Two Different Bacterial Community Types Are Linked with the Low-Methane Emission Trait in Sheep

The potent greenhouse gas methane (CH4) is produced in the rumens of ruminant animals from hydrogen produced during microbial degradation of ingested feed. The natural animal-to-animal variation in the amount of CH4 emitted and the heritability of this trait offer a means for reducing CH4 emissions by selecting low-CH4 emitting animals for breeding. We demonstrate that differences in rumen microbial community structure are linked to high and low CH4 emissions in sheep. Bacterial community structures in 236 rumen samples from 118 high- and low-CH4 emitting sheep formed gradual transitions between three ruminotypes. Two of these (Q and S) were linked to significantly lower CH4 yields (14.4 and 13.6 g CH4/kg dry matter intake [DMI], respectively) than the third type (H; 15.9 g CH4/kg DMI; p<0.001). Low-CH4 ruminotype Q was associated with a significantly lower ruminal acetate to propionate ratio (3.7±0.4) than S (4.4±0.7; p<0.001) and H (4.3±0.5; p<0.001), and harbored high relative abundances of the propionate-producing Quinella ovalis. Low-CH4 ruminotype S was characterized by lactate- and succinate-producing Fibrobacter spp., Kandleria vitulina, Olsenella spp., Prevotella bryantii, and Sharpea azabuensis. High-CH4 ruminotype H had higher relative abundances of species belonging to Ruminococcus, other Ruminococcaceae, Lachnospiraceae, Catabacteriaceae, Coprococcus, other Clostridiales, Prevotella, other Bacteroidales, and Alphaproteobacteria, many of which are known to form significant amounts of hydrogen. We hypothesize that lower CH4 yields are the result of bacterial communities that ferment ingested feed to relatively less hydrogen, which results in less CH4 being formed.


Introduction
The forestomachs of ruminant animals contain a great diversity of prokaryotic and eukaryotic microorganisms that together break down and ferment the feed ingested by the host animal. Volatile fatty acids (VFAs), such as acetate, propionate and butyrate, are formed, together with varying amounts of hydrogen (H 2 ). Methanogenic archaea in the rumen use H 2 to gain energy, producing methane (CH 4 ) in the process. CH 4 is of no nutritional value to the animal, and is eructed and exhaled into the atmosphere, where it acts as a potent greenhouse gas. This CH 4 also represents a major loss of energy to the animal [1,2]. To reduce CH 4 emissions from enteric fermentation, and increase animal productivity, a number of different mitigation strategies have been tested, e.g., feed supplementation with lipids [3,4,5], inhibition of enzymes involved in CH 4 formation [6,7], depletion of ciliate protozoa [8], or vaccination against methanogens (for a recent review see Wedlock et al. [9]). Another potentially very effective way to reduce CH 4 emissions from ruminant animals is to specifically select naturally low-CH 4 emitting animals for breeding and to avoid proliferation of high-CH 4 emitting animals. Measurements of CH 4 emissions from individual sheep in highly-sensitive open-circuit respiration chambers showed that animals in the same flock, even though feeding on the same diet, varied significantly and consistently in their CH 4 yields, measured in g CH 4 per kg of dry matter intake (DMI; [10]). Some individuals have a naturally lower CH 4 yield (low emitters) than others (high emitters). The genetics of the low CH 4 trait, including estimates of heritability, repeatability and genetic correlations with productive traits, are starting to be better understood [11]. Of increasing interest are the underlying factors, both genetic and non-genetic, that explain the observed natural differences in CH 4 yields between individual animals. It is assumed that certain hostrelated characteristics, such as genotype, physiological state, or development of the animal, influence CH 4 yields by controlling the presence and/or abundance of certain microbial populations in the rumen. Studies that analyze the microbiota of ruminants that naturally vary in the amount of CH 4 produced have so far been missing from the literature, and detailed microbial analyses of hundreds of samples have only become possible with the development of next generation sequencing technologies. Understanding the differences in rumen microbial community structure between low-and high-emitting animals will point to those microbial groups that play key roles in the expression of the host trait or that have adapted to it. Isolation and cultivation efforts can then be made to study these particular taxa in greater detail in the future. Knowledge on the physiology of these groups may be useful for targeted modification of rumen microbial communities and promotion of the low-CH 4 trait, or help understand the circumstances that lead to a low-CH 4 trait and any production benefits or tradeoffs. Here, we applied high-throughput barcoded 454 Titanium amplicon sequencing of bacterial, archaeal, and eukaryotic marker genes to determine correlations between rumen microbial community structure and CH 4 yields of 60 high-and 58 low-emitting sheep.

CH 4 yields from sheep are subject to significant natural variation
Individual CH 4 yields from 340 sheep, in four cohorts, were measured using open-circuit respiration chambers. Two full-day measurements were made on each animal on two consecutive days (round [a]), and then again at least 15 days later (round [b]). The CH 4 yields (g CH 4 /kg of dry matter intake [DMI]) were expressed as the means of the two days of each round, giving two values for each animal, one for round [a] and one for round [b]. High-(Hi) and low-(Lo) CH 4 emitting animals were identified from each cohort for rumen microbial community analysis, based on their average CH 4 yields over the two measuring rounds. This grouping, Hi or Lo, is referred to as CH 4 group (Table 1). In total, 118 animals were selected for study as Hi (60 animals) or Lo emitters (58 animals), and displayed highly significant differences in CH 4 yields (p,0.001, Student's t-test). These differences were stable over the two measuring rounds for each animal (p,0.001; Table 2). These two sets of animals with distinctly different CH 4 yields were used to determine if there were parallel differences in rumen microbial community structure based on bacterial, archaeal, and ciliate small subunit rRNA genes and anaerobic fungal internal transcribed spacer 1 (ITS1) genes. Two rumen samples, one from each of the two CH 4 measurement rounds [a] and [b], were investigated from each of the 118 animals, resulting in microbial community structure data from up to 236 rumen samples per microbial group (Table 2). An overview of microbial community composition and diversity across all analyzed samples is shown in Figure S1 and Figure S2, respectively. Each sample was associated with a CH 4 yield for the particular sheep and round from which it came.

High and low CH 4 sheep harbor similar numbers of methanogenic archaea
We enumerated archaeal 16S rRNA genes by quantitative PCR to find out if differences in CH 4 yield correlated with differences in methanogen densities, on the assumption that all archaea present at any significant levels in the rumen are methanogens [12,13]. However, archaeal 16S rRNA gene abundance was not significantly different in rumen samples from 12 Hi and 12 Lo emitters in both measuring rounds of cohort 1 (p = 0.91 and p = 0.14 for rounds [a] and [b], respectively; Wilcoxon rank-sum test; Figure 1).
Two different bacterial community types are linked to lower CH 4

yields in sheep
Differences in bacterial community composition between animals ranked as high-CH 4 emitters and those ranked as low-CH 4 emitters was explored using correspondence analysis (CA; Figure 2A). The resultant plot shows a continuous pattern of samples in a right angle arrangement, indicating a gradient of bacterial community structure across all samples. While samples obtained from animals ranked as high-CH 4 emitters clustered predominantly in and around the center of the graph, samples from animals ranked as low-CH 4 emitters appeared to predominantly cluster along the horizontal and vertical branches. Canonical discriminant analysis (CDA) corroborated the finding of significant differences between bacterial communities in the rumens of low-and high-CH 4 emitting animals (p,0.001; Figure  S3A). Based on results obtained from CA, samples grouping on the horizontal branch (samples 1-67; average CH 4 yield: 14 Figure 2B).
We confirmed this finding by performing principal coordinate analysis (PCoA) on the same data, which revealed the same pattern of sample relationships ( Figure S4A). As found in the CA, samples towards the extremes of the branches were associated with low CH 4 yields and samples at the apex of the angle were associated with high CH 4 yields ( Figure S4B). The relative placements of the samples using CA and PCoA were highly correlated (R 2 = 0.94), showing that both ordination methods gave highly comparable results ( Figure S4C). We then used Partitioning Around Medoids to naïvely divide the samples in the PCoA into three clusters based on minimization of dissimilarity of all samples in a cluster to the center of that cluster ( Figure S4A). The resulting clusters 2 (14.862.6 g CH 4 /kg DMI) and 3 (14.161.4 g CH 4 /kg DMI) were characterized by significantly lower average CH 4 yields than cluster 1 (15.662.1 g CH 4 /kg DMI; Student's t-test cluster 2 versus cluster 1: p = 0.02; Student's t-test cluster 3 versus cluster 1: p,0.001). These analyses all show that at least two different bacterial community types are associated with low-CH 4 yields and one bacterial community type is associated with high-CH 4 yields. Clear linear relationships between CH 4 yield and bacterial community structure were not found (Table S1). This is because key low-CH 4 taxa can be either abundant or rare in a sample depending on which of the two low-CH 4 community types was present. Based on the results from the CA, which are supported by the PCoA, we categorized samples 1-67, based on their CA coordinates, as low-CH 4 -associated CA type LM1, samples 68-193 as high-CH 4 -associated CA type HM, and samples 194-230 as low-CH 4 -associated CA type LM2 (Table 1). In addition to lower CH 4 yields, LM1 samples were also associated with significantly lower acetate to propionate ratios (LM1 average acetate to propionate ratio: 3.760.4, HM: 4.360.5, LM2:4.460.7; Student's t-test LM1 versus HM: p,0.001, Student's t-test LM2 versus HM: p = 0.83).
Notably, the three different CA community types LM1, LM2, and HM were each characterized by a higher relative abundance of a number of key bacterial taxa ( Figure 2C, Figure S5). ANOVA based on the three CA types revealed the taxon composition of samples belonging to each category (Table 3), and three ''ruminotypes'' were defined (Table 1; for ANOVA results based on PAM clustering see Table S2). Of the taxa that represented on average $1% of the bacterial community in at least one of the three ruminotypes, larger abundances of species belonging to the Ruminococcaceae, Clostridiales, Lachnospiraceae, Prevotella, Bacteroidales, Alphaproteobacteria, Catabacteriaceae, Coprococcus, and Ruminococcus (in order of significance) were characteristic of ruminotype H (H = HM-associated). Species of the genus Quinella were most abundant in LM1-associated ruminotype Q (Q = Quinella ovalis). Sequences belonging to the genus Quinella were initially assigned to the genus Selenomonas with unknown species affiliation due to the lack of a reference sequence for the genus Quinella in the greengenes database that we used. A BLAST search against the GenBank database and comparison of the Selenomonas-assigned sequences to the 16S rRNA gene sequence of Quinella ovalis (Quin's oval) in ARB revealed their true affiliation with the genus Quinella. Quite different taxa (Sharpea azabuensis, Olsenella, Fibrobacter succinogenes, Prevotella bryantii, Kandleria vitulina, and Fibrobacter intestinalis) contributed predominantly to LM2-associated ruminotype S (S = Sharpea azabuensis).

Relative abundance of Methanosphaera spp. is increased in LM samples
Correspondence analysis did not reveal any specific clustering pattern based on archaeal community structure ( Figure S6A); however, CDA suggested significant differences between archaeal communities in animals ranked as low and high CH 4 emitters (p, 0.001; Figure S3B). To obtain a better understanding of which taxa, if any, were significantly correlated with CH 4 yield, relative abundances of archaeal taxa were compared with CH 4 yield in the individual measuring rounds using Spearman's Rank correlations. This analysis indicated that the relative abundance of Methanosphaera spp. was negatively and significantly correlated with CH 4 yield in several -but not in all -of the eight measuring rounds (Table S1). This finding was confirmed by the results obtained from plotting Methanosphaera relative abundance against the ranking of each sample derived from CA on the bacterial data set ( Figure 2D). Samples categorized as LM1 and LM2 harbored significantly higher relative abundances of Methanosphaera than samples belonging to the HM type ( Eukaryotic community structure appears to be independent of CH 4

yield
The composition of eukaryotic communities in the rumen samples was explored using CA and CDA, but neither ciliate nor anaerobic fungal communities revealed any obvious clustering by CH 4 group of the animals that could have been attributed to differential CH 4 yields ( Figure S3C, Figure S3D, Figure S6B, Figure S6C). Spearman's Rank correlations suggested that few eukaryotic taxa were correlated with CH 4 yield in the different measuring rounds (Table S1). Overall, however, no consistent trends were observed, and for the majority of taxa there was no correlation at all.

Discussion
This work demonstrates that the natural variation in the CH 4 emission trait is reflected in the composition of the microbial community in the rumen of the host animal. We analyzed bacterial, archaeal, and eukaryotic communities in rumen samples obtained from sheep identified as naturally high or low emitting animals using the most accurate method of CH 4 yield measurement to date (respiration chambers). We reasoned that, if there were differences in the microbial communities of sheep emitting different amounts of CH 4 , these differences should be most apparent in comparisons of the highest and lowest emitting animals, and that they should be consistently found in different groups of animals. Our data on the rumen microbial communities from the 60 highest and 58 lowest CH 4 emitters from four different cohorts of sheep suggest that differences in bacterial and archaeal community structures are associated with a naturally lower CH 4 yield and are consistent with consequent differences in rumen fermentation.

Linking bacterial community structure to CH 4 yields
Profiling of 16S rRNA genes revealed gradients in bacterial community structure across all analyzed rumen samples, and suggested the existence of at least three idealized bacterial community types, referred to as ruminotypes Q, S, and H. Ruminotype H was linked with samples associated with higher than average CH 4 yields (HM samples) while ruminotypes Q and S were linked to lower than average CH 4 yields (LM1 and LM2 samples, respectively). Such intestinal bacterial community types that appear to be specifically associated with a certain host trait have previously been referred to as ''enterotypes'' in humans [14,15], chimpanzees [16], and mice [17]. However, we observed that the rumen microbiota across samples categorized as LM1, LM2 and HM displayed smooth abundance gradients of key genera without discrete clustering of samples. For convenience, we have adopted the term ''ruminotype'' for generalized community types, recognizing that there is so far no clear consensus of the enterotype concept [18].
VFA ratios in the rumen samples suggested that proportionally more propionate was present in LM1 samples. This observation is consistent with the finding that LM1-associated ruminotype Q has a higher relative abundance of organisms closely related to Quinella ovalis (Quin's oval; [19,20]). Quin's oval appears to ferment sugars to equimolar acetate and propionate [21], which is associated with lower H 2 formation than other fermentation pathways [22]. LM2-associated ruminotype S harbored significant proportions of species characterized as lactate producers such as Sharpea azabuensis [23], Kandleria vitulina [24,25], and Olsenella spp. (O. uli; [26]) and succinate producers such as Fibrobacter succinogenes, F. intestinalis [27] and Prevotella bryantii [28]. Formation of these products is generally associated with no or low H 2 production. In contrast, HM-associated ruminotype H was characterized by higher abundances of bacteria characterized as H 2 producers, notably members of Ruminococcus and other Ruminococcaceae, Coprococcus, Lachnospiraceae, and other Clostridiales [29,30].
Linking archaeal community structure to CH 4 yields Methanogen densities were not significantly different in rumen samples from Hi and Lo emitters of cohort 1 across both CH 4 measuring rounds. These results corroborate previous studies that also showed that densities of methanogens were not significantly different between two groups of feedlot bulls [31] and two groups of lambs [32] that produced significantly different amounts of CH 4 . In contrast to those studies, where differences in CH 4 emission resulted from the different diets being administered, Linking Methane Yield with Microbial Ruminotype PLOS ONE | www.plosone.org differences in CH 4 yields in our study were due to natural animalto-animal variation. It has been reported that cattle with naturally high and low feed conversion efficiency vary considerably in the amounts of CH 4 produced [8,33]. But, similar to the sheep in our study, the densities of methanogens in the rumen fluid did not vary in animals with different feed conversion efficiencies [34]. Hence, natural differences in CH 4 yields between individual sheep were not due to naturally differing densities of methanogenic archaea in the rumen, i.e., greater densities of methanogens in high CH 4 animals and lower densities in low CH 4 animals.
Archaeal community structure, however, was significantly different in rumen samples associated with different CH 4 yields.
Members of the genus Methanosphaera were significantly more abundant in LM1 and LM2 samples as opposed to HM samples.
Recent measurements suggest that the rumens of low CH 4emitting sheep are significantly smaller than those of high CH 4emitting sheep [35]. Both LM-associated ruminotypes, Q and S, were associated with bacteria that would ferment the ingested feed to less H 2 than the HM-associated ruminotype H. Fermentation of feed to less H 2 in low CH 4 animals would support a smaller population of hydrogenotrophic methanogens (like Methanobrevibacter spp.), although in a smaller rumen. This could explain why the density of methanogens was not different. In addition to hydrogenotrophic methanogens, other methanogens like Methanosphaera spp. use methyl-groups derived from methoxyl-substit- uents of plant material as a major CH 4 precursor [36,37]. The amount of methyl-group derived CH 4 is probably limited by the availability of the methyl-donors, which is constant in the feed. This methanogenesis from methyl-groups would therefore support similar numbers of Methanosphaera spp. in animals of both CH 4 traits, although they could be expected to be denser in animals with smaller rumens. This would account for the increased relative abundance of Methanosphaera spp. in samples related to LM yields in our study.

No apparent eukaryotic community structure link to CH 4 yield
Reports on the eukaryotic community composition in the rumen and its correlation with CH 4 are scarce, and only very recently has a methodology been developed to analyze not only bacterial and archaeal but also ciliate and anaerobic fungal communities in the rumen using next generation sequencing [38]. Although it has been suggested that certain genera of ciliate protozoa may be linked to higher or lower CH 4 emissions [39,40], we did not find significant and consistent indications that ciliate community composition contributed to the natural differences in the animals analyzed in our study. Similarly, anaerobic fungal community structure did not appear to be linked to CH 4 trait. These findings do not rule out potential differences in ciliate and anaerobic fungal communities on a lower taxonomic, the (meta)genomic or transcriptional level.

Possible synthesis of findings
Our data show that there is a significant correlation between microbial community structure and natural variations in CH 4 emissions from sheep fed the same diet. The full set of factors that caused the observed differences in rumen microbial community structure in animals with high-and low-CH 4 yields remains to be elucidated. Genetic, epigenetic or environmental factors are likely to drive the differences in the rumen microbial communities in naturally high-and low-CH 4 emitting animals. We postulate that the differences in CH 4 emissions are a result of such factors selecting for different microbial communities, which in turn form different amounts of H 2 and so different amounts of CH 4 per unit of feed. High-CH 4 emitting sheep have larger rumens than low-CH 4 emitting sheep [35], and correspondingly lower passage rates [35,41]. Lower passage rates are associated with lower dissolved H 2 concentrations which select for H 2 producing bacteria and hence more CH 4 formation [22]. Genes of the CH 4 forming pathways of methanogens were found to be more highly expressed relative to gene copy in high-CH 4 emitting sheep, which was interpreted as a response to lower H 2 concentrations in high-CH 4 emitting sheep [42].
Follow-on studies looking at correlations between microbial community structure and animal genetics (e.g., single nucleotide polymorphisms on the host genome) or physiological and anatomical data (e.g., rumen morphology, pH, passage rate, etc.) can begin to determine which factors drive the microbial community structure and ultimately result in different CH 4 yields. It is possible that the two different LM-associated ruminotypes are controlled by different factors, and it will be of interest to breeders to evaluate whether one of the two bacterial community types, or even the HM-associated ruminotype, has better production characteristics. If microbial communities are controlled by host genetic factors, then selective breeding can steer ruminant populations towards decreased enteric CH 4 formation. If the pivotal factors originate in the animal's environment (e.g., behavior, handling, etc.), then animal management may be adopted as a means to maximize the low-CH 4 trait.

Measurement of CH 4 yields from sheep
The use of animals, including welfare, feeding, experimental procedures, and the collection of rumen samples used for this study, was approved by the AgResearch Grasslands Animal Ethics Committee (Application number 11975), and complied with the institutional Codes of Ethical Conduct for the Use of Animals in Research, Testing and Teaching, as prescribed in the New Zealand Animal Welfare Act of 1999 and its amendments. CH 4 measurements were conducted on a total of 340 New Zealand sheep separated into four cohorts (or lots; [11]). Two independent measuring rounds, each over two days, were made for each individual, producing a total of four days of CH 4 yield data (g CH 4 /kg DMI). The larger experiment of which these animals were a part is described in detail by Pinares-Patiño et al. [11]. For more information also see Text S1.

Rumen sampling and sample processing
A rumen sample was collected via stomach tubing and immediately stored at 220uC at the end of each of the two CH 4 measurement rounds (17-18 h after the last feeding). Animals within each separate cohort were ranked on the basis of their 4day average CH 4 yield, and rumen samples of in average high-(Hi) and low-CH 4 emitting animals (Lo) were selected for analysis of rumen microbial community structure (Table 1). In cohort 1, 15 of the 17 highest CH 4 emitting animals and 13 of the 15 lowest CH 4 emitting animals were used. In cohort 2, the 15 highest and 15 of the 16 lowest CH 4 emitting animals were used. Rumen samples from the 15 highest emitting animals and the 15 lowest emitting animals in cohorts 3 and 4 were used. This gave a total of 236 rumen samples from 118 animals ( Table 2; for metadata of individual samples refer to File S1). All rumen samples were freeze-dried, homogenized, and stored at 220uC. VFA concentrations were determined from subsamples of rumen contents by gas-liquid chromatography ( [43]; Text S1).

Extraction and amplification of nucleic acids
Nucleic acids were extracted from freeze-dried rumen contents using a combined bead-beating, phenol-chloroform and column purification protocol using the QIAquick 96 PCR purification kit (Qiagen, Hilden, Germany; Text S1; [44]). The abundances of archaeal 16S rRNA genes were quantified in randomly selected rumen samples of 12 Hi emitters and 12 Lo emitters of cohort 1 as described previously (Text S1; [45]).

Microbial community analyses
PCR amplification of bacterial and archaeal 16S rRNA genes, ciliate 18S rRNA genes, and the anaerobic fungal internal transcribed spacer 1 (ITS1), was performed using methods modified from Kittelmann et al. (described in detail in Text S1; [38]). Pyrosequencing was performed on a 454 Life Sciences Genome Sequencer FLX machine (Center for Genome Sciences and Systems Biology, Washington University of St. Louis, USA).
Sequence data were processed and analyzed following the procedure described by Caporaso et al. [46]. Sequence data were phylogenetically assigned using BLAST against reference databases (for more detailed information see Text S1; databases are available from the authors upon request). These sequence data have been submitted to the EMBL database under the study accession numbers ERP003779 (Bacteria), ERP003773 (Archaea), ERP003772 (ciliate protozoa), and ERP003764 (anaerobic fungi). All analyses of high-throughput pyrosequencing data were carried out using the QIIME pipeline [46]. Resulting text files were imported into Excel (Microsoft Corp., Redmond, WA, USA) and R (www.r-project.org) for further statistical evaluation. Detailed descriptions of the statistical analyses performed to evaluate potential differences of microbial community structure in animals with contrasting CH 4 traits are given in Text S1. Table S1 Correlations between microbial taxa and CH 4 ranks of sheep. Spearman's rank correlation factors (r) and corresponding p-values are shown for those microbial groups that represented at least 1% of the bacterial community in at least one rumen sample in the analyzed measuring round and that were found significantly related to either high-CH 4 or low-CH 4 in at least one measuring round (rank 1 = sample with highest CH 4 yield in each measuring round). Taxa that showed a positive correlation in one measuring round and a negative correlation in another were regarded to have no clear correlation with CH 4 (None). The table is sorted by the number of measuring rounds in which the taxa were found to be significantly correlated to CH 4 rank (last column (Number of rounds); shaded from dark orange (large number of measuring rounds that showed significant correlation) to light yellow (small number)) and by the type of correlation (second to last column (High/Low; for correlation to high-CH 4 or low-CH 4 yields). Here, the taxon Eubacterium refers to members of the genus Eubacterium which belong to the family Ruminococcaceae (according to the greengenes taxonomy). (DOCX) File S1.