Analyses of Dry Horse Feed Formulations by Nuclear Magnetic Resonance Spectroscopy (NMR): Improving the Ability to Identify Differences in Formulations

HOIJEMBERG, PABLO ARIEL; PELCZER, ISTVÁN; RALSTON, SARAH L.

Leipzig

Workshop; European Workshop of Equine Nutrition; 2014

Institut für Tierernährung, Universität Leipzig

Introduction: The wet chemical extraction of nutrients and other organic components from horse feeds has been employed historically for a variety of reasons such as ration balancing, detection of contaminants or deviations in formulas. Analyses, however, are most commonly limited to the detection of crude protein (CP, Kjeldahl analysis), fat (EE, Ether extract), starch, water soluble carbohydrates (WSC) and fiber (NDF, ADF, Neutral and acid detergent fiber). More in depth analyses of components such as individual amino acids, fatty acids and lipids is cumbersome and expensive. The analysis of such extracts of dry feeds by nuclear magnetic resonance (NMR) spectroscopy can supply a much larger quantity of data on the nutrient/chemical content at a very reasonable price (USD ~10.00/sample). Several NMR techniques are used to enrich the data stemming from standard one dimension (1D) 1H spectra, such as relaxation filtered, diffusion edited and two dimension (2D) J-resolved spectra, and specific 2D correlation techniques. The more extensive identification of the extracted components makes comparison of different feeds with similar wet chemical analyses possible. Multivariate data analysis (MVDA) is able to detect even small and/or elusive fluctuations in components in similar extract sources if a large enough data set is used. STOCSY analysis (Statistical TOtal Correlation SpectroscopY [Cloarec et al, 2005]) aids in the identification of the components by revealing correlations among the hundreds of peaks found in the complex mixtures. Using STOCSY, peaks corresponding to the same component molecules are easily recognized, and the components responsible for the separation in groups that appear can be identified. It was hypothesized that one could discriminate between two proprietary commercial feed samples that did not differ significantly in standard wet chemistry analyses by use of 1D NMR spectra of serial samples of extracts obtained at increasing soaking times. Materials and Methods: ?Old? and ?New? proprietary pelleted feed formulations were obtained from Standardbred breeding farm that had just switched to a new formula. The ?old? feed formula had been used on the farm for over 8 years, during which time metabonomic analyses of young horses to which it was fed revealed distinct metabolic profiles that differed consistently between horses that had developed hock Osteochondrosis Dissecans (OCD) lesions and those that had not (Ralston et al, 2011). The ?new? feed had been formulated to address the deviations seen in the OCD metabolic profiles. Samples of the feed were submitted to a commercial feed analysis laboratory (Equi-Analytical, Ithaca, NY) for complete wet chemical analysis and to the authors for NMR analyses. The samples for NMR were first ground, then suspended in D2O (with ~0.5 mM sodium azide) and left agitating on an orbital shaker. Aliquots of supernatant were retrieved from each sample vial at increasing soaking times (2, 4 and 26 hours), subsequently centrifuged, and the supernatant transferred to NMR tubes for acquisition of 1D 1H NMR spectra. Spectra were processed according to standard metabonomic procedures (zero-filling, apodization, phase correction, baseline correction, calibration, alignment, suppression of the water region and normalization). The entire set was then subjected to MVDA using principal component analysis (PCA), projection to latent structures discriminate analysis (PLS-DA) and orthogonal PLS-DA (O-PLS-DA) (SIMCA, v 13.0, Umetrics, Umeå, Sweden). Smaller sets of data were analyzed to evaluate the improvement of statistically significant parameters when bigger data sets were available. Loadings/coefficients plots were created with a color coded projection of the (absolute) weight that specific peaks contributed to class separation. STOCSY analysis was performed over the whole set to aid in the identification of key compounds that differed between samples. The color coded scale in the STOCSY plots represents the degree of correlation of the chemical shifts to the chemical shift chosen as the ?driver peak?. Results and Discussion: The difference between the wet chemical analyses of Old versus New dry feed matter showed < 10% difference in CP (16.7% vs 16.0), WSC (7.8% vs 9.7), and Starch (22.5% vs 22.6%) and < 20% difference in ADF(18% versus 21%) and NDF 38.4 vs 34%). However there was clear 2-class separation of the proprietary horse feeds? NMR spectra, validated with only a three-fold increment in the data set size, i.e. 3 soaking times. In the composite of Figure 1, 1, 1+2 and 1+2+3 soaking times were used to build the PLS-DA models to distinguish between ?New? (white circles) and ?Old? (black triangles) feed formulas. Corresponding validation by the permutations test is shown on Figure 2 with Q2 linear regression having a y-intercept below 0.05 for the most complete data set. The NMR peaks with high contribution (weight) to the separation for the ?Old? and ?New? formulations in the corresponding O-PLS-DA model are depicted in coefficients plots (Figure 3 (top)), with positive peaks in higher in the ?Old? formulation than in the ?New? and negative peaks reflecting components higher in the ?New?. The final phase was obtaining STOCSY plots for components of interest (from the coefficients plots), to help identify the compounds that differ between samples (and those that don?t). A comparison of the spectra of the two feeds is shown in Figure 3 (Bottom) with some of the most relevant differences labelled. Further analysis, by means of more sophisticated magnetic resonance heteronuclear correlation techniques, will be focused on the identification of the molecules to which unknown peaks of importance belong.Discussion: The methodology described here addresses a simple method for the unbiased differentiation of the component profiles of two types of dry horse feed. The procedure is not limited to only two samples and it is transferable to any other type of dry food or products of interest. The key is the multiple sampling at incremental soaking times which allowed the procurement of an acceptable amount of data that yielded statistically significant results. This permits a proper distinction between two dissimilar samples of horse feed by multivariate data analysis, instead of a tedious and largely subjective, pairwise visual comparison of spectra. Moreover, these data were further exploited by correlation statistics to attain a more detailed description of the differing components between these samples, having the potential to identify compounds that are in similar or differing concentrations. Reference: 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.K., 2005. Statistical Total Correlation Spectroscopy: An Exploratory Approach for Latent Biomarker Identification from Metabolic 1H NMR Data Sets. Analytical Chemistry, 77, 1282-1289. Figure 1: Soaking times and dataset sizes increase from left to right (time: 2, 2+4 and 2+4+26 hours; dataset size: 1×, 2× and 3×). Left column: Scores Plots for PLS-DA analysis of horse feeds. Figure 2: Permutation tests for validation of PLS-DA models from left to right (time: 2, 2+4 and 2+4+26 hours; dataset size: 1×, 2× and 3×). Points with x=1 are R2 correlation and Q2 predictive values for the original models, while all points to the left belong to models with random permutations of Y (dependent variables) and should ideally produce worse models, with lower R2 and Q2 values, and with y-intercept for linear regression on the Q2 values expected to be below 0.05 for a valid model (x values for all points denote correlation to the original model). Figure 3: Spectral plots of New versus Old feed extracts with peaks of known compounds labelled. Unknown peaks at 0.86 and 1.92 and their related peaks identified by STOCSY also contributed significantly to the separation. Beta-Ala=Beta-Alanine, Lys=Lysine.