@phdthesis{3007071,
  abstract     = {{Sub-acute ruminal acidosis (SARA) is a well-recognized digestive disorder that is an increasing health problem in dairy herds. Dairy cows experiencing SARA often do not exhibit any clear, overt clinical symptoms and symptoms are often delayed from the time of incidence. The only reliable and accurate diagnostic test for SARA is measuring ruminal fluid pH, which is based on invasive techniques. Rumen pH parameters regularly used to diagnose SARA are time pH < 5.6 or time pH < 5.8. High-concentrate diets can lead to SARA and are known to result in changes of the ruminal fermentation pattern and mammary secretion of fatty acids. However, relationships between milk fatty acids (including odd and branched chain fatty acids (OBCFA) and C18 biohydrogenation intermediates) and rumen pH have not yet been determined in experiments designed to induce SARA. As such, milk fatty acids were proposed in this PhD thesis as a new noninvasive diagnostic tool of SARA. 
The objectives of this PhD thesis were 1) to establish relationships between commonly used SARA indicators and milk fatty acids for 6 acidosis induction experiments in dairy cows in which rumen function was challenged by either increasing amounts of QFCH or decreasing physically effective NDF, 2) to present the logistic curve as a new method to describe the daily rumen pH pattern and to estimate the relationship between milk OBCFA and C18 biohydrogenation intermediates and parameters describing these logistic rumen pH curves, 3) to explore the potential to classify acidotic cases based on OBCFA and C18 biohydrogenation models through linear discriminant analysis, 4) to determine whether a non-linear approach is required to diagnose SARA and predict rumen pH parameters from milk FA patterns and whether milk OBCFA and C18 biohydrogenation intermediates are sufficient in diagnosing SARA and predicting rumen pH parameters. 
In Chapter 1, a literature review on the causes, consequences and diagnostic tools of SARA was given. Milk fatty acids and their origin were introduced and current state of the art related to their link with SARA was specified. In Chapter 2, six acidosis induction experiments in dairy cows were presented. In all experiments, rumen pH was measured continuously by an indwelling pH probe. Milk samples of the evening and the morning after the pH registration day were sampled and pooled for further analysis in each experiment. The acidosis induction protocols used in the experiments consisted either of increasing the amount of starch and/or decreasing the physically effective fiber content in the ration of dairy cows. In three experiments, different treatments were applied either to avoid the onset of acidosis (simultaneous application of both the acidosis induction protocol as well as the treatment) or to treat acidosis (treatment was applied after the acidosis induction). Three to twelve animals were monitored per experiment and this led to a global dataset of 277 to 331 control observations and 111 to 165 acidotic observations, depending on the acidosis threshold value applied. The acidosis induction experiments lasted six to nine weeks. 
Parameters commonly used for acidosis diagnosis do not describe diurnal pH variation. Hence, in Chapter 3, the logistic curve was developed as a model to describe the complete daily rumen pH pattern. Based on this logistic curve, two new parameters were introduced related to average rumen pH (β1) or to the rumen pH range (β0). Differences in the milk fatty acid profile of observations with either a low or high average rumen pH were observed and agreed with previously reported milk fatty acid changes, e.g. decreasing iso-branched chain fatty acids and increasing C18:1 trans 10 concentrations as rumen pH declines. Moreover, observations with either a limited or a broad rumen pH range also differed in milk fatty acid profiles. However, milk fatty acids depending on rumen pH range were not always similar across experiments. Moreover, certain milk fatty acid changes were dependent on both average rumen pH as well as rumen pH range. 
Exploration of potential for milk fatty acids to classify acidotic cases in dairy cows was based on linear discriminant analysis. A short introduction on linear discriminant analysis and some evaluation measures were given in Chapter 4A. Results on the linear discriminant analysis for all experiments were described in Chapter 4B. Three different threshold values of SARA were applied based on criteria described in literature: time rumen pH < 5.6 = 180 min/d; time rumen pH < 5.6 = 283 min/d and time rumen pH < 5.8 = 475 min/d. Although the main discriminating milk fatty acids differed across experiments and across threshold values, iso-branched chain fatty acids and C18 biohydrogenation intermediates (C18:1 trans 10 and C18:2 cis 9 trans 11) were selected in ten of the twelve discriminant models. The selected milk fatty acids were correlated with both average rumen pH and/or rumen pH range. 
In Chapter 5, support vector machines (SVM) were applied in order to diagnose SARA in dairy cows and predict rumen pH parameters (time pH < 5.6; time pH < 5.8; β1; β0). Due to analytical limitations and costs related to analyzing the milk fatty acid profile, the use of only a limited number of milk fatty acids, e.g. odd and branched chain and C18 biohydrogenation intermediates, instead of the whole milk fatty acid profile for diagnosing SARA and predicting rumen pH parameters was assessed. Hence, the features of the support vector machines were either all milk fatty acids or milk OBCFA and C18 biohydrogenation intermediates. Both a linear as well as a non-linear (radial-basis) kernel were used for both classification and regression models. SVM classification models performed well with an area under the ROC curve of above 0.7. The linear models with odd and branched chain fatty acids and C18 biohydrogenation as features performed as good as the radial models with all fatty acids as features or differences were minor. Hence, in case of classification, the least time consuming model can be applied. In case of the SVM regression models, the radial model with all milk fatty acids as features performed the best compared to the other regression models, e.g. linear model with all fatty acids as features and the linear and radial model with OBCFA and C18 biohydrogenation intermediates as features. 
The classification and regression models also performed well when combining all experiments and the training set proportionally consisted of data from all experiments. If the training set consisted only of data of three datasets and the model was tested based on the remaining dataset, performance of the models dramatically decreased. As such, the experiment effect is not negligible and should be taken into account in further exploration of prediction models for subacute acidosis in dairy cows. 
To conclude, milk fatty acids are related to changes in both ruminal pH as well as in the microbial population. Milk fatty acids hold potential as biomarkers of subacute ruminal acidosis in dairy cows. However, effects of both cow as well as experiments did influence the predictive models and these effects should be taken into account when new acidosis induction experiments are implemented and new models are evaluated.}},
  author       = {{Colman, Ellen}},
  isbn         = {{9789059895539}},
  keywords     = {{rumen pH,modelling,dairy cows,Milk fatty acids,subacute ruminal acidosis}},
  language     = {{eng}},
  pages        = {{V, 278}},
  publisher    = {{Ghent University. Faculty of Bioscience Engineering}},
  school       = {{Ghent University}},
  title        = {{Milk fatty acids as biomarkers of subacute ruminal acidosis in dairy cows}},
  year         = {{2012}},
}