Development of a Near Infrared Spectroscopy Model for Prediction of Fibre Compounds in Alfalfa

Abstract: Background: This project investigates if it is possible to develop a calibration model from near infrared (NIR) spectroscopic measurements, for determination of the amount and type of fibre fractions within protein powder produced from the legume alfalfa, without performing wet experiments. Alfalfa is also known as Medicago sativa and lucerne, but is in this project further referred to as alfalfa. Such a model would be applicable as a protein powder production process control, by scanning a small amount of sample during the production process, immediately resulting in a fibre content value. With this result, one will know when the process should be stopped by means of nutritional values. Except from fibres, alfalfa contains large amounts of nutrients, for example essential amino acids. The advantageous amino acids are thus extracted from the fibrous alfalfa during the protein powder production process. The alfalfa protein powder is produced from stems, leaves and flowers of intact, freshly harvested alfalfa plants. The raw alfalfa was frozen during storage, then thawed and wetted prior to the first press, which is resulting in a protein rich green juice, and a fibrous pulp. The pH of the green juice is decreased to precipitate proteins. The green juice is then centrifuged resulting in a pellet consisting of the total water soluble solid content extracted from alfalfa. The pellet is freeze dried into a protein powder in order to concentrate the protein content. This process is performed 10 times in total, the first time untreated raw wetted alfalfa is pressed into green juice as mentioned, the following nine times the fibrous pulp from the prior press is wetted and pressed into new samples of green juice. The aim of re-pressing the fibrous pulp is to extract the highest total amount of protein from one batch of alfalfa. This protein powder production from raw untreated alfalfa to protein powder, does increase human digestibility of alfalfa by increasing the amount of protein per weight. Protein powder derived from each of the 10 presses was collected in separate fractions to determine to which extent the fibre profile is changing using an enzymatic gravimetric method. The amounts of protein, insoluble dietary fibre (IDF), soluble dietary fibre (SDF), total dietary fibre (TDF), available carbohydrates (ACH) and ash were determined, since NIR spectra are affected by all compounds of the protein powder. NIR spectra from all 10 presses are related directly to the determined TDF contents, which are used as reference values in order to calibrate a partial least squares (PLS) model that produces predicted TDF values. Attempts were also made to conduct NIR spectra earlier in the protein powder production process, from the green juice prior to centrifugation and from the pellet prior to freeze drying. A cellulose gluten powder dilution series comparable to the 10 presses of protein powder was prepared, to test if a calibration model could be developed from NIR spectral data of powder containing cellulose as one of the main components. The cellulose gluten spectra were also compared with protein powder spectra during spectral compound analyses. Results: The nutrient profile determination resulted in a total decreasing amount of protein from 43.12% w/w for press 1 to 37.84% w/w for press 10. The TDF content increased from 22.80% w/w for press 1 to 47.47% w/w for press 10. ACH decreased from 5.43% w/w for press 1 to 1.10% w/w for press 10, while the amount of determined ash decreased from 8.24% w/w for press 1 to 2.70% w/w for press 10. Usable and promising NIR spectra were conducted from all measured protein and cellulose gluten powder samples. A calibration model predicting TDF contents for each of the 10 presses was developed with a wavenumber range from 6,800 cm-1 to 4,100 cm-1 and R2 = 0.98. For all 10 presses, the mean deviation from the reference TDF contents was 0.76% w/w. NIR spectra from the green juice and pellet could not be conducted with the available NIR instrument and presetting options. Conclusions: It is challenging to convert complex NIR spectra into usable information. Since a broad wavenumber spectrum was chosen for the model development, it was easy to fit the spectra to almost any kind of reference values, even though the spectra do not describe those reference values. It also has to be kept in mind that the model is not validated. Therefore it is hard to draw conclusions regarding the model quality. It can be concluded though, that NIR spectra obtained from the protein powder of alfalfa look promising for further investigation, since a good correlation between the TDF amounts and NIR spectra could be seen. Of future work the first priority should be to validate this produced model. If that looks promising, both a new independent validation set and a larger data set to produce a new calibration model is required to further test the model robustness.