ABSTRACT
Beneficial modulation of the gut microbiome has high-impact implications not only in humans, but also in livestock that sustain our current societal needs. In this context, we have engineered an acetylated galactoglucomannan (AcGGM) fibre from spruce trees to match unique enzymatic capabilities of Roseburia and Faecalibacterium species, both renowned butyrate-producing gut commensals. The accuracy of AcGGM was tested in an applied pig feeding trial, which resolved 355 metagenome-assembled genomes together with quantitative metaproteomes. In AcGGM-fed pigs, both target populations differentially expressed AcGGM-specific polysaccharide utilization loci, including novel, mannan-specific esterases that are critical to its deconstruction. We additionally observed a “butterfly effect”, whereby numerous metabolic changes and interdependent cross-feeding pathways were detected in neighboring non-mannolytic populations that produce short-chain fatty acids. Our findings show that intricate structural features and acetylation patterns of dietary fibre can be customized to specific bacterial populations, with the possibility to create greater modulatory effects at large.
Prebiotic strategies use dietary fibre to manipulate gut microbiota and promote specific populations to improve gut function in humans and production animals. Prebiotics by definition are not broadly metabolized, but rather elicit a targeted metabolic response in specific indigenous microbiota that confers health and nutrition benefits to their host. This in itself presents a challenge; as many newly identified target organisms, such as beneficial butyrate-producing Roseburia and Faecalibacterium spp.1,2, have broad metabolic capabilities that are shared with the vast majority of fibre-fermenting microbiota in the gut ecosystem. Nevertheless, recent studies have revealed intimate connections between the enzymatic and mechanistic features of microorganisms and the glycan structures of the fibres they consume, which creates new conceptual prebiotic targets. This is exemplified by discoveries of sophisticated polysaccharide-degrading apparatuses that enable certain microbiota to consume fibre in a “selfish” manner, whereby complex glycan-structures (such as β-mannans) are cleaved into large oligosaccharides at the cell surface, which are subsequently transported into the cell and depolymerized into monomeric sugars3–5. Such a mechanism restricts the release of sugars into the ecosystem for neighboring scavenging populations, thus giving a selective metabolic advantage to the selfish-degrader in the presence of these highly complex glycans.
Beta-mannans are present in human and livestock diets, and depending on their plant origins, can be heavily decorated with varying amounts of acetylation that protect the fibre from enzymatic degradation6. We recently demonstrated that the human gut commensal Roseburia intestinalis encodes a mannan-specific polysaccharide utilization locus (PUL), and “selfishly” converts highly complex mannan substrates to butyrate4. Within this mannan PUL, a carbohydrate esterase (RiCE2) removes 3-O-, and 6-O-acetylations on mannan, whereas a novel RiCEX removes oriented 2-O-hydroxyl acetylations6, which are distinctive features found in limited mannan moieties and inaccessible to most esterases present in the gut microbiome. Closer genome examinations have revealed that putative CE2/CEX-containing mannan PULs are infrequently distributed amongst common fibre-degrading gut microbiota, yet they are prominent within many butyrate-producers including Roseburia spp., Faecalibacterium prausnitzii, Ruminococcus gnavus, Coprococcus eutactus and Butyrivibrio fibrisolvens4,7. It is well known that the metabolic attributes of these populations are highly desirable in the gastrointestinal tract, and that their depletion is implicated in colorectal cancer, Crohn’s disease, inflammatory bowel syndrome, ulcerative colitis, forms of dermatitis and several other diseases8,9. These collective findings thus raised the question: could a custom fibre that was engineered to match these specialized enzymatic capabilities be harnessed to selectively engage butyrate-producers in a complex microbiome ecosystem?
2-O-acetylated mannans are not commonly found in western dietary fibre sources, however 2-O-acetylations are present in acetylated galactoglucomannan (AcGGM), which is the main hemicellulose in the secondary cell wall of Norway spruce (Picea abies)10. Here, we have utilized controlled steam explosion (SE), followed by ultrafiltration (UF) and fractionation to extract from spruce wood an unadulterated complex AcGGM fibre with a high degree of 2-O-, 3-O- and 6-O- acetylations11, which is amenable to inclusion in animal feed production. To test (1) if our AcGGM fibre could specifically target Roseburia and Faecalibacterium species within a complex microbiome, and (2) if health benefits could be conferred to the host, we produced diets containing 0%, 1%, 2% and 4% AcGGM and fed semi-ad libitum to four cohorts of twelve weaning piglets. Host and gut microbiome effects were monitored temporally over a 28 day period, and metagenomics used to phylogenetically and functionally resolve the genomes of indigenous microbiota. Finally, quantitative metaproteomic analysis was used to monitor the metabolic and enzymatic response of the different microbiota to the varying AcGGM exposure. This approach deciphered how specific beneficial microbiota can be targeted and metabolically stimulated in complex gut microbiome ecosystems, with broader implications towards the evolving strategy of gut microbiome manipulations.
RESULTS AND DISCUSSION
Steam explosion allows scalable production of highly complex dietary mannan fibres from wood
Spruce galactoglucomannan consists of a backbone of β-(1,4)-linked mannose and glucose residues, decorated with α-(1,6) linked galactose branching’s, and a large degree of esterification of the mannose residues by 2-O- and 3-O-, and 6-O-acetylation’s10 (Fig 1a). A crucial part of this study was the development of an efficient, large-scale extraction process entailing SE as well as ultra- and nano-filtration, which ultimately provided high quantities at high purity whilst not damaging the complexity of the AcGGM fibre (Fig. 1b-c). A total of 700kg of dry Norway spruce chips was processed using SE at conditions corresponding to a combined severity factor (R’0) of 1.70. We produced 50 kg of oligo/polysaccharides for feed production (Fig. 1c-e), with a monosaccharide (Man: Glc: Gal) ratio of 4:1:0.6, which was in the form of β-mannooligosaccharides with DP of 2 to 10 and manno-polysaccharides (DP of ≥11), and had degree of acetylation (DA = 0.36). This complexity matched the enzymatic capabilities of mannan PULs encoded in human gut Roseburia and Faecalibacterium spp.4,7 and was predicted to match representatives of the same populations that are indigenous to porcine gut ecosystems12,13 (Fig. 1f).
Spruce AcGGM altered the gut microbiome of weaned piglets in a dose-response manner
While pure culture4, in vitro7 and germ-free mice studies containing “mini-microbiota”4 have shown that varieties of AcGGM can be metabolized by butyrate-producers, we wanted to test the accuracy of AcGGM to elicit a specific response in indigenous representatives of our target populations within a highly complex and competitive environment. To assess the potential benefit of our fibre to the host, we chose to introduce AcGGM to weaning piglets immediately after they transitioned from sow’s milk to solid food. Weaning elicits a rapid, diet driven shift in the gut microbiome, which puts the animals at high risk of infection by intestinal pathogens such as enterotoxigenic Escherichia coli and Salmonella enterica14. These issues have been exacerbated by the banning of antibiotic use (Regulation No. 1831/2003) as growth promoters and prophylactics15, meaning there is added urgency to develop a prebiotic compound to improve health and welfare of the animals during this crucial stage of development16. In this study, four separate cohorts of twelve weaned piglets were given a pelleted feed semi-ad libitum, which contained either 0% (control), 1%, 2% or 4% AcGGM to additionally determine the level necessary to elicit an effect on both the host and its microbiome. Fecal samples as well as animal health and growth performance metrics were taken before AcGGM administration (when piglets were assigned to pens), and subsequently at days 7, 14, 21 and 27 during the feeding trial. At day 28, the piglets were sacrificed and host gut tissue and digesta samples taken from the entire regions of the digestive tract (duodenum, jejunum, ileum, cecum and colon) for down-stream analysis. Despite widespread changes in the microbiome resulting from AcGGM inclusion, surprisingly no significant effects were observed on the host’s physiology, with the average weight, feed conversion ratio, blood cell composition, T cell population, and colon morphology not differing between the control and AcGGM treatments (Supplementary Fig. 1, Supplementary Table 1-2).
Spatial and temporal microbiome changes were monitored using 16S rRNA gene analysis over the month-long trial and showed archetypical patterns with structural features of the gut microbiome varying depending on the specific gut region (Supplementary Fig. 2a). Inclusion of AcGGM into the piglets feed caused a pronounced shift in the microbiome structural composition from the 21st day of the trial onwards (Supplementary Fig. 2b, Supplementary Fig. 3). As expected, the AcGGM-effect was more pronounced in the fibre-fermenting distal regions (cecum, colon) of the gut, where the relative abundance of hundreds of phylotypes was observed to change (adjusted p<0.05) in response to varying inclusion levels (Fig. 2a, Supplementary Table 4). Our target butyrate-producing populations produced mixed results, whereby the relative abundance of Faecalibacterium affiliated phylotypes increased in response to increasing levels of AcGGM (Fig. 2b, Supplementary Fig. 3), whereas Roseburia affiliated phylotypes seemingly decreased (Fig. 2b, Supplementary Fig. 3). Reputable fiber-fermenting populations affiliated to Prevotella, also showed varying responses, with individual phylotypes increasing from 4% to 12% between the control and 4% AcGGM inclusion in both colon and cecum (Fig. 2b, Supplementary Fig. 3). Interestingly, phylotypes affiliated to non-fibre degrading taxa, such as Catenibacterium17 and Dialister18 demonstrated some of the highest dose-dependent increases in relative abundance in response to AcGGM (Fig. 2b), indicating that other underlying factors are likely dictating microbiome structure, besides fibre degradation.
Targeted mannan PULs and butyrate-producing pathways are actively detected in Faecalibacterium- and Roseburia-affiliated populations within the colon of AcGGM-fed pigs
To determine the effect AcGGM had on microbiome function, we analyzed 211.36 Gbps of Illumina HiSeq sequencing data obtained from the colon samples of each pig fed the control and 4% AcGGM diets (average: 8.8 Gbps per sample) (Supplementary Table 3). The metagenomic data was assembled into 355 metagenome-assembled genomes (MAGs), of which 145 had >90% completeness and were considered high quality according to the Genomics Consortium Standards19 (Supplementary Fig. 4a). Phylogenetic relationship of the MAGs were inferred from a concatenated ribosomal protein tree (Newick format available in Supplementary Dataset 1) that were constructed using MAGs from this study and 293 closely related reference genomes. Because our primary goal was to elucidate if our target butyrate-producing populations were activated in response to AcGGM, we conducted metaproteomic analysis on randomly selected colon samples from four control and four 4% AcGGM fed pigs, and mapped 8515 detected protein groups back against our MAG database to identify functionally active populations (Fig. 3, Supplementary Table 5). Community-wide analysis of the MAG genetic content (Supplementary Fig. 4b) from each sample, and distribution of their detected proteins (Fig. 3, Supplementary Fig. 4c), further supported our 16S rRNA gene analysis, reiterating that the microbiomes from piglets fed the control and 4% AcGGM diets were distinct.
Our MAG-centric multi-omic approach gave clear indications as to what effect the AcGGM fibre had on putative butyrate-producing Roseburia and Faecalibacterium populations in the distal gut of pigs. Ten MAGs clustered with representative Roseburia spp. genomes (Fig. 3), which reflected the multiple Roseburia-affiliated phylotypes that were predicted with our 16S rRNA gene analysis. (Supplementary Fig. 3). For the most part, a lower number of Roseburia-affiliated proteins were detected in AcGGM-fed pigs (Fig. 3), reintegrating our initial observations that AcGGM negatively affected Roseburia populations (Fig. 2b). However, within one specific Roseburia-affiliated (MAG041), we detected a higher number of total affiliated protein groups in the 4% AcGGM pig samples compared to the control (avg=94 v 53, Supplementary Table 5). Closer examination of MAG041 revealed a putative CE2/CEX-containing mannan-degrading PUL that was absent in the other Roseburia-affiliated MAGs and was differentially expressed in the AcGGM diet (Fig. 4). Importantly, the MAG041 mannan PUL encoded gene synteny to the R. intestinalis strain L1-82 PUL we recently biochemically characterized in detail4 (Fig. 4). The predicted multi-modular mannanase (CBM27-GH26-CBM23) in the MAG041 mannan PUL is homologous to the GH26 in R. intestinalis L1-82 (48% identity over 87% of the sequence), and can be presumed to fulfill the same function – “selfishly” breaking down AcGGM fibres at the cell surf prior intracellular transport. Besides the detection of GH26 and esterases in AcGGM-fed pigs, other mannan-specific enzymes also responded to the dietary shift within the MAG041 mannan PUL, including a phosphoglucomutase, a multiple-sugar binding protein, GH130.1 4-O-β-D-mannosyl-D-glucose phosphorylase and a GH130.2 β-1,4-manno-oligosaccharide phosphorylase (Fig. 4).
In contrast to Roseburia-affiliated MAGs, only one MAG clustered with F. prausnitzii (MAG243, Fig. 3), inferring that the multiple phylotypes that were predicted with our 16S rRNA gene data (Supplementary Fig. 3) encode high genome similarity and likely co-assembled into a representative population-level MAG. Our metaproteomic analysis predicted that MAG243 was more active in the distal gut of pigs fed 4% AcGGM (avg=32 vs 77 detected protein groups, Fig. 3, Supplementary Table 5), as was its CE2/CEX-containing mannan PUL, which was broadly detectable in the presence of AcGGM but absent in the control samples (Fig. 4, Supplementary Table 6). While, the MAG243 mannan PUL contained two GH130 manno-oligophosphorylases, a mannose 6-phosphate isomerase, phosphoglucomutase and two carbohydrate esterase (CEX and CE2), it lacked a GH26 mannanase representative, which suggests that F. prausnitzii is likely preferentially targeting the shorter acetylated manno-oligosaccharides that form part of the AcGGM creation (Fig. 1). In addition to the mannan-PULs of MAG041 and MAG243 being activated in AcGGM-fed pigs, their butyrate-producing pathway were also detected at high levels, based on label-free quantification (LFQ) scores of detected proteins (Fig. 5, Supplementary Table 6), suggesting that both populations can convert mannan to butyrate (Supplementary Table 6).
Removal of 2-O-, 3-O- and 6-O-mannose acetylations are the key enzymatic activities required for accessing AcGGM
A crucial step in the utilization of mannans as an energy source is the deacetylation of 2-O-, 3-O- and 6-O-mannose residues, which subsequently grants access to the sugar-containing backbone of the fibre. In R. intestinalis L1-82, AcGGM deacetylation occurs via the synergistic actions of two carbohydrate esterases (RiCE2 and RiCEX) that exert complementary specificities6. MAG041 and MAG243 both encoded CE2 homologues within their mannan PULs, sharing 63% and 31% identity (respectively) to RiCE2, which has demonstrated activity on 3O-, (4O-) and 6O-acetylations, and is mannan specific6. For CEX, MAG041 and MAG243 homologs shared 65% and 46% identity (respectively) with RiCEX, including the active site residues and the aromatic stacking tryptophan (Trp326), which in RiCEX are associated with 2-O-acetylation specificity6. Broader screens of our MAG data revealed other CE2/CEX-containing PULs within Firmicute-affiliated MAGs from the pig colon microbiome (Supplementary Fig. 5), however aside from MAG041 and MAG243, they originated from populations that were not as metabolically detectable via metaproteomics in any of the control or AcGGM diets (Fig. 3). Finally, the differential proteomic detection of MAG041 and MAG243 CEs in pigs fed AcGGM diets (Fig. 4), strengthened our hypothesis that both these populations can accommodate the unique features of the AcGGM fibre and are actively engaging in its utilization in vivo.
Despite the apparent specificity of AcGGM, we observed a pronounced “butterfly effect”
Besides the activation of specific butyrate-producers that encoded CE2/CEX-mannan PULs, the AcGGM diet also altered protein expression in multiple populations within the distal gut of weaned piglets. For the most part, proteins originating from the fibre-degrading Prevotella were more detectable in AcGGM-fed pigs (Fig. 3), with MAG191 in particular accountable for the highest levels of detectable proteins in our datasets (Supplementary Table 5). Pathway annotation of abundantly detected Prevotella populations (such as MAG191, MAG196, MAG285, see Fig. 5, Supplementary Table 6) indicated active metabolism of dietary fibres such xylans, starch, cellobiose, α-galactans and mannose sugars as well as acetate, succinate and/or propionate production, which were all detected with higher LFQ scores in AcGGM-fed pigs (Fig. 5, Supplementary Table 6). Several mannan-targeting PULs were identified in Prevotella-affiliated MAGs that were configured in an archetypical “Bacteroidetes-format”, which combines outer-membrane transport and carbohydrate-binding lipoproteins (SusC/D-like) as well as CAZymes20 (Supplementary Fig. 5a). In particular, a PUL recovered from MAG196 encoded predicted SusC/D-like lipoproteins, mannanases (GH26, GH5_7), mannosyl-phosphorylases (GH130) and an esterase, although neither the mannanases nor the esterase were detected in the metaproteomes recovered from the AcGGM-fed pigs (Fig. 5, Supplementary Fig. 5, Supplementary Table 6). In addition, we speculate that MAG196 and MAG191 are perhaps capable of metabolizing elements of the AcGGM fibre such as the α-galactose side-chain or deacetylated manno-oligosaccharides, which was inferred via detected GH36 and GH130 representatives (Fig. 5, Supplementary Fig. 5, Supplementary Table 6).
In spite of the specificity of the AcGGM fibre to match selected mechanistic features of our target populations, our evidence suggests the effect of the AcGGM dietary intervention reverberated further down the microbial trophic networks that support conversion of dietary fibre into keystone short-chain fatty acids (SCFAs) that are of nutritional value to the host animal. Mirroring our 16S rRNA gene analysis, MAGs affiliated to the genera Dialister (MAG150), Catenibacterium (MAG048), Lactobacillus (MAG013) and Megasphaera (MAG053) demonstrated the largest transformation in response to the AcGGM diet, although none were found to encode CE2/CEX-containing mannan PULs (Fig. 2-3, Fig. 5, Supplementary Table 6). However, the MAG048 proteome increased in detection by ~4 fold in AcGGM-fed pigs, which included a putative sugar phosphotransferase system (PTS) and GH1 phospho-β-glucosidases (EC 3.2.1.86) that are predicted to catalyze the phosphorylation of di-oligosaccharides (such as cellobiose and mannobiose) and hydrolyze the PTS-transported sugars into D-glucose and D-glucose 6-phosphate. Concomitantly, glycolysis and acetate-producing pathways from MAG048 were also highly detected in AcGGM-fed pigs (Fig. 5, Supplementary Table 6), suggesting this population is advantageously consuming oligosaccharides that have either been generated via the actions of other fibre-degrading populations or have become available via new ecological niches that have been created via the AcGGM-derived structural shifts in the microbiome. Non-fibre degrading populations also reacted to the AcGGM diet, with both MAG053 and MAG150 predicted via our multi-omics approach to metabolize SCFAs such as lactate and succinate that were generated “in-house” by Lactobacillus- and Prevotella-affiliated populations, and produce butyrate and propionate, respectively (Fig. 5, Supplementary Table 6).
CONCLUSION
Mannan from woody biomass has great potential in functional diets for both human and livestock alike. It is cheap, renewable, does not compete with food sources, and common extraction methods such as SE and UF can be readily adapted to industrial scale production. The AcGGM fibre described here could be produced from forestry by-products such as sawdust or lumber waste, while the solid biomass discarded in this production process contains valuable, steam-exploded cellulose, suitable for inclusion in other processes such as the production of biofuels and platform chemicals via enzymatic treatment and fermentation.
Here, we characterize the impact of the AcGGM structural configuration on microbial uptake and metabolism within the distal regions of the digestive tract, with the key driver of AcGGM selectivity being the presence of acetylations of mannan, as well as carbohydrate composition and size21. Preserving the complexity of AcGGM resulted in a highly specific, dose-dependent shift in the composition of the colon microbiome from weaned piglets, with no diverse effect on host growth performance or health status. Our integrated multi-omics analysis, showed that the AcGGM fibre activated a metabolic response in specific Roseburia and Faecalibacterium populations in vivo, as it did in the previous in vitro experiments4,7, with both populations expressing proteins from highly sophisticated CE2/CEX-containing mannan PULs that are homologous to a biochemically characterized representative in R. intestinalis L1-82. In conclusion, our data provide a foundation for modulatory strategies to design and match custom dietary fibres to unique enzymatic features of their target organisms, although they bring awareness to the fact that the greater network of interconnected metabolic exchanges and trophic structures inherent to the gut microbiome are highly susceptible to minor dietary interventions.
DATA AVAILABILITY
All sequencing reads have been deposited at the NMBU sequence read archive under BioProject PRJNA574295, with specific numbers listed in Supplementary Table 3. All annotated MAGs are publicly available via doi: 10.6084/m9.figshare.9816581. The proteomics data has been deposited to the ProteomeXchange Consortium (http://proteomecentral.proteomexchange.org) via the PRIDE partner repository22 with the dataset identifier PXD015757.
AUTHOR CONTRIBUTIONS
L.M., J.C.G, P.B.P. and B.W. conceived the study, performed the primary analysis of the data and wrote the paper (with input from all authors). L.L. and M.Ø. designed, performed and analyzed the animal experiments. S.L.L.R., L.H.H., M.Ø.A. and J.D. generated the data and contributed to the data analyses. N.T., V.L. and B.H. annotated and curated the MAGs and identified carbohydrate-active enzymes.
COMPETING INTERESTS
The authors declare there are no competing financial interests in relation to the work described.
MATERIALS AND METHODS
Animals, diets and experimental design
Animal care protocols and experimental procedures were approved by the Norwegian Animal Research Authority, approval no. 17/9496, FOTS ID 11314 and treated according to institutional guidelines. A total of 48 cross bred piglets (Landrace x Yorkshire), 24 male and 24 female, with an average initial body weight (BW) of 9,8 ± 0,5 kg, weaned at 28 days of age were sorted by litter, sex and weight and randomly divided into 12 groups of four animals each (one diet per pen), but pigs were housed individually during mealtime. The animals were housed in an environmentally controlled facility with plastic flooring and a mechanical ventilation system. The temperature of the room was maintained at 22°C.
Piglets were fed cereal-based diets containing increasing levels of AcGGM in the diets (1, 2 and 4%). Diets were pelleted with a 3 mm diameter feed formulated to meet the requirements for indispensable amino acids and all other nutrients (NRC, 2012). The composition of diets is listed in Supplementary Table 7. Pigs were fed semi-ad libitum twice a day at a feeding level equal to about 5% of body weight. To evaluate growth performance, the BW of each pig was recorded at the beginning and once a week. Feed consumption were recorded on an individual pig basis during the experiment to calculate individual weight gain and feed intake. After each meal, feed leftovers were registered, dried and subtracted from the total feed intake.
Production of AcGGM
AcGGM oligosaccharides for the feeding trial were produced from Norway spruce chips milled with a hammer mill to <2 mm size. Wood chips were then steam-exploded on a small pilot scale steam explosion rig (100L reactor vessel) at the Norwegian University of Life Sciences (NMBU). The steam explosion was conducted in batches of approximately 6kg dry matter, 14.5 bar pressure (equivalent to 200° C), with 10 minutes residence time. The pH in the collected biomass slurry after the steam explosion was ~3.7, which corresponds to a combined severity factor R’0=1.70 for the process. The severity was calculated by R′0 = (10−pH) × (t × e(Texp−100)/14.75)23. Steam exploded wood was collected in 50 L plastic buckets that were topped up with hot (~70° C) water. The slurry was transferred to a 60Lcider press (Speidel, Germany) and the liquid fraction was pressed out. Milled wood was collected, soaked in hot water again, and pressed for the second time. The liquid fraction was collected and recirculated through a bag filter 50µm pore WE50P2VWR (Allied filter systems, England) partly filled with the wood particles as a filter aid. Once free of floating wood particles, the liquid fraction of hemicellulose was filtered through a 5-kDa spiral wound Polysulphone/polyethersulphone ultrafiltration membrane, GR99PE polyester (Alfa Laval, Denmark) that was deliberately fouled to prevent larger oligosaccharides from running through the permeate, using a GEA pilot-scale filtration system Model L (GEA, Denmark). The fraction retained by the membrane was concentrated by nanofiltration using a TriSep XN 45, which had a higher efficiency for permeating water. The filtrate was further concentrated by vacuum evaporation (set to 65 °C) and the concentrate was freeze-dried and homogenized with a grain mill. The final product consisted of 0.9% rhamnose, 2.7% arabinose, 13.7% xylose, 58.9% mannose, 14.9% glucose and 9.4% galactose (determined by gas chromatography as alditol acetates after sulfuric acid hydrolysis as described previously24). AcGGM contained 0.73 % ash and 2.4% protein (quantified from total nitrogen by the Kjeldahl method). The Man:Glc:Gal ratio in the mannan was 4:1:0.6, and the DA=0.36 (determined by acetate release from NaOH treated AcGGM by the same method as described in the SCFA section below). The dry matter content was determined by drying 0.2g of sample at 105°C for 20 hours. The remaining sample was burned at 600°C for 24 hours in an oven (Carbolite, Sheffield, England) to determine ash content. All measurements were performed in triplicates.
Fecal scoring
During the experiment, fecal consistency was assessed using a scoring system developed by Pedersen and Toft25 to improve and help standardize current protocols for clinical characterization of fecal consistency. The scoring was based on the following 4 consistency categories: score 1 = firm and shaped, score 2 = soft and shaped, score 3 = loose and score 4 = watery. Samples with score 3 or 4 are considered diarrheic. Daily fecal scores for each pen were recorded throughout the trial.
pH measurements
The pH of digesta samples from duodenum, jejunum, ileum, cecum and colon were measured immediately after slaughter. Samples were placed in universal containers and pH measurements made using an Inolab pH7110 pH meter (WTW, Germany).
Blood sampling and flow cytometry
Blood samples were collected from the same six piglets per diet at 0, 7 and 27 feeding days. The blood samples were taken 1-2 hours post-prandial by venipuncture in the jugular vein while pigs were kept on their backs. Non-heparinised and K3EDTA vacuum tubes (Beckman Dickson Vacutainer System) were used to recollect serum and whole blood. Serum was isolated immediately by centrifugation at 1,500 x g at 4°C for 15min. Serum samples were split in PCR-tubes (200 µL) and stored at −80°C until analysis. Hematological and clinical analyses were performed with an Advia® 2120 Hematology System using Advia 2120 MultiSpecies System Software and clinical chemistry analyses were performed with Advia 1800 Chemistry System (both from Siemens AG Healthcare Sector).
For flow cytometry analysis, whole blood was diluted 1:1 in RPMI 1640 and kept on ice until single cells isolation. For the isolation of peripheral blood mononuclear cells (PBMCs) blood was purified by centrifugation in a Ficoll gradient (Kreuzer et al. 2012). Then, isolated PBMCs were incubated with Fixable Yellow Dead Cell Stain Kit (Life Technologies, Thermo Fisher Scientific Inc.) followed by primary monoclonal antibodies (mAbs), brief incubation with 30% normal pig serum to block Fc-receptors, and finally fluorescence-labeled secondary antibodies (Abcam plc, UK). To detect the intracellular CD3 epitope, surface-labeled cells were permeabilized with Intracellular Fixation and Permeabilization Buffer Set (eBioscience, Affymetrix Inc.) according to the manufacturer’s instructions. Labeled cells were analyzed on a Gallios Flow Cytometer (Beckman Coulter, Inc.) and data were processed using Kaluza 1.5 software (both Beckman Coulter, Inc.). Cell gates were designed to select for single and viable mononuclear cells. Defined markers were used to identify the different immune subpopulations. For monocytes, antibodies against CD45, CD3, CD14, CD163 and MHCII were used. To analyze regulatory T cells (T reg) the following antibodies were used: CD45, CD3, TCR γ/δ, CD4, CD8, FOXp3 and CD25, while CD45, CD8, NKp46, CD4, CD8, Ki67 and CD27 were used to identify T and NK cells.
Analysis of Serum Cytokines: MULTIPLEX
Expression of GMCSF, IFNG, IL-1A, IL1B, IL-1RA, IL-2, IL-4, IL-6, IL-8, IL-10, IL-12, IL-18 and TNFα were measured in serum samples using MILLIPLEX MAP Porcine Cytokine and Chemokine Magnetic Bead Panel - Immunology Multiplex Assay (Merck Millipore) following the manufacturer instructions. The measurement was performed using a Bio-Plex MAGPIX Multiplex Reader (BIO-RAD).
Small Intestine Morphology
The samples of the small intestine were collected on day 0 and 28 for determination of intestinal morphology and integrity. Intestinal morphological measurements included the following indices: villus height (VH), crypt depth (CD) and VH:CD. Mean values of VH, CD and their ratio were calculated. Histology evaluation was performed by the Veterinary Histophalogy Center, VeHiCe, Chile.
Microbial Sampling
Fecal samples were collected from 6 piglets per experimental group (n=24) at days 0, 7, 14, 21, and 27 post-weaning. At the end of the trial, all piglets (n=48) were sacrificed, and samples were collected from the lumen of the duodenum, jejunum, ileum, cecum, and colon. Samples were obtained within the first 15 minutes after the piglets were sacrificed and the samples were flash-frozen in liquid nitrogen and stored at −80°C until DNA extraction.
DNA Extraction
DNA was extracted with a MagAttract PowerMicrobiome DNA/RNA Kit (MO BIO Laboratories Inc., Carlsbad, CA, USA) according to the manufacturer instructions, except for the bead beating step where we used a FastPrep-96 Homogenizer (MP Biomedicals LLC., Santa Ana, CA, USA) at maximum intensity for a total of 2 minutes in 4 pulses of 30s with a 5 minute cooling period between each pulse. A KingFisher Flex DNA extraction robot was used for the automated steps of the protocol. The extracted nucleic acids were quantified with a Qubit Fluorimeter and the Qubit dsDNA BR Assay Kit (Thermo Fisher Scientific, Waltham, MA, USA) and stored at −80°C.
16S Amplicon Sequencing and Analysis
16S amplicon sequence data was obtained for all fecal and intestinal samples. The V3-V4 region of the 16S rRNA gene was PCR amplified using the primers Pro341F (5’-CCT ACG GGN BGC ASC AG-3’) and Pro805R (5’-GAC TAC NVG GGT ATC TAA TCC-3’), to which the MiSeq adaptors were additionally incorporated on the 5’ ends26. The 25 µL PCR reactions consisted of 1X iProof High-Fidelity Master Mix (Biorad, Hercules, CA, USA), 0.25 µM primers, and 5 ng template DNA. PCR thermal cycling began with a hot start step at 98 °C for 180 s and was followed by 25 cycles of 98 °C denaturation for 30 s, 55 °C annealing for 30 s, and 72 °C extension for 30 s, followed by a final, 300 s extension step at 72 °C. Amplicons were individually purified with AMPure XP beads (Beckman Coulter, Indianapolis, IN, USA) and indexed with the Nextera XT Index Kit v2 (Illumina, San Diego, CA, USA) according to the Illumina protocol for 16S metagenomic sequencing library preparation. Next, equal volumes from each indexing reaction were pooled together, and the pool was purified with AMPure XP beads. The purified amplicon pool was then quantified with a Qubit Fluorimeter, diluted, mixed with 15% PhiX Control v3 (Illumina), and denatured according to the aforementioned Illumina protocol. The denatured library was sequenced on the Illumina MiSeq platform using the MiSeq Reagent Kit v3 (600 cycle). Data were output from the sequencer as demultiplexed FASTQ format files.
Processing of the data was done with a combination of standalone programs, QIIME27 MOTHUR28 and the R package Phyloseq29. To process the data, the paired end reads for each sample were merged with PEAR30, specifying a minimum assembly length 400, maximum assembly length 575, minimum overlap 50, and no statistical test. Then, PRINSEQ31 version 0.20.4 was used to filter low quality reads by requiring a minimum quality score of 10 for all bases and a minimum mean quality of 30. Primer sequences were trimmed in MOTHUR version 1.36.1, and chimeric sequences were identified and filtered out using QIIME version 1.9.1. Next, open reference OTU0.97 clustering32 was performed with VSEARCH33 version 2.3.2 and the Silva database34 release 128 as the taxonomy reference. Then, the QIIME core diversity analyses script was run. Differentially abundant phylotypes were identified in both cecum and colon for the control vs. 4% AcGGM samples using both the MetagenomeSeq fitZIG and DESeq2 negative binomial algorithms via the QIIME wrapper. The OTU table, phylogenetic tree, representative sequences, and taxonomy from QIIME were incorporated along with the sample metadata into a Phyloseq version 1.22.3 object in R for data exploration and visualization.
Whole Metagenome Sequencing and Analysis
Whole metagenome sequencing was performed at the Norwegian Sequencing Centre on 2 lanes of the Illumina HiSeq 4000 to generate 2 × 150 paired-end reads. TruSeq PCR-free libraries were prepared for 12 control and 12 AcGGM (4%) samples from the colon. All 24 samples were run in both lanes to eliminate the potential for lane-specific sequencing bias. FASTQ format files were received from the sequencing center, and prior to assembly, these were quality filtered with Trimmomatic35 version 0.36 whereby TruSeq adaptor sequences were eliminated, sequences were required to have an average quality score above 20, leading and trailing bases with quality below 20 were removed, sequences with average quality score below 15 in a 4-base sliding window were trimmed, and the minimum read length was required to be 36 bases. Individual sample assembly was accomplished with metaSPAdes36 version 3.11.1. MegaHIT37 version 1.1.3 was used for co-assembly of all 24 samples together as well as co-assembly of the 12 control samples together and the 12 4% AcGGM samples together. MetaBAT38 version 0.26.3 was used to bin the assemblies, and dRep39 was used to dereplicate the multiple assembly and binning combinations to produce an optimal set of MAGs. MASH40 version 2.0 used to compare the similarity of the 24 metagenomes by calculating pairwise Jaccard distances which were imported into R for NMDS ordination and visualization. Completeness and contamination was determined for each MAG using CheckM41 version 1.0.7. Feature and functional annotation were completed with the Prokka pipeline42 version 1.12, and the predicted protein sequences from all 355 MAGs were concatenated to create the metaproteomics reference database. Resulting annotated open reading frames (ORFs) were retrieved, further annotated for CAZymes using the CAZy annotation pipeline with libraries from July 2018 database release43,44, and subsequently used as a reference database for the metaproteomics (with the exception of glycosyltransferases).
Metaproteomics
Proteins were extracted from each sample in quadruplicate by the following method. An aliquot (1 g) of colon digesta from pigs fed either a control diet or a diet supplemented with 4% β-mannan was dissolved 1:1 (w/v) in 50 mM TrisHCl, pH 8.4.
Lysis was performed using a bead-beating approach whereby glass beads (size ≤ 106 µm) were added to the colon digesta slurry and cells were disrupted in 3 x 60 second cycles using a FastPrep24 (MP Biomedicals, Santa Ana, CA, USA). Debris were removed by centrifugation at 16.600 x g for 20 minutes and proteins were precipitated overnight in 16% ice-cold TCA. The next day, proteins were dissolved in 100 μL 50 mM TrisHCl, pH 8.4 and concentration was determined using the Bradford protein assay (Bradford Laboratories, USA) using bovine serum albumin as a standard. Fifty milligrams of protein was prepared in SDS sample buffer, separated by SDS-PAGE using an Any-kD Mini-PROTEAN gel (Bio-Rad Laboratories, Hercules, CA, USA) and stained using Coomassie Brilliant Blue R250. The gel was cut into 6 slices and reduced, alkylated and digested as described previously45. Prior to mass spectrometry, peptides were desalted using C18 ZipTips (Merck Millipore, Darmstadt, Germany) according to the manufacturer’s instructions.
The peptides were analyzed by nanoLC-MS/MS as described previously, using a Q-Exactive hybrid quadupole orbitrap mass spectrometer (Thermo Scientific, Bremen, Germany)46, and the acquired raw data was analyzed using MaxQuant47 version 1.4.1.2. Proteins were quantified using the MaxLFQ algorithm48. The data was searched against a sample-specific database (602.947 protein sequences), generated from the 355 metagenome assembled genomes (MAGs), and against the pig genome (Sus scrofa domesticus). In addition, common contaminants such as human keratins, trypsin and bovine serum albumin were concatenated to the database as well as reversed sequences of all protein entries for estimation of false discovery rates. Protein N-terminal acetylation, oxidation of methionine, conversion of glutamine to pyro glutamic acid, and deamination of asparagine and glutamine were used as variable modifications, while carbamidomethylation of cysteine residues was used as a fixed modification. Trypsin was used as digestion enzyme and two missed cleavages were allowed. All identifications were filtered in order to achieve a protein false discovery rate (FDR) of 1% using the target-decoy strategy. For a protein to be considered valid, we required the protein to be both identified and quantified in both replicates, and in addition, we required at least one unique peptide per protein and at least two peptides in total for every protein. The output from MaxQuant was further explored in Perseus version 1.6.0.7 where filtering, data transformation, and imputation were performed, and visualizations including heatmaps, hierarchical clustering, and volcano plots (for identification of differentially abundant proteins between the mannan and control groups) were made.
Genome tree
Phylogenetic analysis was performed using a block of 22 universal ribosomal proteins (30S ribosomal protein L1, L2, L4-L6, L10, L11, L14, L15,L18 and 50S ribosomal protein S3, S5, S7-S13, S15, S17, S19)49,50. In addition to the MAGs, we recruited 239 reference genomes for phylogenetic resolution. These genomes were selected based on preliminary examination of the assembled metagenome using metaQUAST51. The selection of reference genomes were annotated using the Prokka pipeline, uniformly with annotation of the MAGs. All identified ribosomal protein sequences were aligned separately with MUSCLE v3.8.3152, and manually checked for duplications and misaligned sequences. Divergent regions and poorly aligned positions were further eliminated using GBlocks53, and the refined alignment were concatenated using catfasta2phyml.pl (https://github.com/nylander/catfasta2phyml) with the parameter ‘-c’ to replace missing ribosomal proteins with gaps (-). The maximum likelihood-based phylogenetic analysis of the concatenated ribosomal proteins was inferred using RAxML version 8.2.1254 (raxmlHPC-SSE3 under PROTGAMMA distributed model with WAG substitution matrix) and support values determined using 100 bootstrap replicates. The tree was rooted to the Euryarchaeota phylum and visualized using iTOL55. Clades of reference genomes with only distant phylogenetic relation to the MAGs were collapsed to refine the final tree in Fig. 3. The complete tree is available in Newick format as Supplementary Dataset 1.
ACKNOWLEDGEMENTS
We are grateful for support from The Research Council of Norway (Bionær program 244259, FRIPRO program, PBP: 250479), as well as the European Research Commission Starting Grant Fellowship (awarded to PBP; 336355 - MicroDE). The sequencing service was provided by the Norwegian Sequencing Centre (www.sequencing.uio.no), a national technology platform hosted by the University of Oslo and supported by the “Functional Genomics” and “Infrastructure” programs of the Research Council of Norway and the Southeastern Regional Health Authorities.