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 tailored an acetylated galactoglucomannan (AcGGM) fibre to match unique enzymatic capabilities of Roseburia and Faecalibacterium species, both renowned butyrate-producing gut commensals. The accuracy of AcGGM was tested within the complex endogenous gut microbiome of pigs, wherein we 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. However, AcGGM-inclusion also manifested 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 potential to create greater modulatory effects at large.
MAIN TEXT
Microbiota-directed foods (MDFs) have emerged as a strategy to modulate the gut microbiome, as diet has distinct and rapid effects on microbiome composition and function1,2. MDFs by definition are not broadly metabolized, but rather elicit a targeted metabolic response in specific indigenous microbiota that confers benefits to their host. This in itself presents a challenge; as many newly identified MDF target organisms, such as beneficial butyrate-producing (i.e. butyrogenic) Roseburia and Faecalibacterium spp.3,4, 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 consume5,6, which creates new conceptual MDF 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 sugars5,7,8. 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 degradation9. 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 butyrate5. Within this mannan PUL, a carbohydrate esterase (CE) family 2 (RiCE2) removes 3-O-, and 6-O-acetylations on mannan, whereas a novel CE family 17 (RiCE17) removes the axially oriented 2-O-hydroxyl acetylations9, which are distinctive features found in limited mannan moieties and inaccessible to most of the characterized bacterial esterases present in the gut microbiome. Closer genome examinations have revealed that putative CE2/CE17-containing mannan PULs are in fact prominent within many butyrate-producers including Roseburia spp., Faecalibacterium prausnitzii, Ruminococcus gnavus, Coprococcus eutactus and Butyrivibrio fibrisolvens5,10. 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 diseases11,12. These collective findings thus raised the question: could a custom MDF fibre that was tailored to match these specialized enzymatic capabilities selectively engage butyrate-producers in a complex microbiome ecosystem?
2-O-acetylated mannans are found in a limited number of characterized western dietary fibre sources (i.e. tomatoes13 and coffee5), 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)14. Here, we have utilized controlled steam explosion (SE), followed by ultrafiltration (UF) fractionation to extract complex AcGGM from spruce wood. Processing conditions were selected to tailor the fibre with a high degree of galactose branching’s and 2-O-, 3-O- and 6-O-acetylations15, which is amenable to inclusion as an MDF in animal feed production. Previously, the MDF concept has matched polysaccharides with Bacteroides-encoded PULs to demonstrate the creation of exclusive metabolic niches1,16. In particular, Shepherd et al engrafted exogenous strains in mice via their rare PUL-encoded enzymatic capabilites16, whereas Patnode et al1 illustrated that bioactive carbohydrates were found to target particular Bacteroides species in a defined consortium of 20 human gut microbial species in gnotobiotic mice. However, what is less understood from MDF studies to date, is (1) can the MDF concept be applied to target indigenous populations within a complex endogenous microbiome, and (2) what are the broader secondary community effects if a targeted population is stimulated, i.e. are new niches created and/or existing ones closed?
Here, we tested whether our AcGGM fibre could specifically target beneficial Firmicutes species Roseburia and Faecalibacterium within a “real-world” gut ecosystem that consists of 100-1000’s of different species. To do this, we fed weaned piglets diets containing varying AcGGM levels over a 28-day period that extended from their first meal after sow separation until an adapted, fibre-degrading microbiome was established. Temporal changes in the microbiome were monitored using metagenomics, which phylogenetically and functionally resolved the genomes of indigenous microbiota. Finally, quantitative metaproteomic analysis and carbohydrate microarrays were used to visualize the metabolic and enzymatic responses of the different microbiota to the varying AcGGM exposure. This approach demonstrates how the activity of specific beneficial microbiota can be directly stimulated while simultaneously deciphering the secondary, trophic effects on other populations and metabolic niches, with both aspects having broader implications for developing strategies to effectively modulate the gut microbiome.
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, and a large degree of esterification of the mannose residues by 2-O- and 3-O-, and 6-O-acetylations14 (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. The degree of polymerization (DP) of the AcGGM fibre ranged from β-mannooligosaccharides with DP of 2-10 to manno-polysaccharides (DP of ≥11), with both exhibiting degree of acetylation (DA = 0.35). Crucially, this DA value was higher than previous iterations of the fibre (DA = 0.28)7, and its acetylation patterns (previously determined9) matched the enzymatic capabilities of mannan PULs encoded in human gut Roseburia and Faecalibacterium spp.5,10. We therefore predicted that our AcGGM fibre would match representatives of the same populations that are indigenous to porcine gut ecosystems17,18 (Fig. 1f).
Spruce AcGGM altered the gut microbiome of weaned piglets in a dose-response manner
We previously demonstrated that varieties of AcGGM can be metabolized by pure cultures of Roseburia intestinalis L1-825, in vitro enrichments with human gut butyrate-producers10 and “mini-microbiota” within gnotobiotic mice5. Here, we wanted to test our AcGGM fibre’s accuracy and ability to elicit a specific response in indigenous representatives of our target populations within a highly complex and competitive endogenous microbiome. In total, four separate cohorts of twelve weaned piglets were given a pelleted basal 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. We chose a dietary supplementation strategy, which is commonly applied to piglets during the post-weaning period to offset their particularly low feed intake and enhance gut health and growth performance20. Cautious measures were made to eliminate potential pen biases (see details in materials and methods), and fecal samples as well as animal 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 all intestinal regions (duodenum, jejunum, ileum, cecum and colon) for down-stream analysis (Supplementary Table 1, Supplementary Dataset 1). Measurements of major short-chain fatty acids (SCFAs) in the cecum and colon showed a trend of incremental increases of absolute butyrate levels as AcGGM levels were increased (Fig. 2). Butyrate is commonly measured in the colon at levels that are approximately one-third of acetate, the most abundant SCFA21. Compared to control animals, the mean molar ratio of butyrate to acetate in samples from pigs fed 4% AcGGM increased in both the cecum (from 0.29:1 to 0.37:1 mM, p-value: 0.03) and colon (from 0.35:1 to 0.45:1 mM, p-value: 0.002) (Supplementary Dataset 1), suggesting that AcGGM inclusion promotes butyrogenic fermentation. While changes in SCFA and microbiome composition (Fig. 2-3) resulted from AcGGM inclusion, we observed no adverse effects 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 Dataset 1).
Monitoring of temporal microbiome changes using 16S rRNA gene analysis over the month-long trial indicated that 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). Spatial changes were also examined at the final sampling day and showed typical patterns, whereby the structure of the gut microbiome varied by gut region (Supplementary Fig. 2a, Supplementary Fig. 3, Supplementary Fig. 4). 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. 3a, Supplementary Dataset 2). To determine the effect AcGGM had on microbiome function, we also closely examined the 355 metagenome assembled genomes (MAGs, Supplementary Fig. 5a) that were reconstructed from metagenomic data generated from the colon samples of each pig fed the control and 4% AcGGM diets (Supplementary Table 2). Taxonomic affiliation of the MAGs was calculated via GTDB-Tk (Supplementary Dataset 3), while phylogeny was inferred from a concatenated ribosomal protein tree (Newick format available in Supplementary Dataset 4) that was constructed using MAGs from this study and 293 closely related reference genomes.
Our target butyrogenic populations produced mixed results, whereby the 16S rRNA gene relative abundance of Faecalibacterium affiliated phylotypes increased in response to increasing levels of AcGGM (Fig. 3c, Supplementary Fig. 3c), whereas Roseburia affiliated phylotypes seemingly decreased (Fig. 3b, Supplementary Fig. 3b). However, detailed analysis of Roseburia-affiliated MAGs showed that specific phylotypes that encoded AcGGM-specific PULs were indeed stimulated by the AcGGM fibre (Fig. 3b). Reputable fibre-fermenting populations affiliated to Prevotella22 also showed varying responses (Fig. 3g), with 16S rRNA gene relative abundance of individual phylotypes increasing from 4% to 12% between the control and 4% AcGGM inclusion in both colon and cecum (Supplementary Fig. 3a). Interestingly, relative abundance estimates of both 16S-rRNA gene OTUs and MAGs indicated that phylotypes affiliated to non-fibre degrading taxa, such as Catenibacterium23, Dialister24 and Megasphaera, demonstrated some of the highest dose-dependent increases in relative abundance in response to AcGGM (Fig. 3d-f), indicating that other underlying factors are likely dictating microbiome structure, besides fibre degradation.
Targeted mannan PULs and butyrogenic activities were detected in Faecalibacterium- and Roseburia-affiliated populations within the colon of AcGGM-fed pigs
Because our primary goal was to elucidate whether our target, butyrogenic populations were activated in response to AcGGM, we annotated MAGs affiliated to butyrate-producers and conducted metaproteomic analysis with label-free quantification (LFQ) on randomly selected colon samples from four control and four 4% AcGGM fed pigs (Fig. 4a-b), and mapped 8515 detected protein groups back to our MAGs to identify functionally active populations (Fig. 4c) (Supplementary Datasets 5-6). Community-wide analysis of the MAG genetic content (Supplementary Fig. 5b) from each sample and clustering analysis of their detected proteins (Fig. 4a-b) 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 butyrogenic Roseburia and Faecalibacterium populations in the distal gut of pigs. Fifteen MAGs clustered with representative Roseburia spp. genomes (Fig. 3b, Fig. 4), which reflected the multiple Roseburia-affiliated phylotypes that were predicted with our 16S rRNA gene analysis (Supplementary Fig. 3b, Supplementary Dataset 2). In general, the relative abundance of Roseburia-affiliated MAGs (Fig. 3b) and detected proteins (Fig. 4c) were observed at either static or lower levels in AcGGM-fed pigs, reiterating our initial 16S rRNA gene observations that AcGGM negatively affected Roseburia populations (Fig. 3b). However, one specific Roseburia-affiliated population (MAG041) was detected at significantly higher abundance levels (p=0.0016) (Fig. 3b), and its detected proteins were enriched (p<0.001) in the 4% AcGGM pig samples compared to the control (Fig. 4b-c, Supplementary Dataset 7).
Closer examination of MAG041 revealed a putative CE2/CE17-containing mannan-degrading PUL that was absent in the other Roseburia-affiliated MAGs and was differentially expressed in the AcGGM diet (Fig. 5). Importantly, the MAG041 mannan PUL exhibited gene synteny to the R. intestinalis strain L1-82 PUL whose biochemical properties we recently characterized in detail5 (Fig. 5). 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 surface prior to 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, a GH130.1 4-O-β-D-mannosyl-D-glucose phosphorylase and a GH130.2 β-1,4-manno-oligosaccharide phosphorylase (Fig. 5).
In contrast to Roseburia-affiliated MAGs, only one MAG clustered with F. prausnitzii (MAG243, Fig. 4, Supplementary Dataset 3), implying that the multiple phylotypes that were predicted with our 16S rRNA gene data (Supplementary Fig. 3c, Supplementary Dataset 2) encode high genome similarity and coverage and thus likely co-assembled into a representative population-level MAG. The relative abundance of MAG243 in the distal gut of pigs fed 4% AcGGM increased approximately 5-fold to ∼1.24% (p<0.001) (Fig. 3c), while metaproteomic hierarchical clustering analysis showed that MAG243 proteins were enriched in the same samples (Fig. 4b-c, Supplementary Dataset 7). Similar to MAG041, MAG243 was found to encode a CE2/CE17-containing mannan PUL, which was broadly detectable in the presence of AcGGM but absent in the control samples (Fig. 5, Supplementary Dataset 6). While the MAG243 mannan PUL contained two GH130 manno-oligophosphorylases, a mannose 6-phosphate isomerase, phosphoglucomutase and two carbohydrate esterase (CE17 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 structure (Fig. 1). In addition to the mannan-PULs of MAG041 and MAG243 being activated in AcGGM-fed pigs, their butyrogenic pathways were also detected at high levels, based on label-free quantification (LFQ) scores of detected proteins (Fig. 6-7, Supplementary Dataset 6), suggesting that both populations can convert mannan to butyrate (Supplementary Dataset 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 allows the subsequent breakdown of 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 RiCE17) that exert complementary specificities9. MAG041 and MAG243 both encoded CE2 homologues within their mannan PULs, sharing 63% and 31% identity (respectively) to RiCE2, which has demonstrated activity on 3-O-, (4-O-) and 6-O-acetylations, and is mannan specific9. For CE17, MAG041 and MAG243 homologs shared 65% and 46% identity (respectively) with RiCE17, including the active site residues and the aromatic stacking tryptophan (Trp326), which in RiCE17 are associated with 2-O-acetylation specificity9. Broader screens of our MAG data revealed other CE2/CE17-containing PULs within Firmicute-affiliated MAGs from the pig colon microbiome (Supplementary Fig. 6), however aside from MAG041 and MAG243, they originated from populations whose MAG relative abundance was very low (<0.05%) and metabolic activity was undetectable via metaproteomics in any of the control or AcGGM diets (Fig. 4). Finally, the differential proteomic detection of MAG041 and MAG243 CEs in pigs fed AcGGM diets (Fig. 5), 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.
Beyond the target populations consuming AcGGM, we also observed a pronounced “butterfly effect”
MDFs studies to date have eloquently highlighted that metabolic symmetry between individual fibres and microbiota can be used to stimulate specific populations1,16. However, deeper microbiome effects that result from a target species being enriched via MDFs are poorly understood. Although the specificity of the AcGGM fibre matched selected mechanistic features of our target populations, our data showed that AcGGM dietary intervention reverberated further down the microbial trophic networks that support conversion of dietary fibre into keystone SCFAs that are of nutritional value to the host animal. In particular, we observed varying effects upon different non-AcGGM-degrading butyrogenic populations belonging to clostridial cluster XIVa (Lachnospiraceae: Roseburia spp., Eubacterium rectale, Butyrivibrio and Pseudobutyrivibrio) and cluster IV (Ruminococcaceae: Faecalibacterium prausnitzii-related bacteria)25 (Fig. 6, Supplementary Dataset 6). Specific populations that were closely related to AcGGM-degrading MAG041 were found to be either unaffected (MAG292) or metabolically suppressed (MAG133) in the presence of 4% AcGGM. Indeed, detected proteins from MAG133 were enriched in pigs fed the control diet (adj. p<0.001, Supplementary Dataset 7), and enzymes associated with butyrate production were largely undetected in 4% AcGGM-fed pigs (Fig. 6, Supplementary Dataset 6). In most cases, non-AcGGM degrading, butyrogenic populations were seemingly utilizing sugars found in starch, arabinoxylan and/or arabinogalactan fibres that were detected in the basal feed components using Micro Array Polymer Profiling (MAPP; Supplementary Dataset 8). Broader analysis of our omic data identified multiple abundant populations with similar metabolic capabilities, suggesting that suppressed butyrogenic populations were possibly being outcompeted by Prevotella-affiliated populations such as MAG285 (see below). While SCFA measurements indicated an increase in relative butyrate levels in AcGGM-fed pigs (Fig. 2b), the indirect effects of our MDF on this dynamic functional group are likely dampening an even greater influence of AcGGM-stimulated Roseburia and Faecalibacterium populations upon the broader microbiome and host animal (Fig. 6).
Besides butyrate-producers, MAG-centric metaproteomic enrichment analysis showed that fibre-degrading Prevotella-affiliated populations experienced contrasting effects as a result of AcGGM inclusion. For example, MAG191 was found to account for the highest levels of detectable proteins in our datasets (Fig. 4c, Supplementary Dataset 6), which were enriched in clusters differently detected in the presence of 4% AcGGM (p<0.001, Supplementary Dataset 6). Pathway annotation of AcGGM-enriched Prevotella populations (such as MAG191, MAG196, MAG285, see Fig. 4c, Fig. 7, Supplementary Dataset 6-7) indicated active metabolism of dietary fibres such arabinoxylans, starch, glucans (e.g. cellobiose), α-galactans and mannose sugars (detected via MAPP analysis, Supplementary Dataset 8) as well as acetate, succinate and/or propionate production, which were all detected with higher LFQ scores in AcGGM-fed pigs (Fig. 7, Supplementary Dataset 6). However, many CAZymes and fermentation enzymes from the aforementioned MAGs were also detected across both diets (albeit at varying LFQ values), while other Prevotella populations were specifically enriched in pigs fed the control diet (i.e. MAG034: adj. p<0.001, Supplementary Dataset 7), suggesting that Prevotella-driven baseline consumption of basal feed fibre was occurring irrespective of AcGGM inclusion (Fig. 7, Supplementary Dataset 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 SusC/D-like proteins as well as CAZymes26 (Supplementary Fig. 5a). In particular, a PUL recovered from MAG196 encoded predicted SusC/D-like proteins, 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. 7, Supplementary Fig. 6, Supplementary Dataset 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. 7, Supplementary Fig. 6, Supplementary Dataset 6).
Dietary inclusion of AcGGM also unexpectedly led to pronounced metabolic shifts in numerous phenotypically and phylogenetically diverse microbial populations that are not normally associated with fibre hydrolysis. Mirroring our 16S rRNA gene analysis, MAGs affiliated to the genera Dialister (MAG150), Catenibacterium (MAG048) and Megasphaera (MAG053) were the three highest enriched populations in expression clusters differentially detected in the AcGGM diet (adj. p<0.001), although none were found to encode CE2/CE17-containing mannan PULs (Fig. 3-4, Fig. 7, Supplementary Dataset 6). In particular, the MAG048 proteome 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 disaccharides (such as cellobiose and mannobiose27,28) and hydrolyze the PTS-transported sugars into D-glucose and D-glucose 6-phosphate. Concomitantly, glycolysis and acetogenesis pathways from MAG048 were also highly detected in AcGGM-fed pigs (Fig. 7, Supplementary Dataset 6). Together, this data suggests that certain Catenibacterium populations are advantageously consuming AcGGM-derived disaccharides that were either available as a fraction of the original AcGGM fibre preparation, have 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.
Both MAG053 (Megasphaera) and MAG150 (Dialister) were predicted via our multi-omics approach to metabolize SCFAs such as lactate and succinate that were generated endogenously by Lactobacillus- (i.e. MAG013) and Prevotella- (i.e. MAG285) affiliated populations, and produce butyrate and propionate, respectively (Fig. 7, Supplementary Dataset 6). Neither of these cross-species relationships had any evidence of being directly influenced by AcGGM and are presumably being driven by underlying availabilities of fermentation intermediates (e.g. lactate and succinate) resulting from as-yet uncharacterized niche fluctuations. Collectively, these data highlight that while a highly tailored MDF can directly activate enzymatically equipped populations in an endogenous microbiome, they can also inadvertently alter the metabolisms of other non-mannolytic populations creating effects that should be quantified to ascertain the true modulation potential of an MDF. The community dynamics that drive fibre-fermentation is not well understood and is likely a contributory factor that restricts broader prebiotic fibre development.
CONCLUSION
Here, we have advanced the MDF concept by showing that the defined relationship between fibre and microbe can be used to directly target specific microbiota in highly complex and competitive gut ecosystems. This was achieved by matching the structural features of our characterized AcGGM fibre to a specific genetic locus in our target microbiota, for which there exists comprehensive biochemical information pertaining to the encoded enzymatic functions.
Using integrated multi-omics analysis, we illustrate the impact of the AcGGM structural configuration on microbial uptake and metabolism within the distal regions of the porcine digestive tract, with the key driver of AcGGM selectivity being the presence of acetylations of mannan, as well as carbohydrate composition and size29. Preserving the complexity of AcGGM resulted in a highly specific shift in the composition of the colon microbiome from weaned piglets, with increases in the relative butyrate:acetate ratios and no adverse effect on host growth performance or health status. We 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 experiments5,10, with both populations expressing proteins from unique CE2/CE17-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. However, they also underscore 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. Ultimately, further research is required to determine whether AcGGM MDFs can confer improvements to animal performance metrics (such as growth and disease resistance) in the context of full-scale livestock production.
DATA AVAILABILITY
All sequencing reads have been deposited at the NCBI sequence read archive under BioProject PRJNA574295, with specific numbers listed in Supplementary Table 2. 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 repository30 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., T.R.H. and J.D. generated the data and/or contributed to the data analyses. W.G.T.W. and C.T-J. generated and analyzed MAPP data. 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 and weaned at 28 days of age, were sorted by litter, sex and weight and randomly divided into twelve pens of four animals, with three pens fed each of the four diets. The piglets in each pen were trained to feed at individual stations containing their respective feed. 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%). Inclusion levels were selected based on in-house experience from previous feeding trials and communication with the feed industry, where an inclusion level of 1-4% of MDF in feed would be realistic. The highest inclusion level at 4% was selected in order to test a dose response. 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 3. 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 was recorded for each individual pig 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 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)31. 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 previously32). 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=0.359 (determined by acetate release from NaOH treated AcGGM by the same method as described in the SCFA section below). Moreover, previous two-dimensional (2D) 1H-13C heteronuclear single quantum correlation (HSQC) NMR analysis of the AcGGM used in this study, showed the distribution of acetylations on mannose units, indicating that spruce mannan hydrolysate without enzyme addition contained prevalently 2-O-, some 3-O-acetylations, and a lower degree of 4-O- and 6-O-acetylations9. 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 Toft33 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. The gating strategy used for flow cytometric data is presented in Supplementary Figure 7.
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.
SCFA analysis
Samples of digesta from duodenum, jejunum, ileum, cecum and colon of individual pigs were collected for SCFA analysis. 250 mg or 250 µL of a sample, depending on the source site were mixed 1:1 with 4mM H2SO4, homogenized by shaking at room temperature for one hour, and centrifuged at 12000xg for 10 minutes. The supernatant was collected with a syringe and filtered through a 0.22 µm pore syringe filter. Samples were stored at −20 °C and centrifuged at 12000xg before transferring aliquots into HPLC vials for analysis. SCFA content was analyzed by HPLC using a REZEX ROA-Organic Acid H+ (Phenomenex, Torrance, California, USA) 300×7.8 mm ion exclusion column, isocratic elution with 0.6 mL/min 4 mM H2SO4 at 65 °C. Eluting analytes were detected by UV at 210 nm. The same HPLC method was used to determine the degree of acetylation, by measuring the acetate released from samples of AcGGM dissolved in 100mM KOH and relating the values to the mannose content of the sample. All data were analyzed using a two-tailed t test.
Micro Array Polymer Profiling (MAPP)
All cell wall preparations were analysed similar to Moller et al34. Cell-wall glycans were sequentially extracted from 10mg of each sample by incubation with 2ug/ml cellulase, pH8.9 in 20mM tris buffer for 16hrs, 45°C to release residual glycans bound to cellulose. Each extract was mixed 50/50 with glycerol buffer (55.2% glycerol, 44% water, 0.8% Triton X-100) and spotted with four 5-fold dilutions and two technical replicates onto nitrocellulose membrane with a pore size of 0.45μm (Whatman, Maidstone, UK) using an Arrayjet Sprint (Arrayjey, Roslin, UK). The arrays were probed using antibodies from the JIM, LM (Plant Probes), BioSupplies, series and a Carbohydrate Binding Module (CBM, NZYTech) detected using anti-rat (JIM, LM), anti-mouse (BioSupplies) or anti-His (CBM) secondary antibodies conjugated to alkaline phosphatase (Sigma Aldrich) as described in Pedersen et al35. The microarrays were developed using an NBT/BCIP substrate containing Nitro Blue Tetrazolium (NBT) and 5-bromo-4-chloro-3-indolylphosphate p-toludine salt (BCIP) in Tris buffer (0.5mM MgCl2, 100 mM diethanol-amine, pH 9.5). The arrays where scanned using a desktop scanner (CanoScan 8800 F, Canon) at 2400 dpi and the intensity of NBT/BCIP stained spots quantified using Array-Pro Analyzer 6.3 (Media Cybernetics). This produces data for the normalised intensity of spots relative to the background intensity of the nitrocellulose membrane (to control for background staining). We used this data to create a heatmap, for average spot intensity across biological replicates and sample dilutions.
Microbial Sampling
Fecal samples were collected from 6 piglets per experimental group (n=12) at days 0, 7, 14, 21, and 27 post-weaning. At the end of the trial, all piglets (n=48) were sacrificed, and distinct 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 30 s 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’ ends36. 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, QIIME37 MOTHUR38 and the R package Phyloseq39. To process the data, the paired end reads for each sample were merged with PEAR40, specifying a minimum assembly length 400, maximum assembly length 575, minimum overlap 50, and no statistical test. Then, PRINSEQ41 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 clustering42 was performed with VSEARCH43 version 2.3.2 and the Silva database44 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 Trimmomatic45 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 metaSPAdes46 version 3.11.1. MegaHIT47 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. MetaBAT48 version 0.26.3 was used to bin the assemblies, and dRep49 was used to dereplicate the multiple assembly and binning combinations to produce an optimal set of MAGs. MASH50 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 CheckM51 version 1.0.7. Taxonomic classifications of MAGs was performed using GTDB-Tk52, and relative abundance estimations were generated using CoverM (https://github.com/wwood/CoverM). Feature and functional annotation were completed with the Prokka pipeline53 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 the July 2018 database release54,55, and subsequently used as a reference database for the metaproteomics (with the exception of glycosyltransferases).
Metaproteomics
Proteins were extracted from each sample 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 × 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 previously56. 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 quadrupole orbitrap mass spectrometer (Thermo Scientific, Bremen, Germany)57, and the acquired raw data was analyzed using MaxQuant58 version 1.4.1.2. Proteins were quantified using the MaxLFQ algorithm59. The data was searched against a sample-specific database (602.947 protein sequences), generated from the 355 metagenome assembled genomes (MAGs), and against the genome of pig, (Sus scrofa domesticus) (40.708 sequences). 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. A total of 8515 protein groups (20 350 shared proteins) were considered valid, which required the protein group to be both identified and quantified in at least two replicates, and in addition, we required at least one unique peptide per protein and at least two peptides in total for every protein group. In cases where a protein group consisted of two or more homologues protein identifications, both unique and shared proteins are indicated (i.e. Fig. 4c and Supplementary Dataset 6). The output from MaxQuant was further explored in Perseus version 1.6.0.7 where filtering, data transformation, and imputation were performed.
For metaproteomic enrichment analysis, missing values in the proteomics data were set to zero, and proteins with zero variance across the eight samples, or with a non-zero value in only one replicate animal in either the control or AcGGM treatment were removed. This resulted in 4562 unique protein groups assigned to 12 535 MAG ORFs (i.e. shared detected proteins). A Principle Component Analysis (PCA) was performed on the 4562×8 matrix using the R function prcomp. Hierarchal clustering dendrograms were constructed using the R functions hclust with Pearson correlation and method = “ward.D”, and a heatmap was drawn using the function heatmap.2 with scale = “row”. The number of protein clusters were selected by visual inspection and the clusters were selected using the cutree function. Enzyme commission (EC) annotation and MAG enrichment analysis was performed using the hypergeometric distribution function phyper in R.
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)60,61. 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 metaQUAST62. The reference genomes were annotated using the Prokka pipeline in the same manner as for the MAGs. All identified ribosomal protein sequences were aligned separately with MUSCLE v3.8.3163, and manually checked for duplications and misaligned sequences. Divergent regions and poorly aligned positions were further eliminated using GBlocks64, 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 phylogeny of the concatenated ribosomal proteins was inferred using RAxML version 8.2.1265 (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 iTOL66. 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, 208674/F50, 270038, 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. We thank Sivert Dæhli for performing the detailed analysis of degree of polymerization of the AcGGM.