Metagenomic library construction and enzyme screening

Purified and size-fractioned DNA was ligated into the pCCFOS fosmid vector and further cloned in Escherichia coli EPI300-T1^R according to the instructions of Epicentre Biotechnologies (WI, USA) and a procedure described earlier [21]. Fosmid clones (12288) harbouring approximately 490 megabasepairs (Mbp) of community genomes were arrayed using the QPix2 colony picker (Genetix Co., UK) and grown in 384-microtitre plates containing Luria Bertani (LB) medium with chloramphenicol (12.5 μg/ml) and 15% (v/v) glycerol and stored at −80°C.

To screen for GH activity, the clones were plated onto large (22.5×22.5 cm) Petri plates with LB agar containing chloramphenicol (12.5 μg/ml) to create an array of 2304 clones per plate. Each library was screened for the ability to hydrolyse p-nitrophenyl (pNP) α-L-arabinofuranoside (pNPαAf), pNP-α-galactopyranoside (pNPαGal), pNP-α-L-rhamnopyranoside (pNPαR) and carboxymethyl cellulose (CMC). For screens based on pNP-like substrates, Induction Solution (Epicentre Biotechnologies; WI, USA) was added after an overnight incubation, as recommended by the supplier, to induce a high fosmid copy number. The CMC-active fosmids were screened on agar plates supplemented with 1% (w/v) substrate and Congo red water solution [19]. The positive clones were selected and fully or partially (after sub-cloning in the pUC19 vector) sequenced at the Göttingen Genomics Laboratory (Germany) or by primer walking from both ends at Secugen S.L. (Madrid).

Cloning, expression, purification and characterisation of plant polymeric substance hydrolases

All genes for recombinant enzymes used in the present study were PCR-amplified using custom oligonucleotide primers and were cloned, expressed and purified as described in the Methods S1.

For the enzyme characterisation, the absorbance was measured using a BioTek Synergy HT spectrophotometer under the following conditions: [E]_o = 0–12 nM, [substrate] ranging from 0 to 50 mM in 100 mM buffer, T =  40°C. For the hydrolysis of the pNP derivatives, the corresponding volume of a pNP derivative stock solution (120 mM) in the appropriate buffer was incubated for 2–10 min (with the exception that 30 s were used for the assay for R_09-02) with 12 nM enzyme diluted in 200 µl of 100 mM buffer and measured at 405 nm in 96-well microtiter plates. The substrates tested included p-nitrophenyl (pNP) α-L-arabinofuranoside (pNPαAf), pNP-α- and pNP-β-galactopyranoside (pNPαGal and pNPβGal), pNP-α- and pNP-β-xylopyranoside (pNPαX and pNPβX), pNP-β-D-glucopyranoside (pNPβG), pNP-β-D-cellobioside (pNPβC), pNP-α-L-rhamnopyranoside (pNPαR) and pNP-α- and pNP-β-arabinopyranoside (pNPαAp and pNPβAp). For oligosaccharides other than the activated pNP derivatives, the level of released glucose was determined using a glucose oxidase kit (Sigma-Fluka-Aldrich Co., St. Louis, MO, USA). For xylo- and arabino-oligosaccharides, the levels of released xylose and arabinose were determined using the D-xylose and lactose/galactose (Rapid) assay kits from Megazyme (Bray, Ireland). The hydrolysis of cinnamates (methyl ferulate and coumarate) was routinely measured, and the kinetic parameters were determined as described previously [22]. The initial rates were fitted to the Michaelis–Menten kinetic equation using non-linear regression to determine the apparent K_m and k_cat; kinetic parameter calculations were performed based on the molecular masses described in Table S1.

The standard GH assay contained [E]_o = 12 nM, pNP derivative or substrate at 10 mM and 100 mM 4-(2-hydroxyethyl)piperazine-1-ethanesulfonic acid (HEPES) in a total volume of 200 μl at the optimal pH and temperature for each enzyme.

The standard feruloyl esterase assay contained [E]_o = 12 nM, methyl ferulate at 1 mM in 100 mM HEPES in a total volume of 200 μl at pH 8.0 and T = 40°C.

The pH and temperature optima were determined in the range of pH 4.0–10.0 and 5–60°C. The following buffers (100 mM) were used: acetate (pH 4.0–6.0), MES (pH 6.0–7.0), HEPES (pH 7.0–8.0), Tris-HCl (pH 8.0–9.0) and glycine (pH 9.0–10.0). pH was always adjusted at 25°C.

All of the values were determined in triplicate and were corrected for the spontaneous hydrolysis of the substrate. The results shown are the averages of three independent assays ± the standard deviation.

In silico analysis of proteins and 3-D modelling

The MetaGeneMark tool with refined heuristic models for metagenomes (http://exon.gatech.edu/GeneMark/metagenome/index.cgi; [23]) was used to predict genes in the cloned DNA fragments (DNA sequences of fosmid clones were deposited with GenBank/EMBL/DDBJ under accession numbers JQ303337-JQ303344). The deduced proteins were analysed using blastp and psi-blast [24] against the non-redundant database sourced from the nucleotide (nr/nt) collection, reference genomic sequences (refseq_genomic), whole genome shotgun reads (wgs) and environmental samples (env_nt). The translation products were further analysed for protein domains using the Pfam-A [25] and Cluster of Orthologous Groups of protein (COG) databases [26]. Multiple sequence alignments were generated using the ClustalW tool (http://www.ebi.ac.uk/clustalw/index.html) integrated into the BioEdit software [27]. Structural alignments of the proteins homologous to GH obtained in this study were generated by GenTHREADER [28] and used to retrieve a model from the Swiss-Model server [29]. The PDB entries used as templates are described in the Text S1.

Library screening and general enzyme characteristics

In the present work, the GHs were named according to the origin (rumen, R), fosmid ID and the number of the corresponding coding sequence (CDS) in the genomic fragment sequenced. The R library (12,288 fosmid clones) was screened for the ability to hydrolyse pNPαAf, pNPαGal, pNPαR and CMC. We identified eight positives (designated as r_01 to r_03 and r_05 to r_09). The fosmids with inserts r_01 (pNPαGal positive), r_02 (pNPαAf pos.) and r_03 (pNPαAf pos.) were fully sequenced, whereas those of r_05, r_06 (CMC pos.), r_07, r_08 (pNPαR pos.) and r_09 (pNPαAf pos.) were first subjected to shotgun sub-cloning. After subsequent activity screening with appropriate substrates, the inserts of positive sub-clones were then fully sequenced (Table S1).

Genes in the fully sequenced fosmid or plasmid clones were predicted using the MetaGeneMark tool [23], and the corresponding gene products were further subjected to a nr psi-blast analysis, which identified 14 GH- and 1 feruloyl esterase-like polypeptides (Figure S1; Tables S2, S3, S4) and 13 additional accessory enzymes acting on carbohydrates (Table S5). We observed a high sequence similarity between the GHs and other putative genes from clones r_01 and r_06 to r_09 and the proteins from organisms of the phylum Firmicutes, although the average GC content in those clones was approximately 60% (Table S2); other clones (r_02, r_03 and r_05) possessed genes whose products were related to the proteins from representatives of the phylum Bacteroidetes (Table S2). Many representatives of the above phyla are culturable microorganisms found in the rumen and other regions of the GI tract and are thought to play key roles in the breakdown of proteins and carbohydrate polymers [10].

From these clones (except for r_05 and r_06, whose gene products could not be expressed in an active form), 7 putative GHs plus an additional esterase were cloned, expressed in E. coli and purified. Furthermore, their activities were tested with a battery of substrates (Tables 1 and S6) under optimal temperatures and pH values (Figure 1, Figures S2 and S3) to determine the substrate(s) that were the most highly degraded. The presence or absence of putative secretion signal peptides, domain organisation (Figure S4) and 3-D models (Figure S5) were also analysed based on the sequence data. Seven of twelve pNP derivatives tested were hydrolysed by rumen community-derived enzymes (Tables 1 and S6), and the sequence analysis of the enzymes showed a similarity with specific protein domains of known GHs and esterases that are multimodular with diverse 3-D structures and substrate specificities (for details, see the Text S1 and References S1). As expected for the screening substrates that were used, the major phenotypes identified were α- and β-galactosidase, α-arabinofuranosidase, α-rhamnosidase, β-xylosidase, β-cellobiase and β-glucosidase. The enzymes were characterised by a wide range of pH values ranging from 5.0 to 9.0, and seven of the enzymes exhibited their highest activity at approximately 50°C and showed a rapid loss of activity above this temperature. The only exception to this was with the R_09-02 enzyme, which was active at temperatures below 35°C. A complete description of the enzyme characteristics is provided in the Text S1.

Among all of the polysaccharide-degrading enzymes investigated, R_09-02 appeared to show atypical characteristics, and the extensive analysis of this enzyme is provided below.

R_09-02 is a GHF43 enzyme with atypical activities

The insert of the r_09 DNA fragment (3264 bp; G+C content of 60.69%) contained three GHs, namely R_09-01, a putative truncated GHF43 protein, a xylosidase/arabinosidase (R_09-02) from GHF43 (most similar to those from Bacteroides capillosus and Clostridium hathewayi) and another truncated GHF1 β-galactosidase (R_09-03) (Figure S1 and Table S2H).

R_09-02 has a deduced molecular mass of 54 939 Da and an estimated pI of 4.96. GHF43 comprises a large number of GHs from different organisms (1590 entries in GenBank, 482 in Uniprot and 33 in the PDB) that are known to act mainly on β-1,4(3)-xylans or α-1,3(5)-arabinans, with a few reported cases of galactosidases/galactanases. The analysis of the pure R_09-02 enzyme (a tetramer of approximately 200 kDa) using activated pNP derivatives revealed β-xylosidase, α-arabinofur(pyr)anosidase, β-galactosidase and, to a lesser extent, α-glucosidase activities (Table 1), a profile that does not resemble the typical activity profile of enzymes from the GHF43 family (Table S7) [30]. In the enzymatic assay, the Michaelis-Menten constant (K_m), the catalytic rate constant (k_cat) and the catalytic efficiency (k_cat/K_m) values were determined (Table 1). In terms of its catalytic efficiency, R_09-02 best hydrolysed pNP-α-arabinopyranoside (pNPαAp), followed by pNP-β-xylopyranoside (pNPβX), pNPαAf and pNP-β-galactopyranoside (pNPβGal): 71-, 43- and 5-fold greater k_cat values for pNPαAp, compared to pNPβGal, pNPαAf and pNPβX, respectively. A weak activity with pNPα-glucopyranoside (pNPαG) and pNPα-maltoside (pNPαMal) was detected, and a reduction in the catalytic efficiency with these substrates was mainly due to a 1771-fold reduction in the k_cat in comparison to pNPαAp.

The activity of the purified R_09-02 protein was further analysed against various oligosaccharides, as described in the Methods section (Table 1): 1,4-β-xylo-oligosaccharides (degree of polymerisation [DP] from 2 to 7), 1,5-α-arabino-oligosaccharides (DP from 2 to 7), maltooligosaccharides (DP from 2 to 7), xyloglucan oligosacharides (DP ∼14), larch arabinogalactan, amyloid xyloglucan, and starch. The enzyme hydrolysed short 1,4-β-D-xylo-oligosaccharides with less than seven units and 1,5-α-L-arabino-oligosaccharides with less than five units. All of the substrates were completely degraded to the monosaccharides (not shown), suggesting an exo-mode of action for this glycosyl hydrolase. A [(k_cat/K_m)]_xylobiose/[(k_cat/K_m)]_xylotriose factor of ∼74/1 was observed due to a 3-fold higher k_cat value coupled with a significantly (25-fold) lower K_m value for the shorter substrate. Similarly, a [(k_cat/K_m)]_arabinobiose/[(k_cat/K_m)]_{arabinotriose} factor of ∼35/1 was observed, as the K_m and k_cat values for the disaccharide were 4-fold lower and 9-fold higher, respectively, when compared to the trisaccharide. As shown in Table 1, xylobiose and arabinobiose were hydrolysed with similar k_cat values, although the former was somewhat preferred at lower substrate concentrations (∼2-fold lower K_m), resulting in a 2-fold catalytic efficiency value. According to these data, the enzyme would be essentially bifunctional for xylobiose and arabinobiose at concentrations over 0.3 mM (more than 10-fold the K_m values). The lower activity with the longer substrates indicates the enzyme preference for shorter xylose- and arabinose-containing molecules.

1,4-α-Linked saccharides, ranging from maltose to maltoheptaose, were also used as substrates, albeit with lower efficiencies (less than 250-fold) when compared to those containing 1,4-β-xylose and 1,5-α-L-arabinose (Table 1). The k_cat/K_m value was the highest for maltotriose, followed by maltotetraose and, to a lesser extent, maltopentaose and maltohexaose, whereas maltose and maltoheptaose were poor substrates. No release of hydrolysis products was observed with substrates longer than maltoheptaose (including soluble starch). This substrate length specificity differs from that for xylose and arabinose-containing molecules for which the disaccharides were the preferred substrates.

We further demonstrated that R_09-02 hydrolysed the α-1,4 glucosidic bond of the disaccharide, lactose, and the α-1,6 bond in the trisaccharide, raffinose, which is the most preferred substrate after 1,4-β-xylobiose (six-fold rel. k_cat/K_m) and 1,5-α-arabinobiose (two-fold rel. k_cat/K_m). The tetrasaccharide, stachyose, was also hydrolysed, but R_09-02 was 74-fold less efficient with this substrate in comparison to raffinose, which was mainly due to a 41-fold increase in the K_m, coupled with an approximately two-fold reduction in the k_cat.

None of the other tested substrates was hydrolysed, suggesting that the natural substrates of R_09-02 are short oligosaccharides containing α-1,5 glucosidic bonds between two arabinoses, containing β-1,4 bonds between two xyloses, containing β-1,4 bonds between one galactose and one glucose, containing α-1,6 bonds between one galactose and one glucose and containing α-1,4 bonds between two glucoses. Altogether, the data confirmed the highly promiscuous behaviour of the R_09-02 protein. To the best of our knowledge, no GH with a similar biochemical profile has been described to date (Table S7) [30]. Therefore, the R_09-02 enzyme should be classified as a multifunctional GHF43 protein with β-xylosidase, α-arabinofur(pyr)anosidase, lactase, raffinase, stachyase, β-galactosidase and α-glucosidase activities.

The optimum activity for R_09-02 was observed within a narrow range of temperatures, with a relative activity higher than 80% of the maximum recorded occurring between 30 and 34°C, and within a narrow pH range (5.0–6.0) (Figure 1, panels A and B). This thermal sensitivity of the R_09-02 protein may explain why the protein was found mainly in inclusion bodies at 37°C and that high levels of the active protein could only be obtained when the expression was performed at temperatures lower than 28°C (Figure S6). The half-life of the enzyme at the optimal temperature of 34°C and optimal pH of 6.0 showed that the enzyme was quite unstable: the t_1/2 was approximately 3.8 min (Figure 1C). For this reason, short incubation times (less than 1 min) were used to determine the kinetic parameters. The activity of R_09-02 was not affected by reducing agents, such as dithiothreitol and 2-mercaptoethanol (Figure 1D), suggesting that this enzyme (with 10 cysteine residues per monomer) does not contain any structurally relevant disulphide bonds. The addition of Mg²⁺ and Mn²⁺, but not Ca²⁺, increased the activity of the enzyme by approximately 1.5-fold. As structural calcium ions have been found in the β-sandwich module of other GHF43 enzymes [31], [32] or as a part of their catalytic sites [33], [34], the possibility that Mg²⁺ and Mn²⁺ may have similar structural roles cannot be ruled out. In fact, the original purified enzyme may contain such trace elements because the presence of the chelating agent, EDTA, at 10 mM inhibited the enzyme activity by approximately 67% (Figure 1D).

3D structural analysis of biochemically characterised GHF43

Most of the GHF43 enzymes analysed to date are either highly specific xylosidases or arabinofuranosidases, with a few cases of bifunctional xylosidases-arabinofuranosidases [35] and one reported galactosidase (see the CAZy database; [30]). The broad spectrum of activities found for R_09-02 led us to perform a phylogenetic comparison of the biochemically characterised GHF43 enzymes to determine the evolutionary relatedness of R_09-02 with counterparts that have different substrate specificities. Different modular arrangements were found that contained either a single catalytic module of approximately 300 amino acids (AA) or an N-terminal catalytic domain and an additional 150 AA, 230 AA or 280 AA–long C-terminal domain. These different modular topologies will be referred here as types A, B, C and D, respectively for simplification (Figure 2). Enzymes that contained more than one catalytic domain were excluded from this analysis. The amino acid sequence alignment was performed using only the corresponding GHF43 catalytic domains, based on the hits predicted by the Pfam database (http://pfam.sanger.ac.uk/). The catalytic domains of the type C enzymes were grouped together by the phylogenetic analysis, whereas types A, B and D apparently evolved independently (Figure 2). Because the phylogenetic clustering relies on modular properties rather than on the taxonomic placement of the organism, the separation of the above types was probably an ancient evolutionary event. According to this classification, R_09-02, R_03-04 and R_03-05 (GHF43 enzymes also identified herein; for details see the text S1) would belong to types B, A and C, respectively. The structural models of R_03-04, R_03-05 and R_09-02 (based on templates with PDB codes 3QED, 2EXI and 3C7G and sequence identities of 23.5%, 18.2% and 19.7%, respectively) revealed that R_03-04 contains a single catalytic module, whereas R_03-05 and R_09-02 contain a C-terminal β-sandwich domain (Figure 3, left panel). This accessory domain would be larger in R_03-05 enzyme, with a loop protruding into the active site. Indeed, the substrate-binding site of the structurally resolved type C enzymes includes residues from this β-sandwich [31], [36], and this may explain why the catalytic domain of these enzymes evolved independently. Most of the type C enzymes are known as xylosidases, whereas types A and B were identified mainly as arabinofuranosidases (Table S7) [30]. However, hydrolases from type C group exhibit also arabinofuranosidase activities, and the A and B types contain some xylosidases, indicating that the conversion of a xylosidase into an arabinofuranosidase and vice-versa is possible in any of these groups (Table S7) [30]. A set of residues that could potentially form hydrophobic or polar contacts with the substrate (W82, H301 and R327 in R-09_02) (Figure 3, right panel) is highly conserved within the GHF43 enzymes; the Arg residue is invariantly found in all of the characterised GHF43 enzymes, whereas, in some cases, the His is absent or the Trp is substituted with other hydrophobic residues (not shown), regardless of the main activity of the enzyme. Additionally, other hydrophobic residues (with a highly heterogeneous distribution among the GHF43 sequences) are found in the catalytic pocket and may contribute to the substrate binding. Either these additional residues or changes in the orientation of the lateral chain of conserved residues may be responsible for the differences in the substrate specificity. Whatever the case, R_09-02 belongs to a phylogenetic cluster that shows a quite divergent biochemical profile. This makes the identification of the motifs responsible for the R_09-02 promiscuity difficult, as it probably results from a combination of multiple sequence divergences. When more biochemical and structural data become available, this issue may be re-addressed.