Mechanistic modelling of in vitro fermentation and methane production by rumen microbiota

Abstract Existing mechanistic models of the rumen ecosystem have proven to be useful to better understand and represent rumen fermentation. Opportunities for improving rumen fermentation models include a better representation of the microbiota, hydrogen dynamics and a mechanistic description of pH. The objective of this work was to include such aspects in the development of a mathematical model of rumen fermentation under in vitro conditions. The developed model integrates microbial metabolism, acid–base reactions and liquid–gas transfer. Model construction was based on an aggregated representation of the hydrolysis of carbohydrates and proteins, and the further fermentation of soluble monomers. The model is a differential algebraic equation model with 18 compartments. One of the main contributions of the model developed here resides in the mechanistic description of pH, the use of biochemical reactions and partition rules to define the stoichiometry of fermentation, the representation of hydrogen metabolism, and the representation of the rumen microbiota into functional groups associated with the utilization of hexoses, amino acids and hydrogen. The model was calibrated with published data from a 2 × 2 factorial experiment devoted to assessing the relative importance of the type of inoculum and substrate on the fermentation pattern. The treatments were the level of concentrate in the substrate (low concentrate vs. high concentrate), and the inocula type (obtained from goats fed at low or high concentrate). The model was implemented in Matlab. The code is available on request for academic purposes. Model evaluation was performed by regression analysis and the calculation of statistical indicators using the model predicted values and observed values. The model was capable to represent in a satisfactory fashion the dynamics of the fermentation, that is the pH, the individual volatile fatty acids and the gas compounds, namely methane, hydrogen and carbon dioxide. The model predictions exhibited high concordance correlation coefficients (CCC). For the pH and the CH 4 , the CCC was of 0.91 and 0.93 respectively. For the other variables CCC >0.96. The model developed was instrumental to quantify the differences of the fermentation pattern between the treatment combinations. These differences were mainly captured by parameters related to the flux distribution and were found to be dependent mainly on the type of inoculum. For instance, the flux towards butyrate production from sugars utilization for the microbiota of the inoculum adapted to high concentrate was about 30% higher than that for the inoculum adapted to low concentrate. This result, however, requires further validation with new data. Further developments are needed to incorporate physiological in vivo factors into our model. Nevertheless, the structure developed here appears to be a promising approach for enhancing the mechanistic description of the rumen microbial ecosystem.