Molecular mechanism by which prominent human-gut Bacteroidetes utilize mixed-linkage beta-glucans, major health-promoting cereal polysaccharides

Summary

Microbial utilization of complex polysaccharides is a major driving force in shaping the composition of the human gut microbiota. There is a growing appreciation that finely tuned polysaccharide utilization loci enable ubiquitous gut Bacteroidetes to thrive on the plethora of complex polysaccharides that constitute “dietary fiber”. Mixed-linkage β(1,3)/β(1,4)-glucans (MLGs) are a key family of plant cell wall polysaccharides with recognized health benefits, but whose mechanism of utilization has remained unclear. Here, we provide molecular insight into the function of an archetypal MLG utilization locus (MLGUL) through a combination of biochemistry, enzymology, structural biology, and microbiology. Comparative genomics coupled with growth studies demonstrated further that syntenic MLGULs serve as genetic markers for MLG catabolism across commensal gut bacteria. In turn, we surveyed human gut metagenomes to reveal that MLGULs are ubiquitous in human populations globally, which underscores the importance of gut microbial metabolism of MLG as a common cereal polysaccharide.

eTOC Blurb

Mixed-linkage β(1,3)/β(1,4)-glucan (MLG) is an important complex dietary polysaccharide (dietary fiber), the degradation of which in the human gut depends on the resident microbiota. Tamura et al. outline the molecular mechanism of MLG utilization by Bacteroides ovatus and reveal that the majority of surveyed humans possess MLG-utilizing Bacteroidetes.

Introduction

The composition and homeostasis of the human gut microbiota has a profound and intimate connection to various aspects of our physiology, health and wellbeing (Littman and Pamer, 2011). Indeed, a multitude of diseases such as type 2 diabetes, inflammatory bowel diseases (IBDs), and cancer have been linked to alterations in the population and proportion of microbes in this highly complex and dynamic ecosystem that exists in our large intestine (Biedermann and Rogler, 2015; Fujimura et al., 2010; Kau et al., 2011; Schwabe and Jobin, 2013). The molecular mechanisms by which the microbiota exerts influence on human health are largely unresolved and undoubtedly complex, yet may hold the key to personalized medicine through therapeutics that target the gut microbial ecosystem (Blanton et al., 2016; Haak et al., 2017; Kootte et al., 2012; Subramanian et al., 2015).

A major factor in shaping the composition and physiology of the gut microbiota is the influx of complex glycans – popularly known as “dietary fibre” – that evade degradation by the limited set of human genome-encoded glycoside hydrolases (Hamaker and Tuncil, 2014; El Kaoutari et al., 2013; Koropatkin et al., 2012). Indeed, regular ingestion of plant polysaccharides is integral to maintaining a healthy balance of microbes in our lower gastrointestinal tract (De Filippo et al., 2010; Sonnenburg and Sonnenburg, 2014). Members of the Bacteroidetes, a dominant phylum in the human gut, possess an arsenal of Polysaccharide Utilization Loci (PUL) to target a wide range of complex glycans (El Kaoutari et al., 2013). Analogous to the archetypal Bacteroides thetaiotaomicron starch utilization system (Sus), a hallmark of all Bacteroidetes PULs is the organization of genes clustered around tandem susC/susD homologs (encoding a TonB dependent transporter, TBDT, and a cell-surface glycan-binding protein, SGBP, respectively) (Martens et al., 2009). Additional co-localized and co-regulated SGBP(s), a cohort of enzymes, and a transcriptional regulator typically make up a machinery that acts in concert to sense, break down, and import complex glycans (Grondin et al., 2017; Hemsworth et al., 2016). Many such PULs, each targeting specific glycan structures, have been identified by genomics and transcriptomics (see, e.g., the seminal study by (Martens et al., 2011), but detailed functional characterization lags severely behind (reviewed in (Grondin et al., 2017; Martens et al., 2014)). Developing a precise understanding of the molecular details of complex glycan utilization by individual members of the microbiota is essential to designing targeted therapies based on prebiotics, probiotics, and synbiotics (Ciorba, 2012; Slavin, 2013), as well as novel therapeutic interventions.

Recently, comprehensive functional analysis has revealed the detailed molecular mechanisms by which individual PULs enable human gut Bacteroidetes to utilize predominant plant polysaccharides, including the matrix glycans, xyloglucan (Hemsworth et al., 2016; Larsbrink et al., 2014; Tauzin et al., 2016), xylan (Rogowski et al., 2015), β-mannan (Bågenholm et al., 2017), and rhamnogalacturonan II (Ndeh et al., 2017). Mixed-linkage β(1,3)/β(1,4)-glucans (MLGs, Fig. 1A) from cereal grains constitute an additional key group of dietary glycans, whose utilization by gut microbes was previously unresolved at the molecular level. MLGs are abundant in the aleurone layer of common cereals, including oats (3–8 % dry weight) and barley (2–20 % dry weight) (El Khoury et al., 2012). Beyond their obvious potential to contribute to energy intake (Cummings and Macfarlane, 1997; McNeil, 1984), MLGs have been linked to a range of health benefits, e.g. promoting healthy cholesterol and blood glucose levels, ameliorating insulin resistance, and mitigating metabolic syndrome (El Khoury et al., 2012). In particular, the cholesterol lowering effect of oat MLG has long been recognized by the United States Food and Drug Administration (FDA) as well as the United Kingdom Joint Health Claims Initiative (JHCI), and been confirmed by subsequent studies (Othman et al., 2011).

Figure 1. Cereal mixed-linkage β(1,3)/β(1,4)-glucan (MLG) and MLG Utilization Locus (MLGUL) structures.

A: Chemical structure of MLG, consisting of a linear glucan chain of β(1,4)-linked cellotriosyl and cellotetraosyl units linked by β(1,3) bonds. MLGs from various sources (barley, oat, lichenin, etc) vary in the ratio of cellotriose to cellotetraose units (Lazaridou et al., 2004). Arrows indicate the specific site of hydrolysis by the vanguard endo-glucanase of the MLGUL, BoGH16_MLG. B: Genetic organization of the B. ovatus MLGUL and syntenic loci in select Bacteroidetes species. Homologous genes are connected by colored bars and the locus tag of the TBDT of each syntenic MLGUL is given on the right. See also Figure S1; Table S1.

The mechanisms behind these systemic benefits of MLG are, however, incompletely understood, in part due to a lack of understanding of MLG metabolism by individual members of the human gut microbiota. Thus, we report here the molecular characterization of a mixed-linkage glucan utilization locus (MLGUL) in the common symbiont B. ovatus. Identifying syntenic MLGUL in other Bacteroidetes revealed that as the archetype, this MLGUL serves as a molecular marker for MLG utilization across the Bacteroidetes phylum, thereby enabling future functional prediction across species.

Results

Identification of a multi-gene locus responsible for MLG utilization by B. ovatus

A putative MLGUL was previously identified in B. ovatus (Fig. 1B) based on the presence of a tandem susC/susD homolog signature (Martens et al., 2009) and high-level expression of select genes in the presence of bMLG (Martens et al., 2011). Individual genes in the locus, BACOVA_02741-02745, were all substantially upgregulated (125 to 298-fold) during growth on bMLG vs. glucose as sole carbon sources (Table S1). BACOVA_02742 and BACOVA_02743 encode the signature TBDT/SGBP pair with 28% and 19% protein sequence identity to SusC and SusD, respectively. The putative MLGUL was additionally predicted to encode a second, non-homologous SGBP (BACOVA_02744), a hybrid two-component sensor/transcriptional regulator (HTCS, BACOVA_02740), and up to three glycoside hydrolases (GHs).

BoGH16_MLG (BACOVA_02741) is a member of Glycoside Hydrolase Family 16 (GH16), a family of endo-β-glucanases in the Carbohydrate Active Enzymes (CAZy) classification (Cantarel et al., 2009). GH16 notably includes canonical bacterial MLG endo-glucanases (endo-MLGase) (Planas, 2000), along with a diversity of endo-glucanases and endo-galactanases (Eklof and Hehemann). BoGH3_MLG (BACOVA_02745) is classified into Glycoside Hydrolase Family 3 (GH3), whose members include exo-β-glucosidases (Fincher et al.). Notably, we have determined that BACOVA_02738, which is predicted to encode a second GH3 exo-β-glucosidase, is unlikely to be part of the MLGUL for three reasons: (1) BACOVA_02738 was not significantly upregulated on MLG (1.6-fold vs. glucose control, Table S1), (2) a corresponding gene is not found among syntenic loci (Fig. 1B), and (3) the encoded enzyme was catalytically feeble compared to BoGH3_MLG on β-glucosides relevant to MLG saccharification (vide infra).

To determine the correlation between the presence of the predicted MLGUL and growth of B. ovatus on MLG, we constructed an isogenic mutant of B. ovatus Δtdk (Larsbrink et al., 2014) in which a contiguous region of DNA encoding genes BACOVA_02738-02745 was deleted (B. ovatus ΔMLGUL). Vis-à-vis the parent strain, the B. ovatus ΔMLGUL was able to grow equally well on glucose as the sole carbon source, however the ability to grow on bMLG was completely abolished (Fig. S1). Thus, the putative MLGUL is necessary to confer B. ovatus the ability to utilize MLG.

Enzymology and structural biology of BoGH16_MLG, the vanguard MLGase

Cellular localization

The GH family membership of BoGH16_MLG suggested a potential role as the vanguard enzyme catalyzing polysaccharide backbone cleavage at the cell surface as the essential first step in MLG utilization. Indeed, the presence of a predicted Type II signal sequence (determined with LipoP 1.0 (Juncker and Willenbrock, 2003)) suggested that the protein is membrane-anchored via lipidation of the N-terminal cysteine residue (Paetzel et al., 2002). To validate this prediction, B. ovatus Δtdk was grown on minimal medium with either glucose or bMLG as a sole carbon source prior to immunolocalization of BoGH16_MLG. As shown in Fig. 2A, BoGH16_MLG was clearly visualized on the outer surface of cells in which the presence of the polysaccharide induced MLGUL expression, but was absent from cells grown on glucose (Fig. S2C, S2F). Further analysis of cellular fractions by Western blotting revealed the presence of BoGH16_MLG in the membrane fraction, corroborating its attachment to the outer membrane (Fig. 2C). Interestingly, BoGH16_MLG was also detected in the lysate supernatant (soluble periplasmic or cytoplasmic proteins) and in the culture supernatant (secreted protein) (Fig. 2C). While the former may represent anchored protein released into the soluble fraction during cell lysis, detection in the culture supernatant could result from packaging and release in outer membrane vesicles, which has previously been observed for other Bacteroidetes glycoside hydrolases (Elhenawy et al., 2014).

Figure 2. Enzyme localization analysis.

Phase contrast microscopy and corresponding fluorescence microscopy images of B. ovatus Δtdk cells grown in minimal medium with bMLG as the sole carbon source probed with custom polyclonal antibodies against recBoGH16_MLG (A) and recBoGH3_MLG (B). C: Western blots of protein collected from the culture supernatant, cell lysate supernatant, and cell lysate membrane fraction of B. ovatus Δtdk cells grown in minimal medium with glucose or bMLG as a sole carbon source. See also Figure S2.

Substrate and product specificity

To verify the leading catalytic role of BoGH16_MLG in MLG utilization and determine the specificity of the enzyme for individual β-glucans, recombinant BoGH16_MLG produced in E. coli (recBoGH16_MLG, Fig. S3A, S3B) was screened for hydrolytic activity against a library of polysaccharides. No activity was observed on tamarind xyloglucan, beechwood xylan, wheat arabinoxylan, carob galactomannan, konjac glucomannan, synthetic carboxymethylcellulose, synthetic hydroxyethylcellulose, Xanthomonas campestris xanthan gum, or Ulva sp. ulvan. In this initial screen, BoGH16_MLG was minimally active on all-β(1,3)-glucans, including Laminaria digitata laminarin, yeast β-glucan, and Alcaligenes faecalis curdlan, but exhibited high specific activity on bMLG. The optimum pH of 6.5 (consistent with function in the distal human colon) and maximum temperature range of 45 to 55 °C was determined using bMLG as substrate (data not shown).

Subsequent Michaelis-Menten kinetic analysis at the pH optimum and 37 °C demonstrated that BoGH16_MLG is a highly predominant mixed-linkage β(1,3)/β(1,4)-glucanase (MLGase), with a 33-fold higher specificity constant, k_cat/K_m, for bMLG (Fig. 1A) over laminarin, an all-β(1,3)-glucan with single β(1,6)-linked glucosyl branches (Fig. 3A, Table S2) (Martin et al., 2007). BoGH16_MLG was even less efficient on the other two all-β(1,3)-glucans for which activity was initially observed: The k_cat/K_m, was 147-fold higher for bMLG than yeast β-glucan (similar in structure to laminarin but with longer β(1,6)-linked glucose branches (Lowman et al., 2011)) and nearly four orders of magnitude higher than that of high curdlan, a 22 kDa, non-branched β(1,3)-glucan (Harada et al., 1968) (Fig. 3A, Table S2).

Figure 3. BoGH16_MLG kinetics and MLGUL GHs product analysis.

A: BoGH16_MLG initial-rate kinetics curves fitted to the Michaelis-Menten equation for β-glucan polysaccharide substrates on which it is active. Laminarin was reduced to laminaritol by sodium borohydride reduction to reduce background in the BCA assay. Curve fitting was done on OriginPro 2015 and error bars represent standard deviations from the mean. B: Chromatograms of bMLG and its hydrolysis products by BoGH16_MLG and BoGH3_MLG separated by HPAEC-PAD. Red: full length bMLG polysaccharide. Dark blue: reaction progress time course and limit digest of bMLG hydrolysis by 10 nM BoGH16_MLG. Cyan: reaction progress time course and limit digest of BoGH16_MLG products hydrolysis by 25 nM BoGH3_MLG. Standards are shown below in black: solid lines are those corresponding to limit digest products and dotted line to intermediate products. See also Figures S3, S4 and S5; Tables S2 and S3.

Detailed product analysis was employed to determine the mode of hydrolysis, endo vs. exo, and linkage specificity of recBoGH16_MLG to gain information on the nature of the MLG cleavage products at the B. ovatus cell surface. HPLC analysis at selected time points in the hydrolysis showed the initial production of very large oligosaccharide fragments, which were progressively converted into shorter species and ultimately to two distinct oligosaccharides in the limit-digest (Fig. 3B). This product evolution indicates that BoGH16_MLG operates through an endo-dissociative mode of action in which the MLG polysaccharide is stochastically cleaved along the backbone.

Comparison with oligosaccharide standards (Fig. 3B) and additional LC-MS analysis (data not shown) revealed that the limit-digest products were the mixed-linkage trisaccharide, G4G3G [Glcβ(1–4) Glcβ(1–3) Glc], and the mixed-linkage tetrasaccharide, G4G4G3G [Glcβ(1–4) Glcβ(1–4) Glcβ(1–3) Glc]. Thus, BoGH16_MLG specifically hydrolyzes β(1,4)-linkages of glycosyl residues that are immediately preceded by a β(1,3)-linked glucosyl residue (toward the non-reducing chain end). This specificity is typical of bacterial endo-MLGases within GH16 (Gaiser et al., 2006; Mcgregor et al., 2017; Planas, 2000).

To provide more refined insight into BoGH16_MLG substrate specificity, Michaelis-Menten kinetics were determined for a series of chromogenic glycosides (Fig. S4A, S4B; Table S3). recBoGH16_MLG had no activity on the ortho-chloro-para-nitrophenyl (CNP) β-glycosides of glucose (G-CNP), cellobiose (G4G-CNP), cellotriose (G4G4G-CNP), nor on para-nitrophenyl (pNP) β-glucoside (G-pNP). Weak activity was observed on the pNP and CNP β-glycosides of laminaribiose (G3G), consistent with a requirement for a β(1,3) linkage spanning the -2 to -1 active-site subsites (GH subsite nomenclature according to (Davies et al., 1997)), as was indicated by the bMLG limit-digest analysis (vide supra). Likewise, G4G3G-CNP and G4G4G3G-CNP were specifically and efficiently hydrolyzed to release the aglycone, with no cleavage of the internal glycosidic bonds. The specificity constants (k_cat/K_m values) for CNP release from these mixed-linkage tri- and tetrasaccharides were 800- and 1500-fold greater than that of G3G-CNP, respectively, which indicate that potential -3 and -4 subsites contribute 17 kJ/mol and 1.6 kJ/mol to transition state stabilization (ΔΔG^‡). Indeed, a very significant contribution from the -3 subsite is a common feature of GH16 endo-MLGases (Gaiser et al., 2006; McGregor et al., 2016; Planas, 2000).

BoGH16_MLG tertiary structure

Three-dimensional structures of recBoGH16_MLG were solved by X-ray crystallography to reveal the molecular basis for MLG recognition and hydrolysis. The apo structure of recBoGH16_MLG was determined to a resolution of 1.8 Å by molecular replacement using the structure of Zobellia galactanivorans laminarinase ZgLamC_GH16-E142S (PDB code 4CRQ) (Labourel et al., 2015) as a search model (See Table S4 for processing and refinement statistics). The crystal contained two polypeptide chains in the asymmetric unit corresponding to residues I35-L271 of wild-type BoGH16_MLG for both chains (residue numbering is from transcriptional start site according to the genomic sequence). No electron density was observed for the N-terminal His₆-tag and subsequent 15 amino acids in either chain of the recombinant protein, which suggests that residues C19-D34 of the wild-type enzyme constitute a flexible linker sequence to distance the catalytic module from the outer membrane surface (residues M1-S18 comprise the predicted signal peptide); the sidechain of C19 would constitute the site of N-terminal lipidation (Paetzel et al., 2002). The overall fold of BoGH16_MLG consists of a β-jelly roll architecture typical of other GH16 members (Davies and Sinnot, 2008): Two antiparallel β-sheets stack on top of each other with the concave face forming the polysaccharide substrate binding cleft. The end-on arrangement of the two chains in the asymmetric unit hinted at the possibility of the formation of a dimer (Fig 4A). Size-exclusion chromatography, however, indicated that BoGH16_MLG exists as a monomer in solution (data not shown) which, together with steric considerations of polysaccharide binding through the active-site cleft, indicates that end-on contacts observed between Chains A and B are artifacts of crystal packing.

Figure 4. BoGH16_MLG structural biology.

A: the overall structure of the BoGH16_MLG:G4G4G3G asymmetric unit containing two polypeptide chains shown from orthogonal views with the bound oligosaccharides in yellow and the transparent surface representation in white. Chain A cartoon is shown in cyan, and chain B cartoon is shown in slate blue throughout the figure. B: Mixed-linkage tetrasaccharide ligand modelled into chain A of BoGH16_MLG with the opaque surface representation in gray and the oligosaccharide colored according to B-factors. The glucose in subsite -4 is outside of the active site cleft and has significantly higher B factor than the glucose units in subsites -1 to -3. C: Tyr-181 rotamers observed in the complex structure with the 2Fo-Fc map of the tyrosines shown contoured at 0.5σ in grey. D: Tyr-181 residues observed in the apo structure with the 2Fo-Fc map of the tyrosines shown contoured at 0.5σ in grey. E: Wall-eyed stero view of the active site of chain A of the BoGH16_MLG:G4G4G3G complex. Hydrogen bonding interactions are shown as dashed black lines, sugars are shown in yellow with its 2Fo-Fc map contoured at 1σ in orange, and the conserved GH16 active site residues shown in purple. Hydrophobic stacking interactions in addition to hydrogen bonds position the mixed-linkage oligosaccharide in the negative subsite of BoGH16_MLG. See also Figures S5 and S6; Tables S4 and S5.

The sidechains of the conserved GH16 catalytic residues (Planas, 2000), comprising Glu-143 (nucleophile), Asp-145 (electrostatic “helper”) and Glu-148 (acid/base) are presented on the same face of one β-strand (β8), pointing into the active-site cleft. Notably, these residues are contained in a EXDXXE consensus sequence that is typical of bacterial GH16 laminarinases (β(1,3)-specific endo-glucanases). The insertion of an extra amino acid (underlined), typically methionine, results in a so-called “β-bulge” secondary structural motif that is not found among canonical bacterial MLGases, which instead possess a regular β-strand (Barbeyron et al., 1998; Michel et al., 2001).

Commensurate with this observation, the closest eight structural homologs identified using the Dali server (Holm and Rosenstrom, 2010) feature a β-bulge active-site motif (Table S5). Specifically, the top match (Z-score = 29.3) was the structure of laminarinase “ZgLamC_GH16-E142S“ from Zobellia galactanivorans (PDB code 4CTE) (Labourel et al., 2015), which has 38% amino acid identity and superimposed with BoGH16_MLG with a root mean square deviation (RMSD) value of 2.0 Å over 211 out of 231 Cα pairs. In comparison, the closest GH16 homolog with a regular active-site β-strand was the lichenase (MLGase) from Paenibacillus macerans (PDB code 1MAC) (Hahn et al., 1995), which has a comparable Z-score of 25.1 and an RMSD value of also 2.0 Å over 200 out of 212 Cα pairs, despite having only 22% amino acid identity with BoGH16_MLG.

Soaking crystals of the wild-type enzyme with G4G4G3G yielded a product complex with 1.8 Å resolution (Table S4). The complete tetrasaccharide was modelled in electron density spanning subsites -1 to -4 in the active-site cleft of Chain A, while the electron density for the fourth glucosyl residue in subsite -4 was not resolved in chain B. This is most likely due to disorder of this residue since the corresponding -4 Glc in Chain A is fully solvent exposed, makes no contact with the enzyme, and has significantly higher B-factors (Fig. 4B). These structural observations are consistent with kinetic data for chromogenic MLG oligosaccharides (Table 1), which likewise suggest the existence of three primary negative subsites, -1 to -3, and a weakly interacting -4 subsite.

Table 1.

Enzyme	Substrate	kcat (s−1)	Km (mM)	kcat/Km (s−1 mM−1)	Assay
BoGH3MLG	β-Glc-pNP	59.5 ± 1.46	2.95 ± 0.14	20.2	pNP
	gentiobiose (G6G)	ND	ND	0.0571	HK/G6PDH
	cellobiose	5.52 ± 0.19	7.47 ± 0.48	0.739	HK/G6PDH
	cellotriose	22.1 ± 0.3	0.859 ± 0.033	25.7	HK/G6PDH
	cellotetraose	17.3 ± 0.5	0.687 ± 0.044	25.2	HK/G6PDH
	cellopentaose	19.4 ± 0.8	0.777 ± 0.060	25.0	HK/G6PDH
	cellohexaose	17.4 ± 0.4	0.747 ± 0.041	23.3	HK/G6PDH
	laminaribiose	28.0 ± 1.1	1.90 ± 0.12	14.7	HK/G6PDH
	laminaritriose	34.2 ± 1.0	0.911 ± 0.052	37.5	HK/G6PDH
	laminaritetraose	31.3 ± 2.3	0.898 ± 0.135	34.9	HK/G6PDH
	laminaripentaose	39.5 ± 3.4	1.27 ± 0.20	31.1	HK/G6PDH
	MLGO3 A (G3G4G)	61.6 ± 1.6	0.997 ± 0.040	61.8	HK/G6PDH
	MLGO3 B (G4G3G)	24.7 ± 1.3	0.521 ± 0.064	47.4	HK/G6PDH
	MLGO4 A (G3G4G4G)	55.7 ± 2.7	1.33 ± 0.12	41.9	HK/G6PDH
	MLGO4 B (G4G4G3G)	30.8 ± 2.0	0.736 ± 0.106	41.8	HK/G6PDH
	MLGO4 C (G4G3G4G)	15.7 ± 0.3	0.601 ± 0.031	26.1	HK/G6PDH
BACOVA_02738 (GH3)	β-Glc-pNP	0.212 ± 0.004	2.53 ± 0.13	0.0838	pNP

ND: not determined (in cases where Michealis-Menten curve fitting was not feasible, individual k_cat and K_m values are not reported and k_cat/K_m value was determined from linear curve fit to initial-rate data in the [S] ≪ K_m(apparent) range). Data are represented as the fitted parameters ± standard deviation. Highlighted in bold are the biologically relevant substrates that BoGH3_MLG encounters in the periplasmic space. See also Figure S4.

In both Chain A and B, the three glucosyl residues spanning subsites -1 to -3 are well defined and virtually identical. The reducing-end glucosyl residue in the -1 subsite is in the β-conformation, with the C1 hydroxyl hydrogen bonded to Tyr-181, which is observed in a dual conformation in both chains of the G4G4G3G-complex (Fig. 4C). Interestingly, this dual conformation is not observed in the apo-form of the enzyme; Tyr-181 is “swung in” to the active site in chain B, while it is “swung out” in chain A, the sidechain from chain A stacking on top of the chain B sidechain (Fig. 4D). The conformation of this sidechain will be key to determining the nature of the positive substrate binding subsites, indeed, comparison with other GH16 endo-glucanases clearly suggests the presence of two positive subsites (Gaiser et al., 2006; Planas, 2000). Whether the dynamics observed for Tyr-181 are an artefact of crystallisation, or perhaps play a role in substrate binding and product release is unclear in the absence of an enzyme-substrate complex spanning the positive subsites.

With regard to specific interactions in the negative subsites, subsite -1 is further characterized by hydrogen bonds between Glu-143 and the C2 hydroxyl, Trp-125 and the C6 hydroxyl, as well as Glu-148 and the ring oxygen and the C1 hydroxyl. This glucose is also positioned by a stacking interaction with Trp-125 and Trp-129 (Fig. 4E), both of which are conserved across all GH16 laminarinases. At subsite -2, highly conserved Arg-97 forms a hydrogen bond with the C6 hydroxyl, and Asn-60 hydrogen bonds to the C2 hydroxyl as well as to the glucosidic bond oxygen between the -1 and -2 sugars. Another conserved residue, Trp-138, serves as a platform that stacks with the subsite -2 glucose. In subsite -3, the main interaction is stacking against Trp-58, which also forms a hydrogen bond to the glucosidic bond oxygen between the -3 and -4 sugars (Fig. 4E). Together, these interactions in subsite -3 are responsible for 17 kJ/mol of transition-state stabilization (vide supra).

Downstream saccharificaiton of mixed-linkage oligosaccharides produced by BoGH16_MLG

To elucidate the mechanism for the downstream conversion of the oligosaccharide products of BoGH16_MLG to glucose for primary metabolism, we examined the biochemistry of the two predicted exo-β-glucosidases, BoGH3_MLG and BACOVA_02738(GH3) associated with the MLGUL.

Cellular localization of BoGH3_MLG and the BACOVA_02738(GH3) gene product

BoGH3_MLG and BACOVA_02738(GH3) were unambiguously predicted by SignalP 4.0 (Petersen et al., 2011) to contain a secretion signal peptide, while LipoP 1.0 (Juncker and Willenbrock, 2003) additionally indicated a Type II lipoprotein signal sequence (Paetzel et al., 2002) in BoGH3_MLG only. The same B. ovatus _tdk cultures used for BoGH16_MLG localization, containing glucose or bMLG as the sole carbon source, were probed using polyclonal antibodies independently raised against recombinant BoGH3_MLG and the BACOVA_02738(GH3) gene product. Neither protein was detected on the cell surface by fluorescence microscopy, especially in the presence of bMLG which induces BoGH16_MLG production (Fig. 2B, Fig. S2A). BoGH3_MLG induction by bMLG was confirmed by a Western blot of cellular fractions, which also confirmed that this enzyme is membrane anchored (Fig. 2C).

In contrast, the BACOVA_02738(GH3) gene product was detected to a higher degree in B. ovatus cells grown in minimal medium with glucose as a sole carbon source compared to cells induced with bMLG (Fig. S2B). The lack of upregulation by bMLG is consistent with transcriptional analysis which showed a limited change in expression in bMLG vs. glucose (1.6-fold), which was two orders of magnitude lower than definitive MLGUL genes (Table S1). The higher detection in uninduced cells is explained by the high basal expression of BACOVA_02738(GH3) (more than an order of magnitude greater than all MLGUL members; Table S1). The lack of detection in minimal medium containing bMLG is due to high amounts of induced MLGUL proteins diminishing the presence of the BACOVA_02738(GH3) gene product when normalized to total protein (Fig. S2B).

Substrate product specificity of BoGH3_MLG and BACOVA_02738(GH3)

Initial activity screening on chromogenic pNP glycosides (see Experimental Procedures) revealed that both recBoGH3_MLG and recBACOVA_02738(GH3) are specific exo-β-glucosidases (activity on other pNP glycosides was undetectable at micromolar enzyme concentrations). However, recBACOVA_02738(GH3) is strikingly feeble compared to recBoGH3_MLG on G-β-pNP (k_cat/K_m values of 0.084 mM⁻¹ s⁻¹ versus 20 mM⁻¹ s⁻¹; Fig. S4C, S4D, Table 1). Further, measuring Michealis-Menten kinetic parameters on cello- and laminari-oligosaccharides was not feasible due to overall poor activity and low production yields (data not shown). These kinetic results corroborate the above comparative genetic and transcriptional analyses, collectively suggesting BACOVA_02738(GH3) is not integral to the MLGUL. Hence, this enzyme was not characterized further.

To investigate oligosaccharide substrate preference of the BoGH3_MLG, we conducted initial-rate kinetics analyses on a series of gluco-oligosaccharides of distinct linkage composition and degrees of polymerization (d.p.). The non-reducing-end glucose was hydrolyzed from all-β(1,4)-linked cello-oligosaccharides (d.p. 2–6), all-β(1,3)-linked laminari-oligosaccharides (d.p. 2–5), and mixed-linkage β(1,3)/β(1,4)-gluco-oligosaccharides (d.p. 3–4, 5 examples) with comparable efficiencies, according to classic Michaelis-Menten saturation kinetics (Fig. S4E, S4F; Table 1). In this series, only cellobiose (G4G) was poorly hydrolyzed by BoGH3_MLG vis-à-vis the β(1,3)-linked congener laminaribiose (G3G) and all other gluco-oligosaccharides (e.g., G4G has a k_cat/K_m value 20-fold lower than G3G, Table 1). The β(1,6)-linked diglucoside gentiobiose (G6G) was also a very poor substrate, with a k_cat/K_m value 260-fold lower than that of G3G. Gluco-oligosaccharides with a β (1,3)-linked glucosyl unit at the non-reducing end all have slightly higher k_cat values compared to those with a β(1,4)-linkage in this position, which typically contributes to higher k_cat/K_m values for the former, when substrates of equal d.p. are compared. However, the magnitude of these differences, which are often less than 2-fold, indicate that BoGH3_MLG is essentially agnostic to β(1,3) versus β(1,4) linkages. These results also suggest that in addition to a single negative subsite (-1) commensurate with its exo-activity, BoGH3_MLG has only two positive subsites that contribute to catalysis: in each gluco-oligosaccharide series, tetrasaccharides and larger are hydrolyzed with identical k_cat/K_m values to the corresponding trisaccharides.

Product analysis following extended incubation of BoGH3_MLG with G4G4G3G and G4G3G demonstrated that BoGH3_MLG completely degrades the BoGH16_MLG limit-digest products to glucose. HPLC also revealed that laminaribiose (G3G) is the only new intermediate formed during the course of hydrolysis (Fig. 3B). This demonstrates that BoGH3_MLG sequentially hydrolyzes one glucose unit at a time from the non-reducing end of MLG oligosaccharides, viz.: G4G4G3G → G + G4G3G (also present in the starting mixture) → G+ G3G → G + G. Notably, the individual k_cat and K_m values for each step are nearly identical (Table 1).

BoGH3_MLG and BACOVA_02738(GH3) primary structures

Despite extensive efforts, we were unable to crystallize the key β-glucosidase BoGH3A_MLG for experimental tertiary structure determination. However, BoGH3_MLG has 63% sequence identity to a B. ovatus β-glucosidase (BoGH3B, PDB code 5JP0 (Hemsworth et al., 2016)), from the xyloglucan utilization locus (Fig. S5A) and, as such, was amenable to tertiary structure homology modelling. Phyre2 (Kelley et al., 2015) utilized PDB code 5JP0 as the sole template and 728 out of 742 residues (98% of the sequence, excluding the signal peptide) were modelled with 100 % confidence. The model suggests that BoGH3_MLG possesses a homologous three-domain architecture with the active site being formed at the interface of the (α/β)₈ TIM barrel and α/β sandwich domains (Fig. S5B). The predicted catalytic nucleophile and acid/base residues are Asp-309 and Glu-453, respectively, based on primary and tertiary alignment (Fig. S5A, S5C). Two tryptophan residues were modelled on opposite sides of the entrance to the active site pocket (Fig. S5D), forming a narrow “coin slot”, which may effect a preference towards β(1,3)- and β(1,4)-linked glucans and poor activity against β(1,6)-linked gentiobiose (Table 1). In constrast, enzymes that lack a homologous Trp-453 have a more open entrance to the active site and show broad activity against β(1,2)- and β(1,6)-linked glucans in addition to β(1,3)- and β(1,4)-linked glucans (Karkehabadi et al., 2014; Pozzo et al., 2010).

In comparison, BACOVA_02738(GH3) possess catalytic residues homologous to BoGH3_MLG and similar GH3 members, despite having only 31% sequence identity to BoGH3_MLG (Fig. S5A). The most similar characterized GH3 member to BACOVA_02738(GH3) among ca. 300 members identified in the CAZy is a Chrysosporium lucknowense β-glucosidase with 39% sequence identity (Dotsenko et al., 2012).

Syntenic MLGUL are molecular markers of MLG utilization across the Bacteroidetes

Refined functional characterization of the catalytic specificity of GH components significantly increases confidence in the use of individual PULs as genetic markers of complex carbohydrate metabolism among Bacteroidetes (Cuskin et al., 2015; Larsbrink et al., 2014; Rogowski et al., 2015; Sonnenburg et al., 2010). The MLGUL characterized here represents the only PUL in B. ovatus that confers growth on MLG from cereals. To understand the wider distribution of MLG metabolic capacity among symbiotic Bacteroidetes in the human gut, we correlated the presence of a syntenic MLGUL with growth on bMLG for 354 individual Bacteroidetes strains representing 29 different species.

A total of 121 strains across just 7 of the species were able to grow on bMLG (Fig. 5). In particular, 33 of 33 B. ovatus strains (including the type strain ATCC 8483) grew well on bMLG, as well as 44 of 45 strains of the closely related B. xylanisolvens. The majority of B. uniformis strains tested (33 out of 35) were also competent bMLG utilizers. The limited penetrance of the MLGUL across the genus clearly demonstrates nutrient-niche specialization among individual Bacteroides species.

Figure 5. Penetrance map of MLG utilization ability across diverse human gut Bacteroidetes.

The phylogenetic tree was constructed from fully sequenced strains of the species shown. The number of strains of each species tested for growth is depicted to scale as a black circle at each leaf. The number of those strains that grew on bMLG as a sole carbon source is shown to scale in red within the black circle.

Comparative analysis of available genomic sequences revealed that strains able to grow on bMLG as the sole carbon source harbor a syntenic MLGUL (Fig. 1B). Previous transcriptional analysis demonstrated that the syntenic MLGUL found in B. cellulosilyticus is also activated during growth on bMLG (McNulty et al., 2013). Concordance between the presence of a syntenic MLGUL and the ability to utilize MLG is further highlighted by the lack of a MLGUL in the B. uniformis ATCC 8492, one of only two strains of B. uniformis that could not grow on bMLG. Based on this analysis, we can also predict MLG utilization ability in two sequenced species of Prevotella, Pr. copri DSM 18205 and Pr. multiformis DSM16608, important members of the Bacteroidetes from the human gut and oral cavity, respectively (Fig. 1B).

Discussion

A molecular model for MLG utilization by B. ovatus

Our current suite of data suggests a model by which the MLGUL gene products work in concert to enable the utilization of MLG (Fig. 6), analogous to that of other PUL-encoded systems (Grondin et al., 2017). Thus, BoGH16_MLG is anchored to the outer membrane where it plays a leading role in fragmenting large MLG polysaccharide chains (typical d.p. 700 – 5000, depending upon the plant species of origin (Lazaridou et al., 2004; Zheng et al., 2011)) into oligosaccharides that can be imported into the periplasm via the TBDT. Notably, the specific limit-digest products of BoGH16_MLG endo-hydrolysis identified here, viz. the trisaccharide G4G3G and the tetrasaccharide G4G4G3G (Fig. 3B), have been shown previously to bind the periplasmic sensor domain of the HTCS encoded by BACOVA_02740 (K_D 300 µM and 400 µM, respectively), while the intact MLG polysaccharide does not (Martens et al., 2011). Monomeric glucose, all-β(1,4)-linked cello-oligosaccharides, and all-β(1,3)-linked laminari-oligosaccharides are also not bound by the HTCS (Martens et al., 2011), indicating that the unique linkages present in MLG are central to inducing the MLGUL system. Thus, there is an essential, yet distant, interplay between the outer-membrane localized MLGase and the HTCS in specific nutrient sensing.

Figure 6. Model of mixed-linkage β-glucan saccharification by the concerted action of the MLGUL machinery.

Gene products are colored analogously to the gene locus in Fig. 1. The cell surface localized endo-MLGase BoGH16_MLG cleave large mixed-linkage β-glucan polysaccharides into shorter fragments which are imported into the periplasm via the TonB dependent transporter, BoTBDT. This glycan capture and transport process at the cell surface is aided by the two surface glycan binding proteins BoSGBP-A and BoSGBP-B. The smaller mixed-linkage β-glucan fragments in the periplasm bind the sensor domain of the hybrid two-component sensor BoHTCS to induce upregulation of the system. Periplasmic exo-β-glucosidases BoGH3_MLG and BACOVA_02738(GH3) act from the non-reducing ends to liberate individual glucose monomers which are imported into the cell and metabolized.

It is therefore likely that the BoGH16_MLG limit-digest products, or minimal repeats of these structures [(G4G4G3G)_m(G4G3G)_n], comprise the main products transported through the TBDT in vivo. Recent studies on inulin (β(2,1)-fructan) utilization suggest that some TBDTs are able to import longer polysaccharide chains (e.g. d.p. >20) (Rakoff-Nahoum et al., 2016). Regardless of length, the efficient exo-hydrolytic activity of BoGH3_MLG in the periplasm is sufficient to completely saccharify all imported oligosaccharides to glucose (Fig. 3B), to feed primary metabolism in the cytosol. In this process, the trisaccharide substrate of the HTCS, G4G3G, will always be generated regardless of the imported saccharide chain length, ensuring continual production of the MLGUL up-regulation signal until substrate is exhausted. Interestingly, BoGH3_MLG will never encounter cellobiose (G4G), towards which it has relatively weak activity (Fig. S4F; Table 1), in this process; the final step of saccharification of MLGOs is the hydrolysis of the competent substrate laminaribiose (G3G).

Structural enzymology reveals complex trajectories for the evolution of MLG activity in GH16

Previous phylogenetic analyses of GH16 have suggested that the evolution of the active-site β-bulge motif EXDXXE, which is widespread among Clan GH-B (comprising GH16 and GH7), to a regular β-strand motif EXDXE is a defining feature that delineates endo-β(1,3)-glucanases (laminarinases, EC 3.2.1.39 and EC 3.2.1.6) from mixed-linkage endo-β(1,3)/β(1,4)-glucanases (licheninases, EC 3.2.1.73), respectively (Barbeyron et al., 1998; Michel et al., 2001). In this context, the observation that BoGH16_MLG is highly specific for MLG, despite having a β-bulge motif, was surprising.

Using the CAZy Database as a starting point (http://www.cazy.org/GH16_characterized.html) together with mining of the primary literature, we generated a contemporary maximum-likelihood phylogeny of all biochemically characterized GH16 members (Fig. S6). This analysis indicates that although canonical, normal β-strand MLGases do form a monophyletic group as previously observed, MLGase activity is in fact widespread among the historical “laminarinase” group, in which BoGH16_MLG is itself positioned. Despite currently limited and disparate kinetic data for individual enzymes, it also appears that it is not possible to define further substrate-specific clades within this group based on molecular phylogeny alone, in light of weak bootstrap support. This precludes defining any single evolutionary event giving rise to unique trajectories for the further diversification of extant laminarinases and MLGases in this clade. Instead, it appears that diverse, subtle mutations have allowed the independent evolution of predominant laminarinase or MLGase activity numerous times across a flat evolutionary landscape. As such, we suggest that this GH16 subgroup should be more generally referred to as the “laminarinase/MLGase group” going forward.

Detailed tertiary structural comparison of 10 β-bulge-containing members of this laminarinase/MLGase group revealed, however, that predominant laminarinases harbor a significantly more protruding loop (which is often, but not always, longer) connecting strands β2 and β3 than predominant MLGases (Fig. S7A, S7B). Structural superposition with the BoGH16_MLG:G4G4G3G complex indicates that this loop in predominant laminarinases would clash with MLG in the negative subsites, instead favoring binding of an all-β(1,3)-glucan that curves away from this loop. Such curvature is exemplified by the ZgLamC_GH16-E142S:thio-β-1,3-trisaccharide structure (Fig. S7A, PDB code 4CTE) (Labourel et al., 2015). Indeed, Ilari et al. observed that shortening the homologous loop in LamA from the archeon Pyrococcus furiosus (Fig. S7A, PDB code 2VY0) by 4 amino acids increased the activity towards MLG by 10-fold (Ilari et al., 2009). Likewise, BglF from Nocardiopsis sp. F96 (Fig. S6B, PDB code 2HYK) and LamR from Rhodothermus marinus (Fig. S7B, PDB code 3ILN), which have a 3.3- and 8.5-fold greater specificity constant and specific activity, respectively, toward MLG than laminarin, also have a smaller loop, similar to BoGH16_MLG, in this position. The canonical, regular-β-strand MLGase from Paenibacillus macerans (Fig. S7C, PDB code 1MAC) and Bacillus licheniformis (Fig. S7C, PDB code 1GBG), similarly have a small loop at this position.

Taken together, these analyses reveal a complex evolutionary landscape that computational phylogenetic analysis fails to resolve. Despite using a manually curated, structure-based sequence alignment as input, the maximum-likelihood numerical approach did not delineate the members of the laminarinase/MLGase group on the basis of the distinct active-site loop differences observed in tertiary structures (Fig. S7). Instead, the phylogeny was likely obfuscated by diverse, random variations in amino acid composition across the entire β-sandwich domain, which clearly limits large-scale, unsupervised phylogenetic analysis of these GH16 members. Moreover, analysis of both MLG and laminarin specificity (as a minimum) for individual members of this group, in light of their tertiary structures, is essential to avoid potential mis-annotation of these enzymes.

Mining metagenomic data reveals the ubiquity of MLG utilization in the human gut and beyond

Using syntenic MLGULs as genetic markers, we surveyed the publicly available metagenome data of 426 adults to understand the capacity of human populations to derive nutrition from cereal MLGs. We were able to distinguish the species of origin based on nucleotide sequence except for MLGULs from B. ovatus and B. xylanisolvens, which were strikingly similar at 97% nucleotide identity. The B. ovatus/B. xylanisolvens and B. uniformis MLGULs are the most prevalent; both are observed in about 70% of the total human cohort (Fig. 7). The Pr. copri MLGUL is more often the sole MLGUL of an individual than that of B. cellulosilyticus when only one is present (Fig. 7, cyan lines), despite the latter being more frequent in total. Overall, 92.5% of the subjects harbor at least one of the five different MLGULs identified in this study, irrespective of nationality or whether they have been diagnosed with IBD. MLGULs are ubiquitously detectable despite variability in sampling depth across different metagenomics sequencing projects (Fig. 7). The prevalence of MLGULs across different nationalities is consistent with MLG from cereal grains being a ubiquitous component of the human diet. Indeed, the importance of MLG utilization is underscored by the upregulation of the MLGUL in the ceca of mice fed a complex plant cell wall diet (Martens et al., 2011). Similar widespread global distribution in human populations has been observed for xyloglucan utilization loci (Larsbrink et al., 2014). These observations are sharply contrasted by the PUL that mediates utilization of the red algal polysaccharide porphyran, which is essentially confined to subjects from Japan, where seaweed is a common part of the diet (Hehemann et al., 2010; Larsbrink et al., 2014). Interestingly, we were unable to detect MLGULs in four unweaned infants sampled in the Japanese metagenome project (data not included in our analysis of adult metagenomes). This is consistent with a dearth of Bacteroidetes in infants who receive the bulk of their nutrition from milk and are not yet consuming plant polysaccharides (Urokawa et al., 2007)

Figure 7. Bacteroidetes MLGULs from a survey of 426 adult human gut metagenomes.

Vertical lines represent the presence (cyan when unique, blue when one of multiple) or absence (black) of a corresponding species-related MLGUL in a single individual. The total number of MLGULs observed in an individual is shown in the bottom row, colored according to the legend in the top left corner. The frequency of MLGUL occurrence across all 426 individuals is shown on the right. Variation in sequencing depth in megabase pair is illustrated in the graph below: grey lines show the depth for individual subjects and black lines show the average depth of each metagenomics project.

Moving beyond the human microbiota, we can likewise predict MLG utilization ability in Dysgonomonas gadei and Pr. oryzae (formerly Xylanibacter oryzae) based on the presence of a syntenic MLGUL. These species are commonly found in the termite hindgut and decomposing rice straw, respectively. This provides direct evidence that an analogous MLG utilization system is employed by Bacteroidetes operating in environments beyond the human gut.

Conclusion

Complex carbohydrates that promote the growth of beneficial microbes in our distal large intestine are a cornerstone of a healthy diet. MLGs in particular have long been known to impart healthful effects (Othman et al., 2011), yet its mechanism of utilization for fermentation by gut microbes was unknown. Our work here sheds light on the fine-tuned mechanism that B. ovatus and other Bacteroidetes has evolved to efficiently utilize MLGs in the highly competitive environment of the human gut microbiota. The finding that a majority of humans possess microbes that can utilize this ubiquitous cereal polysaccharide highlights the relevance of potential therapeutic interventions based on MLG utilization to the general population. The present study also sets the stage for future work to understand the quantitative contributions of individual members of the microbiota and their cognate enzymes to MLG utilization in the human gut (Patrascu et al., 2017; Zhong et al., 2015).

Experimental Procedures

Microbiology

B. ovatus gene deletions were constructed using allelic exchange as previously described (Koropatkin et al., 2008). Anaerobic growth profiles were measured as previously described (Martens et al., 2011). Details of localization analysis by immunofluorescence microscopy and immunoblotting are provided in the Supplemental Experimental Procedures.

Cloning, expression, and purification of recombinant enzymes

The genes encoding BoGH16_MLG, BoGH3_MLG, and BACOVA_02738(GH3) were cloned into expression vectors for recombinant production in E. coli. Details of cloning strategies, production, and purification are provided in the Supplemental Experimental Procedures.

Enzyme kinetics and product analysis

Thorough kinetic analysis was conducted on a panel of polysaccharides, oligosaccharides, and chromogenic substrates. Products of enzymatic reactions were analyzed by HPAEC-PAD and HILIC-MS. Details of enzymatic assays, analytical methods, as well as sources of commercial substrates are provided in the Supplemental Experimental Procedures.

X-ray crystallography

Crystals of BoGH16_MLG were screened and optimized by sitting drop vapor diffusion method. The structures of the apo- and G4G4G3G-BoGH16_MLG were solved by molecular replacement. Details of crystallization, data collection, and structure solution are provided in the Supplemental Experimental Procedures.

Bioinformatics

Phylogenetic anlaysis of select GH16 and GH3 sequences was conducted based on structure-guided alignment. Metagenomic survey was carried out by nucleotide BLAST of MLGUL sequences against various metagenome sequence data. Details are provided in the Supplemental Experimental Procedures.

Statistical analysis

All kinetic assays were done in triplicate. Michaelis-Menten parameters are reported as fitted values ± standard deviation throughout the article. All growth experiment results are averages of two biological replicates.

Highlights.

• The molecular mechanism of MLG utilization by Bacteroides ovatus is presented.

• BoGH16_MLG possesses structural features suited for efficient MLG degradation.

• MLG utilization loci (MLGULs) serve as genetic markers for MLG catabolism.

• Bacteroidetes MLGULs are ubiquitous in the gut microbiota of human populations.

Note: MLG may be expanded to “mixed-linkage beta-glucan” for clarity, if the character count restriction can be relaxed.