Latent Biochemical Relationships in the Blood–Milk Metabolic Axis of

Feb 8, 2013 - of Dairy Cows Revealed by Statistical Integration of 1H NMR. Spectroscopic Data ... for health and commercially relevant traits in dairy...
0 downloads 0 Views 2MB Size
Download PDF
Article pubs.acs.org/jpr

Latent Biochemical Relationships in the Blood−Milk Metabolic Axis of Dairy Cows Revealed by Statistical Integration of 1H NMR Spectroscopic Data Anthony D. Maher,*,† Benjamin Hayes,†,‡,§ Benjamin Cocks,†,‡,§ Leah Marett,∥ William J Wales,∥ and Simone J. Rochfort†,‡,§ †

Biosciences Research Division, Department of Primary Industries, Bundoora, Victoria 3083, Australia Dairy Futures Cooperative Research Centre, Victoria 3083, Australia § La Trobe University, Bundoora, Victoria 3083, Australia ∥ Future Farming Systems Research Division, Department of Primary Industries, Ellinbank Centre, Ellinbank, Victoria 3820, Australia ‡

S Supporting Information *

ABSTRACT: A detailed understanding of the relationships between the distinct metabolic compartments of blood and milk would be of potential benefit to our understanding of the physiology of lactation, and potentially for development of biomarkers for health and commercially relevant traits in dairy cattle. NMR methods were used to measure metabolic profiles from blood and milk samples from Holstein cows. Data were analyzed using PLS regression to identify quantitative relationships between metabolic profiles and important traits. Statistical Heterospectroscopy (SHY), a powerful approach to recovering latent biological information in NMR spectroscopic data sets from multiple complementary samples, was employed to explore the metabolic relationships between blood and milk from these animals. The study confirms milk is a distinct metabolic compartment with a metabolite composition largely not influenced by plasma composition under normal circumstances. However, several significant relationships were identified, including a high correlation for trimethylamine (TMA) and dimethylsulfone (DMSO2) across plasma and milk compartments, and evidence plasma valine levels are linked to differences in amino acid catabolism in the mammary gland. The findings provide insights into the physiological mechanisms underlying lactation and identification of links between key metabolites and milk traits such as the protein and fat content of milk. The approach has the potential to enable measurement of health, metabolic status and other important phenotypes with milk sampling. KEYWORDS: metabolomics, metabonomics, nuclear magnetic resonance (NMR), statistical heterospectroscopy (SHY), partial least-squares regression (PLS), dairy, bovine, milk

■

INTRODUCTION

These metabolites are all secreted into milk by the mammary gland, and composed from precursors derived from the blood.2

Milk is a multiphasic biological mixture produced by mammals for infant nutrition and farmed on an industrial scale for human consumption. The complex nature of milk reflects the requirement to supply the infant with a complete food containing diverse biochemical products from small metabolites through to macromolecular fat globules and proteins. The major small metabolite in bovine milk is lactose, but many more metabolites are present in milk such as citric acid cycle intermediates, amino acids, organic acids and ketone bodies.1 © 2013 American Chemical Society

It is known that the composition of blood and milk of dairy cows can be influenced by nutritional intake, disease status and stage of lactation, but detailed knowledge of the specific metabolic relationships between these compartments is limited. Received: November 7, 2012 Published: February 8, 2013 1428

dx.doi.org/10.1021/pr301056q | J. Proteome Res. 2013, 12, 1428−1435

Journal of Proteome Research

Article

Metabolomics (also known as “metabonomics”) employs mass spectrometry (MS) or nuclear magnetic resonance (NMR) technologies to measure the metabolic complement of biological fluids, tissue biopsies or cell extracts and/or to understand the multifactorial response of an organism to a pathophysiological stimulus or genetic modification. 3−5 Although MS is in general of higher sensitivity, measuring the metabolome by NMR has many advantages, including higher analytical reproducibility.6 The resonance intensities in NMR spectral data are directly proportional to the concentrations of the metabolites that give rise to them, a fact has been exploited to recover biological information using advanced statistical methods. So-called Statistical Total Correlation Spectroscopy (STOCSY) was introduced in 20057 and is essentially the construction of a vector or matrix of correlation coefficients to identify statistical relationships within or between metabolomic (or other high dimensional -omic) data sets to recover chemical or biological information. When applied between different data sets, this approach is known as Statistical HeterospectroscopY (SHY), and has been used to extract chemical and biological information from integrated UPLC-MS-NMR data,8,9 and to recover biological information from combined blood plasmacerebrospinal fluid metabolomic data.10 Several studies have applied metabolomics to explore the metabolic content of biofluids from dairy cattle. Boudonck et al. used an MS-based approach to explore the metabolic variation resulting from different processing methods, identifying metabolic signatures that discriminated between different farming methods.11 Klein et al. quantified 44 metabolites in bovine milk using NMR and MS-based metabolomics. By analyzing concentration ranges, they found some metabolites such as lactose were tightly controlled, while others, such as choline, were highly variable between different animals.12 A study by Sundekilde et al. used principal components analysis of NMR-based metabolomics data to identify global metabolic profile differences between breeds and coagulation properties from 14 dairy cows.13 Recent work by Ilves et al. employed principal components analysis to separately analyze blood and milk metabolomes from five cows and found systematic changes over the course of lactation, but observed little correlation between these biofluids using this statistical approach.14 Here we have used NMR to measure blood and milk metabolomes from a cohort of Holstein dairy cows maintained under the same feeding conditions and chosen to have a range of residual feed intake (RFI), a measure of feed conversion efficiency (FCE). Using partial least-squares regression, and then SHY as a sensitive encompassing approach for statistical data integration, we quantified the metabolic relationships between milk and blood to test the hypothesis that signatures of animal metabolic status or valuable complex phenotypes might be measurable in milk.

■

These cows were part of a larger project investigating feed conversion efficiency.15 All procedures were conducted in accordance with the Australian Code of Practice for the Care and Use of Animals for Scientific Purposes (National Health and Medical Research Council, 2004). Approval to proceed was obtained from the DPI Agricultural Research and Extension Animal Ethics Committee. Aliquots of fresh milk samples were collected as part of daily milking schedule and immediately frozen on dry ice. Blood was collected on the same morning into heparin-containing tubes from the tail vein of each cow. Blood samples were centrifuged for 5 min at 3000g and aliquots of the upper layer (plasma) were immediately frozen. Samples were stored at −80 °C. We are not aware of any studies that have investigated the effects of storage temperature on the metabolic profiles of milk stored at −80 °C, but given the relatively short storage time (less than 3 months), we consider it reasonable to assume these samples were metabolically stable at this temperature. Sample Preparation

Blood plasma samples were prepared by adding 300 μL of thawed sample to 300 μL of D2O. This was centrifuged for 5 min at 13000g and 550 μL of this mixture was added to a 5 mm diameter NMR tube. Milk samples were first centrifuged at 4000g for 15 min to separate the less dense cream (upper layer) from the skim milk. Then, 300 μL of skim milk was carefully aspirated from the lower layer and added to 300 μL of D2O, and 550 μL of the supernatant was added to 5 mm diameter NMR tubes for analysis. Nuclear Magnetic Resonance

All NMR experiments were acquired on a Bruker AvanceIII spectrometer operating at 800.13 MHz equipped with a CryoProbe (Bruker Biospin, Karlsruhe, Germany). Data were acquired with a pulse sequence equivalent to the first increment of a NOESY 2D experiment with presaturation applied immediately prior to the first pulse for 2 s at the frequency of the water resonance and during the mixing time (100 ms), and a Carr−Purcell−Meiboom−Gill (CPMG) with presaturation during the recycle delay (2 s) and a total echo time of 153.6 ms. The time domain was 128k and the sweep width was 20 ppm. Data were Fourier transformed with exponential line broadening of 0.5 Hz in the frequency domain. Spectra were phased and baseline corrected in Topspin 3.0 (Bruker Biospin, Karlsruhe, Germany) before being imported into Matlab (Mathworks, Natick, MA) at full resolution. Data Analysis

After import into Matlab, plasma NMR data were referenced to the anomeric doublet resonance of α-glucose (δ5.233), while milk samples were aligned to the lactose signal at the same chemical shift. Local alignment was performed using correlation optimized warping of selected peaks to account for peak position variation caused by natural differences in pH or ionic content of samples.16 Nonquantitative spectral regions containing the residual water peak were removed (δ4.6−δ5.05) before normalizing to most probabilistic quotient.17 Principal components analysis (PCA) and partial least-squares (PLS) regression models were constructed using in-house software based on the NIPALS algorithm.18 To reduce the risk of overfitting PLS models, the Q2 parameter was calculated using the formula

MATERIALS AND METHODS

Materials

Deuterium oxide (D2O, 99%) was purchased from Cambridge Isotope Laboratoies (Andover, MA). All other chemicals were purchased from Sigma (North Ryde, NSW, Australia) Sample Collection

Samples were collected from 54 Holstein cattle from the Department of Primary Industries (DPI) Ellinbank Centre, Victoria, Australia (latitude 38°14′S, longitude 145°56′E). 1429

dx.doi.org/10.1021/pr301056q | J. Proteome Res. 2013, 12, 1428−1435

Journal of Proteome Research

Article

⎛ ∑ (y − y )̂ 2 ⎞ ⎟ Q2 = 1 − ⎜ SSY ⎠ ⎝

where SSY is the sum of the squares of the Y matrix, and ŷ is the predicted value of the response y, calculated from 7-fold internal cross validation. Estimates of the significance of the Q2 values were obtained by response permutation where, for each PLS model, several hundred parallel models were constructed from the same data but with randomly reordered Y data. The “permuted” Q2 values were then regressed against the correlation of the Y data with the permuted Y data to assist in the identification of models with predictive potential. To assist biological interpretation, PLS models were visualized by plotting model loadings as a function of chemical shift after multiplying by the standard deviation of the X matrix. The square of the loadings were then projected as a color onto this plot thus facilitating interpretation since the loadings resemble NMR spectra.19,20

Figure 1. Typical 1H NMR spectra from blood (red spectra) and milk (blue spectra) sampled from the same cow. For both biofluids, 1 “1D” (acquired using the first increment of a NOESY pulse sequence with presaturation at the water resonance frequency) and a relaxationedited (“RE”, acquired using a CPMG pulse sequence) are displayed. Some of the major metabolites have been annotated.

Statistical Heterospectroscopy (SHY)

SHY correlation matrices were calculated as previously described.8,10 Briefly, a matrix of correlation coefficients was constructed from every column of the milk data matrix to every column of the blood plasma data matix. P-values were also calculated from every correlation and those less than 0.001 were excluded from the matrix, equivalent to a Bonferroni correction of 50 independent tests. The resultant matrix was then plotted as a 2D array with each data point color-coded according to its correlation coefficient. The identification of genuine correlations was further refined by excluding data points that were clearly caused by spectral noise; these are readily identifiable as isolated pixels or odd-shaped patterns such as lines and smears that may be caused by artifacts from local peak alignment or other data curation steps. We can then interpret the data in the same manner as a 2D NMR spectrum, except that each “peak” is a statistical correlation between a given metabolite in the blood with a corresponding metabolite in the milk data.

metabolically distinct from the blood samples, as would be expected, with complete separation of the two biofluids along PC1. These data also suggest that the blood metabolome form a “tighter” group, metabolically, compared to milk, as judged by the increased dispersion of the milk samples along the PC2 dimension. This implies that the blood metabolomes of these cows are under tighter homeostatic control compared to milk, but further quantitative analyses of individual metabolites would be needed to prove this. Further PCA and PLS models were constructed separately from milk or blood metabolic data. The samples in this study were collected from animals selected as part of a larger study investigating feed conversion efficiency (FCE), so initial analyses tested the hypothesis that efficient cows were metabolically distinct from inefficient cows. Interestingly, no quantitative relationship could be established that separated these animals using either PCA or PLS-DA for high versus low FCE for data from either biofluid (Figures S2−S4). This implies that the underlying metabolic basis for FCE is too multifactorial or “upstream” of blood and milk in these cows, and that selection of cows for higher FCE will not have a substantial effect on the metabolic content of milk produced. Further PLS analyses were used to identify the existence of quantitative relationships between blood or milk metabolic data and commercially relevant traits. Phenotypic (i.e., trait) data such as weight gain, milk yield, percent lactose, protein and fat were measured for each animal. A PLS regression model was constructed between each of these phenotypes and spectroscopic data sets to establish the extent to which of these phenotypes could be predicted from the metabolic profiles, and to identify biomarker signatures underlying those relationships. We computed the Q2 value after 7-fold internal cross validation, and if it was above 0.3; we further tested its robustness by 1000fold recalculation of the PLS model by random permutation of the Y data. If the Q2 from the original Y data was greater than those from the permuted Y, the model was considered significant. Under these criteria, three phenotypes were found to be predictable from the milk metabolome: milk protein percentage, milk fat percentage, and milk yield. Figure 2 shows the PLS regression loadings plot from a PLS model constructed between the milk metabolomic data and the protein percentage in milk. The PLS loadings plot (Figure 2A)

■

RESULTS AND DISCUSSION Figure 1 shows an example of the four types of spectral data acquired from the blood and milk samples collected from an individual cow. The data acquired using a standard pulse-andacquire pulse sequence are labeled “1D”, and those from the CPMG (or relaxation-edited) experiment are labeled “RE”. Spectra from blood plasma samples (upper two spectra in red) resemble those observed for human blood plasma, and many resonance assignments can be made from known chemical shift and coupling patterns.21 The “1D” spectra from skim milk are dominated by broad signals from proteins and lipids, many of which are also visible in the relaxation edited despite the 153.6 ms total echo time, suggesting these signals are from a broad range of mobile macromolecules. Because of the enhanced NMR “visibility” of small molecules through reduced overlap in the CPMG experiments, we performed our subsequent statistical analyses on these data. Principal Components Analysis and Partial Least-Squares Regression

Initially, a PCA model was constructed from a concatenated matrix consisting of the blood and milk NMR digitized spectra to investigate the overall variation in the combined data. The scores plot in Figure S1 shows the milk samples were 1430

dx.doi.org/10.1021/pr301056q | J. Proteome Res. 2013, 12, 1428−1435

Journal of Proteome Research

Article

Figure 2. Results from a partial least-squares regression model for quantification of the relationship between the milk NMR metabolomic data against the percentage milk protein. (A) PLS loadings displayed as the covariance of the NMR variables with protein percentage, color-coded by the square of the correlation coefficient. (B) Permutation testing of the Q2 value for the PLS model. The Q2 for the unpermuted response variable can be seen on the right-hand side of the graph (at r2 = 1).

Figure 3. Two-dimensional statistical heterospectroscopy (SHY) “spectrum” expanded around the region between δ2.6 and δ3.0. Each data point has been color-coded according to its correlation coefficient indicated by the colorbar on the right-hand side. The projections to the left and above the main plot show the stacked plot of all blood and milk spectra in the respective data sets. The insets expand around two of the key correlations discussed in the text.

makes up approximately 5% of nonprotein nitrogen in cows milk, and our results show that this is inversely related to protein and fat percentage in milk. Lactose was also found to be inversely related to the fat and protein percentages in milk, while being correlated positively with milk yield (Figure S5A,B). This result would be expected since lactose production in the Golgi vesicles of the mammary tissue essentially

showed that higher milk protein percentage was associated with lower levels of orotic acid, lactose and citrate; milk yield was associated with higher lactose levels (Figure S5A,B), while a similar metabolic pattern to that found for protein percentage was observed for fat percentage (Figure S5C,D). Orotic acid is an intermediate in pyrimidine biosynthesis, and cow’s milk is the major source of orotic acid in the human diet.22 Orotic acid 1431

dx.doi.org/10.1021/pr301056q | J. Proteome Res. 2013, 12, 1428−1435

Journal of Proteome Research

Article

Figure 4. Scatter plots of selected correlations from the SHY analyses displayed in Figure 3. (A) Scatter plot showing correlation between blood and milk for the resonance at δ2.93. (B) Scatter plot showing correlation between blood and milk for the resonance at δ2.73. (C) Scatter plot showing correlation between the intensity of the peak at δ2.73 in blood and days in milk for each cow.

(DMA). Further investigations revealed this particular animal was a recent calver, and at the time of sampling was only 25 days into lactation, compared to the group median of 168 days. In Figure 4C, we have plotted the number of days in milk against the intensity of this peak in blood serum NMR spectra, showing it was very much an outlier with respect to this parameter. We cannot conclude from these data that blood DMA is inversely proportional to lactation stage (this would require further longitudinal studies), but it is noteworthy that this subtle metabolic relationship was unlikely to have been detected by statistical methods other than SHY. For example, PCA is usually used as a first pass data analysis tool to detect outliers. We have plotted the scores from a PCA model from the milk NMR data in Figure S6A. When we label the datum from the animal in question (red circle), we see it was potentially an “outlier”; however, the loadings (Figure S6B) were dominated by variables representing lactose, making it unlikely that more subtle metabolic effects would have been uncovered. Other significant correlations discovered using SHY have been plotted as scatter plots in Figure 5. A metabolite at δ3.15, assigned to dimethylsulfone (DMSO2), was found to be highly correlated across the biofluids (r = 0.69, p = 9.5 × 10−9). Interestingly, this is consistent with this metabolite levels being highly correlated across the blood−CSF metabolic axis in human subjects infected with HIV-1.10 DMSO2 in human plasma is known to arise from either dietary sources or from gut microbial production.23 In this instance, further identifying the original source of this metabolite may give insights into

determines the volume of water in milk, which proportionally dilutes the protein and fat concentrations.1 Statistical Heterospectroscopy To Determine Relationship between Metabolites in Blood and Milk

Statistical Heteroscpectroscopy (SHY) is a powerful approach to recovering latent biological information in spectroscopic data sets from multiple complementary samples. A matrix of correlations was constructed between each variable in the milk NMR spectral data with every variable in the blood NMR spectral data. After removing statistically insignificant correlations, regions of high positive and negative correlations were color-coded with red and blue pixels, respectively, and are suggestive of a biological relationship between the respective metabolic components. Figure 3 plots a subregion of the SHY matrix, expanded about the regions between δ2.6 and δ3.0. This plot resembles a 2D NMR spectrum where each “peak” is a statistical correlation between resonance intensities in the two series of spectra. The two insets further expand on the regions around the two highest correlating peaks in this region, those at δ2.73 and δ2.93. The presence of a correlation can be visualized by plotting the intensities of these peaks against each other in a scatter plot. The scatter plot for the peak at δ2.93 (Figure 4A) shows a high, systematic correlation for this metabolite between these biofluids. We have assigned this to trimethylamine (TMA). However, the peak at δ2.73 (Figure 4B) was seen to have a single cow with high serum and milk levels of this metabolite, while little correlation existed for the remainder of the cows, thus the high correlation coefficient (r = 0.59, p = 3.8 × 10−6). This metabolite has been assigned to be dimethylamine 1432

dx.doi.org/10.1021/pr301056q | J. Proteome Res. 2013, 12, 1428−1435

Journal of Proteome Research

Article

Figure 5. Scatter plots of selected metabolic relationships identified from SHY.

serum acetoacetate (r = 0.49, p = 0.00017), alanine (r = 0.39, p = 0.0041) and formate (r = 0.55, p = 1.8 × 10−5), but the biological significance of these correlations cannot be interpreted at this stage.

rumen function and its effect on milk composition in these animals. We also identified inverse relationships between serum valine and milk fumarate (r = −0.46, p = 0.00058), and milk glycerophosphocholine and serum trimethylamine-N-oxide (r = −0.46, p = 0.00045). Fumarate is a citric acid cycle intermediate and valine is usually incorporated into this pathway for degradation. It is known that uptake of valine and other amino acids can significantly exceed output in the milk,24 being used as an additional N and C source.25 While there is not enough information to deduce the mechanisms explaining this correlation, the results may indicate there is variation in the valine uptake and utilization by the mammary gland which may potentially be a significant determinant of blood valine levels. Positive relationships were found for an unidentified milk metabolite at δ2.14, tentatively assigned to an N- or O-linked acetylated moiety on a mobile region of a glycoprotein,21 with

■

CONCLUSIONS This paper establishes the first comprehensive analysis of the metabolic correlations between blood and milk in dairy cows; relationships that potentially have important biological and economic relevance. For example, our discovery of the correlation between TMA in blood and milk, the confirmation of the importance of lactose, and the link of orotic acid in milk to higher milk protein and fat, may have immediate application for identification of cows that can produce higher levels of quality milk. This may be particularly useful if measured metabolites in the blood of young heifers not yet milking can be predictive, and further studies are required to demonstrate this. 1433

dx.doi.org/10.1021/pr301056q | J. Proteome Res. 2013, 12, 1428−1435

Journal of Proteome Research

■

Some of the milk properties measured and correlated to specific metabolites in this study are relevant to nutrition and the suitability of milk for processing,13 providing knowledge on mechanisms controlling milk quality. In principle, the same type of analysis could be applied in humans to extend our understanding of the blood-milk metabolic interactions for the benefit of infant nutrition. These results also exemplify the utility of statistical integration techniques such as SHY for identifying correlations across different data sets, given an appropriate experimental design, and may help explain why other studies that combined blood and milk metabolomics did not observe similar types of correlations. For example, the recent study by Ilves et al. reported “little correlation between the molecular composition of blood and milk” metabolomes over the course of lactation in dairy cows, but the experimental design was qualitatively different to ours in that there were only 5 cows in total.14 In our study, rather than taking a longitudinal approach, we have taken a single time point with sufficient numbers of samples to measure robust correlations using SHY. The study confirms that milk is largely a separate metabolic compartment to blood when examined for natural metabolites, but relationships can be measured and milk has the untapped potential for rich information about cows’ performance and health. The metabolites identified here as being linked across the milk−blood axis may provide the basis for further studies to see if they could be predictive for animal status and performance. Testing milk and blood for metabolic surrogates for complex phenotypes like FCE might be possible; however, in this initial study, we were not able to measure a clear link of metabolite signature for this complex trait. It may be possible to do this with an increased animal sample size; however, it may be difficult because FCE is a systemic trait contributed to incrementally by many genes and physiological processes.26 It also may be possible that the biochemical information in blood plasma collected from the tail vein of cows may not be metabolically “close” enough to key organs contributing to FCE such as the rumen to recover biological information correlating with FCE. For example, blood collected from the portal vein could be expected to be more representative of the metabolism adjacent to the rumen and gut, increasing the chances of observing a significant relationship. However, the difficulty of collecting blood from the portal vein might limit the applicability of this idea in practice. Despite this, a number of the metabolites measured in milk such as orotic acid, TMA and DMSO2 are microbial products, indicating the potential of the approach to gain information about rumen status in future studies. Some differences discovered between animals in steady-state metabolite levels may be indicative of fundamental physiological differences, given the animals in this study are in the same environment. Some of these are directly or indirectly linked to protein synthesis. The metabolic analysis confirms the importance and independence of the mammary gland in maintaining milk as a reliable source of sustenance for sucking young and provides a basis for more targeted investigations into important and complex metabolic differences.

■

Article

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Phone: +61-3-9032 7014. Fax: +61-3-9032-7158. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS Funding for this research was provided by the Dairy Futures Cooperative Research Council, Gardiner Foundation, Victoria and Department of Primary Industries, Victoria, Australia.

■

REFERENCES

(1) Linzell, J. L.; Peaker, M. Mechanism of Milk Secretion. Physiol. Rev. 1971, 51 (3), 564−597. (2) Shennan, D. B.; Peaker, M. Transport of milk constituents by the mammary gland. Physiol. Rev. 2000, 80 (3), 925−951. (3) Fonville, J. M.; Richards, S. E.; Barton, R. H.; Boulange, C. L.; Ebbels, T. M. D.; Nicholson, J. K.; Holmes, E.; Dumas, M.-E. The evolution of partial least squares models and related chemometric approaches in metabonomics and metabolic phenotyping. J. Chemom. 2010, 24 (11−12), 636−649. (4) Fiehn, O. MetabolomicsThe link between genotypes and phenotypes. Plant Mol. Biol. 2002, 48 (1−2), 155−171. (5) Nicholson, J. K.; Connelly, J.; Lindon, J. C.; Holmes, E. Metabonomics: A platform for studying drug toxicity and gene function. Nat. Rev. Drug Discovery 2002, 1 (2), 153−161. (6) Keun, H. C.; Ebbels, T. M.; Antti, H.; Bollard, M. E.; Beckonert, O.; Schlotterbeck, G.; Senn, H.; Niederhauser, U.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Analytical reproducibility in 1H NMR-based metabonomic urinalysis. Chem. Res. Toxicol. 2002, 15 (11), 1380− 1386. (7) Cloarec, O.; Dumas, M. E.; Craig, A.; Barton, R. H.; Trygg, J.; Hudson, J.; Blancher, C.; Gauguier, D.; Lindon, J. C.; Holmes, E.; Nicholson, J. Statistical total correlation spectroscopy: an exploratory approach for latent biomarker identification from metabolic 1H NMR data sets. Anal. Chem. 2005, 77 (5), 1282−1289. (8) Crockford, D. J.; Holmes, E.; Lindon, J. C.; Plumb, R. S.; Zirah, S.; Bruce, S. J.; Rainville, P.; Stumpf, C. L.; Nicholson, J. K. Statistical heterospectroscopy, an approach to the integrated analysis of NMR and UPLC-MS data sets: application in metabonomic toxicology studies. Anal. Chem. 2006, 78 (2), 363−371. (9) Crockford, D. J.; Maher, A. D.; Ahmadi, K. R.; Barrett, A.; Plumb, R. S.; Wilson, I. D.; Nicholson, J. K. 1H NMR and UPLC-MS(E) statistical heterospectroscopy: characterization of drug metabolites (xenometabolome) in epidemiological studies. Anal. Chem. 2008, 80 (18), 6835−6844. (10) Maher, A. D.; Cysique, L. A.; Brew, B. J.; Rae, C. D. Statistical integration of 1H NMR and MRS data from different biofluids and tissues enhances recovery of biological information from individuals with HIV-1 infection. J. Proteome Res. 2011, 10 (4), 1737−1745. (11) Boudonck, K. J.; Mitchell, M. W.; Wulff, J.; Ryals, J. A. Characterization of the biochemical variability of bovine milk using metabolomics. Metabolomics 2009, 5 (4), 375−386. (12) Klein, M. S.; Almstetter, M. F.; Schlamberger, G.; Nurnberger, N.; Dettmer, K.; Oefner, P. J.; Meyer, H. H.; Wiedemann, S.; Gronwald, W. Nuclear magnetic resonance and mass spectrometrybased milk metabolomics in dairy cows during early and late lactation. J. Dairy Sci. 2010, 93 (4), 1539−1550. (13) Sundekilde, U. K.; Frederiksen, P. D.; Clausen, M. R.; Larsen, L. B.; Bertram, H. C. Relationship between the metabolite profile and technological properties of bovine milk from two dairy breeds elucidated by NMR-based metabolomics. J. Agric. Food Chem. 2011, 59 (13), 7360−7367.

ASSOCIATED CONTENT

S Supporting Information *

Additional information as noted in text. This material is available free of charge via the Internet at http://pubs.acs.org. 1434

dx.doi.org/10.1021/pr301056q | J. Proteome Res. 2013, 12, 1428−1435

Journal of Proteome Research

Article

(14) Ilves, A.; Harzia, H.; Ling, K.; Ots, M.; Soomets, U.; Kilk, K. Alterations in milk and blood metabolomes during the first months of lactation in dairy cows. J. Dairy Sci. 2012, 95 (10), 5788−5797. (15) Williams, Y. J.; Pryce, J. E.; Grainger, C.; Wales, W. J.; Linden, N.; Porker, M.; Hayes, B. J. Variation in residual feed intake in Holstein-Friesian dairy heifers in southern Australia. J. Dairy Sci. 2011, 94 (9), 4715−4725. (16) Tomasi, G.; van den Berg, F.; Andersson, C. Correlation optimized warping and dynamic time warping as preprocessing methods for chromatographic data. J. Chemom. 2004, 18 (5), 231− 241. (17) Dieterle, F.; Ross, A.; Schlotterbeck, G.; Senn, H. Probabilistic quotient normalization as robust method to account for dilution of complex biological mixtures. Application in 1H NMR metabonomics. Anal. Chem. 2006, 78 (13), 4281−4290. (18) Geladi, P.; Kowalski, B. R. Partial least-squares regression: a tutorial. Anal. Chim. Acta 1986, 185, 1−17. (19) Cloarec, O.; Dumas, M. E.; Trygg, J.; Craig, A.; Barton, R. H.; Lindon, J. C.; Nicholson, J. K.; Holmes, E. Evaluation of the orthogonal projection on latent structure model limitations caused by chemical shift variability and improved visualization of biomarker changes in 1H NMR spectroscopic metabonomic studies. Anal. Chem. 2005, 77 (2), 517−526. (20) Maher, A. D.; Cloarec, O.; Patki, P.; Craggs, M.; Holmes, E.; Lindon, J. C.; Nicholson, J. K. Dynamic biochemical information recovery in spontaneous human seminal fluid reactions via 1H NMR kinetic statistical total correlation spectroscopy. Anal. Chem. 2009, 81 (1), 288−295. (21) Nicholson, J. K.; Foxall, P. J.; Spraul, M.; Farrant, R. D.; Lindon, J. C. 750 MHz 1H and 1H-13C NMR spectroscopy of human blood plasma. Anal. Chem. 1995, 67 (5), 793−811. (22) Indyk, H. E.; Woollard, D. C. Determination of orotic acid, uric acid, and creatinine in milk by liquid chromatography. J. AOAC Int. 2004, 87 (1), 116−122. (23) Engelke, U. F.; Tangerman, A.; Willemsen, M. A.; Moskau, D.; Loss, S.; Mudd, S. H.; Wevers, R. A. Dimethyl sulfone in human cerebrospinal fluid and blood plasma confirmed by one-dimensional 1 H and two-dimensional 1H-13C NMR. NMR Biomed. 2005, 18 (5), 331−336. (24) Trottier, N. L.; Shipley, C. F.; Easter, R. A. Plasma amino acid uptake by the mammary gland of the lactating sow. J. Anim. Sci. 1997, 75 (5), 1266−1278. (25) Richert, B. T.; Goodband, R. D.; Tokach, M. D.; Nelssen, J. L. In vitro oxidation of branched chain amino acids by porcine mammary tissue. Nutr. Res. 1998, 18 (5), 833−840. (26) Pryce, J. E.; Arias, J.; Bowman, P. J.; Davis, S. R.; Macdonald, K. A.; Waghorn, G. C.; Wales, W. J.; Williams, Y. J.; Spelman, R. J.; Hayes, B. J. Accuracy of genomic predictions of residual feed intake and 250-day body weight in growing heifers using 625,000 single nucleotide polymorphism markers. J. Dairy Sci. 2012, 95 (4), 2108− 2119.

1435

dx.doi.org/10.1021/pr301056q | J. Proteome Res. 2013, 12, 1428−1435