Prediction of polyethylene glycol (PEG) concentration in ovine faeces by near infrared reflectance spectroscopy (NIRS)

In studies oriented to estimate pasture intake in grazing/browsing herbivores, diet digestibility and total faeces production should be determined. Digestibility can be estimated through in vitro techniques and faeces production with total collection bags, a very impractical method under most circumstances, or by indigestible external markers. One of such markers is polyethylene glycol (PEG), obtained by polymerization of ethylene oxide. PEG can be determined by different methods that require reagents and are time and labour consuming. Therefore, near infrared reflectance spectroscopy (NIRS) appears as an interesting alternative method. The objective of this work was to evaluate the suitability of NIRS to predict the PEG content in faecal samples of sheep grazing a naturalised permanent pasture in Southern Chile.
Faeces recently voided were collected from paddocks grazed by sheep at the Agricultural Experimental Station of Universidad Austral de Chile, at Valdivia, Región de los Ríos. Samples were pooled, dried, ground and homogenised. Known amounts of PEG were added to sub-samples, to obtain final concentrations in the range of 0 to 10% PEG (dry matter basis) at increments of 0.2 percentage units. Mixtures were prepared in triplicates with 150 total sub samples.
Visible-NIR spectra (400-2500 nm) were collected in reflectance (NIRSystems 6500 monochromator) with samples in circular cells inserted in a spinning module. Software WINISI II v 1.02 was employed to collect spectral data and develop calibrations. Modified partial least squares (MPLS) regression models were adjusted among reference (% PEG) and spectral values, applying different mathematical treatments (none, first or second subtraction order, different smoothing intervals) and applying or not a scatter correction treatment of standard normal variate (SNV) and detrending. Cross-validation was employed to test the calibrations developed and indicators of certainty (coefficient of determination for cross validation, R2CV) and uncertainty (root mean square of cross validation, RMSECV) were obtained. The best equation was selected according to RMSECV (lowest) and a higher RPD, that is, the relationship between the standard deviation of the reference values and RMSECV.
The selected equation included eight PLS terms and a 2,8,8 treatment, i.e., a second derivative (second subtraction order) of the spectra across intervals of eight data points and smoothing segments of eight data points. SNV was also applied. Statistical terms for the best equation (R2CV, SECV and RPD) were 0.99, 0.26% and 10.93. These results indicate that the prediction equation obtained exhibits a close relationship of the spectral data to the reference values and an excellent adjustment of predicted versus reference values of PEG content. An RPD ratio over three is considered adequate for a calibration to perform discriminant analysis. Results for this work, with a SECV (0.26%) less than one tenth of the standard deviation of the reference values (2.84%), imply that the equation could be used with confidence to predict PEG content in sheep faeces.
It is concluded that NIRS can be employed as fast and accurate method to predict PEG concentration in ovine faeces and therefore has a direct application in research involving pasture intake in grazing sheep.