ABSTRACT
Bacterial communities resident in the hindgut of pigs, have profound impacts on health and disease. Investigations into the pig microbiome have utilized either culture-dependent, or far more commonly, culture-independent techniques using next generation sequencing. We contend that a combination of both approaches generates a more coherent view of microbiome composition. In this study, we surveyed the microbiome of Tamworth breed and feral pigs through the integration high throughput culturing and shotgun metagenomics. A single culture medium was used for culturing. Selective screens were added to the media to increase culture diversity. In total, 46 distinct bacterial species were isolated from the Tamworth and feral samples. Selective screens successfully shifted the diversity of bacteria on agar plates. Tamworth pigs are highly dominated by Bacteroidetes primarily composed of the genus Prevotella whereas feral samples were more diverse with almost equal proportions of Firmicutes and Bacteroidetes. The combination of metagenomics and culture techniques facilitated a greater retrieval of annotated genes than either method alone. The single medium based pig microbiota library we report is a resource to better understand pig gut microbial ecology and function. It allows for assemblage of defined bacterial communities for studies in bioreactors or germfree animal models.
INTRODUCTION
The microbiome in the hindgut of mammals has been associated with feed conversion efficiency (Singh et al. 2014), pathogen exclusion (Piewngam et al. 2018) and the production of metabolites that directly influence host signaling pathways (Byndloss et al. 2017). It has become clear in recent years, that the microbiome has a drastic impact on host health. Many current methods to study the swine microbiome are based upon dietary intervention (Metzler-Zebeli et al. 2015; Hedegaard et al. 2016). That is, a dietary substrate is introduced to the animal and an effect on microbiome composition, typically 16S rRNA analysis, is measured. As sequencing costs have decreased, more studies have relied upon culture-independent methods to explore the microbiota of pigs. However, culture-independent experiments cannot provide mechanistic information about the factors influencing bacterial communities. We attempt to bridge the schism between culture-dependent and independent methods by incorporating both to investigate the microbiomes of Tamworth and feral pigs: both greatly underrepresented in the literature.
Currently there are an estimated 6 million feral pigs in the United States (USDA 2018). Feral pigs were first introduced in the early 1500s by Spanish settlers and cause significant ecological damage. It has been shown that feral pigs decrease the amount of plant litter and cover in areas they feed (Siemann et al. 2009). Yet, with the ecological and economic toll feral pigs exert, little study has been conducted to elucidate the structure of their microbiome. The Tamworth breed is thought to be descended from the Old English Forest pig and has not been crossed or improved with other breeds since the late 18th Century (British Pig Association 2019). The breed is not a traditional animal used in high production agriculture, bred instead for its tolerance to cold weather and ability to forage. Additionally, the Tamworth breed is under watch by the Livestock Conservancy, after previously being designated as threatened, and the microbiome composition has yet to be characterized.
In this study, we characterized the microbiomes of Tamworth and feral pigs using a combination of culture-dependent and independent techniques on direct colon and cecum contents. To date, modern high throughput culture efforts have been reserved almost exclusively to human fecal samples. Here we extend such methodology to pigs. The culture strategy employs a single medium, yBHI, with various selection screens to shift taxa retrieval. A single medium isolation strategy will facilitate downstream defined community studies. For example, simple to complex bacterial communities can be assembled in bioreactors to study the mechanisms of pig gut microbiome succession (Auchtung, Robinson and Britton 2015). Similarly, colonization of such defined communities constituted could reveal how gut bacterial species or combinations impact gut development and immunity (Goodman et al. 2011). Availability of a well characterized strain library with genome information will facilitate future studies to better understand the role of pig gut microbiome in health and disease.
MATERIALS AND METHODS
Sample collection and preparation
Permission was granted from purchasers of three Tamworth pigs to obtain colon and cecum samples immediately following slaughter. The Tamworth pigs sampled here were not given any antibiotics or growth promoters and could freely graze. Small incisions were made into either the colon or cecum with a sterile disposable scalpel. Lumen contents were gently squeezed into sterile 50 mL tubes, mixed with an equal proportion of 40% anaerobic glycerol (final concentration 20% anaerobic glycerol), and immediately snap frozen in liquid nitrogen. For culture preparation, samples were pooled under anaerobic conditions in a vinyl chamber (Coy Labs, USA). Feral samples were kindly provided by boar hunters in Texas, US. A similar procedure was followed where colon and cecum samples were taken immediately following evisceration, mixed with anaerobic glycerol and frozen.
Metagenomics
DNA was extracted from gut samples using the DNeasy PowerSoil kit (Qiagen, Germany) following the provided kit protocol. After extraction, Microbial DNA was enriched with the NEBNext® Microbiome DNA Enrichment Kit (New England Biolabs, US) to remove host DNA present after DNA extraction. Metagenomic sequencing was conducted on the Illumina MiSeq platform utilizing V2 (250 bp) paired-end sequencing chemistry. Raw sequencing reads were quality controlled using the read-qc module in the software pipeline metaWRAP (Uritskiy, DiRuggiero and Taylor 2018). Briefly, reads are trimmed to PHRED score of > 20 and host reads not removed by enrichment were removed by read-mapping against a reference pig genome (GCF_0 00003025.6). Resultant reads from read-qc are hereby referred to as high-quality reads. High-quality reads were passed to Kaiju (Menzel, Ng and Krogh 2016) for taxonomy annotation against the proGenomes database (http://progenomes.embl.de/, downloaded March 1, 2019). Kaiju was run in default greedy mode and resultant annotation files were parsed in R (R Core Team 2019). Mash (Ondov et al. 2016) was run to estimate the Jaccard distance between samples. A total of 10 000 sketches were generated for each sample and the sketches were compared using the dist function provided in the Mash software.
Antimicrobial resistance (AMR) genes were predicted from metagenomics assemblies. High-quality sequencing reads were assembled into contigs using the assembly module in metaWRAP; metaSPAdes (Nurk et al. 2017) was the chosen to assemble the reads: contigs greater than 1000 bp were retained. Prodigal (Hyatt et al. 2010) was run to predict open reading frames (ORF) using the metagenomic training set. Abricate (Seemann 2018) was then run to annotate the ORF against the NCBI Bacterial Antimicrobial Resistance Reference Gene Database (https://www.ncbi.nlm.nih.gov/bioproject/PRJNA313047, downloaded April 22, 2019).
Contigs were gathered into bins using three methods: MetaBAT2 (Kang et al. 2019), MaxBin2 (Wu, Simmons and Singer 2016) and CONCOCT (Alneberg et al. 2014). Contig bins were kept if the contamination was less than 5% and bin completeness was greater than 85% as determined by CheckM (Parks et al. 2015). Bins from the three methods used were refined into a coherent bin set using the bin_refinement module in metaWRAP. Refined bins were reassembled with a minimum contig length of 200 bp and the same contamination and completeness parameters as initial bin construction. Metagenomic bin and pure isolate phylogeny was generated using UBCG (Na et al. 2018) to identify and align 92 marker genes. Tree construction was conducted using RAxML (Stamatakis 2014): GTR+G4 nucleotide model. To identify KEGG homologues, ORF were identified in metagenomic assemblies, bins and culture genomes using Prodigal. The resultant ORF were annotated against the KEGG database using KofamKOALA (Aramaki et al. 2019) run locally.
Culturomics
Colon and cecum samples were pooled respective to feral and Tamworth samples before culture experiments. All culture experiments, including pooling, were conducted under anaerobic conditions inside an anaerobic chamber (Coy Labs, USA). Samples were serially diluted in sterile anaerobic PBS and spread plated onto the media conditions listed in Supplemental Table 1. Plates were inoculated at 37°C for 48 hours before initial colony selection. 25 colonies were non-selectively sub-cultured from the initial plate to yBHI plates. The procedure was repeated after 72 hours for a total of 50 colonies per media condition. Colonies were primarily identified using MALDI-TOF (Bruker, Germany). MALDI-TOF scores greater than 2.0 were considered a positive species identification. Scores between 1.7 and 2.0 were taken as positive genus identification. Colonies without a positive MALDI-TOF identification were identified by sequencing the 16 s rRNA gene. Briefly, DNA was extracted from colonies using the DNeasy Blood and Tissue Kit (Qiagen, Germany) following the manufacturer's protocol. 16 s rRNA sequence was amplified using 27F and 805R primers. The primer sequence is listed in Supplemental Table 1. Genomes of the selected strains were sequenced on the MiSeq platform utilizing paired-end v3 chemistry (300 bp). Sequencing reads from individual strains were assembled with Unicycler (Wick et al. 2017): minimum contig length of 200 bp. The raw sequencing reads from the culture isolates and metagenomic samples are hosted at NCBI under the BioProject ID PRJNA555322.
RESULTS
Metagenomics describes community composition
The gut microbiota of Tamworth breed and feral pigs are underrepresented in the literature. As such, the pigs provide a unique model to study the implementation of culture-dependent and independent techniques. To begin the investigation, colon and cecum samples were metagenomically sequenced from both breeds. Figure one shows the taxonomic annotation of the metagenomic reads respective to the source of isolation (Fig. 1A and B). Shotgun metagenomic sequencing provides an accurate description of the community outside the bias of culture-dependent methods. By incorporating sequencing, the efficiency of the culture method can be elucidated. The phylum Bacteroidetes represents nearly 53% of all classified reads in Tamworth pigs, compared to 29% in feral pigs. The abundance of Firmicutes in Tamworth samples is lower than feral samples at 15% and 28% respectively. Additionally, nearly 10% more of the feral reads were unclassified compared to Tamworth (37%, 28%). Turning to the genus level, the large increase of Bacteroidetes in Tamworth pigs is primarily composed of the genus Prevotella, Fig. 1 B (38%, feral 11%). Remarkably, the genus Bacteroides showed almost identical distribution between the feral and Tamworth pigs (7.6% and 7.6%, respectively). The increase of Firmicutes in feral samples is due to an increase in several genera such as Ruminococcus, Clostridium and Eubacterium. This corresponds with significantly higher Shannon diversity index values in feral samples compared to Tamworth (p = 0.0024, Mann-Whitney U test). Full phylum and genus annotation tables are provided in Supplemental Table 2.
Metagenomic analysis of feral and Tamworth colon and cecum samples.(A)(B) Relative abundance of major phyla and genera annotated from sequencing reads respective to isolation source. (C) Matrix depicting the MASH distance between feral and Tamworth samples. Clusters 1 and 2 are defined by kmeans clustering. (D) Principal component analysis of the taxon abundance obtained from the read annotation. Samples are colored respective to isolation source.
Another advantage of implementing culture-independent methods is the speed and ease with which samples can be clustered. To cluster the sequencing results from the microbiome samples, Mash (Ondov et al. 2016) was used to sketch the reads sets and compile a distance matrix (Fig. 1C). Within the matrix, both Kmeans clustering and hierarchical clustering (average-linkage) readily separated the input samples according to isolation source. Mash provides a method to compare metagenomes that is not subject to annotation bias. Principal component analysis (PCA) of the OTU tables was the second method employed to cluster the metagenome results. Again, two distinct groups corresponding to Tamworth and feral samples are seen in the plot (Fig. 1D). Interestingly, the Tamworth samples are more homogenous in both the Mash and PCA methods. All pigs were taken from the same farm which may account for the lower inter-animal microbiome divergence. The herd status of the feral pigs is unknown.
With metagenomic sequencing, the AMR homologues from a bacterial community can be identified. Briefly, the sequencing reads are assembled into contigs, open reading frames are predicted, and annotated against relevant databases of AMR genes. AMR homologues were identified in the gut samples as the culturing scheme depends on antibiotic selection to shift bacterial diversity. High loads of antibiotic resistant bacteria in fecal samples would inhibit the culture scheme. The gut samples from Tamworth pigs’ all yielded at least eight AMR homologues (Fig. 2). Additionally, four AMR genes: cfxA, lnu(AN2), mef(En2) and tet(40) were identified in every Tamworth sample. In general, the level of AMR genes in Tamworth samples was higher than feral samples. No common pattern is apparent for Feral samples; however, tet(Q) is found in 5 of 9 feral samples. The full result of the AMR gene query is listed in Supplemental Table 3.
Antimicrobial resistance (AMR) homologues annotated from metagenomic samples. Columns depict individual samples and rows correspond to AMR homologues. Blue color depicts the presence and white color corresponds to absence. AMR homologues were considered present if the coverage value was greater than 90% and a percent identity value greater than 70%.
Selective screens shift plating diversity
High throughput anaerobic culturing was the second method employed to study the microbiota of Tamworth and feral pigs. We implemented culture-dependent methods, in addition to sequencing methods, as we believe the two will provide a more coherent view of a microbiomes structure and function. The culture sampling strategy utilized is as follows: a base medium (yBHI, or close derivatives) had various selective screens (antibiotics, heat, bile, etc.,) applied to it. A finite growing surface is available for colonization and some species will grow more rapidly and subsequently outcompete others. If appropriate selective pressure is applied, we hypothesized that interspecies selection would decrease allowing for taxa not retrieved in plain medium conditions to grow. The approach is similar to one previously used to culture strains from human fecal samples (Rettedal, Gumpert and Sommer 2014). One major difference is said work used multiple media compositions, rather than one as in our study. Ten media conditions were used for both Tamworth and feral samples and are listed in Supplemental Table 1. 25 colonies were picked at 48- and 72-hours post inoculation, for a total of 50 colonies per condition. In total, 1000 colonies were selected from plates, of which 884 were successfully identified. Selective screens shifted the taxa retrieved (Fig. 3). Fig. 3 depicts the number of isolates per media condition with a bar plot depicting the total number of isolates retrieved. Lactobacillus sp. was the most abundant organism retrieved (166 isolates) followed by Escherichia coli (86), Lactobacillus mucosae (74) and Streptococcus hyointestinalis (64). The top ten isolates cultured are listed in Table 1. One case of selection completely changing plate diversity compared to plain media is that of heat shock treatment. As expected, many spore forming genera including Bacillus and Clostridium were only able to grow when the inoculum was heated to kill vegetative cells. The selective screens placed upon yBHI not only shifted the taxa retrieved from each plating condition as shown in Fig. 3, but also shifted species richness and evenness (Fig. 4). The most diverse plating condition (Shannon Index) for both Tamworth and feral samples was obtained from plain yBHI: showing as a log-normal community distribution. Similar log-normal community structures are observed for BSM (Tamworth only), Erythromycin and heat shock treatments. Bile treatments and chlortetracycline exhibited strong selective pressure shown as geometric series in the species-rank abundance plots (Fig. 4). Most of the taxa retrieved from the bile condition were identified as Proteobacteria, indicating that the dosage of bile (1 g / L) was too high.
Bacteria isolated from various media conditions. Columns represent individual media conditions and row correspond to bacterial taxa retrieved, cells are colored respective to the number of isolates cultured per media condition. The corresponding bar plot to the left of the matrices shows the total number of isolates retrieved per isolation source.
Rank abundance curves of the various media conditions. The community evenness of the various media conditions is shown respective to the isolation source. The inlay plot depicts the Shannon Index respective to the isolation source.
The culture strategy did not recapitulate the community in the inoculum as defined by metagenomics. In both Tamworth and feral samples, a high number of Firmicutes and Proteobacteria were isolated, compared to the metagenomic sampling where Bacteroidetes was the most abundant phylum for both sources. If we disregard the bile conditions, which were dominated by Proteobacteria, yBHI clearly selects for common Firmicutes genera including: Lactobacillus, Streptococcus and Bacillus. While the screens were successful in increasing the total number of species retrieved, no condition matched the inoculum in form. Prevotella for example, the most abundant genus in Tamworth pigs, was only retrieved seven times from 500 colonies. Taken together, the strategy was successful in gathering many isolates that can grow on a common medium but failed in that the most abundant taxa were not retrieved in proportion to the inoculum.
Culturing captures genomic information not captured in metagenomics
The sampling strategy employed did not recapitulate the inoculum community. However, one of the main reasons we chose to culture was that we believed rare taxa would provide information that would be loss to metagenomics. To examine this, we sequenced selected isolates and generated 81 high quality metagenomic bins (completeness > 85%, contamination < 5%). The phylogeny of the metagenomic bins and culture genomes was estimated (Fig. 5). Consistent with read taxonomy, many of the bins constructed from both Tamworth and feral samples were annotated to the phylum Bacteroidetes. The phyla Firmicutes, Proteobacteria and Actinobacteria were comprised almost entirely of isolate genomes. Isolate genomes not only populated clades of the tree missed by metagenomic bins, but provided genes not observed in metagenomic assemblies nor bins (Fig. 6). Open reading frames (ORF) were predicted from metagenomic assemblies, metagenomic bins, and culture isolate and were annotated against the KEGG database. Fig. 6 shows the abundance (natural log) of KEGG homologues respective to the source of the ORF. The full KEGG annotations from the bins, isolates, and metagenomes are provided in Supplemental Table 4. Metagenomic bins contained less information than the metagenomic assemblies. This is expected as the bins are derived from contigs in the assemblies and not all the contigs will be gathered into bins. The isolates however provided KEGG homologues that were completely missed through culture-independent methods. Thus, culture and culture-independent methods can augment a microbiota analysis providing information that the other method cannot capture.
Maximum-likelihood tree of metagenomic bins and culture genomes. Tree was constructed from a nucleotide alignment of 92 single-marker genes. General time reversible (GTR) was chosen as the substitution model in tree construction.
KEGG annotation of open frames from metagenomic assemblies, bins and culture genomes. Rows and columns are clustered using an average linkage method. KEGG annotations counts are represented as the natural log to increase the clarity of the figure.
DISCUSSION
In this work, we establish a methodology that provides a rigorous examination of the pig gut microbiome. Instead of viewing culture-dependent and independent methods as a binary decision, we contend that both techniques complement one another. Community composition and gene content is readily identified by metagenomics. In addition, metagenomic techniques require less labor than culture-dependent ones. As a result, current research on the pig gut microbiota has almost exclusively focused on culture-independent methods (Xiao et al. 2016; Tan et al. 2017; Wang et al. 2019). Metagenomic sequencing however cannot establish the role specific taxa play in microbial communities. Recent works on the gut microbiome of humans have shown the value of culture-dependent techniques, especially in identifying mechanisms that govern bacterial communities (Gutiérrez and Garrido 2019; Poyet et al. 2019). Incorporation of culture-techniques would greatly increase our understanding of how microbial communities form and persist in pigs.
Despite the high abundance of Prevotella in both Tamworth and feral pigs, the culture sampling strategy we employed only yielded seven Prevotella isolates from Tamworth samples (7/500, 1.4%). No Prevotella was isolated from the feral inoculum. In contrast, several genera including Lactobacillus, Escherchia, Streptococcous and Bifidobacterium were overrepresented in culture samples as compared to metagenomic sequencing. Our culture results align with an early culture examination of the pig microbiome. In that study, the two most abundant isolates cultured were gram-positive cocci and Lactobacillus (Russell 1978). Both our work and the earlier work relied upon complex media derived largely of peptone digests. As Prevotella is associated with an increase of dietary fiber, work will be needed to develop a defined media that is not based upon peptides such as yBHI. Previous studies have been wildly successful in culturing many bacteria that were previously thought to be ‘unculturable’ (Browne et al. 2016; Lagier et al. 2016). In those works, samples were plated onto multiple media formulations to encourage a broad growth of bacterial species. While the multiple media approach generates a higher number of taxa, one study isolated over 1300 species (Lagier et al. 2016), creating multiple media formulations is expensive and time-consuming. Additionally, bacteria isolated from different media formulations may not grow together on a common media, forfeiting any combined in vitro experimentation. The approach we implemented ensures all isolates can grow on a common media source.
Recent studies have proposed metagenomic binning as a culture-independent method to extract genomes from samples (Albertsen et al. 2013; Tully, Graham and Heidelberg 2018; Pasolli et al. 2019; Wang et al. 2019). However, one of the main pitfalls of metagenomic binning is that metagenomic assemblers struggle to assemble contigs of closely related taxa, especially if the organisms are found in low abundance (Ayling, Clark and Leggett 2019). With knowledge now that strain-level variation occurs in species of the microbiome (Lloyd-Price et al. 2017), targeted culture efforts are needed to confirm that strain variation observed in metagenomic data is not simply due to assembler bias. Also, culture isolates provided genetic information that was not captured in the metagenomic sequencing.
ACKNOWLEDGEMENTS
Computations supporting this project were performed on High-Performance Computing systems managed by Research Computing Group, part of the Division of Technology and Security at South Dakota State University. This work was in part supported by the grants from the South Dakota Governors Office of Economic Development (SD-GOED) and the United States Department of Agriculture (grant numbers SD00H532-14 and SD00R646-18) awarded to JS. We gratefully thank Tom Harrison, Jeff Hopper and Brett Grogan for their help in collecting feral pig gut microbiota samples. We are saddened by the untimely passing away of Tom Harrison during the course of this study. We dedicate this manuscript to Tom's memory.
Conflicts of interest. None declared.
