Lacto-N-biosidase Encoded by a Novel Gene of Bifidobacterium longum Subspecies longum Shows Unique Substrate Specificity and Requires a Designated Chaperone for Its Active Expression*

Background: Phenotypically lacto-N-biosidase-positive Bifidobacterium longum JCM1217 does not possess a gene homologous to previously identified lacto-N-biosidase. Results: Hypothetical proteins BLLJ_1505 and BLLJ_1506 encode lacto-N-biosidase and its designated chaperone, respectively. Conclusion: The enzyme showed unique and unexpected substrate specificity. Significance: The enzyme is important for understanding how B. longum consumes human milk oligosaccharides and also may serve as a new tool in glycobiology. Infant gut-associated bifidobacteria possess species-specific enzymatic sets to assimilate human milk oligosaccharides, and lacto-N-biosidase (LNBase) is a key enzyme that degrades lacto-N-tetraose (Galβ1–3GlcNAcβ1–3Galβ1–4Glc), the main component of human milk oligosaccharides, to lacto-N-biose I (Galβ1–3GlcNAc) and lactose. We have previously identified LNBase activity in Bifidobacterium bifidum and some strains of Bifidobacterium longum subsp. longum (B. longum). Subsequently, we isolated a glycoside hydrolase family 20 (GH20) LNBase from B. bifidum; however, the genome of the LNBase+ strain of B. longum contains no GH20 LNBase homolog. Here, we reveal that locus tags BLLJ_1505 and BLLJ_1506 constitute LNBase from B. longum JCM1217. The gene products, designated LnbX and LnbY, respectively, showed no sequence similarity to previously characterized proteins. The purified enzyme, which consisted of LnbX only, hydrolyzed via a retaining mechanism the GlcNAcβ1–3Gal linkage in lacto-N-tetraose, lacto-N-fucopentaose I (Fucα1–2Galβ1–3GlcNAcβ1–3Galβ1–4Glc), and sialyllacto-N-tetraose a (Neu5Acα2–3Galβ1–3GlcNAcβ1–3Galβ1–4Gal); the latter two are not hydrolyzed by GH20 LNBase. Among the chromogenic substrates examined, the enzyme acted on p-nitrophenyl (pNP)-β-lacto-N-bioside I (Galβ1–3GlcNAcβ-pNP) and GalNAcβ1–3GlcNAcβ-pNP. GalNAcβ1–3GlcNAcβ linkage has been found in O-mannosyl glycans of α-dystroglycan. Therefore, the enzyme may serve as a new tool for examining glycan structures. In vitro refolding experiments revealed that LnbY and metal ions (Ca2+ and Mg2+) are required for proper folding of LnbX. The LnbX and LnbY homologs have been found only in B. bifidum, B. longum, and a few gut microbes, suggesting that the proteins have evolved in specialized niches.

Infant intestinal microflora develops from a tripartite relationship between mother's milk, infant, and bacteria, and it is generally accepted that the microbiota in the intestines of breast-fed infants are rich in particular bifidobacterial species, such as Bifidobacterium breve, Bifidobacterium bifidum, and Bifidobacterium longum subsp. longum/infantis (collectively termed infant gut-associated bifidobacteria) (1,2). These species/subspecies dominate in the gut ecosystem within a week after birth and continue their predominance until weaning, when the gut microflora become adult-like (1). For this reason, human milk has long been considered to contain bifidogenic compounds (3)(4)(5)(6).
In contrast to the two species mentioned above, the other infant gut-associated bifidobacteria, B. longum subsp. longum (referred to as B. longum) and B. breve have received far less attention, perhaps because they grow poorly on HMO-containing medium (OD 600 Ͻ 0.3) (12,26). Because more than 70% of HMOs are fucosylated (i.e. derived from secretor and Lewispositive donors), the poor growth ability of the two species may be attributable to a lack of 1,2-␣-L-fucosidase and 1,3-1,4-␣-Lfucosidase (5). Accordingly, when B. breve JCM1192 T and B. longum JCM1217 T are inoculated in HMO-containing media, they consume LNT only (12). In the spent medium of B. breve JCM1192 cultures, no degradation products (mono-, di-, and trisaccharides) of LNT have been detected, whereas in the culture supernatant of B. longum JCM1217, Lac appears transiently as LNT decreases. Thus, whereas B. breve JCM1192 imports intact LNT as does B. infantis, LNT consumption by B. longum JCM1217 should involve secretory LNBase and the subsequent GNB/LNB pathway (supplemental Fig. S1). Indeed, LNBase activity has been previously detected in B. longum strains JCM1217 and JCM7054 (17).
The genomic sequence of B. longum JCM1217 was made publicly available by Fukuda et al. in 2011 (27). Interestingly, we found that this strain possesses no gene homologous to the previously identified LNBase (glycoside hydrolase family 20, GH20) from B. bifidum (17). Although the B. longum JCM1217 genome encodes a GH20 enzyme, the protein obviously lacks the sequence motifs to distinguish LNBase from ␤-N-acetylhexosaminidase (another GH20 member) (28). Actually, the GH20 enzyme of B. longum JCM1217 was recently found to be ␤-N-acetylglucosaminidase active on lacto-N-triose II (GlcNAc␤1-3Gal␤1-4Glc) and on chitin oligosaccharides but not on LNT (29). These findings prompted us to identify a potentially novel LNBase gene encoded in the genome of B. longum.
Here, we found that the locus tags BLLJ_1505 and BLLJ_1506, neither of which has been functionally annotated, constitute LNBase. The enzyme body is the product of BLLJ_1505, whereas the product of BLLJ_1506 acts as a designated chaperone for the BLLJ_1505 protein (LNBase). The substrate specificity of this novel LNBase was found to be quite different from that of GH20 LNBase. Interestingly, it liberated GalNAc␤1-3GlcNAc from the p-nitrophenyl sugar. The disaccharide structure is present in the O-glycans of ␣-dystroglycan (30). BLLJ_1505 and BLLJ_1506 homologs were found in the genomes of B. longum, B. bifidum, and a few gut microbes, suggesting that this LNBase has uniquely evolved in particular species.
described previously (18,31). p-Nitrophenyl (pNP)-␤-lacto-Nbioside I (LNB-␤-pNP) was obtained from Sigma-Aldrich or synthesized essentially as described previously (32). Briefly, hepta-Oacetyl-LNB was prepared by treating LNB with acetic anhydride and pyridine. Following chlorination with hydrogen chloride in acetic acid, p-nitrophenol was glycosylated in dimethylformamide. O-Deacetylation was carried out using sodium methoxide in methanol. The synthesized LNB-␤-pNP was used for screening a genomic library of B. longum subsp. longum JCM1217 (described below). GalNAc␤1-3GlcNAc␤-pNP and GalNAc␤1-4GlcNAc (LacdiNAc)␤-pNP were purchased from Tokyo Chemical Industry Co. (Tokyo, Japan). Other reagents of analytical grade were obtained from commercial sources. The oligonucleotides used in the plasmid and strain constructions are listed in supplemental Table S1.
Bacterial Strains and Media-The strains used in this study were B. longum subsp. longum JCM1217 T and 105-A (33) and Escherichia coli DH5␣ and BL21 ⌬lacZ (DE3)/pRARE2 (23). Bifidobacteria were grown in GAM broth (Nissui Pharmaceutical, Tokyo, Japan) or basal medium (17) at 37°C under anaerobic conditions. Chloramphenicol and spectinomycin were added at final concentrations of 4 and 30 g/ml, respectively, as required. E. coli strains were grown in LB broth, and when necessary, ampicillin, chloramphenicol, and spectinomycin were added at concentrations of 100, 30, and 30 g/ml, respectively.
Construction of a Genomic Library of B. longum JCM1217-The genome of strain JCM1217 was extracted as described previously (13) and partially digested with Sau3AI. The 8 -10-kbp DNA fragments were recovered and ligated with the BamHIdigested pBR322. The resulting reaction mixture was used to transform E. coli DH5␣ to obtain the genomic library, which was then screened for LNB-␤-pNP hydrolysis. Note that E. coli DH5␣ exhibits neither ␤-1,3-galactosidase nor LNBase activity.
Gene Disruption and Complementation Analysis of B. longum-A targeted gene disruption in B. longum was carried out as follows. The region corresponding to nucleotide numbers 1794468 -1795807 (see Fig. 1) was amplified by high fidelity PCR using PrimeStar MAX DNA polymerase (Takara Shuzo, Shiga, Japan). BamHI sites were attached at both ends (supplemental Table S1). The amplified fragment (a partial lnbX gene) was ligated with the 1.9-kbp BamHI fragment (carrying ColE1 ori and the spectinomycin resistance gene) of pBS423 (34) to generate a suicide plasmid. The resulting plasmid was introduced into B. longum 105-A by electroporation and integrated into the genome by a single crossover event at the lnbX locus. Spectinomycin-resistant colonies were isolated, and the disruption of the lnbX gene was confirmed by genomic PCR analysis (data not shown).
Complementation analysis was performed using E. colibifidobacteria shuttle vector pTK2064, in which ColE1 ori (high copy) of plasmid pBS423 (34) was replaced with pSC101 ori (low copy) (supplemental Table S1). Exchanging the ori ensured stable maintenance of the target genes (ϳ6.3 kbp, GC content 66%) in cells. The lnbX and lnbXY genes (1792583-1798023 and 1792583-1798862 nt, respectively, in Fig. 1) were amplified by high fidelity PCR and inserted into the NdeI site of pTK2064 by the In-Fusion methodology (Clontech) to generate pTK2193 (lnbX ϩ ) and pTK2174 (lnbX ϩ Y ϩ ), respectively. A plasmid (pTK2241) carrying the lnbY gene under the control of the lnbX promoter was constructed by inverse PCR using pTK2174 as a template, in which nucleotides 1793160 -1797962 ( Fig. 1) were eliminated (supplemental Table S1). All of the PCR-amplified fragments were sequenced to ensure that no base changes other than those designed had occurred.
Expression and Purification of Recombinant Proteins-Recombinant LnbX and LnbY were expressed in E. coli and purified as both non-tagged and hexahistidine (His)-tagged forms. The expression vectors contained the DNAs coding for amino acid residues 31-1573 of LnbX and amino acid residues 30 -280 of LnbY, following genetic removal of the signal peptide and membrane anchor of LnbX and the signal peptide of LnbY. The non-tagged and C-terminal His-tagged LnbX proteins were expressed using pET3a and pET23b (Novagen, Darmstadt, Germany), respectively (supplemental Table S1). The PCR-generated fragments were inserted into the NdeI-BamHI and NdeI-NotI sites of pET3a and pET23b, respectively, by the In-Fusion methodology. To express the non-tagged and N-terminal His-tagged LnbY, the amplified genes were first inserted into the NdeI-BamHI site of pET3a, and the resulting T7 expression cassettes were moved to a ColE1-compatible plasmid pCDF (Novagen) (supplemental Table S1). The entire fragments used for later manipulation were sequenced.
The non-tagged LNBase (LnbX) was purified from the cellfree extract of E. coli cells co-expressing the lnbX and lnbY genes. The cells were grown in 2 liters of LB medium, and when the optical density at 600 nm reached 0.5, isopropyl-␤-D-thiogalactopyranoside was added to induce protein expression. The cells were harvested, suspended in 20 mM Tris-HCl buffer (pH 8.0), and disrupted by sonication. The supernatant was applied to a DEAE-Sepharose fast flow column (GE Healthcare), and proteins were eluted by a linear gradient of 1 M NaCl. Each fraction was tested for LNB-␤-pNP-hydrolyzing ability. Active fractions were combined, dialyzed to 50 mM sodium phosphate buffer (pH 7.0) containing 0.8 M ammonium sulfate, and applied to a butyl-Sepharose 4 column (GE Healthcare). Elution was performed by decreasing the concentration of ammonium sulfate to 0 M. The protein was further purified by Mono Q 5/50 GL (0 -1 M NaCl in 20 mM Tris-HCl buffer (pH 8.0)) (GE Healthcare), Poros HP (0.8 -0 M ammonium sulfate in 50 mM sodium phosphate buffer (pH 7.0)) (PerSeptive Biosystems), and Superdex 200 10/300 GL (20 mM Tris-HCl buffer (pH 8.0) containing 150 mM NaCl) (GE Healthcare) column chromatography. The purified protein was concentrated using an Amicon Ultra 100K concentrator (Millipore, MA). His-tagged LnbX was similarly expressed and purified by nickel-nitrilotriacetic acid-agarose (Qiagen, Hilden, Germany), Mono Q 5/50 GL, and Superdex 200 10/300 GL column chromatography.
Non-tagged LnbY was purified similarly, using the same columns. The purity of LnbY was assessed by SDS-polyacrylamide gel electrophoresis of each fraction. His-tagged LnbY was purified by nickel-nitrilotriacetic acid-agarose and Superdex 75 10/300 GL column chromatography. The purified protein was concentrated using an Amicon Ultra 10K concentrator.
Protein concentrations were determined using a Bradford protein assay kit (Bio-Rad). Bovine serum albumin was used as a standard.
Inductively Coupled Plasma Emission Spectroscopy-Prior to the analysis, the purified, non-tagged LnbX and LnbY were extensively dialyzed against 5 mM Tris-HCl (pH 8.0) in the presence and absence of 0.2 mM EDTA. The content and concentration of metal ions in the proteins were determined by inductively coupled plasma emission spectroscopy using ICPS-8100 (Shimadzu, Kyoto, Japan), with dialysis buffer as a control. The standard was ICP multielement standard solution IV.
Edman Degradation Analysis-The N-terminal amino acid sequence of the non-tagged LnbX was determined by Edman degradation using a PPSQ-33A protein sequencer (Shimadzu).
Enzyme Assay-The standard reaction mixture contained 50 mM MES buffer (pH 5.4). pNP-sugars and oligosaccharides were used as the substrates. The reaction was initiated by adding the enzyme, and the reaction mixture was incubated at 25°C for an appropriate time, in which the linearity of the reaction rate was observed. The enzyme concentrations used in the kinetic analysis were 0.90 nM (LNB-␤-pNP), 1.8 nM (GalNAc␤1-3GlcNAc␤-pNP and LNT), and 140 nM (LNFP I). The substrate concentrations were varied from 0.3 to 2 times the respective K m values. The reaction was terminated by adding 1 M sodium carbonate (for pNP-sugars) or by heating (for oligosaccharides). The amount of liberated p-nitrophenol was determined from the absorbance at 400 nm. Oligosaccharide hydrolysis was monitored by high performance anion exchange chromatography with pulsed amperometric detection (HPAEC-PAD) using a Dionex ICS 3000 system, equipped with a CarboPac PA-1 column (Dionex, Sunnyvale, CA). The elution was performed under constant flow (0.25 ml/min) with a linear gradient of 0-330 mM sodium acetate in 125 mM NaOH at 30°C for 20 min. Standard curve was constructed from known concentrations of Lac. The reaction products of the LNFP I hydrolysis were treated with 1,2-␣-L-fucosidase (13) prior to HPAEC-PAD analysis to remove the fucosyl residue from the 2Ј-fucosyl-LNB (Fuc␣1-2Gal␤1-3GlcNAc) (reaction product), which co-elutes with Lac. The kinetic parameters were calculated by curve-fitting the experimental data to the Michaelis-Menten equation, using KaleidaGraph version 4.0 (Synergy Software). One unit of enzyme activity was determined as the amount of enzyme required to produce 1 mol of p-nitrophenol or Lac under the specified conditions.
The physicochemical properties of the enzyme were determined using 1 mM LNB-␤-pNP as a substrate. To estimate the optimal temperature, the reaction mixture was incubated at various temperatures for a short period (Ͻ1 min). The optimal pH was determined using the following buffers (100 mM): citrate-NaOH (pH 4.0 -5.0), MES (pH 5.0 -6.5), MOPS (pH 6.5-8.0), and Tris-HCl (pH 8.0 -9.0). The temperature and pH stabilities were determined from the residual activities following enzyme incubation in 50 mM MES (pH 5.4) for 30 min at various temperatures and overnight dialysis against various 10 mM buffers at 4°C, respectively.
Determination of Stereochemical Course of Hydrolysis-The reaction mixture containing 50 mM MES buffer (pH 5.4), 2 mM LNB-␤-pNP, and the enzyme (7 milliunits) was incubated at 30°C in a total volume of 60 l. The mixture was sampled at specified times and immediately injected into a high performance liquid chromatography (HPLC) system equipped with TSKgel Amide-80 (4.6 ϫ 250 mm) (Tosoh, Tokyo, Japan). Elution was carried out using a solvent system of acetonitrile/water of 65:35 at a flow rate of 1 ml/min and was monitored at 214 nm.
Electrospray Ionization-MS-For oligosaccharide analysis, mass spectra were obtained on an LCMS-2020 system (Shimadzu) in positive ion mode. The samples were dissolved in 0.1% formic acid/acetonitrile (1:1 by volume) and injected at 3 l/min with a microsyringe pump.
Western Blot Analysis Using Anti-His Tag Antibodies-Cellfree extracts (0.6 mg) of E. coli cells expressing His-tagged LnbX in the presence and absence of co-expression of nontagged LnbY were separated by size exclusion chromatography (Superdex 200 10/300 GL). Each fraction (0.5 ml) was assayed for protein concentration, LNBase activity, and LnbX content. LnbX was detected using anti-His-tag antibodies conjugated with horseradish peroxidase (Qiagen), following SDS-polyacrylamide gel electrophoresis and membrane transfer (Immobilon-P, Millipore). Chemiluminescence was detected using ECL Western blotting detection reagents (GE Healthcare Life Sciences) and a LAS-3000 imaging system (Fujifilm, Tokyo, Japan).
In Vitro Refolding of LnbX-Denatured LnbX was refolded in the presence and absence of LnbY and metals (Ca 2ϩ and Mg 2ϩ ). This experiment was carried out on C-terminal His-tagged LnbX and N-terminal His-tagged LnbY. The purified LnbX (final concentration 1 M) denatured in 6 M guanidine HCl was diluted 100-fold by adding 50 mM HEPES buffer (pH 7.0) containing various concentrations of LnbY (0, 1, 5, 20, and 50 M) and metal ions (0, 0.01, 0.1, 1, and 5 mM CaCl 2 and MgCl 2 ) in a total volume of 500 l. The mixture was immediately placed in the dialysis cassette (Slide-A-Lyzer G2, Thermo Scientific, MA), and dialyzed against HEPES buffer containing or not containing metal ions. The samples were removed from the dialysis cassettes at specified times and monitored for hydrolytic activity toward LNB-␤-pNP. Enzymes were assayed within 2 min of incubation to avoid refolding during activity determination.

RESULTS
Identification of BLLJ_1505 and BLLJ_1506 as the Constituents of LNBase from B. longum-A genomic library of B. longum JCM1217 was constructed using E. coli DH5␣, and the transformants were screened for their ability to liberate p-nitrophenol from LNB-␤-pNP. Five of 1920 transformants tested positive for the activity. Sequence analysis revealed that the DNA fragment in the plasmid isolated from the clones covers a particular genomic locus. The shortest fragment comprises genomic nucleotides 1792199 -1800198 (Fig. 1). This region includes two complete open reading frames (ORFs) designated BLLJ_1505 (1599 aa) and BLLJ_1506 (280 aa). Neither ORF had been functionally annotated; nor did they share sequence similarities with any characterized proteins. The insert also contained two truncated ORFs (BLLJ_1504 and BLLJ_1507), which were close homologs of bacterial ␥-glutamylcysteine synthetase and 2-hydroxyglutaryl-CoA dehydratase activator, respectively. Pfam analysis indicated a right-handed ␤ helix region (␤ helix, PF13229) at amino acid residues 158 -315 of BLLJ_1505 and an uncharacterized sugar-binding domain (FIVAR, PF07554) at residues 1435-1487, whereas BLLJ_1506 lacks any functional motifs (35). The SignalP 4.1 and PSORT servers predicted the presence of a signal peptide (1-30 aa) and a membrane anchor (1577-1593 aa) in BLLJ_1505 and a signal peptide (1-29 aa) in BLLJ_1506 (36). BLLJ_1505 and BLLJ_1506 were separated by 3 bp, and an inverted repeat was found immediately downstream of BLLJ_1506 (1798848 -1798889 nt). Homologs of these proteins in the database are introduced below.
We first disrupted locus tag BLLJ_1505 by a homologous recombination event. This manipulation was performed on B. longum strain 105-A (LNBase ϩ ) rather than JCM1217, because the latter is not amenable to standard genetic manipulation (33). Sequence identity of 99% was observed between the genes from both strains (data not shown). Disruption of the corresponding gene in strain 105-A resulted in complete loss of LNT-hydrolyzing ability, as revealed by TLC analysis (Fig. 2, lane 3). Activity was not restored by introducing BLLJ_1505 via a shuttle vector (Fig. 2, lane 5). BLLJ_1506 alone similarly failed to rescue the LNBase Ϫ phenotype (Fig. 2, lane 6); however, when both BLLJ_1505 and BLLJ_1506 were introduced, the activity resumed (Fig. 2, lane 7). These results indicate that LNBase of B. longum JCM1217 consists of BLLJ_1505 and BLLJ_1506 (hereafter referred to as lnbX and lnbY, respectively).
Besides B. longum, both lnbX and lnbY were required for the active LNBase expression in E. coli. In this experiment, the signal peptide and membrane anchor of LnbX and the signal peptide of LnbY were genetically removed, and the corresponding genes were expressed in separate plasmids (see "Experimental Procedures"). The cell-free extract of the E. coli strain carrying both lnbX and lnbY showed LNB-␤-pNP-hydrolyzing activity (9.5 units/mg), whereas strains expressing either lnbX or lnbY did not (Table 1). Interestingly, mixing the cell-free extracts of E. coli strains individually expressing lnbX and lnbY induced low level activity (0.38 unit/mg). When added prior to mixing, EDTA completely quenched the activity, whereas after mixing, it produced no effect on activity (0.34 unit/mg). Similarly, EDTA did not alter the activity in the cell-free extract of E. coli cells co-expressing lnbX and lnbY (9.5 versus 9.6 units/mg).
Purification and Physicochemical Characterization-We first purified non-tagged recombinant LNBase from E. coli cells co-expressing LnbX and LnbY. Purification was carried out by monitoring the LNB-␤-pNP-hydrolyzing activity. The purification process is summarized in

Novel Lacto-N-biosidase from B. longum
sis with an apparent molecular mass of 161 kDa, which agrees with the calculated mass of LnbX (164 kDa) (Fig. 3a). Edman degradation analysis revealed the N-terminal amino acid sequence as MQSAT, consistent with the designed construct for expressing LnbX (31-1573 residues). The addition of His tag to the C terminus of the protein did not alter the enzymatic properties of LnbX (discussed below), and the tagged protein was easily purified by Ni 2ϩ affinity chromatography. Moreover, LnbX migrated to a virtually identical position in the SDS-polyacrylamide gel electrophoresis regardless of whether LnbY was co-expressed or not (see Fig. 7a). The addition of purified LnbY to purified LnbX (correctly folded form, described below) produced no effect on LNBase activity (data not shown). These results indicated that the active LNBase molecule from B. longum is a single gene product (that of lnbX). The native molecular mass of LNBase (LnbX) deduced from size exclusion chromatography was 356 kDa, indicating that the enzyme forms a dimer in solution (Fig. 3b). Inductively coupled plasma emission spectroscopy revealed that LnbX contains 1 Ϯ 0.1 molecule of magnesium, 3 Ϯ 0.6 molecules of calcium, and 0.4 Ϯ 0.1 molecule of zinc. Magnesium and calcium ions are probably tightly bound to or buried in the protein because the addition of EDTA to the dialysis buffer did not alter the metal contents; however, zinc ion was removed from the protein by adding EDTA. Recall that EDTA does not affect the specific activity of the purified enzyme.
The optimal pH and temperature of the enzyme for LNB-␤-pNP hydrolysis was 5.4 (MES buffer) and 60°C, respectively. The enzyme was stable in the pH range 4.5-9.5 and below 45°C for 30 min. The purified recombinant enzyme maintained its activity for at least 3 weeks (data not shown).
Kinetic analysis revealed that the catalytic efficiency of LnbX toward LNB-␤-pNP and LNT is 5-and 4-fold higher, respec- Novel Lacto-N-biosidase from B. longum AUGUST Fig. 4e). The activity of the enzyme on LST a was too low to determine the kinetic parameters.
Stereochemistry of Hydrolysis-Stereochemical course of the reaction was determined by monitoring the anomeric configuration of LNB released from the pNP-substrate (Fig. 6). A normal phase HPLC was employed for this purpose. LNB appeared with a ratio of ␣-anomer to ␤-anomer of 25:75 and 35:65 after 1and 10-min incubation, respectively. The peak area for LNB-␤-pNP decreased, whereas those of p-nitrophenol and LNB increased as the reaction proceeded (10 and 30 min), and the ␣/␤-ratio of LNB had almost reached the equilibration of mutarotation (60/40) at 120 min. These results showed that LnbX is a retaining enzyme.
Role of LnbY in the Folding Process of LnbX-To examine the folding state of LnbX, we performed gel filtration analysis using the cell-free extracts of E. coli cells expressing His-tagged LnbX with and without LnbY (non-tagged), followed by SDS-poly-acrylamide gel electrophoresis and Western blot analysis using anti-His tag antibodies (Fig. 7, a and b). The protein concentration and specific activity (from 7.5-to 13-ml elution volume) were also determined for each fraction. The quantity of Histagged LnbX obtained in a soluble fraction was consistent and independent of LnbY co-expression (Fig. 7a). LnbX eluted in a void volume in the absence of LnbY, whereas it eluted at 10.5-11.0 ml (molecular mass 399 -316 kDa) in the presence of LnbY. No LNBase activity was detected in any fractions of cellfree extract of the strain expressing LnbX alone, whereas the activity exactly coincided with the LnbX protein peak of the cell-free extract of the strain co-expressing LnbY in the chromatography.
In vitro refolding experiment of LnbX was also performed, in which the renaturation process was represented by the enzymatic activity on LNB-␤-pNP (see "Experimental Procedures"). When the denatured LnbX was incubated and dialyzed alone, no detectable activity was observed throughout the incubation periods (Fig. 7c). No precipitation occurred during dialysis to reflect that the protein formed a soluble aggregate in the cellfree extracts of E. coli cells expressing LnbX alone (Fig. 7b, left). In contrast, when the refolding solution contained LnbY (nondenatured), activity gradually recovered. Efficient refolding occurred when Ca 2ϩ and Mg 2ϩ were supplemented together with LnbY. As mentioned above, both metal ions are contained in the native enzyme. Magnesium ions were found to be more important than calcium ions in the refolding process. When Mg 2ϩ alone was added, the activity was over 90% of that obtained in the presence of the two metal ions, whereas Ca 2ϩ The reactions were carried out in 50 mM MES buffer (pH 5.4) (for LnbX) and citrate-phosphate buffer (pH 4.5) (for LnbB) containing 1 mM (for pNP-sugars) and 5 mM (for oligosaccharides) substrates. Mixtures were incubated overnight at 25°C in the presence and absence of the enzyme (6 milliunits toward pNP-LNB). The reaction products were analyzed by thin layer chromatography (see Fig. 4). Neither LnbX nor LnbB acted on pNP-monosaccharides (L-Arap␣-pNP, Gal␤-, GalNAc␤-, Glc␤-, GlcNA␤-, GlcUA␤-, Man␤-, Xyl␤-), Lac␤-pNP, and LacdiNAc␤-pNP. ϩϩϩ, complete hydrolysis (100%); ϩϩ, significant hydrolysis (Ͼ50%); ϩ, partial hydrolysis (Ͻ50%); Ϫ, no hydrolysis. b The kinetic parameters of LnbB are taken from our recent report (28). c ND, not determined. alone failed to attain 70% activity (data not shown). Refolding also depended on the concentration of the metal ions; 0.1 mM was found to optimize refolding within the tested range (10 M, 0.1 mM, 1 mM, and 5 mM; data not shown). Interestingly, the metal ions by themselves could slightly stimulate the refolding of denatured LnbX, and very low but non-negligible activity was detected as the incubation continued (Fig. 7c). Refolding efficiency varied with the molar ratio of LnbY added to denatured LnbX. When the ratios were increased to 5 and 20 from equimolar, LNB-␤-pNP-hydrolyzing activity was elevated by 1.2-and 1.5-fold, respectively; however, a ratio of 50 caused a decline in the activity (Fig. 7d). The highest activity attained in the refolding experiments (6.3 units/mg) was one-sixth that of the native recombinant enzyme (37.2 units/mg), indicating a renaturation efficiency of about 17%.

DISCUSSION
The present study revealed a novel LNBase in B. longum that completely differs from hitherto identified GH families with respect to amino acid sequence, substrate specificity, maturation process, and possibly structure.
Physicochemical Properties-Initially, we considered that LNBase is a heterodimer comprising the LnbX and LnbY subunits because LNB-␤-pNP-hydrolyzing activity occurred when the cell-free extracts of E. coli strains individually expressing LnbX and LnbY were mixed and incubated (Table 1). However, the subsequent experiments revealed that the purified LNBase consists solely of the lnbX gene product. The C terminus of the protein could be truncated to 1431 amino acid residues (immediately upstream of the FIVAR domain; Fig. 1) without loss of activity (data not shown).
Sequence Features-In the complementation analysis using lnbX-deficient B. longum, plasmid-mediated introduction of the lnbX gene failed to restore LNBase activity, although the genome of the deficient strain contained an intact lnbY. Introduction of both lnbX and lnbY genes by plasmid recovered the LNBase ϩ phenotype of the parental strain (Fig. 2). These results indicate that the lnbX and lnbY genes constitute an operon, and the failure of the phenotypic change in the strain carrying plasmid-borne lnbX is due to a polar effect. The two genes were separated by a mere 3 bp, and a promoter-like sequence was not found upstream of the ORF of lnbY; instead, a 42-bp inverted repeat was found immediately downstream of lnbY.
In the genomes of B. longum JCM1217 and B. longum BBMN68, the lnbXY gene is located between the ␥-glutamylcysteine synthetase and 2-hydroxyglutaryl-CoA dehydratase activator genes (99% identity in nucleotide sequence over the region covering the four genes) (supplemental Fig. S2). In   Table 3.
the genomes of B. longum strains DJ010A, F8, JDM301, KACC91563,andNCC2705(possiblyLNBase Ϫ ),the␥-glutamylcysteine synthetase gene is located just downstream of the 2-hydroxyglutaryl-CoA dehydratase activator gene (KEGG database). We currently lack knowledge of how strains JCM1217 and BBMN68 acquired the lnbXY gene at this locus. No insertion sequence was found at either end of the gene (IS Finder). Interestingly, however, the genomes of B. bifidum strains BGN4, PRL2010, and S17 contain lnbX and lnbY homologs with 50 -60% amino acid sequence identity, which are also located near the ␥-glutamylcysteine synthetase and 2-hydroxyglutaryl-CoA dehydratase activator genes (supplemental Fig. S2). In these genomes, the lnbX and lnbY homologs are separated by the ␥-glutamylcysteine synthetase gene, and their sense strands are opposed. Note that in B. bifidum, either the LnbX homolog or the LnbY homolog lacks a signal peptide; therefore, although both proteins are expressed in the cells, they cannot associate to assist the folding of LnbX. The LnbX homologs have been found in the human gut microbial genomes of Ruminococcus lactaris ATCC29176 (RUMLAC_00599), Clostridium nexile DSM1787 (clonex_ 02958), and Clostridium celatum DSM1785 (HMPRF0216_ 02974). The amino acid identity between these homologs and B. longum LnbX is about 40%. Among these bacteria, R. lactaris ATCC29176 alone possesses the LnbY homolog (27% identity, RUMLAC_00598), but the protein lacks a signal peptide. These results suggest that the gene(s) has specifically evolved within the gut niche. Given the length and sequence diversity among the few known occurrences, the gene(s) might still be evolving. Further elucidation requires characterization of each of these homologs.
Substrate Specificity-Besides sequence, the substrate specificity of LNBase from B. longum significantly differs from that of GH20 LNBase from B. bifidum. Whereas GH20 LNBase (including one from Streptomyces sp.) acts only on unmodified LNB structures (Fig. 4) (17,37), the novel LNBase from B. longum was able to act on LNFP I and LST a, which have Fuc and Neu5Ac residues at the O2 and O3 positions of the distal Gal residue, respectively. Considering the high activity of LnbX on LNT, followed by LNFP I and LST a, and that the enzyme is inactive on GlcNAc-pNP, binding of the substrate at subsite Ϫ2 might be indispensable for the enzyme to exert its activity. The subsite Ϫ2 is slightly widened at the O2 position of Ϫ2 Gal, enabling LnbX to efficiently hydrolyze GalNAc␤1-3-GlcNAc␤-pNP (Table 3). GalNAc␤1-3GlcNAc␤ disaccharide structure has been found in the O-mannosyl glycans of ␣-dystroglycan, and ␤1-3GalNAc transferase with acceptor specificity toward GlcNAc (B3GALNT2) has been identified in humans and mice (38). Recently, Stevens et al. (39) reported that mutations in B3GALNT2 cause congenital muscular dystrophy-dystroglycanopathy, characterized by brain and eye anomalies. GalNAc␤1-3GlcNAc is a linkage isomer of LacdiNAc (N,NЈdiacetyllactosediamine, GalNAc␤1-4GlcNAc) found in the N-glycans of glycoproteins such as lutropin (a hormone produced by gonadotroph cells) and tenascin-R (an extracellular matrix exclusively expressed in the central nervous system) as well as in glycoproteins from parasites (40,41). The LacdiNAc structure in the leukemia inhibitory factor receptor is known to influence self-renewal maintenance of mouse embryonic stem cells (42). Considering that LnbX does not act on LacdiNAc␤-pNP and that GalNAc␤1-3GlcNAc␤-pNP hydrolysis is not inhibited by the presence of LacdiNAc␤-pNP, the enzyme may serve as a new tool for examining the glycan structures of glycoconjugates and for distinguishing between GalNAc␤1-3-GlcNAc and LacdiNAc structures. However, we have not tested whether the enzyme can liberate GalNAc␤1-3GlcNAc from natural substrates. With respect to the unique substrate specificity of B. longum LNBase, we also emphasize that B. longum JCM1217 grown in HMO medium induces a slight but continuous decrease of LNFP I (from 0.67 to 0.24 g/liter) (12). We did not identify 2Ј-fucosyl LNB in that study because we lacked a standard compound, and we had not elucidated the unique substrate specificity of LnbX. Interestingly, the HPLC profile in that paper displays a small peak between Lac and 2Ј-fucosyllactose (12).
LnbY as a Designated Chaperone for LnbX-In both bifidobacteria and E. coli, LnbY is required for active expression of LnbX (i.e. LNBase activity (Table 1 and Fig. 2)). In the absence of LnbY co-expression, LnbX formed a soluble aggregate and eluted in the void fraction in the gel filtration analysis with no detectable activity (Fig. 7b, left). In the presence of LnbY, active LnbX was expressed and eluted at the anticipated fraction (Fig.  7b, right). Also, as mentioned above, no apparent post-translational modification occurred on LnbX. These results strongly suggest that the LnbY protein is involved in the folding process of LnbX. The denatured LnbX gained considerable activity during the incubation with LnbY and removal of guanidine HCl by dialysis (Fig. 7c). The presence of two metal ions (Ca 2ϩ and Mg 2ϩ , especially Mg 2ϩ ) was important for efficient refolding (Fig. 7c). The addition of EDTA to mature LnbX had no effect on enzymatic activity (Table 1) and did not sequester the ions during dialysis, indicating that the two metal ions were tightly bound by the protein and stabilized the correct folding of LnbX. Indeed, the two metals slightly stimulated correct folding of LnbX even in the absence of LnbY (Fig. 7c). The results in Table  1 also indicate that LnbY can partially renature a soluble aggregate form of LnbX. To our knowledge, this is the first glycosidase that requires a designated chaperone for its active expression. Mucin type core 1 (T antigen)-synthesizing ␤1-3Gal transferase (C1␤Gal-T) is known to require its designated chaperone, Cosmc (43). Although the C1␤Gal-T and Cosmc genes are located on different chromosomes, Cosmc appears to be specialized for C1␤Gal-T folding, because the activity of other glycosyltransferases is normal in Cosmc-null Jurkat cells. Interestingly, both C1␤Gal-T and Cosmc are secretory proteins and form homodimers and thus share a functional similarity with LnbX and LnbY. Another functional homolog is seen in the synthetic pathway of the glycosylphosphatidylinositol anchor. PGI-X has been identified as an important component for active expression of PIG-M, the catalytic subunit of glycosylphosphatidylinositol mannosyltransferase I (44). Interestingly, these chaperone proteins are of similar size (Cosmc (318 aa), PGI-X (252 aa), and LnbY (280 aa)) despite their sequence dissimilarity.
Physiology in HMO Assimilation-By identifying the lnbXY gene, we unequivocally revealed that B. longum JCM1217 extracellularly decomposes LNT by this novel LNBase. The produced LNB and Lac should be imported into the bifidobacterial cells by the GNB/LNB transporter and the Lac transporter, respectively, for further degradation (19,20,21) (supplemental Fig. S1). Among the infant gut-associated bifidobacteria, B. bifidum and some strains of B. longum specifically degrade type-1 HMOs by LNBase (LnbB and LnbX, respectively). B. longum LNBase showed lower K m and higher k cat values for LNT than B. bifidum LNBase. Therefore, assuming that the LnbX homolog of B. bifidum does not function within cells (as described above), B. longum can exploit LNT over B. bifidum. This might be an important survival strategy for gut- The reaction mixture containing 50 mM MES buffer (pH 5.4), 2 mM LNB-␤-pNP, and the enzyme (7 milliunits) was incubated at 30°C and sampled at the indicated times to be analyzed by an HPLC system equipped with TSKgel Amide-80 (4.6 ϫ 250 mm). Elution was carried out at room temperature using acetonitrile/water (65:35) flowing at 1 ml/min and was monitored by measuring the absorbance at 214 nm.
dwelling B. longum because this subspecies essentially assimilates LNT only among HMOs due to the lack of ␣-Lfucosidase (12). B. longum might employ its highly active LNBase to sequester LNTs formed from LNFP I and lacto-N-difucohexaose I (the main HMO species) by B. bifidum 1,2-, and 1,3/4-␣-L-fucosidases in the gut (13,15). In this sense, it is interesting that B. bifidum grown in HMO medium ignores the degradation products of HMOs in the spent medium even during the exponential growth phase (12). B. infantis probably circumvents the interception by ingesting all HMOs as intact forms and digesting them intracellularly (23,24,25).
Conclusions-Hitherto uncharacterized protein sequences of unknown function, namely BLLJ_1505 and BLLJ_1506, are found to constitute LNBase, an enzyme exhibiting a unique substrate specificity and maturation process. These findings are important not only for comprehensive understanding of HMO assimilation by infant gut bifidobacteria but also in