Publication Details

Residual feed intake (RFI) is a livestock feed efficiency measure defined as the difference between an animal’s actual feed intake and its expected feed requirements for maintenance and growth over a specific time period. RFI is independent of growth characteristics such as body weight (BW) and average daily gain (ADG) [

However, because RFI measurements are expensive and time-consuming, they are performed only on a small percentage of the cattle population. Simpler or cheaper proxies for measuring RFI are clearly desirable. Because RFI is a measure of metabolic efficiency, it has been proposed that metabolomics or metabolite measurements of bovine biofluids may offer a lower cost alternative to manual RFI measurement. Several metabolomics studies have been conducted in beef cattle to explore the relationship between RFI and metabolite levels [

Serum metabolomic data were obtained from 15 HRFI and 10 LRFI young Angus bulls using three metabolomics platforms including NMR, LC-MS/MS, and ICP-MS. A total of 145 metabolites were identified and quantified in each serum sample (

Using a combination of NMR and LC-MS/MS, a total of 58 water-soluble organic compounds were identified and quantified in bovine serum. The most abundant water-soluble organic compounds in serum were lactate (5393 ± 2341 µM), glucose (4115 ± 326 µM), and urea (1389 ± 266 µM). The lowest concentration that could be reliably detected in serum was 0.035 ± 0.021 µM for putrescine.

The TMIC Prime assay (a locally developed LC-MS/MS assay) provided quantitative results for 74 lipids or lipid-like compounds including 10 phosphatidylcholines (PCs), 14 lysophosphatidylcholines (LysoPCs), 5 sphingomyelins (SMs), 5 hydroxysphingomyelins (SM(OH)s), and 40 acylcarnitines (ACs) in bovine serum. Note that some LysoPC and PC species identified by the TMIC Prime assay correspond to multiple (ranging from as few as 2 to as many as 24) possible unique lipid structures. In our study, SM(16:0) (69 ± 10 µM) and C14:2-OH (hydroxytetradecadienylcarnitine) (7.5 ± 1.1 nM) were the most and least abundant lipid-like compounds identified in serum, respectively.

ICP-MS also provided quantitative results for 13 trace minerals in bovine serum. The most abundant elements identified and quantified by ICP-MS were sodium (134 ± 16 mM), potassium (4.3 ± 0.3 mM), calcium (2.2 ± 0.2 mM), and phosphorus (1.3 ± 0.2 mM). While the least abundant metals quantified by ICP-MS were cesium (1.6 ± 0.2 nM), barium (190 ± 40 nM), and strontium (940 ± 140 nM).

Using univariate analysis, we compared the serum metabolite profile of those young Angus bulls identified as being HRFI with those identified as being LRFI. The most significantly different metabolites (

Principle component analysis (PCA) showed moderately separable clustering between HRFI and LRFI animals (

From the significant metabolites identified via our univariate and multivariate analyses, we used logistic regression to generate two optimal models for distinguishing HRFI from LRFI animals. One biomarker panel uses only NMR-acquired data while the second uses only LC-MS/MS acquired data. The NMR model used two metabolites that are easily measured by NMR, i.e., formate and leucine (with an AUROC of 0.92 and a

As noted above, the best performing panel was the NMR-based test, which included formate and leucine. A logistic regression equation for these two candidate biomarkers was used to calculate the receiver operating characteristic (ROC) curve and to calculate the area under the ROC curve or AUROC (

The second best performing RFI prediction panel included two metabolites that could only be measured by LC-MS/MS, i.e., C4 (butyrylcarnitine) and LysoPC(28:0). A logistic regression equation for these two candidate biomarkers was used to generate a model with a final AUROC of 0.89 (

The main objective of this study was to identify candidate serum biomarker metabolites that could successfully discriminate HRFI cattle from LRFI cattle. To optimize the likelihood of identifying robust RFI biomarkers we used a combination of three quantitative metabolomics platforms (NMR, LC-MS/MS, and ICP-MS). Using these three platforms, we were able to identify and quantify a total of 145 metabolites, including 58 water-soluble organic compounds, 74 lipid-like compounds, as well as 13 metal ions. Overall, we found a very good agreement between the results of these 145 experimentally quantified metabolites with those of reported elsewhere (available in

Of course, there were a few exceptions to this rule. The most variable metabolite reported in serum was betaine. The value of betaine reported by our study ranged from 131 to 205 µM, and the literature-reported values ranged from 14 to 26 µM [

To date, there have been four other published metabolomic studies that have attempted to identify relationships between blood metabolite levels and bovine RFI [

Another notable difference was found for blood glucose concentrations between our study and the values reported by Fitzsimons et al. [

While other studies have identified possible associations between blood metabolites and bovine RFI, as yet, no published study has attempted to develop quantitative metabolite biomarker panels to predict RFI in cattle. Using logistic regression models, two categorical predictive biomarker panels were developed from this study to categorically predict RFI and to distinguish HRFI animals from LRFI animals.

The best performing panel was an NMR-based, two-metabolite model that included formate and leucine. The second-best performing panel was an LC-MS/MS based two-metabolite model that included C4 (butyrylcarnitine) and LysoPC (28:0). Both panels have high sensitivity and specificity (AUROC > 0.85), making them good candidates to distinguish or predict HRFI animals from LRFI animals. Because these panels consist of just two metabolites, it is possible to construct very fast (<5 min/sample) and inexpensive (<$10) NMR or MS-based assays that could be used to perform bovine RFI characterization.

Basarab et al. [

As noted in the Methods section, the serum samples used to perform these metabolomic assays were collected at 15 months (shortly before the cattle were slaughtered at 17 months). Beef cattle produced in the United States and Canada can be slaughtered at any time from 12 months to 24 month of age, with the highest quality beef coming from those slaughtered under 24 months of age and the most tender meat found in animals slaughtered between 12–18 months of age [

It is also important to note that other physiological factors certainly play a role in the composition of the bovine metabolome (and therefore the biomarker panel parameters described here), including physical maturity, sex, and castration status. These physiological and age-dependent differences would be expected to lead to changes in the optimal cut-off concentrations. Typically, bulls are castrated at three to six weeks of age to become steers [

In addition to working with animals covering a wider set of ages (to ascertain the RFI biomarker age-range), it would also have been useful to perform further validation of these biomarkers on a “hold out” set of animals. These hold-out animals would have ideally raised elsewhere or at a different time using similar feeding, housing, and animal management conditions. However, the high costs of measuring RFI, the length of the study (almost two years) and the costs of maintaining the animals for 17 months make these sorts of studies prohibitively expensive, especially given the limited resources for this sort of discovery-based study.

Our study identified a number of significantly different metabolites that seemed to drive the observed differences in RFI, i.e., C4 (butyrylcarnitine), LysoPC(28:0), formate, and leucine. Each of these compounds plays an important role in bovine metabolism. C4 (butyrylcarnitine) is an acylcarnitine formed when fatty acyl-coenzyme A (fatty acyl-CoA) enters through the carnitine shuttle into the mitochondria for β-oxidation and the tricarboxylic acid (TCA) cycle to produce ATP [

LysoPC(28:0) belongs to the lysophosphatidylcholine family of lipids which are derived by partial hydrolysis of phosphatidylcholines by removing one of the fatty acid groups, via the action of phospholipase A2 (PLA2) [

As comprehensively reviewed by Herd and Arthur [

The other two serum metabolites that were most differentiating between HRFI and LRFI animals included formate and leucine. The concentration of formate and leucine was higher in HRFI animals and lower in LRFI animals. Formate participates in NADPH synthesis and catalyzes the conversion of fumarate into succinate in the TCA cycle [

Leucine is a branched-chain amino acid and its catabolism generates succinyl-CoA and acetyl-CoA, both of which can upregulate the activity of the TCA cycle [

We also performed a further study to understand if the variations in the concentration of C4 (butyrylcarnitine), LysoPC(28:0), formate, and leucine between HRFI and LRFI bulls correlated with the concentration of these metabolites in their rumen. This was done to explore whether these metabolite difference may be associated with differences in ruminal activity or rumen microbial activity. However, we found no such correlation (data not shown).

The collection and analysis of bovine serum in this study was approved by the University of Alberta’s Animal Care Committee (Animal Use Protocol [AUP] 1129) under the auspices of the Canadian Council of Animal Care [

Twenty-five purebred Angus bulls, raised on the University of Alberta’s Roy Berg Kinsella Research Ranch (Kinsella, AB, Canada), were used in this study. After weaning, bulls were fed and managed according to industry standards for production of potential replacement yearling bulls in Alberta until their RFI test at approximately 13 months of age [

From the end of May 2015 until mid-August 2015, bulls were tested for RFIf (RFI that was adjusted for rib fat thickness at the end of feedlot test) at approximately 13 to 16 months of age using the GrowSafe

The end of the RFI test weight was estimated from the slaughter weight. Rib fat thickness measurements (12/13th rib fat depth and LT area) were also determined at end of test, using an Aloka SSD-210 portable ultrasonographic scanner (Aloka Co., Tokyo, Japan). The initial BW at the start of the test and ADG were derived from a linear regression of the serial BW measurements against time (day). Then, the metabolic BW (MWT) in kg was calculated as midpoint BW

Blood samples (10 mL) were collected in the morning (just before feeding) at 15 months of age from a jugular vein using vacutainer serum collection tubes (Becton Dickinson, Mississauga, ON, Canada). Blood samples were kept in a cooler on ice, transferred to the laboratory within 3 h after collection, and centrifuged at 2000×

Three metabolomics platforms, including NMR, LC-MS/MS, and ICP-MS, were used to identify and quantify a total of 145 metabolites in each bovine serum sample. Using NMR, LC-MS/MS, and ICP-MS, 42, 116, and 13 metabolites were identified and quantified, respectively, of which 26 metabolites were common between NMR and LC-MS/MS. Details of sample preparation along with how the samples were run on each metabolomics platform have been previously described in detail by Foroutan et al. [

A targeted, quantitative LC-MS/MS metabolite profiling approach was employed that combined reverse-phase liquid chromatography and mass spectrometry (RPLC-MS) with direct flow injection (DFI) mass spectrometry (DFI-MS) (RPLC-DFI-MS/MS). LC-MS/MS was employed to determine the concentrations of up to 143 compounds (including amino acids, biogenic amines, glucose, organic acids, acylcarnitines, PCs, LysoPCs, SMs, and SM(OH)s) using an in-house quantitative metabolomics assay (TMIC Prime) [

Data analysis was performed using MetaboAnalyst 4.0 according to previously published protocols [

Multivariate statistics, including PCA, PLS-DA, and ROC curve analysis, were performed using MetaboAnalyst 4.0. The data was scaled and normalized using a cube root transformation and auto scaling, which generated a clear Gaussian distribution plot prior to multivariate analysis. A permutation test involving 2000 randomized datasets was implemented to minimize the possibility that the observed separation of the PLS-DA was due to chance (a valid model should have a

ROC curves were calculated by MetaboAnalyst 4.0 to evaluate the predictive ability of potential metabolic biomarkers using a logistic regression model. The area under the ROC curve (AUC or AUROC) was used to interpret the performance across the two different biomarker models to determine the best cut-off point for maximal sensitivity and specificity. A ROC curve plots the false-positive rate (1-specificity) on the X axis versus sensitivity on the Y axis. On the one hand, sensitivity (or recall) is defined as the number of true positives divided by the sum of the true positives and false negatives. On the other hand, specificity is defined as the number of true negatives divided by the sum of the true negatives and false positives. In a ROC curve, the accuracy of a test for correctly distinguishing one group from another, such as HRFI bulls from LRFI bulls, is measured by the area under the ROC curve (AUROC). The AUROC equal to 1 is the highest value indicating a perfect discriminating test, which is obtained when all positive samples are ranked before negative ones. A permutation test involving 1000 randomized permutations was implemented to validate (a valid model should have a

In this study we evaluated the effectiveness of using multi-platform, quantitative metabolomics to identify candidate serum biomarkers that can easily distinguish HRFI animals from LRFI animals. LC-MS/MS, NMR, and ICP-MS were used to identify and quantify 145 serum metabolites in an effort to maximize our chances to identify and develop a suitable set of metabolite RFI biomarkers. We successfully identified two significant candidate biomarkers panels (AUROC > 0.85) that can predict RFI categorically. These include a two-metabolite model (formate and leucine) that is compatible with NMR analysis and a two-metabolite model (C4 (butyrylcarnitine) and LysoPC(28:0)) that is compatible with LC-MS/MS analysis. These results suggest that serum metabolites could be used to categorically predict RFI (early on) and inexpensively distinguish HRFI cattle from LRFI cattle.

While the results we obtained are very statistically significant and appear to be consistent with other reported studies on bovine RFI, the main limitation in this study was the small sample size (15 HRFI vs. 10 LRFI cattle). Given the significant costs and time associated with performing RFI measurements on cattle, this is a limitation that is difficult to overcome. Another limitation lies in the fact that the study was conducted on only a single sex (bulls), from a single breed (Angus cattle), consuming the same diet. However, it is important to note that we demonstrated that the data we measured in this study was broadly consistent with data collected for other beef cattle RFI studies. This gives us reason to believe that the results presented here will be shown to be largely reproducible elsewhere. Nevertheless, in order to properly confirm the robustness of these serum biomarkers as proxies to distinguish between divergent RFI cattle, further validation studies using a larger cohort of cattle with more diverse genetic backgrounds and from different management settings will be needed.