op5Man5 homotrimer is similar to that of the GH42 β-galactosidases, the active sites are more diverse. Besides the trimeric β-galactosidases, the only other enzymatic activity found in family GH42 is the α-L-arabinopyranosidase from Bifidobacterium animalis (BlAr-ap42B; PDB code 5XB7 (Viborg et al., 2017), also a homotrimer, and the enzyme displays structural similarity to Op5Man5 comparable to that of the GH42 β-galactosidases: 2.2 Å for 261 (of 349) aligned Cα positions (22.5% sequence identity) for domain A; and 2.4 Å for 165 (of 204) aligned Cα positions (15.2% sequence identity) for domain B.
Additionally, the exo-β-1,4-mannosidase Bs164 (PDB code 6T7G (Armstrong and Davies, 2020) from the human gut symbiont Bacteroides salyersiae is a homotrimer with striking similarity to the GH42 β-galactosidases. Bs164 has been classified as a member of family GH164, but regardless, the enzyme features the topology and tertiary structure of the GH42 β-galactosidases ( Figure S6 and Figure S7). As in the case of RaBGal, the sequence identity shared between Bs164 and Op5Man5 is low, and structural comparison gives an overall r.m.s.d. value for the monomer of 2.5 Å for 424 (of 553) aligned Cα positions ( The structure of the exo-β-mannosidase CaMan5 from Cutibacterium acnes has been reported (PDB code 6GVB (Reichenbach et al., 2018). Like Op5Man5, this enzyme belongs to GH5, but is active as a homodimer and has been assigned subfamily 18 of the GH5 sequences. CaMan5 hydrolyzes the β-1,4 mannosyl linkage in Man-β-1,4-GlcNAc of the N-glycan core. The CaMan5 monomer is not modular in architecture, and includes only the catalytic TIM-barrel domain. Structural Indeed, the low sequence similarity to structurally characterized enzymes makes a detailed comparison difficult. By considering only the catalytic TIM-barrel domain, a reasonable structure-based sequence alignment can be generated for the GH5 exo-β-1,4-mannosidase with known structure, i.e., Op5Man5, CaMan5, BlMan5B, CmMan5A and RmMan5B ( Figure S8). Including the GH2, GH1 and GH164 enzymes was not considered meaningful because of too pronounced structural differences. The sequence alignment shown in Figure S8 was created by performing a structural superimposition of the TIM-barrel domains to align equivalent residues spatially. This results in a sequence identity of 24% between Op5Man5 and the best matching homologs (CaMan5 and BlMan5B), and lower percentages for CmMan5A (16.8%) and RmMan5B (16.4%). Thus, even the catalytic TIM-barrel, which is the most similar region of these GH5 exo-β-1,4-mannosidase, show very low similarity. Considering that all of these enzymes display exo-β-1,4-mannosidase, there are very few conserved residues in the active site, and identities across all five enzymes are limited to the catalytic glutamate residues Glu156 and Glu281, Arg58, Asn155, His240 and Trp317 (Op5Man5 numbering). The homodimeric exo-β-1,4-mannosidases CaMan5 and BlMan5B are active on the β-1,4-linkage in Man-GlcNAc of the chitobiose core of N-glycans. Despite the overall structural differences between Op5Man5 and these enzymes, their active sites show higher resemblance to that of Op5Man5 than do CmMan5A and RmMan5B. The entrance of the substrate-binding pocket in Op5Man5, CaMan5, and BlMan5B shows a similar shape, that differs from that of CmMan5A and RmMan5B ( Figure S8).
Using BLASTP (https://blast.ncbi.nlm.nih.gov/Blast.cgi) to search the database of non-redundant protein sequences for sequences similar to that of Op5Man5 returned only sequences of uncharacterized GH enzymes, many of which contain the β-galactosidase trimerization domain. As expected, the closest related sequences were from the Opitutaceae family of the Verrucomicrobia phylum ( Figure S9). Possible homologs of Op5Man5 are also found in Streptomyces and Bacillus species, but with considerably lower sequence similarities.

Conclusions
In this work we have reported on the characterization of a new hemicellulose-degrading enzyme, Op5Man5, from the bacterial symbiont Opitutaceae bacterium strain TAV5 thriving in the hindgut of the notorious termite pest R. flavipes. We used single particle cryo-EM to generate an initial 3.7-Å model molecular model for further phasing of synchrotron intensity data at 2.2 Å resolution using molecular replacement. The structure reveals that Op5Man5 forms a 200-kDa homotrimer, which represents the first example of a GH5 enzyme with a GH42β-galactosidase-type homotrimeric assembly. Interestingly, within a narrow range around pH 7.5, the homotrimer is observed to be in equilibrium with a hexameric or heptameric state, which could also be observed in small numbers in the cryo-EM micrographs. While the opit5_10225 gene had been annotated as an endo-β-mannanase in the sequence databases, our work establishes that the enzyme is an exoβ-1,4-mannosidase with high specificity for the terminal β-1,4-mannosidic linkages in mannooligosaccharides and linear, mainly unsubstituted, β-mannans such as ivory nut mannan. Furthermore, in the The mannotriose molecule in the exo-β-1,4-mannosidase RmMan5B occupies subsites -1 to + 2, and in the endo-β-mannanase TfMan, mannotriose is bound at subsites -2 to -4. The loops forming the steric block that closes off the negative subsites in the exo-acting enzymes Op5Man5 and RmMan5B are indicated by a gray circle. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) presence of alcohols, Op5Man5 catalyzes simultaneous hydrolysis and transglycosylation reactions of mannobiose into mannooligosaccharides and alkyl β-mannosides, compounds that are attractive for the production of food additives, animal feed, detergents, and personal care products. The abundance of GH39, GH148, GH78, and GH5 genes in the genome of Opitutaceae bacterium strain TAV5 strongly suggests that this verrucomicrobial member is specialized towards degradation of the hemicellulose fractions of wood-based termite feed.
Site-directed mutagenesis to generate the active-site mutant Op5Man5 variant E156Q/E281Q was performed using the following forward and reverse primers: The wild-type gene sequence was used as template for generating the single mutants E156Q and E281Q whereas E156Q sequence was used as a template for generating the double mutant E156Q/E281Q. Mutagenesis was performed using the QuikChange II site-directed mutagenesis kit (Agilent Technologies) as described by the manufacturer. The mutated gene sequences were confirmed by DNA sequencing, and the plasmids with confirmed mutations were transformed into E. coli BL21 (DE3)-T1R for protein production.

Production and purification of Op5Man5 and active-site mutants
The insert sequence was verified, and the plasmid transformed into the BL21 (DE3) T1R strain. The cells were cultured overnight in (50 ml) Terrific Broth (TB) medium with kanamycin (50 μg ml − 1 ) and 0.4% glycerol (v/v) at 30 • C, 175 r.p.m. A small volume (1%) overnight culture was used to inoculate 2 L of TB medium supplemented with 0.8% glycerol (v/v) and kanamycin (50 μg ml − 1 ). The culture was incubated at 37 • C, 180 r.p.m. until OD 600 2, after which the temperature was lowered to 18 • C. Protein expression was induced at OD 600 3 with the addition of IPTG to a final concentration of 0.5 mM, and the culture was incubated overnight at 18 • C. The cells were harvested, and the pellet resuspended in lysis buffer (100 mM HEPES, 500 mM NaCl, 10% glycerol, 10 mM TCEP 25 mM, pH 8.0) with one tablet of cOmplete TM protease inhibitor cocktail (Roche). The suspended pellet was homogenized using an AVESTIN Emulsiflex-C3 system, and further disrupted by pulsed sonication (4 s/4s 3 min, 80% amplitude).
The lysate was centrifuged (20 min at 49000 × g) using a Beckman Ti 45 rotor, and the resulting supernatant filtered through a 0.45-μm filter. The resulting supernatant was filtered through a 0.45-μM filter. The resulting supernatant was loaded onto a Ni-NTA HisTrap HP column (5 ml, GE Healthcare) using an Ä KTA start system (GE Healthcare, Uppsala, Sweden). The columns were washed with 20 mM HEPES buffer (pH 7.5), 500 mM NaCl, 50 mM imidazole, 10% (v/v) glycerol, 0.5 mM TCEP. Bound protein was eluted with buffer containing 500 mM imidazole. Protein samples were concentrated using Vivaspin20 centrifugal concentrators (polyethersulfone filter, molecular weight cut-off, MWCO, 30 kDa), and loaded onto a HiLoad 16/60 Superdex 200 prep grade column (GE Healthcare Life Sciences) equilibrated with 50 HEPES buffer (pH 7.5) 300 mM NaCl, 10% (v/v) glycerol and 2 mM TCEP. The fractions corresponding to the middle of the eluted peak were pooled and checked for purity by SDS-PAGE and liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis. The fractions containing Op5Man5 were recovered and pooled, and concentrated using a Viva-spin20 concentrator (MWCO 30 kDa) to a final concentration of 5-10 mg ml − 1 . The Op5Man5 mutant (E194Q/E294Q) was produced as described for the wild type.

Substrate screening
Different aryl glycosides were screened to identify suitable substrates. The reaction mixture contained 0.2 ml of 1 mM pNP substrate in 50 mM sodium phosphate (pH 6.0) and 5 μl of enzyme (2 mg ml − 1 ).
After incubation at 40 • C for 10 min, the reaction was stopped by adding 20 μl 2 M Na 2 CO 3 . The released p-nitrophenol product was measured using a spectrophotometer at 410 nm. Enzyme and substrate free controls were also incubated and measured under the same conditions. All enzyme assays were performed in triplicate and one unit of enzyme activity (U) was defined as the amount of enzyme that produces one μmol of pNP per minute.
Substrate screening of oligosaccharides and polysaccharides was performed using the GOD-POD assay. Unless otherwise indicated, assay mixtures contained substrate and appropriately diluted enzyme in 50 mM sodium phosphate buffer (pH 6.0) at 40 • C. Briefly, appropriate concentration of Op5Man5 was mixed with 500 μl of polymeric substrate (5 mg ml − 1 ) or 100 μl oligosaccharides (5 mM) and the reaction was terminated by heating the sample at 100 • C for 5 min. One unit of enzyme activity was defined as the amount of protein that released 1 μmol min − 1 of glucose.

Effect of pH and temperature on enzyme activity
The effects of pH on Op5Man5 activity were tested using 50 mM sodium acetate buffer (pH 4.0-5.5), sodium phosphate buffer (pH 6.0-7.5), and tris buffer (pH 8.5-9.0) at 40 • C using pNP-βMan in sodium phosphate buffer (pH 6.0). The effect of temperature (T) on activity was investigated in temperature range 20 • C to 70 • C. The residual activity was expressed as a percentage of the highest activity. The stability of activity as a function of T was determined by incubating Op5Man5 (0.02 mg ml − 1 ) at T ( • C) = 40, 50, 60, and 70. Aliquots were taken at different time points (samples were taken at the time points: 0, 5, 10, 20, 30, 40, 60, 80 min; and 2, 4, 6, 12, 18, 24 h) and assayed as described above using pNP-βMan as substrate. Data were collected in triplicate, and the half-time of irreversible thermal inactivation (a t ) was calculated by non-linear regression using the exponential decay equation in SigmaPlot (Systat Software, San Jose, CA): a t = a 0 ( − t*ln 2

Effect of metal ions
The effect of metal ions (NiCl 2, FeCl 3, CaCl 2 , CuCl 2 , MgCl 2 , ZnCl 2, CrSO 4 and MnCl 2 ) on Op5Man5 activity was investigated with pNP-βMan as substrate using the standard assay described above. Before starting the reaction, Op5Man5 (2 mg ml − 1 ) was pre-incubated in 50 mM sodium phosphate (pH 6.0) in the presence of either 2 or 5 mM metal ion for 5 min at 40 • C. The reaction was started by adding 0.2 ml of 1 mM pNP-βMan in 50 mM sodium phosphate (pH 6.0). The remaining activity was measured after incubation for 10 min at 50 • C. The residual activity was determined with reference to the standard reaction without added metal ions set to 100%. The negative control contained Op5Man5 that had been treated with 1 mM EDTA. All reactions were performed in triplicate with the standard deviation errors calculated.

Determination of kinetic parameters for pNP-βMan and MOSs
The kinetic parameters, Michaelis-Menten constant (K m ), maximum velocity (V max ), and turnover number (k cat ), were determined using pNP-βMan as substrate. Reaction mixtures (200 μl) contained 0.5-3.0 mM (at the points 0.025, 0.5, 1.0, 1.5, 2.0, 2.5 and 3.0 mM) pNP-βMan 50 mM sodium phosphate buffer (pH 6.0), to which 5 μl Op5Man5 (2 mg ml − 1 ) was added to start the reaction (40 • C for 10 min). The reaction was stopped by adding 20 μl 2 M Na 2 CO 3 . All enzyme assays were performed in triplicate. Values of K m , V max and k cat were calculated by nonlinear regression using Michaelis-Menten equation fit using the GraphPad Prism 5 kinetic analysis package (GraphPad Software, San Diego, CA).
For the MOSs (M4-M6), initial velocity of each reaction was measured within 12 h using high-performance anion exchange chromatography (HPAEC) and the k cat /K m values were derived from a linear fit v = [E 0 ][S 0 ](k cat /K m ). HPAEC (Dionex ICS5000, Sunnyvale, CA, USA) equipped with a Carbo-Pac PA200 column (3 μm × 250 mm) and a pulsed amperometric detector (PAD) was used to analyze hydrolysis products. The hydrolysates were diluted in deionized water (1:100 v/v), and 10 μl of sample was injected in a PA200 column with an isocratic flow of 0.5 ml⋅min − 1 of 100 mM NaOH at 30 • C for 35 min. Peak assignment was performed by comparison of the retention time with a series of standard MOSs. Calibration was performed with standard solutions at concentrations in the range 0.01 to 0.2 mM. The peak area was used for quantification.

TLC analysis of products from MOS hydrolysis
Hydrolysis products obtained from MOSs (M2, M3, M4, M5, and M6) were analyzed by TLC. Op5Man5 (1 μl of 2 mg ml − 1 ) was added to buffered MOS solutions (50 μl of 10 M2-M6), and the reaction allowed to proceed for 72 h at 40 • C. Samples were withdrawn at different time points (0, 12, 24, 48, 72 h), and the reactions terminated by heating the samples in a 100 • C water bath for 10 min. After centrifugation (10,000 × g, 10 min), the samples were spotted on Silica gel 60 F 254 plates (Merck Millipore) and air-dried. TLC was performed using a mobile solvent consisting of butanol:propanol:ethanol:water (2:3:3:2, v/ v), and the silica plates were stained with thymol and heated for 10 min at 120 • C. The control samples were treated identically, with the exception of using heat-inactivated enzyme (100 • C, 10 min).

Transglycosylation of mannobiose and alcohols by Op5Man5
To demonstrate whether alcohols can act as acceptors for glycosyl moieties, various alcohols (500 mM) were used as acceptors and mixed with mannobiose (5 mM) followed by incubation with 5 μl (2 mg ml − 1 ) Op5Man5 in 50 mM phosphate buffer (pH 6.0) at 40 • C for 24 h. The transfer products were analyzed using TLC as described above.

Crystallization and X-ray intensity data collection
Crystals were obtained at room temperature from a solution containing 20 mg ml − 1 Op5Man5 in 0.1 M phosphorus citrate (pH 4.2), 30% (v/v) ethanol, and 5% (w/v) polyethylene glycol 1000 using the vapor diffusion method and sitting drops. Crystals were subjected to soaking with 1 mM UO 2 (NO 3 ) 2 with the aim to obtain heavy-atom phases. Synchrotron data were recorded on crystals vitrified in liquid nitrogen (100 K) at Diamond Light Source beamline I03 to 2.2 Å resolution, and scaled using the XDS software package (Kabsch, 2010) (Table S2). The crystals belong to space group P2 1 2 1 2 1 with the cell dimensions a = 102.14 Å, b = 162.45, c = 168.18 Å, α = β = γ = 90 • with three molecules (one homotrimer) in the asymmetric unit.

Grid preparation, cryo-EM and image processing
Aliquots of 3 µl of purified Op5Man5 (0.12 mg ml − 1 ) were applied to glow-discharged (60 s, 20 mA using Quorum GloQube), 300-mesh (copper) C-Flat (R2/2) carbon grids (EMS, USA). Using a FEI Vitrobot Mark IV (Thermo Fischer Scientific) operating at 4 • C and 100% humidity, grids were blotted for 3 s before being flash-frozen into liquid ethane. Microscopy data was collected using a Talos Arctica electron microscope (Thermo Fischer Scientific) operated at 200 kV equipped with a Falcon III direct electron detector (Thermo Fischer Scientific). A total of 338 movies were acquired automatically with EPU (Thermo Fischer Scientific), using single-electron counting mode at a nominal magnification of 150,000 × (0.99 Å pixel − 1 ), a flux of 0.56 e Å 2-1 s − 1 and a total dose of 60 e Å 2-1 over a total of 24 frames with defoci ranging from − 0.5 μm to − 3 μm.
All of the following data processing steps were performed in cry-oSPARC version 2.12 (Punjani et al., 2017). Frames were aligned, averaged and dose-weighted with Patch motion (Rubinstein and Brubaker, 2015) and the contrast transfer functions for each micrograph was estimated with Patch CTF. 67 micrographs with a CTF fit quality above 5 µm were excluded. Template-based particle picking yielded 179,684 particles of which 22,453 spurious particles were removed by 2D classification, yielding a dataset of ~ 157231 particles. Through abinitio reconstruction and heterogeneous refinement, a class with 85,735 particles could be identified and was subsequently refined via homogeneous and non-uniform refinement to 3.7 Å, for which per particle defoci were optimised and higher order aberrations corrected (Zivanov et al., 2019). The reported resolution is based on the Fourier shell correlation between two independently refined half maps (Scheres and Chen, 2012) using a 0.143 criterion (Rosenthal and Henderson, 2003), and the FSC-curve was corrected for the convolution effects of a soft mask using high-resolution noise substitution (Chen et al., 2013). The density map was locally filtered and sharpened by applying a negative Bfactor (-220 Å 2 ) that was estimated using automated procedures (Rosenthal and Henderson, 2003).

Crystal-structure determination and model refinement
A partial model was produced from the 3.7-Å cryo-EM potential map using the model-building option (Map to model, 3 building cycles) in the Phenix suite (Adams et al., 2010). The partial cryo-EM model was used for molecular replacement against the 2.2-Å X-ray amplitudes with Phenix Phaser (Adams et al., 2010). A clear solution was found for the three monomers of the homotrimer in the asymmetric unit of the P2 1 2 1 2 1 cell (313.8, TFZ 20.1). The MR phases were further subjected to improvement by density modification and three-fold averaging using Phenix Resolve. The improved phases were of high quality and allowed building of a nearly complete model (1651 residues of 1755) using automatic building with Phenix AutoBuild, which generated a model with 1626 residues and R and R free values of 0.205 and 0.243, respectively. The Op5Man5 model was refined with phenix.refine and adjusted iteratively by hand with Coot (Emsley and Cowtan, 2004) with guidance from σ A -weighted 2F o -F c electron-density map at 2.2-Å resolution. The final model contained 1637 residues and 1056 solvent molecules (R/ R free 0.201/0.237). A comparison between the initial 3.7 Å cryo-EM map and the 2.2 Å 2F o -F c electron-density map is shown in Figure S11. The atomic coordinates and structure factor amplitudes have been deposited in the Protein Data Bank (www.rcsb.org) with the accession code 7BOB.

Sequence analyses
Analysis of signal peptides was performed using SignaIP-5.0 (Almagro Armenteros et al., 2019). Structure-based sequence alignment was performed manually by interactively analyzing the overlaid protein structures using Coot (Emsley and Cowtan, 2004). The aligned sequences were visualized using ESpript 3 (espript.ibcp.fr (Robert and Gouet, 2014). For generating the phylogenetic tree, BLASTP was first used to identify and align the amino-acid sequences, after which maximum likelihood phylogenetic analysis was performed using PhyML (www.phylogeny.fr (Dereeper et al., 2008). The phylogenetic tree was generated using Interactive Tree Of Life (iTOL) v 5.7 (Letunic and Bork, 2019).

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.