The Incidence of Oscillatory Behavior in the Continuous Fermentation of Zymomonas mobilis P. James McLellan,* Andrew J. Daugulis, and Jinghong Li Department of Chemical Engineering, Queen’s University, Kingston, Ontario, Canada K7L 3N6 The incidence of oscillatory behavior in the continuous culture of Zymomonas mobilis has been examined using a combination of experimental investigations and a predictive model. The tendency to oscillatory behavior was assessed by perturbing the feed substrate concentration and dilution rate in a continuous fermentation starting from a number of distinct initial conditions. The entire range of qualitative dynamic behavior was observed: overdamped, underdamped, and sustained oscillatory responses. The predictive capabilities of a model previously proposed by our research group were confirmed over this range of operation. A key component of this model is the inclusion of a dynamic specific growth rate term which accounts for the inhibition associated with historical ethanol concentration change rate. Parameters and relationships estimated for the model were used to identify key characteristics leading to oscillatory behavior. In particular, differences in the sensitivities of ethanol production rate versus specific growth rate to ethanol concentration and its rate of change dictate whether sustained oscillations will occur. Experimental evidence indicates that a change in morphology is associated with oscillatory behavior. The change in morphology to a more filamentous form may explain the change in specific growth and product formation characteristics. Introduction Zymomonas mobilis has been promoted as a more promising microorganism than yeast for the industrial production of ethanol. However, one disadvantage as- sociated with continuous fermentation of Zymomonas mobilis is the incidence of oscillatory behavior in which biomass, product, and substrate cycle under certain fermentation conditions (e.g., Bruce et al., 1991; Ghom- midh et al., 1989; Lee et al., 1980; Lee et al., 1979). Ethanol productivity decreases for periods of time during oscillations, leading to high levels of residual substrate which are subsequently lost from production. Previous work in our laboratory (Li et al., 1995; Li, 1996) has determined that the ethanol change rate history, and not specifically ethanol concentration, is a major factor inhibiting cell-specific growth rate. The concept of a dynamic specific growth rate has been proposed in a previous paper (Daugulis et al., 1997) to explain the inhibition of the instantaneous specific growth rate due to ethanol concentration change rate history. The dynamic specific growth rate refers to a situation in which the cells exhibit a growth rate which is less than or equal to the growth rate which would otherwise be characterized by instantaneous culture conditions. On the basis of this concept, a dynamic model has been proposed to describe the transient behavior of biomass, ethanol, and substrate concentration (Daugulis et al., 1997). This model has been used to describe experimental results presented by other researchers for Zymomonas mobilis fermentations, as well as the results of forced oscillation experiments (Daugulis et al., 1997). There are two major approaches to the modeling problem. The first is the use of a wholly mechanistic model, in which the variables correspond directly to physical quantities. The major examples of this approach are the use of a viable/nonviable/dead cell compartment model (Gommidh et al., 1989; Jarzebski, 1992) and the use of a metabolic compartment model (Veeramallu and Agrawal, 1990; Jobses et al., 1985) representing groups of macromolecules. Researchers have also used the terms “structured” and “unstructured” to describe models that do or do not contain compartments associated with specific physiological entities. The difficulty posed by these models lies in measuring values for the compart- ment variables. If these variables cannot be measured directly, it is still possible to estimate values given a rich enough model structure; however it is more difficult to conclude definitively that the model structure is correct. Compartment models also require knowledge of the initial conditions of the compartment variables, in addi- tion to kinetic parameters. Thus, to apply such models to the prediction of dynamic behavior a priori, one must know these conditions. The second approach is a phenomenological one, in which functional relationships are used to describe the observed behavior. These functional relationships reflect a portion of the underlying physical structure, but not to the same extent as a mechanistic model. For example, in the case of ethanol inhibition of specific growth, we have used two differential equations to reflect the memory component of the organism with respect to ethanol concentration change rate. This is on the basis of previous experiments, which clearly identified the ethanol concentration change rate as a key factor in inhibition. The use of two differential equations reflects the observation that there is a certain lag at which * To whom correspondence should be addressed. Telephone: (613) 533-6343. Fax: (613) 533-6637. E-mail: mclellnj@ chee.queensu.ca. 667 Biotechnol. Prog. 1999, 15, 667-680 10.1021/bp990070d CCC: $18.00 © 1999 American Chemical Society and American Institute of Chemical Engineers Published on Web 07/01/1999