The Structure Of A SusD Homolog, BT1043, Involved In Mucin O-Glycan Utilization In A Prominent Human Gut Symbiont

Abstract

Mammalian distal gut bacteria have an expanded capacity to utilize glycans. In the absence of dietary sources, some species rely on host-derived mucosal glycans. The ability of Bacteroides thetaiotaomicron, a prominent human gut symbiont, to forage host glycans contributes to both its ability to persist within an individual host and to be transmitted naturally to new hosts at birth. The molecular basis of host glycan recognition by this species is still unknown, but likely occurs through an expanded suite of outermembrane glycan-binding proteins that are the primary interface between B. thetaiotaomicron and its environment. Presented here is the atomic structure of the B. thetaiotaomicron protein BT1043, an outermembrane lipoprotein involved in host glycan metabolism. Despite a lack of detectable amino acid sequence similarity, BT1043 is a structural homolog of the B. thetaiotaomicron starch-binding protein SusD. Both structures are dominated by tetratrico peptide repeats that may facilitate association with outer membrane β-barrel transporters required for glycan uptake. The structure of BT1043 complexed with N-acetyl lactosamine reveals that recognition in mediated via hydrogen-bonding interactions with the reducing end of β-N-acetyl glucosamine, suggesting a role in binding glycans liberated from the mucin polypeptide. This is in contrast to CBM 32 family members that target the terminal non-reducing galactose residue of mucin glycans. The highly articulated glycan-binding pocket of BT1043 suggests that ligand binding to BT1043 relies more upon interactions with the composite sugar residues than upon overall ligand conformation as previously observed for SusD. The diversity in amino acid sequence level likely reflects early divergence from a common ancestor while the unique and conserved α-helical fold the SusD family suggests a similar function in glycan uptake.

The distal gut microbial community (microbiota) of humans and other mammals is one of the most densely populated ecosystems on Earth. It is dominated by members of Bacteria but also contains members of Archaea and Eukarya (1, 2). The microbiota, whose acquisition begins at birth (3), plays pivotal roles in a number of facets of human physiology, including development of the immune system (4, 5), and acquisition of nutrients (6). Despite the essential role of the microbiota, it is not clear how this community retains its functional stability over time with constant changes in diet and in the face of continuous and rapid replacement of the intestinal epithelium and its overlying mucus layer.

While the composition and availability of dietary glycans changes over time, the endogenous pool of host glycans, including glycosaminoglycans, mucin O-linked glycans and N-linked glycans, offers a more consistent food source for gut microbial species. Members of the bacterial phylum Bacteroidetes, one of two dominant phyla in the human and other mammalian gut microbiotas (2), are well-characterized consumers of complex glycans, including those emanating from the host mucosa (7). Correspondingly, these species have large collections of glycoside hydrolases and polysaccharide lyases in their genomes (8), a feature that likely contributes to this phenotype.

Bacteroidetes thetaiotaomicron is a member of the Bacteroidetes phylum, whose genome has been sequenced (9). It has served as a model organism for identifying some of the molecular mechanisms deployed by human gut Bacteroidetes for sensing, recognizing, importing, and processing, dietary and host carbohydrates (10-13). Whole genome transcriptional profiling of B. thetaiotaomicron isolated from the intestines of adult gnotobiotic mice fed a diet of simple sugars, as well as gnotobiotic mice surveyed during the suckling period of postnatal development, revealed increased expression of glycoside hydrolases such as hexosaminidases and fucosidases compared to gnotobiotic mice consuming a standard polysaccharide-rich chow diet (12, 14). These findings suggested increased degradation of host glycans by this bacterium during periods when dietary polysaccharides were scarce. This ability of B. thetaiotaomicron to selectively forage on host glycans was shown to be important for fitness in vivo. A mutant strain deficient in mucin O-glycan utilization is unable to effectively compete with an isogenic wild-type strain in the intestines of gnotobiotic mice (13). This same mutant was also defective in the ability to be transmitted from a colonized mother to newborn pups in the days immediately following their birth. Together, these observations provide evidence that host glycan metabolism contributes to both the ability of this species to persist in a given host and to be transmitted naturally to new hosts.

The ability of Gram-negative, gut Bacteroides spp. such as B. thetaiotaomicron to forage upon a wide variety of glycans in the gut environment is attributed to the expression of similarly patterned polysaccharide utilization loci (PUL) (13, 14). PULs encode cell envelope-associated protein complexes consisting of one or more glycolytic enzymes, and homologs of two proteins, SusC and SusD, that are involved in glycan recognition and import (8). The starch utilization system (Sus) of B. thetaiotaomicron was the first such PUL described (15), although 87 additional Sus-like PULs have since been identified in this species (13), and several hundred more in other gut Bacteroides species (8). The Sus system is comprised of eight genes (susRABCDEFG) with five of its protein products, SusCDEFG, located in the outer membrane. In response to maltose, the SusR regulator drives expression of the sus operon, which is required for growth on amylose, amylopectin, pullulan and maltooligosaccharides (16, 17). Large starch molecules are hydrolyzed by the outer membrane α-amylase SusG and imported into the periplasm through SusC, a predicted TonB-dependent β-barrel porin (18-20). While SusD has a critical role in starch and maltooligosaccharide binding (17), the roles of SusE and SusF are unclear (19, 20). Two periplasmic glycoside hydrolases, SusA and SusB, a neopullulanase and a α-glycosidase respectively, further degrade periplasmic oligosaccharides imported by SusC (18).

SusD is a starch-binding protein with a novel α-helical fold, unlike any other carbohydrate binding module (CBM) studied to date (17). SusD preferentially binds helical amylose as suggested by the atomic structure of SusD complexed with α-cyclodextrin and the imposition of a similar geometry on bound maltoheptaose. Isothermal titration calorimetry confirmed that SusD binds cyclic oligosaccharides with higher affinity than linear forms, suggesting that recognition is dominated by the three-dimensional conformation of oligosaccharide ligands, rather than specific interactions with the composite sugars. Finally, a deletion mutant that only lacked production of SusD was unable to grow on starch molecules composed of ≥ 6 glucose units, indicating an essential role for this protein in starch uptake. Based on recent analysis of SusD and the notable expansion of SusD homologs among Bacteroides spp. (8), it is likely that other members of this protein family have similar roles in glycan recognition and uptake.

BT1043 from Bacteroides thetaiotaomicron is believed to play a role similar to SusD in a PUL that likely targets mucin O-glycans. Its expression is dramatically upregulated in vivo, ~190-fold, in mice consuming glycan-deficient diets (12, 14), and in vitro, ~101-fold, in host glycan fractions enriched for mucin O-glycans (13). In contrast to SusD, that recognizes the α-1,4-glucosidic linkages contained in starch, it is hypothesized that BT1043 recognizes linkages that are specific to mucin O-glycans. In order to elucidate the molecular determinants of host glycan recognition by BT1043, its atomic structure in the presence and absence of N-acetyl lactosamine (LacNAc), a common disaccharide component of host glycans. While BT1043 does not share any direct sequence homology to SusD, it has a remarkably similar protein fold including four conserved tetratrico peptide repeats (TPRs). The TPR units provide a scaffold for the rest of the protein structure, and may mediate protein-protein interactions with other members of the outermembrane glycan-binding complex. Extensive hydrogen-bonding interactions with the reducing end of the β-N-acetylglucosamine moiety of LacNAc suggest that BT1043 recognizes free mucin O-glycans, although its cognate glycan remains to be determined. The glycan-binding pocket in BT1043 has a homologous shape and location as that found in SusD. However, while the amylose-binding site of SusD presents a relatively smooth, shallow cleft on the protein surface for interactions with the glycan backbone, the binding pocket of BT1043 is more articulated and likely makes more specific interactions with the composite sugar moieties themselves. To assess the generality of these findings, the structures of SusD and BT1043 are compared to the crystal structure of another putative mucin O-glycan-binding SusD homolog, BT3984. Despite diversity at the amino acid sequence level, a likely consequence early divergence from a common ancestor, SusD-like proteins have a conserved α-helical fold, suggesting a homologous function in glycan uptake.

MATERIALS AND METHODS

Heterologous protein expression

The BT1043 gene (residues 18 – 546) was amplified by PCR from genomic DNA prepared from Bacteroides thetaiotaomicron ATCC 29148 (also known as VPI-5482). The amplicon was cloned into pET28rTEV where the thrombin cleavage site of pET-28a (Novagen) has been modified to a tobacco etch virus (TEV) protease cleavage site. pET28rTEV-bt1043 was transformed into Rosetta (DE3) pLysS (Novagen) for protein expression. Cells were grown in TB medium at 37°C with shaking (225 rpm) until they reached an O.D. of ~0.4, at which time the temperature was adjusted to 22°C. Once the cultures reached an O.D. of ~0.8, cells were treated with 0.2 mM ITPG to induce BT1043 expression, and allowed to grow for 16 h at 22°C. Cells were subsequently harvested by centrifugation, frozen in liquid nitrogen, and stored at -80°C. Selenomethionine-substituted protein was produced via the methionine inhibitory pathway (21), as previously described (22).

Purification of native and selenomethionine-substituted BT1043

Both native and selenomethionine-substituted BT1043 were purified using a 5 ml Hi-Trap metal affinity cartridge according to the manufacturer’s recommendations (GE Healthcare). The cell lysate was loaded onto the column in His Buffer (25 mM NaH₂PO₄, 300 mM NaCl, 10 mM imidazole pH 8.0) and BT1043 was eluted using an imidazole (10 – 300 mM) gradient. The His-tag was removed by incubation with rTEV (1:100 molar ratio relative to BT1043) at room temperature for 16 h. The cleaved protein was then dialyzed against His Buffer and passed over the affinity column to remove the His-tagged rTEV and undigested BT1043. Purified BT1043 was dialyzed against 20 mM HEPES/100 mM NaCl (pH 7.0) and concentrated to ~9.5 mg/ml for crystallization.

Crystallization and Data Collection

Initial crystallization conditions for apo-BT1043 were determined via hanging drop using the Hampton Screen (Hampton Research). Large single crystals of SeMet BT1043 were grown at 4°C in batch plates by seeding small crystals into mother liquor that contained ~5 mg/ml BT1043, 6-7% polyethylene glycol 6000, 50 mM NaCl, and 50 mM 2-(cyclohexylamino)-ethanesulfonic acid (CHES), pH 9.0. Crystals were of the space group P2₁2₁2₁ with unit cell dimensions of a=60.28 Å, b=102.23 Å, c=175.75 Å, and contained two molecules in the asymmetric unit. Crystals of native BT1043 complexed with N-acetyl lactosamine were grown at room temperature via hanging drop. BT1043 (OD₂₈₀ ~25) was mixed with solid N-acetyl lactosamine (LacNAc) to a final concentration of 120 mM, and equilibrated against mother liquor containing 12-14% poly(ethylene) glycol 4000, 50 mM Na/K phosphate, and 100 mM succinate pH 5.0. Crystals of BT1043 with LacNAc were cubic P2₁3, with a = b = c = 159.06 Å, and contained two molecules per asymmetric unit.

Both crystal forms were serially transferred to final cryoprotectant solutions containing 20% ethylene glycol and 15-21% of the appropriate PEG in synthetic mother liquor prior to flash freezing with liquid nitrogen. SeMet BT1043 diffraction maxima were collected on a 3 × 3 tiled “SBC3” CCD detector at the Structural Biology Center 19-BM beamline, and the BT1043/LacNAc data were collected similarly at the Structural Biology Center 19-ID beamline (Advanced Photon Source, Argonne National Laboratory, Argonne, IL). X-ray data were processed with HKL3000 and scaled with SCALEPACK (23). Data collection statistics are shown in Table 1.

Table 1. Data Collection Statistics.

Values in parentheses represent the highest resolution shells.

	Peak	Inflection	Remote	N-acetyl lactosamine
Space Group	P212121			P213
Unit Cell (Å)	a = 60.28, b = 102.23, c = 175.75			a = b = c = 155.99
Wavelength (Å)	0.97934	0.97951	0.97167	0.97167
Resolution (Å)	50.0 – 1.96 (2.03 – 1.96)	50.0 – 1.96 (2.03 – 1.96)	50.0 – 1.94 (2.01 – 1.94)	50.0 - 2.75 (2.80 – 2.75)
# independent Reflections	77053 (7003)	77151 (6885)	78668 (7058)	32650 (1478)
Completeness (%)	97.3 (90.1)	97.1 (88.3)	96.9 (88.1)	98.4 (90.6)
Redundancy	4.1 (3.7)	4.1 (3.6)	4.1 (3.5)	8.3 (4.2)
Avg I/Avg σ(I)	35.4 (14.8)	35.0 (12.7)	32.2 (8.5)	20.5 (3.4)
Rsym (%)	8.6 (12.7)	6.8 (11.4)	6.2 (13.9)	7.8 (25.2)

X-ray structure determination

The structure of BT1043 was solved via MAD phasing from the x-ray data collected using the selenomethionine-substituted crystals. The program SOLVE (24) was used to determine and refine the initial positions of the selenomethionines, and RESOLVE (25) was then applied for solvent flattening and initial model building. Alternate cycles of manual model building in O (26), with maximum-likelihood refinement with CNS (27), was used to build and refine the 2.0 Å selenomethionine-substituted BT1043 (R_work =17.1%, R_free= 20.7%). The structure of BT1043/LacNAc was determined via molecular replacement using the program AMoRe (28) from the CCP4 suite of programs (29) with the SeMet BT1043 structure as a search model. Alternate cycles of manual model building in O and refinement using CNS were combined to complete the model. Relevant refinement statistics are presented in Table 2. Ramachandran plot analysis using PROCHECK revealed that the selenomethionine substituted 2.0Å apo structure of BT1043 contained 90.9% of residues in the most favored regions, 8.5% in the additionally allowed regions, 0.3% in the generously allowed regions, and 0.2% in the disallowed regions (30). In chains A and B, T59 assumes a ϕ,ψ angle of 40,-117 as part of a turn between a 3₁₀ helix and α2, which lines part of the glycan-binding pocket. The electron density in this region is unambiguous. The 2.8Å structure of BT1043 with N-acetyl lactosamine contained 84.4% of residues in the most favored regions, 15% in the additionally allowed regions, and 0.6% in the generously allowed regions with no residues in the disallowed regions. In the structure of BT1043 with N-acetyl lactosamine, T59 is in close proximity to the ligand, and the ϕ, ψ angles of this loop are slightly altered as a result of glycan binding.

Table 2. Refinement Statistics.

Values in parentheses represent the highest resolution shells

	Apo, Semet	N-acetyl lactosamine
PDB code	3EHM	3EHN
Resolution	50.0 – 2.0 2.01 – 2.0	50.0-2.80 2.82-2.80
# protein atoms	8065	8106
# hetero-atoms	670	178
Rwork	17.1 (18.2)	19.6 (28.7)
Rfree	20.7 (21.8)	24.6 (35.6)
# reflections	65281 (1226)	27651 (3080)
Avg B values (Å2)
Protein Atoms	18.9	35.6
Ligand	N/A	45.4
Solvent	25.445	22.1
RMS deviations
Bond Length	0.0052	0.0069
Bond Angles	1.26	1.33

Isothermal Titration Calorimetry

ITC measurements were carried out using a MicroCal VP-ITC titration calorimeter (MicroCal, Inc.). BT1043 was dialyzed overnight against a solution containing 20 mM HEPES (pH 7.0), 100 mM NaCl, prior to the experiment. A solution of N-acetyl lactosamine (Sigma-Aldrich) was prepared using dialysis buffer. BT1043 concentrated to 0.247 mM was placed in the reaction cell and the reference cell was filled with deionized water. After the temperature was equilibrated to 25°C, 25 successive 10 μl injections of 20 mM N-acetyl lactosamine were made while stirring at 460 rpm and the resulting heat of reaction was measured. The isotherm observed suggested weak binding, with a Kd of ~12 mM for n=1 binding sites.

Glycan Array Screening

BT1043 was fluorescently labeled using the Alexa Fluor 488 Protein labeling kit (Molecular Probes, Inc.) according to the manufacturers’ instructions. The protein was then sent to the Consortium for Functional Glycomics Protein-Glycan Interaction Core (H) facility at Emory University for glycan array screening (www.functionalglycomics.org). Glycan binding was assessed on a printed array (version 3.1) of 377 mammalian glycans using labeled BT1043 at 0.2mg/ml. No significant binding was detected above background. Because binding may have been affected by the fluorophore that reacts with exposed lysine residues, screening was attempted a second time using His-tagged BT1043 (0.2mg/ml) with fluorescently-labeled anti-His antibodies, provided by the facility. Again, glycan binding was not detected. Failure to detect binding may have been due to the absence of ligands specific for BT1043 or Kd values greater than the limit of detection (high micromolar to low millimolar).

RESULTS

The x-ray crystal structure of the soluble portion of BT1043 (residues 18 – 546) was determined using MAD phasing methods with selenomethionine-substituted protein crystals. The resulting 2.0Å structure (R_work = 17.1%, R_free = 20.7%) was refined using the remote wavelength data. BT1043 is a monomer that has an α-helical fold consisting of 18 α-helices, nine 3₁₀ helices, three two-stranded anti-parallel β-sheets, and multiple reverse turns (Figure 1A). Most striking is the presence of four tetratrico peptide repeats (TPR), each consisting of a helix-turn-helix (Figure 1). These helix-turn-helix pairs are defined by α1 (residues 39-50) and α4 (residues 99-125) as TPR1, α5 (130-154) and α6 (residues 174-193) as TPR2, α7 (residues 211-230) and α8 (residues 234-245) as TPR3, and α12 (residues 358 – 371) and α13 (378-393) as TPR4. Together these TPR units create a right-handed superhelix along one side of the protein, and cradle the rest of the structure, with α15 (443–459) and α16 (463–473) packing against these TPR units. BT1043 shares significant structural homology with SusD as well as BT3984 from B. thetaiotaomicron (pdb 3CGH); a predicted mucin glycan-binding SusD homolog (13). Each of these proteins contains 4 TPR units in the same relative positions and creating a right-handed superhelix along one side of the structure (Figure 1B). Beyond the TPR domain these structures deviate substantially, although they are nearly entirely α-helical. BT3984 can be superimposed onto BT1043 with an RMSD of 1.37Å over 373 Cα (34% sequence identity by structural alignment over 373 Cα). Interestingly, SusD can be superimposed onto BT1043 with an RMSD of 1.78Å over 207 Cα atoms (14.5% sequence identity by structural alignment over 207 Cα) even though BLASTP failed to detect any amino acid sequence homology between BT1043 and SusD. This suggests that proteins in the SusD family may display a wide range of sequence variability despite having similar three-dimensional folds.

Figure 1. The 2.0 Å structure of selenomethionine-substituted BT1043.

(A) Cartoon representation of the apo BT1043, colored blue to red from N- to C-terminus. The N-acetyl lactosamine ligand (magenta spheres) has been modeled in to show the location of the glycan-binding pocket. The 18 α-helices and 6 β-strands are labeled as shown; 3₁₀ helices are not labeled. Four helix-turn-helix pairs, or tetratrico peptide repeats (TPRs), are labeled with α1 & α4 as TPR1, α5 & α6 as TPR2, α7 & α8 as TPR3, and α12 & α13 as TPR4. (B) Overlay of BT1043, shown in blue, SusD (PDB 3CK7), shown in green, and BT3984 (PDB 3CGH) shown in red, in a similar view as in panel A to highlight to similarities in the TPR domain.

Like SusD, BT1043 was not predicted to have a TPR domain since it lacks the traditional TPR consensus sequence (W4-L7-G8-Y11-A20-F24-A27-P32) (31). TPR domains are typically involved in mediating protein-protein interactions. Most often, TPR containing proteins recognize a peptide from another protein via the concave interior surface created by the right-handed superhelical twist of several TPR units. In BT1043, two antiparallel β-strands, β2 (residues 260-262) and β6 (residues 353-365) are recognized in this concave area via hydrophobic interactions with the N-terminal halves of α5 and α7, while α16 makes mostly hydrogen-bonding and electrostatic interactions with the C-terminal portion of these same α-helices. In a similar manner, α15 has some hydrophobic and some hydrogen-bonding interactions with α12 and α13 of TPR4. Thus, in the structure of BT1043 as well as other SusD family proteins, this concave surface acts as a scaffold by cradling the rest of the protein structure. It was speculated that SusD homologs interact with their predicted TonB-dependent porins via these conserved TPR units, perhaps via the solvent exposed loops connecting adjacent helices and TPR units. This edge-on mode of binding between a TPR domain and target protein was observed between the TPR domain of p67^phox, and Rac, two subunits of NAPDH oxidase (32). In this complex, two loops that connect TPR units 1 and 3 in p^67phox create the binding surface for Rac-GTP, resulting in recognition at the edge of the TPR bundle rather than the concave surface. While the crystal structures of some TPR proteins display large movements of the TPR helices and/or disruption of individual TPR bundles when crystallized under difference conditions (31, 33), the TPR domains of SusD and BT1043 are invariant even when the proteins are crystallized in different space groups. The stability of the TPR arrangement in the SusD family of proteins is likely a result of the multiple interactions these units make with the rest of the protein, resulting in much less flexibility than seen in other TPR-containing proteins.

Previous transcriptional analysis revealed that the PUL operon encoding BT1043 (BT1042-45) is upregulated ~101-fold during growth in vitro on porcine mucin glycan fractions enriched for mucin O-glycans, as opposed to glycosaminoglycans or N-glycans (13). This pool of glycans is estimated to have as many as 79 different oligosaccharides species of varying lengths (34). To identify the repertoire of oligosaccharides recognized by BT1043, glycan binding was examined by the Consortium for Functional Glycomics Protein-Glycan Interaction Core (H) facility at Emory University. Glycan binding to 377 natural and synthetic mammalian glycans on a printed array (version 3.1) was assessed using both fluorescently-labeled BT1043 and with fluorescently-labeled antibodies to the N-terminal His-tag of BT1043. Neither attempt identified a potential glycan. This was not altogether surprising since even SusD had a weak binding affinity for even a closely related analogue of its natural ligand. In order to define the structural elements involved in glycan recognition, the atomic structure of BT1043 was determined in the presence of N-acetyl lactosamine (LacNAc), a common disaccharide component (β-d-galactosyl-1,4-β-d-N-acetylglucosamine) of mucin O-glycans that could be obtained in reasonable quantities for biochemical studies (35). Transcriptional analysis indicates that the BT1042-1045 locus of B. thetatiotaomicron is upregulated ~4 fold in response to LacNAc, suggesting that LacNAc may represent a part of a larger glycan more readily recognized by BT1043 (13). In addition, isothermal titration calorimetry (ITC) revealed that BT1043 binds LacNAc with a Kd ~12 mM (data not shown).

While LacNAc may not be an ideal analogue of BT1043’s cognate glycan, the results with SusD lend support to using these small oligosaccharides to identify at least some of the key elements involved in recognition of the larger glycans (17). ITC studies demonstrated that SusD binds linear maltoheptaose with a Kd ~1 mM, but binds the circular analogues α-, β-, and γ-cyclodextrin with a Kds of 0.089, 0.065 and 0.146 mM respectively. Co-crystal structures of the SusD with various maltooligosaccharides revealed that the mode of binding was the same in each case. Indeed, even though ITC failed to detect binding of maltotriose to SusD, it was possible to co-crystallized maltotriose complexed with SusD and this small oligosaccharide bound in an identical manner as the trisaccharide component of the larger oligosaccharides. A similar binding trend is emerging on work with the galactomannanan/glucomannan-binding SusD homolog from Bacteroides ovatus (unpublished results).

There is an interesting parallel between this family of glycan binding proteins and the capsid proteins of the Calicviruses. Norwalk virus is able to bind both A and H blood group antigens and this interaction is thought to greatly enhance subsequent interactions with the viral receptor. The structure of a portion of the capsid protein complexed with both polysaccharides was made only possible by the addition of very large amounts of glycans during crystallization (36), as was done in these studies. They found that while the H blood group polysaccharide interacts with the terminal α-fucose, β-galactose moieties, the A type blood group antigen interacts only via the terminal α-GalNAc moiety. Nevertheless, the protein residues involved in hydrogen binding and van der Waal contacts were common to both glycans. Therefore, the protein contacts observed in the BT1043/LacNAc complex are more than likely to be biologically relevant, but it also seems likely that, in-vivo, additional contacts are made with longer polysaccharides. Also, as was proposed with SusD (17), the authors suggested that the relatively weak protein/glycan interactions might be overcome by multivalent interactions.

In order to ensure saturation of BT1043 with ligand, ~120 mM LacNAc was added to BT1043 during crystallization and freezing; lower amounts of ligand were not attempted. The 2.8Å structure of BT1043 with LacNAc (R_work = 19.6%, R_free = 24.6%) revealed one molecule of LacNAc clearly bound in the glycan-binding site predicted by the structure of SusD (Figure 2A). The apo and LacNAc-bound structures of BT1043 can be superimposed with an RMSD of 0.4Å, revealing virtually no structural changes upon glycan binding. The shallow glycan binding pocket is shaped by α2 and α9, in close proximity though not interacting with residues at the N-terminus of α4 that comprises TPR1. LacNAc is anchored to the binding pocket through interactions at the reducing end of β-N-acetylglucosamine (GlcNAc) (Figure 2B). The O1 of the GlcNAc anomeric carbon is positioned within ~3.0 Å and 2.5 Å of the side chain carboxylic oxygens of E283, and 2.7 Å from the N^ε1 atom of W87. Similarly, the acetyl O atom of GlcNAc is positioned 3.1 Å from the N^ε1 atom of W87 and 3.5 Å from the carboxamide N of N91. The amide N of GlcNAc points towards the side chain carboxamide oxygen of Q67, 3.3 Å away, and the hexose ring is stacked against the phenolic side chain of Y281 at a distance of ~4Å. The galactose moiety does not have any significant interactions with the protein, with a single hydrogen bond between O4 and the carboxamide O of N64. The extensive interactions between GlcNAc and BT1043 suggest that the cognate ligand for BT1043 contains GlcNAc, although it may be preceded by one or more different monosaccharides and/or by different linkages (i.e. β1-3 versus β1-4). It is likely that the cognate ligand of BT1043 is much longer than a disaccharide with sugars towards the non-reducing end anchoring it to the protein since the porcine O-glycan fraction that stimulated the BT1042-1045 locus by 101-fold was comprised of longer (9-14 sugars) oligosaccharides (34). In addition, the glycan-binding pocket of BT1043 contains a number of possible hydrogen-bonding acceptors and donors at the reducing end of LacNAc that may play a role in binding a longer glycan. Since SusD prefers maltooligosaccharides ≥6 glucose units that allows the ligand to adopt a helical conformation, it is likely that BT1043 prefers a somewhat longer oligosaccharide with an overall shape that can better complement the binding pocket. The fact that N-acetylglucosamine is anchored to BT1043 via its reducing end suggests that the protein recognizes O-glycans liberated from the mucin protein, as opposed to those attached to the polypeptide and therefore lacking an exposed reducing sugar (37).

Figure 2. N-acetyl lactosamine (LacNAc) binding in the 2.8Å structure of BT1043.

(A) Omit map (3σ) of native BT1043 complexed with LacNAc. The reducing O1 oxygen of N-acetylglucosamine (GlcNAc) and the non-reducing O4 oxygen of galactose are labeled, along with the α-helices that shape the glycan-binding site. (B.) Close-up view of LacNAc binding to BT1043. Coordinating and nearby residues are displayed, with distances labeled in angstroms (Å).

Based upon on the structural alignment of BT1043 and SusD, the general features of the glycan-binding sites of BT1043 and SusD were compared (Figure 3A). The SusD glycan binding site has similar size and shape. However, the residues lining the glycan binding pocket are quite different. The SusD binding pocket is unique in that the hydrophobic residues that dominate starch binding (W98, W320 and Y296) form an arc that complements the shape of helical α-amylose. Shorter to medium length oligosaccharides, including maltoheptaose, bind to SusD with much less affinity than the cyclodextrins that have a fixed, curved geometry (38). Since BT1043 binds to different oligosaccharides, it is not surprising that BT1043 does not have this arc of hydrophobic residues. The most conservative difference between the two proteins is the substitution of Y281 in BT1043 for W320 in SusD. The conservation of R118 in SusD with R101 in BT1043 and W121 in SusD with Y104 in BT1043 is interesting (Figure 3A), although neither of these residues was observed to interact with glycans in the various SusD complexes (38).

Figure 3. Overlay of the SusD and BT1043 glycan-binding pockets.

(A). Close-up view of the amino acids lining the glycan binding pocket of BT1043 (blue) with that of SusD (pink). Starch-binding residues in SusD are overlayed with the equivalent residues in BT1043. The ribbon representation of the glycan binding pocket is shown to highlight the similarities in the overall shape and size of site, despite differences in amino acid sequence. (B) Close-up view of the BT1043 (blue-purple) LacNAc binding site with the equivalent glycan-binding site in BT3984 (yellow-orange).

The structure of the SusD homolog BT3984 (PDB accession ID, 3CGH) from B. thetaiotaomicron was recently deposited by the Joint Center for Structural Genomics (JCSG). Like BT1043, BT3984 is also highly upregulated in the mouse distal gut when the bacterial food source is limited to host mucin glycans (13). Interestingly, expression of these two proteins is stimulated differently during in vitro growth on host glycan fractions that are enriched for O-glycans, suggesting that they recognize and respond to different host glycan cues. Unlike the locus containing BT1043, the PUL carrying BT3984 (BT3983-88) is slightly upregulated (~10-fold) during growth on mucin core 1. A comparison of the BT1043 glycan-binding site with the equivalent site in BT3984 shows that nearly half of the residues lining this shallow pocket are conserved, including Y281 and Q67 (BT1043 numbering) that help coordinate GlcNAc in BT1043 (Figure 3B). In BT3984, F85 replaces W87 in BT1043 eliminating the hydrogen-bonding donor provided by the indole N atom. In BT1043, LacNAc binds on one side of the glycan binding pocket, and is not within reasonable hydrogen-bonding or hydrophobic stacking distance with R101, N68, or Y104. Interestingly, it is this side of the binding pocket that shows the most variability between BT1043 and BT3984, with R101, N68 and Y104 in BT1043 being replaced by W98, F65 and V101 in BT3984, respectively (Figure 3B). While the significance of these differences is unknown, it is certainly a possibility that these residues aid in binding branched versus linear O-glycan structures (37).

Although there are similarities in the overall fold and glycan-binding pocket architecture, electrostatic surface renderings of SusD, BT1043 and BT3984 highlight less obvious yet important differences among these proteins (Figure 4). For binding maltooligosaccharides, the SusD glycan binding pocket features a shallow, flatter surface, comprised of an arc of aromatic residues that complements the shape of helical starch. Thus, in SusD, glycan recognition appears to rely more on the shape (e.g. van der Waal interactions) of the substrate, and less upon electrostatic or hydrogen-bonding interactions between the protein and glucose residues. In contrast, the binding pockets of BT1043 and BT3984 are more articulated and dominated by polar residues that can provide a network of hydrogen-bonding interactions. This may suggest that oligosaccharide recognition by O-glycan targeting SusD family members is dominated by electrostatic and hydrogen-bonding interactions with individual monosaccharide components, as opposed to interacting mainly with the backbone the glycan.

Figure 4. Comparison of glycan binding pockets of BT1043, BT3984, and SusD.

Shown here are stereo pairs of surface representations generated in the program CHIMERA (46) of the glycan binding surfaces of these three polysaccaride binding proteins colored according to the electrostatic potential (red for negative and blue for positive) calculated in the program GRASP (47) using the DelPhi (48) algorithm. The ball and stick figure in the BT1043 model represents the structure of the bound LacNAc and α-cyclodextrin in the case of the SusD surface representation.

DISCUSSION

The atomic structures of the three SusD family proteins determined to date (BT1043, SusD and BT3984) display a unique α-helical fold. By comparison, the carbohydrate-binding modules (CBMs) associated with cellulases and other glycoside hydrolases, and lectin proteins are dominated by β-sheets (39). The most conserved feature of the SusD family is the presence of four TPR units that form a right-handed superhelix along one side of the protein and provide a scaffold upon which the rest of the α-helical structure is packed. BLAST searches with SusD, BT3984, and BT1043 amino acid sequences failed to detect the presence of the TPR domain since they all lack the archetypal consensus sequence; W4-L7-G8-Y11-A20-F24-A27-P32 (31). Proteins containing TPR units are typically involved in protein-protein interactions (31), and it is possible that the conserved TPR units of the SusD family are required for associating with the other members of the extracellular protein complex, particularly SusC, a predicted TonB-dependent β-barrel porin that is a defining feature of the Sus-like PULs. Previous studies demonstrated that both SusC and SusD are required for starch binding to the cell surface. Further, since they were found to co-purify, they likely form a functional oligomer that recognizes starch (19, 20). The most variable part of the SusD family structures contains a series of loops and α-helices that lie opposite the TPR superhelix (Figure 1B). The biological significance is unknown, but could have implications for the way different SusD family proteins interact with other accessory proteins or glycolytic enzymes in a potential complex. Note that while a few residues from these variable regions span one side of the glycan binding pockets in these proteins, the size and shape of this site is relatively conserved in all three structures. In the crystal structure of SusD complexed with α-cyclodextrin, two molecules of SusD were wrapped around the oligosaccharide, suggesting that tighter binding of the glycan may be mediated through avidity effects (17). Based on the close structural similarity of BT1043 and BT3984 with SusD, it is possible that these proteins also associate with their respective SusC homologs to create multivalent, high affinity binding sites for mucin glycans.

All gut Bacteroides PULs defined to date contain susC/susD homologs that are usually found in combination with one or more genes encoding glycolytic enzymes. These clusters are often linked to genes encoding extracytoplasmic function sigma factors (ECF-σ)/anti-σ factor pairs, hybrid two component systems or other types of sensor/regulators (8, 13). BT1043 was initially identified as a SusD homolog based upon an iterative BLAST search within the B. thetaiotaomicron genome, using each low-scoring hit as a query sequence to identify more divergent SusC/SusD paralogs. Genes included as susD homologs in this search also had to meet the criteria of inclusion in a gene pair downstream of an independently identified susC homolog (8). The striking expansion of genes related to susC and susD in the genomes of human gut Bacteroidetes is likely due, at least in part, to gene duplication and diversification events (8). Thus, the variation in both amino acid sequence and substrates specificity of these factors likely represents a case of divergent evolution in which, over time, these PULs have diversified and evolved towards different glycan specificities. BT1043 is one of the most divergent of SusD homologs, sharing only 6.7% sequence identity with SusD over all Cα atoms (based upon the structural alignment of the TPR domain), nearly that expected for two random sequences (~5.6%), whereas most structural homologues share at least 8-9% sequence identity (40). This implies that SusD and BT1043 (and BT3984) diverged very early in the evolution of the Bacteroides PULs.

To identify the repertoire of glycans that BT1043 can recognize, the protein was examined using the glycan array screening facility at the Consortium for Functional Glycomics Protein-Glycan Interaction Core (H) facility at Emory University. BT1043 was screened both by covalent labeling of the protein with a fluorescent tag, and indirectly using fluorescently-labeled antibodies to the N-terminal His-tag of BT1043. Neither attempt identified a potential cognate glycan. However, this was not altogether surprising in light of previous analysis of SusD ligand binding (17). Maltooligosaccharide binding to SusD is dominated by the three dimensional conformation of maltooligosaccharides and glycan backbone interactions rather than specific interactions with the composite glucose residues. Isothermal titration calorimetry confirmed that a circularized starch analog such as α-, β- and γ-cyclodextrin bound with nearly 20-fold greater affinity than similarly-sized linear sugars even though the crystal structures suggested a similar mode of binding with little or no change in the number of hydrogen-bonding or hydrophobic stacking interactions. Furthermore, the ϕ,ψ angles of maltotriose or maltoheptaose bound to SusD closely approached those of double helical amylose, suggesting that SusD recognizes the native conformation of the polysaccharide. Therefore it is possible that glycan recognition within the SusD family of proteins is driven by the native structure of the polysaccharide that is imposed by the linkages between composite sugars and not simply the stereochemistry of individual sugars. If this is the case, ligand binding to BT1043 or BT3984 will be dependent not only on the monosaccharide composition, but also on the context of these sugars in terms of their linkages to each other and the overall size/shape of the glycan. Isothermal titration calorimetry experiments suggest that N-acetyl lactosamine binds to BT1043 weakly with a Kd ~12 mM. Consistent with this observation, the BT1042-45 locus is upregulated ~4 fold in response to N-acetyl lactosamine (13). The structure of BT1043 complexed with LacNAc demonstrates that the GlcNAc moiety is anchored by hydrogen-bonding interactions at its reducing end, and by aromatic stacking with the phenolic side chain of Y281. While it is likely that the cognate ligand of BT1043 contains GlcNAc, it is difficult to predict the preceding glycan linkages and monosaccharides that comprise the full length oligosaccharide since mucin O-glycans can be quite variable in both length and composition. In addition, it is difficult to make predictions about glycan binding since BT1043 is markedly different than other CBMs or lectin structures.

It is quite possible that some structural rearrangement occurs in the glycan-binding pocket upon longer oligosaccharide binding to BT1043. In the structure of SusD with maltoheptaose, two small loops (residues 70-77 and 292-296) undergo rearrangement allowing the ligand to bind, whereas maltotriose, bound in an identical position as three of the glucose residues of maltoheptaose, does not require such a structural perturbation for binding (17). Even though the binding pockets of BT1043 and BT3984 do not appear to have a similarly positioned flexible loop, it is possible that some rearrangement occurs to accommodate the larger, cognate glycan.

The lack of interactions between BT1043 and the galactose of LacNAc suggest that either galactose is not the preceding monosaccharide in the cognate ligand, or that the glycan linkage is different. It is possible that a β1,3-linkage off of GlcNAc, as opposed to the β1,4 linkage present in LacNAc, may better span the glycan binding site and facilitate hydrogen-bonding interactions with R101, Y104 and N64 in addition to a potential hydrophobic stacking interaction with Y281 (Figure 2B). β1,3-linked glycans are quite common in mucin O-glycan core structures and therefore are a likely possibility in the cognate glycan for BT1043 (34, 37). If the linkage were β1,3, or perhaps any other linkage besides β1,4, then GalNAc may be a reasonable substitute for GlcNAc as the reducing sugar. In the complex of BT1043 with GlcNAc, the bridging oxygen of the β1,4 linkage is within 3.3 Å of the Y281. However, it is not in an ideal orientation for hydrogen bonding, and therefore may or may not play a role in dictating glycan specificity.

The hypothesis that BT1043 binds a mucin glycan containing a GlcNAc moiety is supported by the function of the other genes found in its PUL. In the majority of Sus-like PULs, one or more glycolytic enzymes contain N-terminal lipidation sites similar to that found in SusD and SusG, making it likely that they are also presented on the outer membrane. These enzymes are believed to begin degradation of the target polysaccharide for import into the cell via the associated TonB–dependent porin much as SusG cleaves large starch molecules into maltooligosaccharides. In the BT1042-45 operon, a component of a larger PUL, BT1044 encodes a predicted CAZy (Carbohydrate-Active enZYmes) glycoside hydrolase family 18 enzyme endo-β-N-acetylglucosaminidase that could cleave adjacent to N-acetyl lactosamine extensions, liberating reducing end N-acetylglucosamine-containing oligosaccharides from glycosylated mucin proteins (13, 41). Likewise, BT1045 encodes a predicted concanavalin A-like lectin/glucanase, that are sometimes found as part of multienzyme mucinase complexes (42). Both BT1044 and BT1045 contained predicted secretion and lipidation signal sequences for tethering to the outermembrane, suggesting that they may assist in processing mucin oligosaccharides prior to import via the SusC homolog, BT1042. Similarly, the BT3983-88 PUL, also encodes a predicted endo-β-N-acetylglucosaminidase (BT3987).

The mode of LacNAc binding to BT1043 is quite different from that observed by other LacNAc and/or O-glycan binding carbohydrate binding modules (CBMs). The crystal structures of the family 32 CBM from Clostridium perfringens N-acetyl-β-hexosaminidase with bound galactose, LacNAc and the type II blood group H-trisaccharide reveal that a series of hydrogen-bonds with the non-reducing end galactose moiety dominate the protein-carbohydrate interface (43). Indeed, recognition of the non-reducing end of galactose was also observed in the crystal structures of the CBM 32 domain from the Micromonospora viridifaciens sialidase (44), and in the CBM 32 domain of the C. perfringens NanJ sialidase (45) with bound galactose. The function of these CBMs is likely to bind the non-reducing terminal end of mucin glycans, and thus allowing attachment to glycans covering the surface of the mucin protein, while BT1043 seems to recognize the reducing end N-acetylglucosamine moiety of mucin O-glycans, suggesting that BT1043 binds free mucin glycans thereby participating more in foraging rather than tissue adhesion.

Mucin glycan foraging by Bacteroides thetaiotaomicron is a significant contributing factor to this species adaptation to, and persistence in, the distal gut (13). Whole genome transcriptional profiling of B. thetaiotaomicron in the intestines of gnotobiotic mice revealed at least 12 Sus-like PULS that are upregulated 83- to 488-fold under conditions when the bacteria are deprived of dietary polysaccharides and thus turn to host glycans as an alternative carbon source (13). Similarly, these and other PULS are also upregulated to varying degrees during growth on glycan fractions that are enriched for mucin O-glycans. In a B. thetaiotamicron mutant with reduced ability to utilize several of its O-glycan targeting PULs, including those harboring BT1043 and BT3984, both persistence in the mouse intestines and mother-to-pup transmission is diminished, demonstrating the importance of this nutrient niche in vivo (13). The bacterium likely devotes multiple PULs towards mucin O-glycan utilization because of the complexity and structural diversity of glycans in this class. Likewise, the abundance of mucin glycan PULs, with 12 systems highly upregulated in vivo, supports the necessity of this nutrient source to this species’ survival in the competitive gut environment.

The abbreviations used are

LacNAc

N-acetyl lactosamine

PUL

polysaccharide utilization loci