A novel method for the sequential experimental design to optimize fed-batch fermentations was applied to a hyaluronidase fermentation by Streptococcus agalactiae. The method employees hybrid models that contain mechanistic, fuzzy and neural network components.
This method is applied to the Hyaluronidase fermentation. The product formation rate was found to be dependent on both growth rate µ as well as the transient rate
c=-dµ/dt.
Streptococcus agalactiae was cultivated discontinuously in a 5-litre fermenter (stirred tank reactor, 34 grdC, pH 7, 300 rpm) without aeration. 12 (=n) fermentation runs were performed with variated complex media consisting of casein-peptone (Fluka resp. Merck), yeast extract (Difco resp. Ohly), Glucose and mineral salts at different concentrations. Glucose (500g/l) was fed in fixed relation to NaOH (20%) that was used for pH control.
For the sequential experimental design a modified E-criterion was formulated that combines optimal parameter identification (using Fisher's information matrix) and optimal productivity.
The specific Hyaluronidase formation rate qp was found to be proportional to the specific growth rate µ under certain conditions:
- The specific growth rate µ must be greater than 0.1 1/h.
- The transient rate dµ/dt have to be greater than a certain, unknown value c.
To identify this critical parameter c a fed batch experiment (n+1=13) was designed using a hybrid (fuzzy and mechanistic) dynamic model which was constructed by data analysis. The data analysis of this 13th experiment with feed of Glutamin results in a modification of the hybrid model. The modified model was used for new experimental design.
The developed concept of sequential experimental design matches a typical situation of bioprocess development: The knowledge is fuzzy and changing step by step. The experiment have to be designed for both, the identification of the structure and parameters of a hybrid dynamic model as well as optimization of the productivity. Thus, instead of the established theory of D-optimal experimental design(1) an modified E-criterion was developed and the approach was modified for hybrid (fuzzy) models.