This study presentsresults concerning the use of FT-MIR spectroscopy to predict a set of plasmaparameters included in the metabolic profiles of horses. To build predictionmodels that accurately predict parameters in unknown samples, the calibrationdataset should contain samples that represent all possible sources of variationthat can be encountered when the prediction model is used on an unknown sample.For this reason, it is very important to select samples that will provide thelargest range of information for building the calibration dataset 7. The horses selected for this studywere all clinically healthy; consequently, the variability observed in the datais representative of normal physiological conditions.
In the absence of clinicalsigns of disease, individuals in a population can be characterized by thepresence of high variability in the concentrations of many blood biomarkers andin the occurrence of subclinical conditions of disease (i.e., the presence of acute-phaseproteins). The concentrations of the blood parameters measured in the presentstudy were within the reference ranges reported for horses without clinicalsigns of disease 21–23.The accuracy of infraredanalysis is affected by the quality of the reference assays used 24. Precision is associated with thevariability in the actual value being measured when an assay is replicated.When a particular assay is run repeatedly on the same sample and the resultsobtained have little variability, the assay is said to have high precision 25.
Large variability in the observedresults indicates low assay precision. In our study the reference assays usedto analyze some blood parameters were characterized by not having optimalrepeatability between runs, with a CV greater than 10% for haptoglobin andbetween 3 and 10% for creatinine and HDL and LDL cholesterol concentrations and for PON,MPO, FRAP, ceruloplasmin, and GGT activityEd1 . For these blood parameters, the predictionmodels where characterized by an RPD lower than 2, which is considered to bethe minimal threshold for acceptable approximate quantitative predictions 20. Among these blood variables thatdo not exhibit optimal repeatability in the reference chemistry, only the predictionmodels of HDL and LDL cholesterol could be considered nearly acceptable. Improvementin the reference assays for all these blood parameters that do not exhibit optimalrepeatability could improve the predictive ability of the prediction modelbased on FT-MIR spectroscopy.With theexception of inorganic P, the prediction models of all the minerals analyzed inthis study were characterized by poor prediction ability. Similar results wereobserved, mainly for electrolytes, in our previous study on plasma from dairycows 8 and in studies on cow milk. Inparticular, the concentrations of Mg, Na and K were poorly predicted by FT-MIRspectroscopy 26.
In contrast, Soyeurt et al. 26 and Toffanin et al. 27 have suggested the potential ofFT-MIR spectroscopy to predict Ca and P content in cow milk. Because thecorrelations between Ca and P concentrations estimated by the FT-MIRpredictions and the known milk components were inferior to the correlationcalculated based on the cross-validation, Soyeurt et al.
26 concluded that these calibrationequations were obtained from a real spectral absorbance. Consequently, theprediction of Ca and P concentrations in milk by using FT-MIR spectra is relatedto the real spectral absorbance and not estimated on the basis of thecorrelation of these concentrations with those of other known milk componentsalso predicted by FT-MIR.Aspreviously observed 8, the current study still confirms thedifficulty in predicting mineral content in plasma with FT-MIR spectroscopy,particularly minerals that are not included in organic compounds. The difficultyof predicting mineral content in blood observed in our study is likely due totwo main reasons: First, among the minerals measured in the blood, a largeproportion were in ionized form and were not included in organic compounds. Approximately50% of total Ca in plasma is in the ionized form, and approximately 45% islinked to protein, while approximately 70% of Mg is ionized 28.
Second, in this study, and aspreviously observed in dairy cow plasma 8, the variability observed for theminerals in our dataset was lower compared to the variability observed forother blood parameters. As previously stated, the development of better predictionmodels requires a dataset containing samples with greater variability, leadingto the improvement of the predictive ability of FT-MIR spectroscopy. Toincrease the variability of these indices, it would be useful to include, inthe population, some animals with clinical diseases or physiological conditionsthat alter the mineral blood concentration.Theprediction models of the enzymatic activities measured in this study had poorpredictive ability. As indicated in our previous work 8, the inadequate estimation wasprobably due to the non-normal distribution of the data, with a very lowproportion of high values in this dataset. The logarithmic transformation didnot improve the predictive ability of the prediction models. A betterassessment of enzymatic activities can be obtained by using a database withgreater variability; in particular, a representative number of samples withhigh values could improve the predictive ability of FT-MIR spectroscopy.
Moreover, the activities of the enzymes were measured with the referencemethods, but the absolute concentrations of the enzymes were not quantified. Itis therefore possible that discrepancies exist between the amount and activityof the enzymes.Excellent prediction ability was obtained for several of the parameters, mainly parametersassociated with energy and protein metabolism. Among the energy-relatedparameters, an excellent calibration curve was obtained for total cholesterol, withRER and RPD values greater than the threshold values (10 for RER and 3 for RPD)suggested for excellent prediction 18,20. Acceptable results were also obtained forcholesterol fractions and for triglycerides. Among the parameters associatedwith protein metabolism, excellent prediction models were obtained for totalprotein, albumin, globulin, and urea. Very high RER and RPD values wereobtained for total protein and protein fractions.
These results confirm thoseobtained in our previous study on the plasma of dairy cows 8. The results obtained with FT-MIR for the totalprotein and protein fractions are particularly interesting given that Fourier-transforminfrared spectroscopy is a widely used tool in many different fields and is evenused to evaluate secondary and tertiary protein structure 29,30, to study differentiation of plasmin andplasminogen in milk 31, and to evaluate secondary and tertiarystructural changes in bovine plasminogen 32, and those in casein 33 and casein fractions in goat milk 34. The ability of FT-MIR to predict the concentrationsof protein fractions (i.e., albumin and globulin) offers the opportunity formore extensive evaluations than just the evaluation of protein metabolism. In fact,plasma albumin concentration is determined by the hepatic synthetic rate, whichis normally in equilibrium with degradation. As indicated by Tennant and Center35, hypoalbuminemia may be caused by defectivealbumin synthesis associated with severe hepatocellular disease or may becaused by increased albumin loss resulting from glomerulopathy (protein-losingnephropathy), severe intestinal inflammation, or inflammatory conditions 36.
In acute or chronic inflammatory conditions, arise in total protein concentrations caused by elevated globulin fractions mayoccur; albumin concentrations are often decreased in these situations. Thecombined effect of these changes is a decrease in the albumin/globulin ratio. Often,the total protein concentration is within the reference range, whereas the albumin/globulinratio is decreased; therefore, the albumin/globulin ratio is of greaterclinical significance than the total protein concentration 37. In severe, chronic hepatopathy, concentrationsof IgM, IgG, and IgA tend to be elevated. Both decreased albumin and increasedglobulin result in a decrease in the albumin/globulin ratio 35.
Therefore, FT-MIR also offers the possibility,by the prediction of the concentrations of protein fractions, to obtaininformation associated with the health status and welfare of animals and todiagnose many different problems, including conditions associated with the liver,kidney and gastrointestinal tract and inflammatory conditions.