OUP user menu

Microbiome analysis of dairy cows fed pasture or total mixed ration diets

Alexandre B. de Menezes , Eva Lewis , Michael O'Donovan , Brendan F. O'Neill , Nicholas Clipson , Evelyn M. Doyle
DOI: http://dx.doi.org/10.1111/j.1574-6941.2011.01151.x 256-265 First published online: 1 November 2011


Understanding rumen microbial ecology is essential for the development of feed systems designed to improve livestock productivity, health and for methane mitigation strategies from cattle. Although rumen microbial communities have been studied previously, few studies have applied next-generation sequencing technologies to that ecosystem. The aim of this study was to characterize changes in microbial community structure arising from feeding dairy cows two widely used diets: pasture and total mixed ration (TMR). Bacterial, archaeal and protozoal communities were characterized by terminal restriction fragment length polymorphism of the amplified SSU rRNA gene and statistical analysis showed that bacterial and archaeal communities were significantly affected by diet, whereas no effect was observed for the protozoal community. Deep amplicon sequencing of the 16S rRNA gene revealed significant differences in the bacterial communities between the diets and between rumen solid and liquid content. At the family level, some important groups of rumen bacteria were clearly associated with specific diets, including the higher abundance of the Fibrobacteraceae in TMR solid samples and members of the propionate-producing Veillonelaceae in pasture samples. This study will be relevant to the study of rumen microbial ecology and livestock feed management.

  • rumen microbiome
  • amplicon sequencing
  • metagenomics
  • 16S rRNA gene


Microorganisms form a dense and active community in the rumen and comprise a diverse array of bacteria, archaea, protozoa and fungi, with a wide variety of metabolic capabilities (Stewart et al., 1997). Rumen microbial community structures are influenced by a range of factors including diet, diurnal variations in animal metabolism (Dehority & Orpin, 1997) and cow breed (Guan et al., 2008). Diet is a key determinant of microbial composition in the rumen, influenced by the complexity of available substrates within feed. Most bovine diets are plant-based and rich in complex polysaccharides that enrich the rumen community for microorganisms capable of breaking down specific polymeric components in the diet (Krause et al., 2003). The end products of this primary degradation then sustain a chain of fermentative microorganisms that ultimately produces a range of organic acids together with hydrogen and carbon dioxide (Stewart et al., 1997). Some organic acids produced, such as propionate and butyrate, can be absorbed directly across the bovine gut wall to serve as an energy source for the animal (Stewart et al., 1997) and the hydrogen produced is used by ruminal archaea for methanogenesis (Boone et al., 1993; Stewart et al., 1997).

Because of the complex and interlinked nature of the rumen microbial community, changing the diet and thus substrates for primary degradation has a cascading effect on rumen microbial metabolism, with consequent changes in both the organic acid profiles and the methane levels produced (Wolin & Miller, 1997) impacting the quality and quantity of animal production. Rumen acidosis and bovine methane production are important examples of the interaction between rumen microbial metabolism and diet influencing animal performance (Johnson & Johnson, 1995; Goad et al., 1998; Owens et al., 1998; Tajima et al., 2000; Khafipour et al., 2009; Fernando et al., 2010). In intensive livestock production systems, diets are often grain based and rich in fermentable carbohydrates, resulting in enhanced microbial production of lactic acid and a consequent lowering of rumen pH with harmful consequences to the animal (Owens et al., 1998). Enteric methane production also represents a loss of energy to the animal (Johnson & Johnson, 1995) and is significant in terms of its impact as a greenhouse gas (Thorpe, 2009). In particular, high levels of dietary fibre are associated with increased methane production (Johnson & Johnson, 1995). A number of dietary based strategies have been used to reduce methane production, including supplementation with plant extracts, dietary lipids, halogenated methane analogues and ionophores (Hegarty, 1999; Mohammed et al., 2004; Beauchemin & McGinn, 2006; Foley et al., 2009; Place & Mitloehner, 2010).

The effect of diet on rumen microbial community structure has been investigated widely using culture-based and standard molecular methods (McAllister et al., 1996; Tajima et al., 2000, 2001; Edwards et al., 2004; McSweeney & Denman, 2007; Wright et al., 2007; Ellis et al., 2008; Hart et al., 2008; Patra & Saxena, 2009; Wanapat & Cherdthong, 2009). Next-generation sequencing has not been widely used in the study of rumen microbial ecology, and yet, such an approach would allow a much greater coverage of the microbial diversity and physiology of a complex environment such as the rumen (Hall, 2007). Next-generation sequencing has been successfully used to characterize the microbial ecology of a range of other complex environments, including the human gut, marine surface waters and deep sea vents (Miller et al., 2009; Poretsky et al., 2009; Wu et al., 2010). The few rumen-based studies that have used this approach include the profiling of the fibre-associated rumen metagenome and their glycoside hydrolases (Brulc et al., 2009), together with two studies focused on 16S rRNA gene amplicon profiles (Callaway et al., 2010; Pitta et al., 2010).

In this study, terminal restriction fragment length polymorphism (TRFLP) and next-generation sequencing have been used to assess the effects of a pasture-based and a total mixed ration (TMR) diet on rumen microbial community structure in dairy cows. TMR is a high-energy diet whose use might increase as EU regulations restrict land use for dairy farming from 2015 (O'Neill et al., 2011), and there are concerns that an increase in the use of TMR-based diets may lead to higher methane emissions. A parallel study has directly examined the effect of these diets on methane emission from cows and demonstrated that a TMR diet results in significantly higher methane emission (O'Neill et al., 2011).

Materials and methods

Experimental design

The experimental design followed that described by O'Neill (2011), except that four cows were rumen cannulated and used for rumen microbial community analysis. In brief, four rumen cannulated cows were randomly arranged in a Latin Square design, with two cows offered a pasture diet for 2 weeks, and then offered a TMR diet for a further 2 weeks. The other two cows were offered TMR for 2 weeks and then changed to pasture for 2 weeks. Cows offered a pasture diet were allowed to graze for 20 h day−1 (this excludes the time spent at milking in the morning and evening). Cows offered a TMR diet were offered a mixed feed each day that comprised (g kg−1 dry matter) 360 maize silage, 410 concentrate blend (375 soya bean meal, 335 rolled barley, 116 maize gluten meal, 115 citrus pulp, 27 mineral/vitamin premix, 22 vegetable fat, 10 limestone – all g kg−1 dry matter), 170 grass silage, 30 molasses and 30 straw. Cows allocated to the TMR group were housed indoors with a cubicle accommodation for the duration of the study. TMR was fed ad libitum, with at least a 5% feed refusal daily. Methane emission measurements were obtained from a parallel set of 48 noncannulated cows, fed identically at the same time (methane measurements cannot reliably be obtained from cannulated cows). These data and the methods used are reported in O'Neill (2011).

DNA extraction

Five hundred milliliters of rumen contents were sampled from each cow on two consecutive days after they had been offered either pasture or TMR for 14 days. Samples were then separated into liquid and solid phases by filtration through muslin. A total of 28 samples were collected, i.e. 4 cows × 2 diets (pasture or TMR) × 2 phases (liquid and solid) × 2 sampling times, −samples (4) from one of the cows fed the TMR diet that became sick during the experiment. Samples were immediately stored at −20 °C. DNA was then extracted from separate solid and liquid samples via a phenol–chloroform-based method (Griffiths et al., 2000) and was stored at −20 °C until required. Extracted DNA was qualitatively assessed and quantified using a Nanodrop spectrometer (Thermo scientific).

Community fingerprinting by TRFLP

Microbial assemblages were determined using TRFLP following a modification of the method of Liu (1997). Briefly, after the extraction and purification of DNA, the bacterial and archaeal 16S rRNA genes were amplified using the primer sets 27F–1492R (Lane, 1991) and arch-8F–1492R (Banning et al., 2005), respectively, whereas the protozoan 18S rRNA gene was amplified using the primers PPSUF–PPSUR (Sylvester et al., 2004). In each case, the forward primer was labelled with 6-FAM (Sigma). PCR reactions were performed in 50-μL volumes containing 10 μL of 5 × Mg-free PCR buffer, adjusted concentrations of MgCl2, 15 pmol of each primer, 200 μM of each dNTP, 25 μg bovine serum albumin, ∼10 ng of extracted total DNA and 2.5 U Go-Taq DNA polymerase (Promega, UK). The thermocycling conditions were as follows: 94 °C for 3 min (one cycle); 94 °C for 1 min, annealing temp, 72 °C for 2 min (26 cycles); and 72 °C for 7 min.

The resulting amplicon mixtures were then digested enzymatically as follows: approximately 50 ng of PCR product was added to a reaction mixture containing sterile Millipore water, 20 U of either restriction endonuclease TaqI (NEB, UK) for archaea or 20 U of Msp1 for the remaining groups, along with 2 μL of the corresponding enzyme buffer. Digests were performed in a final volume of 20 μL and incubated at either 60 or 65 °C for 12 h. Digests were desalted by precipitation in 100% (v/v) ethanol and 0.25 M sodium acetate, and left overnight at −20 °C. Tubes were then centrifuged at 14 000 g at 4 °C for 15 min, washed with 70% (v/v) ethanol and resuspended in nuclease-free water.

TRFLP fragment separation

All fragment lengths were determined by electrophoresis using an AB3031 automated sequencer (Applied Biosystems). Electrophoresis was carried out on a 36 cm capillary and fragments were separated at 60 °C and 4 kV for 120 min. Analysis of fragment profiles was performed using genemapper (Applied Biosystems) software. Fragments with relative peak heights <1% were regarded as a background and excluded from analysis. Fragments that differed by <0.5 bp in different profiles were considered identical. In addition, each sample was extracted in triplicate and fragments occurring in two of the three samples were included and merged. Fragments occurring only once in the three samples were excluded from further analysis. Statistical analysis followed (Dobretsov et al., 2005; Muckian et al., 2007; Scallan et al., 2008) using Bray–Curtis as a measure of similarity and with square root transformation of the data.

High-throughput, deep sequencing of bacterial 16S rRNA gene amplicons

Amplicon sequencing followed the methodology of Martinez (2009). The region extending from the V1 to the V2 variable regions of the 16S rRNA gene was amplified using primers containing the Roche-454 A or B Titanium sequencing adapters (shown in italics), followed by a unique eight-base barcode sequence (B) and finally the 5′ end of primer A-8FM, 5′-CCATCTCATCCCTGCGTGTCTCCGACTCAGBBBBBBBBAGAGTTTGATCMTGGCTCAG and B357R, 5′-CCTATCCCCTGTGTGCCTT-GGCAGTCTCAGCTGCTGCCTYCCGTA-3′.

PCR reactions from 28 samples were quality controlled for saturation by gel electrophoresis and quantified using genetools software (Syngene, Cambridge, UK). Amplicons from all samples were pooled in equal amounts, gel purified and quantified using picogreen (Invitrogen, Carlsbad) and a Qubit fluorimeter (Invitrogen), followed by sequencing using the Roche-454 Titanium platform.

Raw sequences were quality-filtered, based on the presence of the complete forward primer and barcode, the presence of ≤2 ‘N’ (where N is equivalent to an interrupted and resumed signal from sequential flows), size between 200 and 500 nt and a quality score ≥20. Following filtering, sequences were trimmed to remove the 3′ adapter and primer sequences and finally sorted according to barcode. Processed sequence files associated with each sample were stored in a database server and are publically available at http://cage.unl.edu. The CLASSIFIER algorithm was used to assign taxonomic status to each sequence (Wang et al., 2007; Liu et al., 2008). Sequences were classified to the lowest taxonomic level until a score of <0.8 was attained. Sequences with similarity scores of <0.8 were considered as ‘unclassified’ at the next taxonomic rank. Each taxon is reported here as its proportion to the total number of sequences in a sample. The proportions of each taxon in each sample were averaged for each treatment and rumen phase.


The effects of two diets (pasture, TMR) were assessed in a Latin Square design; two cows were offered a pasture diet for 2 weeks and then switched to TMR, with the two other cows started on TMR and then offered pasture. Microbial community structure was assessed in both the liquid and the solid phases of rumen contents, collected at the end of each diet (pasture or TMR) regime. Initially, TRFLP was used to assess changes across a number of microbial groups, including bacteria, archaea, and protozoa. Nonmetric multidimensional scaling (nMDS) ordination was used to visualize differences in the community structure between the diets and the rumen content phases (Fig. 1a–c). nMDS clearly showed a separation between both the diet and the rumen phase in bacterial communities. In archaeal communities, permanova statistics indicated that there was a clear separation between rumen liquid and solid phases, but separation was less obvious between diets. Neither the diet nor the rumen phase appeared to affect protozoal communities. permanova analysis was carried out to determine the strength of the effects noted by nMDS (Table 1). Diet had a significant effect on both bacteria (P=0.001) and archaea (P=0.001), but did not significantly affect protozoa.


nMDS plots of TRFLP profiles generated from amplified bacterial and archaeal 16S rRNA gene (a and b, respectively) or the protozoal 18S rRNA gene (c) from rumen DNA. Each cow is represented by number and liquid and solid refer to liquid and solid rumen content.

View this table:

permanova values for TRFLP analysis of the bacterial, archaeal and protozoal rumen communities

GroupP-value dietP-value form
  • Diets are pasture and TMR and forms are liquid or solid rumen content.

TRFLP analysis combined with permanova showed that bacterial communities were clearly affected by diet. On this basis, combined with the view that rumen bacteria have the greatest metabolic diversity, high-throughput sequencing was used to quantify changes in the bacterial community and to identify which particular bacterial groups responded to diet. Deep sequencing of 16S rRNA gene amplicons using the 454 Titanium platform was used. Around 400 000 good-quality sequences were obtained, and the sequence length varied between 192 and 492 bp. Sequences were uploaded at the Core for Applied Genomics database (http://gutmicro.unl.edu/ClientLogin/login.php), where they are publically available. Rarefaction analysis (Fig. 2) indicated that bacterial diversity was higher in TMR samples compared with pasture and also higher in liquid compared with the solid phase.


Rarefaction analysis of rumen 16S rRNA gene amplicons of each treatment and condition based on clusters at 0.03 and 0.06 nucleotide identity. PSTR and TMR represent pasture and TMR diets; liquid and solid refer to liquid and solid rumen content.

nMDS of 16S amplicons was once again used to visualize the effects of diet and rumen phase on bacterial community structure, as assessed by high-throughput sequencing, at different taxonomic levels (Fig. 3). At the phylum level, there was an obvious separation between communities in liquid and solid phases, but the separation between diets was less evident. At lower taxonomic levels (class, order, family), similar trends were observed, but with much clearer separation for both diet and rumen phase. This was again reinforced by permanova analysis (Table 2), with both diet and rumen phase having significant effects at all taxonomic levels. This effect was slightly less pronounced at the phylum level, with a lower significance for diet (P=0.04).


nMDS plots of bacterial 16S rRNA gene sequence tags generated from amplified bacterial 16S rRNA gene at phylum (a), class (b), order (c) and family (d) levels. Each cow is represented by a number and liquid and solid refers to liquid and solid rumen content.

View this table:

permanova values for amplicon sequencing analysis of the rumen bacterial communities at different taxonomic levels

GroupP-value dietP-value form
  • Diets are pasture and TMR and forms are liquid or solid rumen content.

At the phylum level, for all samples, around 90% of the sequences could be classified, with Firmicutes and Bacteroidetes dominating, typically together representing around 80% of the total sequences (Fig. 4a). 14 bacterial phyla were detected overall. Among other phyla, Fibrobacteres and Spirochaetes were considerably more prevalent in the solid phase, whereas Actinobacteria was much more evident in the liquid phase. SIMPER analysis revealed that the overall dissimilarity was 10.5% between pasture and TMR and 14.9% between liquid and solid phases. Comparing individual phyla between pasture and TMR diets, Bacteroidetes and Firmicutes were the main phyla responsible for the level of dissimilarity observed (19.3% and 18.7%, respectively), followed by Fibrobacteres (14.3%) and Proteobacteria (13.7%). The situation was slightly different when comparing the rumen phases, with Fibrobacteres being responsible for the highest dissimilarity (20.9%), followed by Bacteroidetes (17.7%), Firmicutes (14.5%) and Spirochaetes (14.2%). Interestingly, Spirochaetes did not contribute to differences between pasture and TMR diets.


Percentage contribution of sequences belonging to each phylum (a), class (b) and family (c) to the total number of sequences in the database. Only classes contributing >0.1% and families contributing >0.2% of the total sequences in at least one treatment were included in the analysis.

When sequences were analysed at the class level (Fig. 4b), Firmicutes were dominated by Clostridia, with Erysipelotrichi also making a significant contribution particularly in cows fed a pasture diet. Among Proteobacteria, the Alpha, Beta and Gamma classes were detected, and among Bacteroidetes, around 20–30% were unclassified. When analysed at the family level, 83 families were detected, with 15 families contributing >0.2% of the total sequences in any single treatment or phase (Fig. 3c). The most prevalent families were the Prevotellaceae, the Lachnospiraceae, the Ruminococcaceae and the Fibrobacteriaceae, together with unclassified groups of Bacteriodales and Clostridiales. Generally, members of the Prevotellaceae were more represented in cows fed a pasture diet, members of the Lachnospiraceae were more prevalent in rumen solid samples from cows fed pasture and both members of the Spirochaetaceae and the Fibrobacteraceae were prevalent in the solid phase. Sequences of the Coriobacteriaceae were most abundant in the liquid phase. Interestingly, in rumen liquid from cows fed pasture, around 15% of the sequences were from the Erysipelotrichaceae; this family was present at <3% in other treatments/phases.


TRFLP analysis revealed that both diets (pasture, TMR) and the rumen phases (liquid, solid) had distinct bacterial and archaeal communities, whereas the protozoal community was little affected by diet or rumen phase. It is not clear why the protozoal community was unaffected by diet or rumen phase. The main factor shaping protozoal community structure was the individual animal itself, with pronounced differences between animals. It is clear therefore that protozoan communities were very similar between rumen liquids and solids and that any response that might have occurred to the change in diets was not pronounced enough to override the differences between the protozoal communities in each cow. This may be because protozoans are known to selectively retain themselves in the rumen (Jouany et al., 1988). The finding that the bacterial community was significantly affected by diet was not surprising, as bacteria are involved in the first steps of biomass breakdown in the rumen as well as in the fermentation of breakdown products (Stewart et al., 1997). The distinct bacterial communities that developed with the two diets probably vary in their metabolic potential, resulting in different metabolites becoming available for downstream methanogenic activity (Stewart et al., 1997; Ellis et al., 2008) and altering methanogen diversity.

In this study, next-generation sequencing approaches proved to be a powerful tool to reveal rumen bacterial diversity and the associated effects of diet on that diversity. It also yielded much greater coverage of bacterial diversity than could be provided by other molecular diversity techniques such as TRFLP or cloning. Importantly, treatment effects on less abundant groups, which may be significant at a functional level, can be assessed. In general, culture-based and standard molecular approaches have indicated that the rumen bacterial community is dominated by Bacteroidetes and Firmicutes (Stewart et al., 1997; Wallace, 2008). Typical rumen bacteria include cellulose/fibre degraders such as Fibrobacter spp. and Ruminoccocus spp., bacteria that degrade more readily metabolizable carbohydrates such as Prevotella spp. and Ruminobacter spp., and those that ferment and convert different types of fatty acids such as members of the genera Megasphaera, Succinovibrio and Treponema (Stewart et al., 1997). Similarly, using high-throughput amplicon sequencing, the most abundant groups observed in this study also belonged to the phyla Firmicutes, Bacteroidetes and Fibrobacteres. Notably, the family Fibrobacteraceae accounted for almost 10% of the sequences in the TMR solid phase, whereas it was only a minor component in the studies by Pitta (2010) and Callaway (2010). Fibrobacter DNA is known to be difficult to amplify (Tajima et al., 2001) and the difference between these studies may reflect PCR bias. Similarly, some genera that were abundant in these other studies were only minor components in this study, such as Rikenella and Tannerella in the study by Pitta (2010) and Roseburia and Enterococcus in the study by Callaway (2010). Overall, such differences probably reflect the inherent differences between cows at different localities with variations in dietary matter and highlight the need for further research documenting the rumen microbial ecosystems in cows of different provenance. Common features between this study and the two other rumen 16S rRNA gene studies included the dominance of Prevotella, and significant abundances of Treponema, Butyrivibrio and Ruminococcus.

Diet had a clear effect in shaping the microbial communities in this study. For example, there was a higher abundance of the Fibrobacteraceae in cows fed TMR. This may reflect the presence of straw in this diet, increasing the levels of lignocellulose, and also the higher digestibility of the pasture-based diet (Connolly et al., 2009). Another family that was affected by diet was the Erysipelotrichaceae, which was significantly more abundant in cows fed pasture. The role of members of this family in rumen microbial fermentation is unknown, with most published data reporting their role as skin pathogens (Verbarg et al., 2004), although 16S rRNA genes for this group have also been reported in the ovine rumen (Perumbakkam et al., 2011).

An interesting aspect of this work is the availability of a parallel study that revealed significant differences in methane emissions between cows fed exactly the same diets as the ones in this study. Methane measurements were not performed on the cows used herein due to concerns that cannulation might have interfered with methane estimation. However, in the parallel experiment conducted at the same time and location and in which identical feed regimes were used, cows fed a pasture-based diet produced 60% less methane per cow, and, when corrected for dry matter intake, an 11% reduction occurred (O'Neill et al., 2011). This study may therefore offer possible links between specific bacterial groups and methane output. In particular, higher abundances of members of the Prevotellaceae, Erysipelotrichaceae and Veillonellaceae were observed in cows fed pasture. Very little is known about the metabolism of the Erysipelotrichaceae (Verbarg et al., 2004); however, both the Prevotellaceae and the Veillonellaceae include members that are known to produce propionate as a major fermentation product (Sijpesteijn & Elsden, 1952; Chen & Wolin, 1981; Counotte et al., 1981; Strobel & Russell, 1991; Strobel, 1992; Kishimoto et al., 2006; Chiquette et al., 2008). The abundance of the Veillonelaceae in particular was around three times higher in cows fed pasture, comprising up to 3% of all sequences. Propionate production is often associated with lower methane emission (Denman et al., 2007; Patra, 2010; Watanabe et al., 2010) and higher abundance of propionate-producing bacteria might have diverted H2 away from methanogenesis reducing methane emission levels, as was observed with pasture-fed cows.

In conclusion, diet had a significant effect in shaping the microbial community in the rumen of dairy cows, in particular that of the bacterial and archaeal communities, whereas the protozoan community was mostly unaffected. Detailed knowledge has been generated about the effects of two widely used and contrasting diets on the rumen bacterial community. This study should be of great interest for researchers investigating rumen function, rumen microbial ecology, livestock methane mitigation strategies and feed management.


This study was supported by the Irish Department of Agriculture Fisheries and Food Research Stimulus Fund (Grant no. RSF 07 517). We would like to acknowledge Dr Bas Boots for advice on statistical analysis.


View Abstract