You are on page 1of 12

Metabolomics (2010) 6:583–594 DOI 10.

1007/s11306-010-0227-6

ORIGINAL ARTICLE

Metabolomics reveals unhealthy alterations in rumen metabolism with increased proportion of cereal grain in the diet of dairy cows
Burim N. Ametaj • Qendrim Zebeli • Fozia Saleem • Nikolaos Psychogios • Michael J. Lewis • Suzanna M. Dunn • Jianguo Xia • David S. Wishart

Received: 4 February 2010 / Accepted: 25 June 2010 / Published online: 14 July 2010 Ó Springer Science+Business Media, LLC 2010

Abstract This study presents the first application of metabolomics to evaluate changes in rumen metabolites of dairy cows fed increasing proportions of barley grain (i.e., 0, 15, 30, and 45% of diet dry matter). 1H-NMR spectroscopy was used to analyze rumen fluid samples representing 4 different diets. Results showed that for cows fed 30 and 45% grain, increases were observed in the concentration of rumen methylamine as well as glucose, alanine, maltose, propionate, uracil, valerate, xanthine, ethanol, and phenylacetate. These studies also revealed lower rumen 3-phenylpropionate in cows fed greater amounts of cereal grain. Furthermore, ANOVA tests showed noteworthy increases in rumen concentrations of N-nitrosodimethylamine, dimethylamine, lysine, leucine, phenylacetylglycine, nicotinate, glycerol, fumarate, butyrate, and valine with an enriched grain diet. Using principal component analysis it was also found that each of the 4 diets could be distinguished on the basis of the measured rumen metabolites. The two closest clusters corresponded to the 0 and 15% grain diets, whereas the 45% barley grain diet was significantly separated from the other clusters. Unhealthly levels of a number of potentially toxic metabolites were found in the rumen of cattle fed 30 and 45%
Electronic supplementary material The online version of this article (doi:10.1007/s11306-010-0227-6) contains supplementary material, which is available to authorized users.
B. N. Ametaj (&) Á Q. Zebeli Á S. M. Dunn Department of Agricultural, Food and Nutritional Science, University of Alberta, Edmonton, AB T6G 2P5, Canada e-mail: burim.ametaj@ualberta.ca F. Saleem Á N. Psychogios Á M. J. Lewis Á J. Xia Á D. S. Wishart Departments of Biological Sciences and Computing Science, University of Alberta, Edmonton, AB T6G 2E9, Canada e-mail: david.wishart@ualberta.ca

grain diets. These results may have a number of implications regarding the influence of grain on the overall health of dairy cows. Keywords Rumen metabolic profile Á NMR Á Dairy cow Á Principal component analysis Á Hierarchical clustering analysis Á Barley grain

1 Introduction Dairy cows have evolved through a close symbiotic relationship with a vast ensemble of rumen microbes (i.e., the microbiota). These microbes give dairy cattle important metabolic capabilities including the ability to digest cellulose-rich feedstuffs and to convert them into a wide range of compounds important for body maintenance and milk production. Keeping a stable and healthy rumen microbiota is also critical for maintaining healthy cows (Nocek 1997; Emmanuel et al. 2008) and high milk quality (Jenkins and McGuire 2006). Today’s intensive dairy management systems encourage the inclusion of large amounts of cereal grains in the diets of dairy cattle to support high milk yields and enhance cost efficiency. Barley grain, a cereal characterized by rapid degradation of its starch in the rumen, is widely used in Western Canada and elsewhere as one of the main energy components in the diets of dairy cows (15–40% in diet dry matter) and beef cattle (40–85%). Yet, these feeding practices often lead to the accumulation of large amounts of short-chained fatty acids (SCFA) and to the acidification of rumen fluid. This often leads to a pathological condition commonly referred to as acute or sub-acute rumen acidosis (ARA or SARA; Nocek 1997). Moreover, feeding diets rich in readily available carbohydrates is associated with

123

584

B. N. Ametaj et al.

major alterations to the rumen microbiota, which are accompanied by significant changes to the rumen metabolic pathways (Tajima et al. 2000; Khafipour et al. 2009). Several lines of evidence support a role for altered rumen metabolism, (due to the feeding of high grain/low fiber diets) in the development of several pathologic conditions including laminitis (Nocek 1997), fatty liver (Ametaj et al. 2005), systemic inflammation (Emmanuel et al. 2008), and milk fat depression syndrome (Jenkins and McGuire 2006; Zebeli and Ametaj 2009). Although the interrelationships of rumen metabolism and host health status are widely accepted, it is not yet clear which of the compounds generated by the activity of rumen microbiota, under conditions of diet-induced stress (i.e., high grain feeding), are significant to a cow’s health. Moreover, little is known with respect to microbial-mediated changes in rumen metabolism during the feeding of graded proportions of cereal grain in the diet. So far, most conventional studies addressing rumen metabolism, have focused on dietary effects on a single class of rumen compounds, such as SCFA, and have not attempted a comprehensive metabolomic characterization of the rumen of dairy cattle. Metabolomics is an emerging field of ‘‘omics’’ science that uses high-throughput approaches, such as 1H-NMR spectroscopy, coupled with multivariate analysis to extract comprehensive metabolic information and measure metabolic phenotypes in mammals, plants, and microbes (Vinayavekhin et al. 2010). Metabolomics has opened new perspectives in the field of nutrition research, allowing scientists to explore the complex metabolic pathways in response to diet (Wishart 2008a). To our knowledge, only a few studies have used modern metabolomics techniques to address nutritional issues in ruminant animals. One study looked at the effects of intra-ruminal infusion of SCFA in non-lactating heifers (Bertram et al. 2005), another study investigated the metabolism of rumen epithelial tissue of newborn calves (Bertram et al. 2009), and a third study used metabolomics to look at the quality of marketed milk (Boudonck et al. 2010). Recent research has also shown that metabolomics can be used to characterize metabolic profiles at the gut level. These studies allow one to explore the symbiotic complexity between the host and its microbiota as affected by diet, and to unravel the effects of diet on the development of metabolic diseases (Dumas et al. 2006). Given this potential we hypothesized that the use of metabolomics would provide a more comprehensive view on how rumen metabolism changes during the feeding of graded proportions of cereal grain. Such a study could improve our understanding of diet-related rumen metabolism and give further insight into metabolite-disease associations in dairy cattle. For this investigation we used 1 H-NMR spectroscopic profiling along with multivariate statistics to determine the influence of feeding graded

amounts of barley grain on rumen metabolite patterns in lactating dairy cows.

2 Materials and methods All experimental procedures were approved by the University of Alberta Animal Care and Use Committee for Livestock, and animals were cared for in accordance with the guidelines of the Canadian Council on Animal Care (1993). Data related to concentration of endotoxin and pH of the rumen fluid, plasma acute phase proteins, and milk production for this experiment have been reported previously (Emmanuel et al. 2008; Zebeli and Ametaj 2009). In this manuscript, we reinterpret previous data collected on rumen pH and endotoxin in the context of the metabolomic profile changes evaluated in this research. 2.1 Animals and diets Eight clinically healthy, rumen-fistulated primiparous Holstein cows (60 ± 15 days in milk) at the University of Alberta’s Dairy Research and Technology Centre were used in this study. The original experimental design of this research was a replicated 4 9 4 Latin square design with four 21-day periods (Emmanuel et al. 2008; Zebeli and Ametaj 2009). However, because of the complexity of the study only the rumen samples collected during the first experimental period were used in this report. The first 11 days were used for adaptation to the experimental diets, and the last 10 days for collection of rumen fluid samples. Based on previous similar feeding experiments, an adaptation period of 11 days was considered sufficient to allow wash-in of the experimental diets before the measurements. Cows were fed one of the total mixed rations (TMR) containing increasing proportions of rolled barley grain (i.e., 0, 15, 30, or 45% on a dry matter basis). Thus, 2 different cows were randomly allocated to each diet. All diets were formulated to meet the nutrient requirements of a 680 kg lactating cow as per NRC (2001) guidelines. Ingredients and nutrient composition of the 4 TMR are presented in the Supplementary Table 1. 2.2 Rumen fluid sample collection For this study, samples from rumen fluid (100 mL) were obtained on days 1, 3, 5, 7, and 10 of the measurement period (i.e., days 12, 14, 16, 18 and 21 of the experiment) at 0800 h. All rumen fluid samples were collected through the cannula using a tube fitted with a strainer and a syringe into a 140-mL plastic container. The pH of the rumen fluid was determined immediately after collection by a mobile pH meter (Accumet AP61, Fischer Scientific, Ottawa, Ontario,

123

Rumen metabolomics

585

Canada). Subsequently, rumen fluid samples were centrifuged at 6,0009g for 15 min (Rotanta 460 R, Hettich Zentrifugan, Tuttlingen, Germany), and the supernatant was stored at -20°C until it was analyzed for metabolite and endotoxin content. 2.2.1 Rumen fluid sample preparation for NMR spectroscopy Rumen samples were thawed at room temperature and centrifuged at 10,000 rpm for 140 min to remove macromolecules (fine feed particles and microbiota). The supernatant was collected and passed through 0.2 lM syringe filters (Fischer Scientific, Fairlawn, NJ). Subsequently, 35 lL of D2O and 15 lL of a standard buffer solution (0.16 mM DSS [disodium-2,2-dimethyl-2-silapentane-5sulphonate], 10 mM imidazole, and 0.02% NaN3 in H2O) were added to 300 lL of filtered rumen sample. Subsequently, these samples (350 lL) were transferred to a standard Shigemi microcell NMR tube for subsequent NMR spectral analysis. 2.3 1H-NMR spectrum acquisition All 1H-NMR spectra were collected on a 500 MHz Inova (Varian Inc., Palo Alto, CA) spectrometer equipped with either a 5 mm HCN Z-gradient pulsed-field gradient (PFG) room-temperature probe or a Z-gradient PFG Varian coldprobe. 1H-NMR spectra were acquired at 25°C using the first transient of the tnnoesy-presaturation pulse sequence, which was chosen for its high degree of quantitative accuracy (Saude et al. 2006). Spectra were collected with 64 transients and 4 steady state scans using a 4 s acquisition time and a 990 ms solvent presaturation. A spectral width of 5,997 Hz was used for collecting 24 k data points. Spectra were processed using the Processor module of Chenomx NMR Suite 6.0. The free induction decays (FIDs) were zero filled to 64 k in the frequency domain, and multiplied by an exponential weighting function corresponding to a line broadening (0.5 Hz) before Fourier transformation. Spectra were referenced to DSS (0.0 ppm) and manually corrected for phase and baseline distortions. 2.4 Sample fitting Processed spectra were imported into the Chenomx NMRSuite 6.0 software for quantification (Weljie et al. 2006; Wishart 2008b). In total, 46 compounds were quantified from each spectrum. Briefly, test spectra were chosen and concentrations for each compound were averaged over the test profiles. These averages were used as the starting concentration vector for fitting or refitting all the

acquired spectra. All spectra were randomly ordered (within acquisition batches) for spectral fitting using Chenomx Profiler. For each spectrum, the compounds identified were sorted by decreasing concentration and then fit for concentration and translation in that order. The three test profiles were compared with the corresponding profiles to test for consistency. Each compound concentration was then normalized by dividing the measured concentration into the total concentration of all metabolites in that sample (Weljie et al. 2006). We also used sample spiking to confirm the identity of several spectral signatures seen in our NMR spectra. This was conducted by adding 50–500 lM of the presumptive compound to selected rumen samples to test if the corresponding 1H-NMR signals changed as expected. 2.5 GC-MS compound identification An additional step for compound identification was performed, by selecting 2 rumen spectra from 0% grain diet for Gas Chromatography-Mass Spectrometry (GC-MS) analysis. The samples were extracted separately to obtain separate pools of polar and lipophilic metabolites using previously reported protocols (Wishart et al. 2008). Derivatized polar extracts and lipophilic extracts were analyzed using an Agilent 5890 Series II GC-MS instrument operating in an Electron Impact (EI) ionization mode. For GC-MS analysis, 2 lL of extract were injected using a split/splitless injector with a split ratio 5:1 onto a DB-5 capillary column (Agilent J&W Scientific, 30 m 9 250 lm 9 0.25 lm). The injector port temperature was held at 250°C and the helium carrier gas flow rate was set to 1 mL/min at the initial oven temperature of 50°C. The oven temperature was increased at 10°C/min to 310°C for a final run time of 26 min. The full scan mode of the quadrupole MS was used at a mass range of 50–500 m/z, with a solvent delay of 6 min. The MS ion source temperature was set to 200°C. The AMDIS spectral deconvolution software (Version 2.62) from NIST (National Institute of Standards and Technology) was used to process the total ion chromatogram and the EI-MS spectra of each GC peak. After deconvolution, the purified mass spectrum of each of the trimethylsilylated metabolites was identified using the NIST MS Search program (version 2.0d) which was linked to the NIST mass spectral library (2008). In AMDIS, each search produces a list of library spectra (‘‘hits’’), which is ranked by the similarity to the target spectrum according to a mathematically computed ‘‘match factor’’. The match factor indicates the likelihood that our spectrum and the reference NIST spectrum arose from the same compound. In the current case, we considered hits with a match factor of [60% and a probability [20%. These cutoff values have been shown to yield consistent results with synthetic

123

586

B. N. Ametaj et al.

mixtures containing known compounds and with samples independently analyzed by NMR. Retention Indices (RIs) were calculated using a C8-C20 alkane standard solution (Fluka, Sigma-Aldrich). 2.6 Rumen endotoxin Concentration of cell-free endotoxin in the rumen fluid supernatant was determined by the pyrochrome Limulus amebocyte lysate assay (Associates of Cape Cod Inc., East Falmouth, MA), as described previously (Emmanuel et al. 2008). Briefly, 10 mL of rumen fluid samples were centrifuged at 6,0009g for 15 min and the supernatant stored at -20°C until analysis. For the assay, the supernatant was thawed and 1.5 mL centrifuged at 10,0009g for 30 min. All samples were tested in duplicate, and the optical density values were read on a microplate spectrophotometer (Spectramax 190, Molecular Devices Corporation, Sunnyvale, CA) at a wavelength of 405 nm. The inter- and intraassay CV were controlled to limits of \10%. 2.7 Statistical analyses Parametric analysis of the data was conducted using the MIXED procedure provided by SAS (SAS Institute Inc., Cary, NC, Version 9.1.3). The model included the fixed effect of treatment and the ‘‘random’’ effect of the individual cow. The measurement day was considered as a repeated measure with a first-order autoregressive covariance structure, which assumes that the covariance of measurements for the same animal decays with time (Littell et al. 1998). Instead of raw P values, the Bonferroni correction and the false discovery rate (FDR) were applied to counter the effect of multiple hypothesis testing, since they are considered more appropriate for analyzing the multivariate results found in metabolomic data (Broadhurst and Kell 2006). For multivariate analysis, data were imported into the MetaboAnalyst software (http://www.metaboanalyst.ca; Xia et al. 2009). Principal component analysis (PCA), which is an unsupervised pattern recognition method, was performed to examine the intrinsic variation in the NMR data set (39 in total; n = 9 for diet 0% grain, and n = 10 for the other dietary treatments), and to reduce the dimensionality of the data. A score plot was used to show the similarities and differences among the data sets. In a score plot the data sets exhibiting similarities are clustered together and those that are different are placed further apart. The loadings plot shows the variables responsible for the variation within the dataset, and the correlations among individual rumen fluid metabolites corresponding to the first two eigenvalues (i.e., PC 1 and PC 2). Sample normalization against baseline concentration at 0% grain diet was done by first creating a dummy baseline sample by

averaging all the nine samples collected at 0%. After this normalization step, all samples were normalized against this reference sample using probabilistic quotient normalization procedures (Dieterle et al. 2006; Xia et al. 2009). The compound normalization was conducted using autoscaling, which made all compound variance equal to 1. Hierarchical clustering analysis (HCA) with Euclidean distance measures and an average linkage method was performed to explore the presence of clustering patterns among the rumen fluid metabolites. Pretreatment of the data in HCA included data normalization (x - m/std), where x denotes given values, m is the mean, and std indicates the standard deviation of a certain variable, as well as conversion to log base 2 ratios. The expression patterns and a heat map of each variable were categorized using an average linkage hierarchical clustering program (Seo 2005).

3 Results 3.1 Outcome of ANOVA tests Results of the ANOVA tests with Bonferroni and FDR corrections with regard to metabolites measured in the rumen fluid are presented in Table 1. Our data showed that feeding dairy cows increasing proportions of barley grain was associated with statistically significant increases of several rumen metabolites including methylamine (MA), glucose, alanine, maltose, uracil, xanthine, and phenylacetate. Moreover, the analysis of variance indicated greater levels of rumen endotoxin and acetate. ANOVA also indicated potential increases in the concentrations of propionate, ethanol (EtOH), valerate, N-nitrosodimethylamine (NDMA), dimethylamine (DMA), lysine, leucine, phenylacetylglycine (PAG), nicotinate, glycerol, fumarate, butyrate, and valine, whereas rumen fluid pH was lower with increasing dietary grain (Table 1). On the other hand, greater proportions of barley grain in the diet resulted in lower concentrations of 3-phenylpropionate (3-PP) in the rumen fluid. 3.2 Multivariate analysis of rumen metabolites Typical 1H-NMR spectra of rumen fluid corresponding to diets with 0, 15, 30, and 45% barley grain are shown in Fig. 1. All NMR spectra were collected on samples obtained on day 18 of the experiment (day 7 of the measurement period). Although there were some day-to-day differences in the responses of rumen metabolomic profile, the day 18 spectra were considered as typical responses of the cows to the different diets due to relatively long time that the cows were under the feeding regimen.

123

Rumen metabolomics

587

Table 1 Concentrations of rumen metabolites and endotoxin measured in rumen fluid of lactating dairy cows fed graded amounts of barley grain Itemb Barley grain proportion (% of diet dry matter) 0 Metabolite concentrations (lM, unless otherwise stated) 1,3-Dihydroxyacetone (1,3-D) 3-Hydroxybutyrate (3-HB) 3-Hydroxyphenylacetate (3-HP) 3-Phenylpropionate (3-PP) Acetate (mM) Acetoacetate Alanine Aspartate Benzoate Butyrate (mM) Cadaverine Caffeine Choline Dimethylamine Ethanol Ferulate Formate Fumarate Glucose Glutamate Glycerol Glycine Histidine Hypoxanthine Isobutyrate Isoleucine Isovalerate Lactate Leucine Lysine Maltose Methanol Methylamine N-nitrosodimethylamine (NDMA) Nicotinate Phenylacetate Phenylacetylglycine (PAG) Proline Propionate (mM) Ribose Succinate Tyrosine Uracil Valerate (mM) Valine Xanthine 15.6 44.8 36.1 446 47.6 49.5 197 157 36.9 7.97 73.2 15.3 17.2 4.36 131 4.41 321 1.64 400 317 93.9 104 31.8 87.9 1280 144 891 124 109 116 49.9 24.1 34.3 67.9 21.5 241 62.1 130 15.0 237 82.0 46.5 92.6 1.41 109 68.7 7.2 60.6 56.7 396 43.8 47.1 212 164 37.4 6.89 56.2 10.8 14.7 7.59 176 11.8 337 5.82 580 342 99.1 105 29.7 134 1283 124 720 122 99.2 105 53.7 22.9 136 49.3 24.7 363 55.0 117 17.9 238 84.5 41.3 149 2.03 109 101 7.9 46.7 45.7 322 41.4 54.1 397 201 53.9 6.42 93.6 15.7 16.9 29.5 203 3.91 320 1.42 720 442 148 158 35.2 219 1216 222 932 228 164 125 51.7 28.0 202 86.8 35.7 371 80.1 165 18.0 298 103.6 52.2 193 2.51 166 137 6.4 77.1 42.9 328 56.5 58.0 415 123 38.2 19.3 110 13.7 19.8 33.7 341 11.9 258 6.44 2090 425 170 150 24.8 220 1430 148 998 133 155 197 124 71.9 608 97.0 32.2 443 88.7 141 27.6 219 80.7 50.6 224 2.70 173 155 5.64 11.8 6.88 29.2 2.04 10.1 26.8 23.1 6.22 1.22 19.7 3.36 5.49 10.8 50.9 2.88 36.7 0.57 120 62.3 26.7 24.9 4.35 59.9 121.8 29.9 98.1 28.5 19.8 24.3 21.1 23.8 45.8 9.87 3.60 33.6 11.4 30.2 2.45 46.4 15.4 8.90 16.2 0.30 25.2 8.79 1.00 0.96 0.51 0.04 \0.01 1.00 0.10 0.65 0.69 0.40 1.00 1.00 1.00 0.77 0.49 0.71 1.00 0.33 \0.01 1.00 0.77 1.00 1.00 1.00 1.00 0.47 0.82 0.34 0.92 0.41 0.19 1.00 \0.01 0.22 0.32 0.01 0.38 1.00 0.12 1.00 1.00 1.00 \0.01 0.26 0.91 \0.01 0.731 0.517 0.499 0.031 0.003 0.914 0.040 0.522 0.396 0.280 0.639 0.928 0.943 0.344 0.357 0.439 0.950 0.255 0.002 0.589 0.410 0.470 0.641 0.538 0.723 0.307 0.539 0.149 0.442 0.246 0.072 0.512 0.001 0.207 0.220 0.010 0.305 0.808 0.100 0.846 0.827 0.902 0.026 0.355 0.291 0.002 15 30 45 SEMc Statisticsa BON FDR

123

588 Table 1 continued Itemb Barley grain proportion (% of diet dry matter) 0 Endotoxin (mg/L) Rumen pH
a b c

B. N. Ametaj et al.

SEMc 45 7.90 6.52 0.80 0.08

Statisticsa BON \0.01 0.32 FDR 0.001 0.210

15 0.81 6.70

30 6.31 6.85

0.64 6.81

BON Bonferroni correction, FDR false discovery rate (Broadhurst and Kell 2006); Significances were considered at p = 0.05 Rumen fluid for the analysis was collected shortly before the morning feeding, and the cows were fed once daily at 0800 h SEM standard error of the mean

Fig. 1 Typical 500 MHz 1H-NMR rumen fluid spectra (0–9 ppm) from dairy cows fed 0% (a), 15% (b), 30% (c), and 45% (d) barley grain. Numbers indicate metabolites, as follows: 1, acetate; 2, propionate; 3, butyrate; 4, valerate; 5, isovalerate; 6, DSS; 7, 3-phenylpropionate; 8, alanine; 9, isoleucine; 10, glucose; 11, proline; 12, isobutyrate; 13, phenylacetate; 14, glycine; 15, cadaverine; 16, aspartate; 17, ethanol; 18, leucine; 19, succinate; 20, glutamate; 21, methylamine; 22, lactate; 23, choline; 24, fumarate; 25, acetone; 26,

isopropanol; 27, 3-hydroxybutyrate; 28, acetoacetate; 29, 3-hydroxyphenylacetate; 30, N-nitrosodimethylamine; 31, methanol; 32, valine; 33, lysine; 34, dimethylamine; 35, glycerol; 36, maltose; 37, imidazole; 38, formate; 39, uracil; 40, nicotinate; 41, tyrosine; 42, benzoate; 43, histidine; 44, hypoxanthine; 45, phenylacetylglycine; 46, xanthine; 47, ribose. The intensity of the aromatic region (5.1–9.0 ppm) is 4 times higher than that of the aliphatic one (0.0–4.2 ppm)

Figure 2a shows the score plot of the first two principal components (PCs), extracted in this study. The 4 ellipses indicate 95% bivariate normal ellipses that summarize the distribution of the principal component scores for the 4 different diets. Inspection of this figure shows clear clusters corresponding to cows fed different amounts of barley grain, which appeared to be well separated by both PC 1 and PC 2. The ellipses corresponding to the responses of diets with 0 and 15% grain strongly overlapped with each other. However the responses of cows fed 45% grain are clustered further apart from those corresponding to diets 0

and 15%, whereas the cluster of cows with 30% grain in their diet is located between the clusters corresponding to diets with 0, 15, and 45% grain (Fig. 2a). As shown in the responses of specific cows at 5 different sampling days, the PCA revealed certain inter-day variation, whereby the responses of the last day (day 10) of the measurement period were clustered closer to each other, near the lower central region of the graph (Fig. 2a). Loading plots, showing the individual rumen fluid metabolites responsible for the variation of the first two eigenvalues (PC 1 and PC 2), are given in Fig. 2b. This

123

Rumen metabolomics

589

gives a graphical representation of the extent to which each factor accounts for the variance in the data and shows the relationship between the different rumen variables. Determinants with the same distance from 0 and with similar directions are positively correlated. Those in the opposite direction are negatively correlated. The first PC essentially shows a contrast between amino acids and glucose degradation products as well as rumen pH. The second PC highlights the contrast between the primary amines and branched chain fatty acids and aspartate (Fig. 2b). The metabolite methylamine (MA) appears to play an important role in the separation along the PC 2. The results of hierarchical clustering analysis (HCA) are presented in Fig. 3. This dendrogram shows the presence of different sub-clusters having different numbers of metabolites with various degrees of similarity. The responses of each variable, to four different diets, are indicated with changes in the color intensity on the heatmap. The variables with higher similarity are positioned near the top of the dendrogram. Interestingly, the HCA revealed the presence of one cluster consisting of the amino acids glycine, valine, leucine, glutamate, and isoleucine as well as lactate and a semi-volatile organic compound such as NDMA. Another cluster included the branched chain fatty acids isobutyrate and isovalerate as well as tyrosine and phenylacetate. A distinct feature of the HCA was that rumen metabolites positioned on the top (i.e., from glycine to succinate) of the heat map showed a much more variable response, whereas the cluster that included metabolites from alanine to butyrate showed a clear increase with increased grain in the diet. Except for 3-PP that decreased with increased dietary grain, the heat map for the remaining metabolites (i.e., from ferulate to rumen pH) seemed to be independent of the amount of dietary grain (Fig. 3).

4 Discussion In this study 1H-NMR-based metabolomics was used to evaluate whether feeding different proportions of grain in the diet would affect rumen metabolism. We also wanted to investigate whether the observed metabolite changes could give further insight into the role of grain feeding on the etiopathology of diet-related diseases in dairy cows. Our results clearly showed that these four different grain diets could be easily distinguished on the basis of rumen metabolites. Our data demonstrated that cows fed on 45% barley grain had greater concentrations of MA, endotoxin, propionate, glucose, and EtOH in their rumen fluid. On the other hand, cows fed 0 and 15% barley grain fell into two similar clusters characterized by lower concentrations of

Fig. 2 Principal component analysis (PCA) of the rumen metabolomic profile. a The PCA score plot distinguishes the metabolic profiles of rumen fluid in cows fed 0% (open triangle), 15% (plus), 30% (times), and 45% (open diamond) rolled barely grain in the diet. The first number in the data points represents diet (0, 15, 30, and 45), the second is about the cow number (1–8 cows), and the third indicates sampling day (1, 3, 5, 7, and 10). b Loading plot showing the correlations among individual rumen fluid metabolites of the first two eigenvalues (PC 1 and PC 2). This gives a graphical representation of the extent to which each factor accounts for the variance in the data and shows the relationship between the different metabolites. Variables with the same distance from 0 with similar positions are positively correlated. Those in the opposite direction are negatively correlated. For abbreviations of rumen fluid metabolites see Table 1

123

590 Fig. 3 Hierarchical clustering analysis for different rumen fluid metabolites measured before morning feeding in dairy cows fed four different diets (0, 15, 30, and 45% barley grain inclusion in dry matter). The expression patterns of each parameter (shown in each row) were categorized by an average linkage hierarchical clustering program (Seo 2005). The bars indicate the normalized fold change levels, with lighter grey boxes denoting an expression ratio greater than the mean and darker grey boxes denoting an expression ratio below the mean. White boxes denote an intermediate level. The tree clusters and their shorter Euclidean distance indicate higher similarities. Similarity between two metabolites is represented by branch height, therefore the lower a node is vertically, the more similar its sub-tree. Minimum similarity is 63%

B. N. Ametaj et al.

PAG, xanthine, NDMA, acetate, and leucine and greater 3-PP. The potential health implications of altered metabolite levels in rumen fluid, due to different proportions of barley grain in the diet, are discussed further below. One of the most interesting observations from this study was a multi-fold increase in the concentrations of methylamine (MA) in the rumen of dairy cows fed with the highest proportions of grain (i.e., 30 and 45%). To our knowledge, this is the first report of such an association between grain consumption and MA levels. Several investigators have shown that bacterial decarboxylation of amino acids such as glycine, tyrosine, and phenylalanine (Davis and De Ropp 1961; Hill and Mangan 1964) and the degradation of plant lipids, rich in choline (Hill and Mangan 1964; Neill et al. 1978), contribute to the production of methylated amines in ruminant animals. In addition, research conducted by Hill and Mangan (1964) indicated that an additional contributing factor in

production of methylated amines is the presence of an acidic media. Indeed, the data from our study showed that cows fed 30 and 45% barley grain had a rumen pH below 5.8 during 6–12 h post-feeding (data not shown). This pH is at levels typically seen during SARA. The importance of methylated amines to the health of dairy cows is related to the fact that if they are absorbed into the blood circulation they might be degraded by semicarbazide-sensitive amine oxidases. This degradation process leads to the production of harmful metabolites such as hydrogen peroxide and formaldehyde, both of which are toxic and both of which are associated with diseases such as diabetes, vascular disorders, heart failure, and Alzheimer’s disease in humans (Yu et al. 2006). To our knowledge, there is no information about the role of methylated amines in the etiopathology of metabolic diseases in dairy cows. It would be of interest to explore their role on health issues of dairy cows. Intriguingly, Hill and Mangan (1964) and Neill et al. (1978)

123

Rumen metabolomics

591

demonstrated that methylated amines are converted into methane by rumen microorganisms. Methane is a greenhouse gas that contributes to atmospheric pollution. It is well-known that a variety of factors, including the level and the nature of dietary carbohydrate fermented in the reticulorumen and the ratio between acetate and propionate, affect the amount of methane emitted from cattle (Johnson and Johnson 1995). Our data suggest that methylated amines might be an additional source of methane that arises from the feeding of high-grain diets. Although the concentration of NDMA was not significantly different, explained by low statistical power, data analysis indicated that cows fed the higher proportions of barley grain (i.e., 30 and 45%) potentially had greater concentrations of NDMA (N-nitrosodimethylamine). NDMA is a semi-volatile compound released in the rumen fluid due to the interaction of ingested nitrites or nitrates with amines such DMA and TMA (Lijinsky 1999). Gastrointestinal NDMA has been reported to be a strong hepatotoxic compound in sheep (Koppang 1964) and is known as a powerful animal carcinogen capable of inducing benign and malignant tumors in a variety of tissues including liver, kidney, lung, and the nasal cavity (Souliotis et al. 2002). The reason for elevated levels of NDMA in the rumen fluid of cows fed higher proportions of grain is not well understood presently; however, our HCA revealed presence of a metabolite cluster that included NDMA and the amino acids glycine, valine, leucine, glutamate, isoleucine as well as lactate. This association suggests that higher concentrations of NDMA might be related to microbial degradation of the aforementioned amino acids and lactate. Indeed, Davis and De Ropp (1961) reported that feeding rats with levels of glycine doubled the amount of MA (a precursor to both DMA and NDMA) excreted in the urine. Interestingly, Hashimoto et al. (1975) reported that only pathogenic bacteria from human, rat, and sheep intestines were able to catalyze the formation of NDMA from DMA and nitrate in vitro. This suggests that the feeding of large amounts of cereal grains to dairy cattle might create favourable conditions for the growth and activity of pathogenic bacteria in the rumen. Our results also indicated a more than 2-fold increase, although not statistically significant, in the amount of EtOH in the rumen fluid when cows were fed 45% barley grain in the diet. This is in agreement with previously reported data on ruminants (sheep) that indicate that overfeeding with wheat grain is associated with EtOH concentrations as high as 8 mM (Allison et al. 1964). Kristensen et al. (2007) also reported rumen EtOH concentrations above 2 mM in dairy cows fed a TMR containing *54% corn silage and *13.5% rolled barley grain. Moreover, Kristensen et al. (2007) also reported that the EtOH concentration in rumen fluid increased 5-fold within 1–3 h after the morning

feeding. In our study the amount of EtOH in the rumen fluid of cows reached the greatest concentration (i.e., 341 lM) on a 45% barley grain diet. The lower concentration of rumen EtOH in our cows, compared to previous reports, might be due to collection of rumen samples prior to the morning feeding. Note that the cows in our experiment were fed only once per day, at 0800. Although there are no indications for the involvement of EtOH in the etiopathology of metabolic disease in dairy cows, human research implicates EtOH in multiple pathological states (Estruch et al. 1993). Intriguingly, Enomoto et al. (2001) showed that chronic exposure to EtOH increases gut permeability and portal vein translocation of endotoxin. Additionally, research conducted by Pruett and Pruett (2006) demonstrated that EtOH was not able to initiate an acute phase response in C3H/HeJ (TLR-4 mutant) mice, indicating that the response is mediated through the endotoxin receptor, TLR-4. We also reported that cows fed with larger amounts of barley grain (30 and 45% diet DM) developed an inflammatory state with greater plasma concentrations of serum amyloid A, lipopolysaccharide-binding protein, and C-reactive protein (Emmanuel et al. 2008). Recently, Khafipour et al. (2009) suggested an unknown physiological factor that might affect gastrointestinal barrier functions and facilitate translocation of endotoxin into blood circulation and initiate an inflammatory response during feeding of high-grain diets. It is postulated that rumen EtOH might be the agent that facilitates translocation of endotoxin through rumen and colon walls contributing to endotoxin-related disorders during feeding of high-grain diets to dairy cows. Another interesting cluster of metabolites that was associated with high-grain diets and deserves further discussion was that of xanthine, uracil, alanine, and endotoxin. All four of these metabolites were elevated with increasing amounts of dietary grain. Interestingly, all four metabolites are degradation products of rumen bacteria. McAllan and Smith (1973) showed that bacterial nucleic acids (i.e., RNA or DNA) incubated with rumen fluid were rapidly converted into xanthine, hypoxanthine, and uracil. It is also known that the death of both Gram-negative and Gram positive-bacteria is associated with the release of endotoxin and alanine (Trent et al. 2006). It is well established that the decrease in rumen pH in dairy cows fed high-grain diets is associated with significant changes in microbiota composition. Many microbial species are not able to survive the stress of a low pH and lack of nutrients (Slyter 1976). For example, Khafipour et al. (2009) reported a decrease in the number of Gram-negative Bacteroidetes in rumen as a consequence of a high-grain diet. Therefore, the increased concentrations of xanthine, uracil, alanine, and endotoxin in the rumen fluid of cows fed highgrain diets appear to be biomarkers of widespread bacterial

123

592

B. N. Ametaj et al.

cell lysis and a fundamental change in the rumen microbiota. There are also two other mechanisms to explain the higher concentration of alanine in rumen fluid. The first one might be related to greater availability of glucose and pyruvate due to the degradation of larger amounts of starch, coupled with higher levels of ammonia. This ¨ rlygsson et al. (1995), assumption is supported by data of O who determined that alanine is produced via an amination of pyruvate via a transamination reaction coupled to alanine aminotransferase. This suggests that the presence of high levels of pyruvate and ammonia favour the production ¨ rlygsson et al. 1995). The of alanine by rumen bacteria (O second reason for higher alanine content might be related to the fact that feeding high amounts of readily available carbohydrates to cattle is associated with lowering of rumen pH and the sporulation of certain rumen bacteria to survive the adverse conditions (Rieu-Lesme et al. 1996). Sporulated bacteria use L-alanine to germinate; however when rumen conditions are not favourable for growth they convert L-alanine to D-alanine to postpone germination. This process is known as autoinhibition (Gould 1970). Dalanine is used by both Gram-negative and Gram-positive bacteria in the biosynthesis of their cell wall components (peptidoglycan and lipoteichoic acid) (Hoogenraad and Hird 1970). It would be of interest to further explore the latter hypothesis. To our knowledge, there are no known health implications for high levels of alanine in the rumen or gastrointestinal tract of dairy cows or other livestock animals. Our study showed opposing responses of 3-PP and phenylacetate to the amount of dietary grain in the diet. Thus, the concentration of 3-PP linearly decreased with increasing proportions of grain in the diet, whereas an opposite effect was observed with regards to rumen phenylacetate. Research conducted in forage-fed sheep has shown that the latter two compounds are the predominant aromatic acids, with 3-PP and phenylacetate accounting for about 50.8 and 13.5% of total aromatic acids in the rumen fluid, respectively (Pagella 1998). Our results demonstrate that the ratio between these aromatic compounds can be changed by increasing the amount of grain in the diet. In the rumen, 3-PP and phenylacetate are generated by the activity of ruminal microbiota on plant constituents, namely via the hydrogenation of phenolic compounds such as p-coumaric, ferulic, and caffeic acids, followed by a dehydroxylation process (Chesson et al. 1999). The concentration of ferulate in the rumen fluid did not change in this study, and was generally low. This could be attributed to the time of sampling, which was always done before the morning feeding. An alternative origin of 3-PP and phenylacetate is also possible via the deamination of aromatic amino acids such as tyrosine and phenylalanine (Martin

1982; Burlingame and Chapman 1983). Interestingly, our HCA revealed the formation of a sub-cluster among phenylacetate, tyrosine, and the branched chain fatty acids isobutyrate and isovalerate. These branched chain fatty acids are also synthesized from deamination of the amino acids valine, isoleucine, leucine, and proline (Andries et al. 1987). On the other hand, 3-PP formed a cluster with benzoate and 3-HP, suggesting the presence of different origins of 3-PP and phenylacetate in the rumen of cows fed with increasing amounts of barley grain. Overall, our data indicated that deamination processes for various amino acids were favoured in the rumen for cows on high grain diets, and this is reflected by a strong positive association between high-grain diets and rumen phenylacetate. The low concentration of 3-PP in a grainenriched diet might also be related to the changes in the rumen microbiota and the availability of other compounds in the rumen fluid. For example, Turlin et al. (2001) demonstrated that the expression of genes coding for key enzymes involved in 3-PP production such as 3-phenylpropionate dioxygenase and 3–phenylpropionate-20 ,30 -dihydrodiol dehydrogenase were strongly repressed in the presence of glucose. Note that we found that glucose was significantly increased in the rumen of cows fed higher amounts of barley grain. Other than a role of 3-PP in the growth of rumen bacteria and their protection from oxidative stress, there are no data to relate changes in the concentrations of 3-PP and phenylacetate with the health of dairy cattle (Turlin et al. 2005). Our study confirms previous findings with regard to higher concentrations of VFA with increasing the proportion of cereal grains in the diet. These VFA observations were made using traditional analytical methods such as GC and HPLC (Iqbal et al. 2009). Previous research has indicated that feeding cattle with diets rich in readily degradable carbohydrates, such as barley grain, is associated with increased concentration of VFA in the rumen fluid. These changes in the VFA profile arise from changes in the propionate and valerate anabolic pathways (Bugaut 1987). Accumulation of VFA in the rumen fluid is known to be associated with the development of ARA and SARA and a chain of other related metabolic disorders in dairy cattle. Interestingly, the rumen concentration of lactate in our study did not differ among diets despite large variations in the amount of readily fermentable carbohydrates. This result also agrees with previous reports (Nocek 1997; Iqbal et al. 2009). Evidently lactate is rapidly converted to propionate via the acrylate pathway, found in several lactateutilizing rumen bacteria (Satter and Esdale 1968). Overall, the consistency of our data with those reported by conventional analytical techniques shows that NMR-based metabolomics is a reliable approach for the evaluation of rumen metabolic events.

123

Rumen metabolomics

593 Wishart, which was supported financially by the Alberta Agricultural Research Institute (AARI; Edmonton, Alberta, Canada), the Alberta Livestock Industry Development Fund (ALIDF; Edmonton, Alberta, Canada), and the Natural Sciences and Engineering Research Council of Canada (NSERC; Ottawa, Ontario, Canada). The technical assistance of D. G. V. Emmanuel, R. P. Pandian, and S. Sivaraman (University of Alberta, Edmonton, Alberta, Canada) is highly appreciated. We also are grateful to the technical staff of Dairy Research and Technology Centre at the University of Alberta for their help and care to the cows used in this study.

The appropriate amount of cereal grains to feed dairy cattle is a hotly debated issue among ruminant nutritionists and health scientists. On one hand, cereal grains are necessary dietary ingredients to support high milk production. On the other hand, large amounts of grain may adversely affect rumen metabolism as well as the health and productivity of dairy cows (Nocek 1997; Iqbal et al. 2009). In this regard, rumen data from this study clearly suggests that diets containing 30 and 45% cereal grain generate a whole variety of metabolites that have been or might be implicated in the etiology and pathogenesis of different metabolic diseases in dairy cows. Interestingly, the diet containing 15% barley grain was shown to supply enough energy and nutrients to support high milk production (28.0 kg fat-corrected milk/d, Zebeli and Ametaj 2009) without adversely affecting rumen metabolism, and the overall metabolic health of dairy cows (Emmanuel et al. 2008).

References
Allison, M. J., Dougherty, R. W., Bucklin, J. A., & Snyder, E. E. (1964). Ethanol accumulation in the rumen after overfeeding with readily fermentable carbohydrate. Science, 144, 54–55. Ametaj, B. N., Bradford, B. J., Bobe, G., Nafikov, R. A., Lu, Y., Young, J. W., et al. (2005). Strong relationships between mediators of the acute phase response and fatty liver in dairy cows. Canadian Journal of Animal Science, 85, 165–175. Andries, J. I., Buysse, F. X., Debrabander, D. L., & Cottyn, B. G. (1987). Isoacids in ruminant nutrition: their role in ruminal and intermediary metabolism and possible influences on performances—a review. Animal Feed Science and Technology, 18, 169–180. Bertram, H. C., Kristensen, N. B., Malmendal, A., Nielsen, N. C., Brod, R., Andersen, H. J., et al. (2005). A metabolomic investigation of splanchnic metabolism using 1H NMR spectroscopy of bovine blood plasma. Analytica Chimica Acta, 536, 1–6. Bertram, H. C., Kristensen, N. B., Vestergaard, M., Jensen, S. K., Sehested, J., Nielsen, N. C., et al. (2009). Metabolic characterization of rumen epithelial tissue from dairy calves fed different starter diets using 1H NMR spectroscopy. Livestock Science, 120, 127–134. Boudonck, K. J., Mitchell, M. W., Wulff, J., & Ryals, J. A. (2010). Characterization of the biochemical variability of bovine milk using metabolomics. Metabolomics. doi:10.1007/s11306-0090160-8. Broadhurst, D. I., & Kell, D. B. (2006). Statistical strategies for avoiding false discoveries in metabolomics and related experiments. Metabolomics, 2, 171–196. Bugaut, M. (1987). Occurrence, absorption and metabolism of short chain fatty acids in the digestive tract of mammals. Comparative Biochemistry and Physiology, 86B, 439–472. Burlingame, R., & Chapman, P. J. (1983). Catabolism of phenylpropionic acid and its 3-hydroxy derivative by Escherichia coli. Journal of Bacteriology, 155, 113–121. Canadian Council on Animal Care. (1993). Guide to the care and use of experimental animals (2nd ed., Vol. 1). Ottawa: CCAC. Chesson, A., Provan, G. J., Russell, W. R., Scobbie, L., Richardson, A. J., & Stewart, C. (1999). Hydroxycinnamic acids in the digestive tract of livestock and humans. Journal of the Science of Food and Agriculture, 79, 373–378. Davis, E. J., & De Ropp, R. S. (1961). Metabolic origin of urinary methylamine in the rat. Nature, 190, 636–637. Dieterle, F., Ross, A., Schlotterbeck, G., & Senn, H. (2006). Probabilistic quotient normalization as robust method to account for dilution of complex biological mixtures. Application in 1H NMR metabonomics. Analytical Chemistry, 78, 4281–4290. Dumas, M.-E., Barton, R. H., Toye, A., Cloarec, O., Blancher, Ch., Rothwell, A., et al. (2006). Metabolic profiling reveals a contribution of gut microbiota to fatty liver phenotype in

5 Concluding remarks In summary, 1H-NMR-based metabolomics and multivariate analysis were used to provide concentration data on 46 rumen metabolites for cattle fed various grain-enriched diets. With increasing amounts of dietary cereal grain we found dramatic increases of rumen MA and endotoxin, enhanced concentrations of glucose, alanine, maltose, propionate, uracil, valerate, xanthine, and phenylacetate as well as potential increases in the concentration of NDMA, EtOH, and DMA. The only metabolite that dropped with increasing amounts of dietary barley grain was 3-PP. The PCA of the rumen data showed 4 distinct clusters of metabolites corresponding to the 4 distinct cereal grain diets. Metabolite levels corresponding to diets with 0 and 15% barley grain were highly similar and mapped closely together, while the two other clusters (from 30 and 45% grain diets) were located far apart. Avoiding digestive and systemic health disturbances of ruminants through a healthy nutrition is critical to ensure that milk and meat are produced from healthy animals. Given that the feeding of large amounts of barley grain to dairy cattle resulted in major increases in a number of potentially harmful rumen metabolites, our results suggest that a systematic evaluation of milk and meat products of cows fed high grain diets should also be performed to detect the presence of these metabolites. Overall, our study suggests that 1H-NMRbased metabolomics could be used as a reliable tool to study the interactions between nutrition and health in dairy cows.
Acknowledgements We acknowledge co-leading of the project entitled ‘Profiling of Dairy Cattle Metabolome’ by Drs. Ametaj and

123

594 insulin-resistant mice. Proceedings of National Academy of Sciences USA, 103, 12511–12516. Emmanuel, D. G. V., Dunn, S. M., & Ametaj, B. N. (2008). Feeding high proportions of barley grain stimulates an inflammatory response in dairy cows. Journal of Dairy Science, 91, 606–614. Enomoto, N., Ikejima, K., Yamashina, S., Hirose, M., Shimizu, H., Kitamura, T., et al. (2001). Kupffer cell sensitization by alcohol involves increased permeability to gut-derived endotoxin. Alcoholism, Clinical and Experimental Research, 25, 51S–54S. ´ s, J. M., Villegas, E., Jonque ´ , A., & UrbanoEstruch, R., Nicola ´ rquez, A. (1993). Relationship between ethanol-related Ma diseases and nutritional status in chronically alcoholic men. Alcohol and Alcoholism, 28, 543–550. Gould, G. W. (1970). Germination and the problem of dormancy. Journal of Applied Bacteriology, 33, 34–49. Hashimoto, S., Kawai, Y., & Mutai, M. (1975). In vitro Nnitrosodimethylamine formation by some bacteria. Infection and Immunity, 11, 1405–1406. Hill, K. J., & Mangan, J. L. (1964). The formation and distribution of methylamine in the ruminant digestive tract. Biochemistry Journal, 93, 39–45. Hoogenraad, N. J., & Hird, F. J. R. (1970). The chemical composition of rumen bacteria and cell walls from rumen bacteria. British Journal of Nutrition, 24, 119–127. Iqbal, S., Zebeli, Q., Mazzolari, A., Bertoni, G., Dunn, S. M., et al. (2009). Feeding barley grain steeped in lactic acid modulates rumen fermentation patterns and increases milk fat content in dairy cows. Journal of Dairy Science, 92, 6023–6032. Jenkins, T. C., & McGuire, M. A. (2006). Major advances in nutrition: impact on milk composition. Journal of Dairy Science, 89, 1302–1310. Johnson, K. A., & Johnson, D. E. (1995). Methane emissions from cattle. Journal of Animal Science, 73, 2483–2492. Khafipour, E., Li, S., Plaizier, J. C., & Krause, D. O. (2009). Rumen microbiome composition determined using two nutritional models of subacute ruminal acidosis. Applied and Environmental Microbiology, 75, 7115–7124. Koppang, N. (1964). An outbreak of toxic liver injury in ruminants. Nord Veterinaermed, 16, 305–322. Kristensen, N. B., Storm, A., Raun, B. M., Røjen, B. A., & Harmon, D. L. (2007). Metabolism of silage alcohols in lactating dairy cows. Journal of Dairy Science, 90, 1364–1377. Lijinsky, W. (1999). N-Nitroso compounds in the diet. Mutation Research, 443, 129–138. Littell, R. C., Henry, P. R., & Ammerman, C. B. (1998). Statistical analysis of repeated measures data using SAS procedures. Journal of Animal Science, 76, 1216–1231. Martin, A. K. (1982). The origin of urinary aromatic compounds excreted by ruminants 3. The metabolism of phenolic compounds to simple phenols. British Journal of Nutrition, 48, 497–507. McAllan, A. B., & Smith, R. H. (1973). Degradation of nucleic acids in the rumen. British Journal of Nutrition, 29, 331–345. Neill, A. R., Grime, D. W., & Dawson, R. M. (1978). Conversion of choline methyl groups through trimethylamine into methane in the rumen. Biochemistry Journal, 170, 529–535. Nocek, J. E. (1997). Bovine acidosis: implications on laminitis. Journal Dairy Science, 80, 1005–1028. NRC. (2001). Nutrient requirements of dairy cattle (7th rev. edn). National Academy of Sciences, Washington, DC. ¨ rlygsson, J., Anderson, R., & Svensson, B. H. (1995). Alanine as an O end product during fermentation of monosaccharides by Clostridium strain P2. Antonie van Leeuwenhoek, 68, 273–280. Pagella, J. H. (1998). Urinary benzylated compounds as potential markers of forage intake and metabolism of their precursors in ruminants. PhD Dissertation, Aberdeen University, UK.

B. N. Ametaj et al. Pruett, B. S., & Pruett, S. B. (2006). An explanation for the paradoxical induction and suppression of an acute phase response by ethanol. Alcohol, 39, 105–110. Rieu-Lesme, F., Dauga, C., Morvan, B., Bouvet, O. M. M., Grimont, ´ , J. (1996). Acetogenic sporulating cocci P. A. D., & Dore isolated from the rumen. Research in Microbiology, 147, 753–764. Satter, L. D., & Esdale, W. J. (1968). In vitro lactate metabolism by ruminal ingesta. Applied Microbiology, 16, 680–688. Saude, E. J., Slupsky, C. M., & Sykes, B. D. (2006). Optimization of NMR analysis of biological fluids for quantitative accuracy. Metabolomics, 2, 113–123. Seo, J. (2005). Information visualization design for multidimensional data: integrating the rank-by-feature framework with hierarchical clustering. Ph.D. Dissertation, University of Maryland. Slyter, L. L. (1976). Influence of acidosis on rumen function. Journal of Animal Science, 43, 910–929. Souliotis, V. L., Henneman, J. R., Reed, C. D., Chhabra, S. K., Diwan, B. A., Anderson, L. M., et al. (2002). DNA adducts and liver DNA replication in rats during chronic exposure to Nnitrosodimethylamine (NDMA) and their relationships to the dose-dependence of NDMA hepatocarcinogenesis. Mutation Research, 500, 75–87. Tajima, K., Arai, S., Ogata, K., Nagamine, T., Matsui, H., Nakamura, M., et al. (2000). Rumen bacterial community transition during adaptation to high-grain diet. Anaerobe, 6, 273–284. Trent, M. S., Stead, C. M., Tran, A. X., & Hankins, J. V. (2006). Diversity of endotoxin and its impact on pathogenesis. Journal of Endotoxin Research, 12, 205–223. Turlin, E., Perrotte, M., Danchin, A., & Biville, F. (2001). Regulation of the early steps of 3-phenylpropionate catabolism in Escherichia coli. Journal of Molecular Microbiology and Biotechnology, 3, 127–133. Turlin, E., Sismeiro, O., Le Caer, J. P., Labas, V., Danchin, A., & Biville, F. (2005). 3-phenylpropionate catabolism and the Escherichia coli oxidative stress response. Research in Microbiology, 156, 312–321. Vinayavekhin, N., Homan, E. A., & Saghatelian, A. (2010). Exploring disease through metabolomics. American Chemical Society Chemical Biology, 5, 91–103. Weljie, A. M., Newton, J., Mercier, P., Carlson, E., & Slupsky, C. M. (2006). Targeted profiling: Quantitative analysis of 1H NMR metabolomics data. Analytical Chemistry, 78, 4430–4442. Wishart, D. S. (2008a). Metabolomics: Applications to food science and nutrition research. Trends in Food Science and Technology, 19, 482–493. Wishart, D. S. (2008b). Quantitative metabolomics using NMR. Trends in Analytical Chemistry, 27, 228–237. Wishart, D. S., Lewis, M. J., Morrissey, J. A., Flegel, M. D., Jeroncic, K., Xiong, Y., et al. (2008). The human cerebrospinal fluid metabolome. Journal of Chromatography B, Analytical Technologies in the Biomedical and Life Sciences, 871, 164–173. Xia, J., Psychogios, N., Young, N. & Wishart, D. S. (2009). MetaboAnalyst: a web server for metabolomic data analysis and interpretation. Nucleic Acids Research, 37(Web Server issue), W652–W660. Yu, P., Xin, H., Lu, L., Fan, H., Kazachkov, M., Jiang, Z. J., et al. (2006). Involvement of semicarbazide-sensitive amine oxidasemediated deamination in lipopolysaccharide-induced pulmonary inflammation. American Journal of Pathology, 168, 718–726. Zebeli, Q., & Ametaj, B. N. (2009). Relationships between rumen lipopolysaccharide and mediators of inflammatory response with milk fat production and efficiency in dairy cows. Journal of Dairy Science, 92, 3800–3809.

123