Fuzzy neural network for the control of high cell density cultivation of recombinant Escherichia coli

A five-layer fuzzy neural network (FNN) was developed for the control of fed-batch cultivation of recombinant Escherichia coli JM103 harboring plasmid pUR 2921. The FNN was believed to represent the membership functions of the fuzzy subsets and to implement fuzzy inference using previous experimental data. This FNN was then used for compensating the exponential feeding rate determined by the feedforward control element. The control system is therefore a feedforward-feedback type. The change in pH of the culture broth and the specific growth rate were used as the inputs to FNN to calculate the glucose feeding rate. A cell density of 84 g DWC/l in the fed-batch cultivation of the recombinant E. coli was obtained with this control strategy. Two different FNNs were then employed before and after induction to enhance plasmid-encoded β-galactosidase production. Before induction the specific growth rate was set as 0.31 h⁻¹, while it was changed to 0.1 h⁻¹ after induction. Compared to when only one FNN was used, the residual glucose concentration could be tightly controlled at an appropriate level by employing two FNNs, resulting in an increase in relative activity of β-galactosidase which was about four times greater. The present investigation demonstrates that a feedforward-feedback control strategy with FNN is a promising control strategy for the control of high cell density cultivation and high expression of a target gene in fed-batch cultivation of a recombinant strain.     

Evaluation of spectrofluorometry as a tool for estimation in fed-batch fermentations

Native culture fluorescence was investigated as an additional source of information for predicting biomass and glucose concentrations in a fed-batch fermentation of Alcaligenes eutrophus. Partial least squares (PLS) regression and a feed forward neural network (FFNN) coupled with principle component analysis (PCA) were each used to model the kinetics of the fermentation. Data from three fermentations was combined to form a training set for model calibration and data from a fourth fermentation was used as the testing set. The fluorescent soft-sensors were compared with a previously developed feed forward neural network soft-sensor model which used oxygen uptake rate (OUR), carbon dioxide evolution rate (CER), aeration rate, feed rate, and fermentor volume to estimate biomass and glucose concentrations. The best model performance for predicting both biomass and glucose concentrations was achieved using the native fluorescence-based models. Real data predictions of the biomass concentration in the testing set were obtained using both the PLS and FFNN PCA modeling utilizing fluorescence measurements plus the rate of change of the fluorescence measurements. Accurate predictions of the glucose concentration in the testing set were obtained using the FFNN PCA modeling technique utilizing the rate of change of the fluorescence measurements. Substrate exhaustion was indicated qualitatively by a first-order PLS model utilizing the rate of change of fluorescence measurements. These results indicate that native culture fluorescence shows promise for providing additional valuable information to enhance predictive modeling which cannot be extracted from other easily acquired measurements.     

Hybrid modeling approach to on-line estimation of yeast biomass concentration in industrial bioreactor

The industrial fed-batch yeast cultivation process has been divided into four different metabolic phases (adaptation, carbon limited, oxygen limited and maturation) by a neuro-fuzzy classification model that consists of 4 applied linguistic rules on 2 state variables: oxygen uptake rate and liquid volume. The membership functions have been automatically adapted by this fuzzy perceptron, i.e., by a supervised learning algorithm initialized by prior operator's knowledge. Process compartmentalization has made easier and more realistic a subsequent state estimation of the biomass concentration with separate artificial neural networks combined with balance equations. Static networks with local recurrent memory structures were used, and the inputs were standard cultivation state variables: respiratory quotient, molasses feed rate, ethanol concentration, etc. This hybrid approach is generally applicable to state estimation or prediction when different sources of process information and knowledge have to be integrated.     

Optimization of fed-batch culture of hybridoma cells using dynamic programming: single and multi feed cases

The optimization of fed-batch culture of hybridoma cells is accomplished on a mathematical model using dynamic programming. Optimal feed trajectories are found using a seventh order model for a single feed stream containing both glucose and glutamine and for two separate feed streams of glucose and glutamine. Compared to a constant feed rate, optimal trajectories can improve the final MAb concentration by 11 % for the single feed case and by 20% for the multifeed case. Higher MAb concentrations can be expected for fed-batch optimization with feed enriched in nutrients.     

Modeling of the pyruvate production with Escherichia coli: comparison of mechanistic and neural networks-based models

Three different models: the unstructured mechanistic black-box model, the input–output neural network-based model and the externally recurrent neural network model were used to describe the pyruvate production process from glucose and acetate using the genetically modified Escherichia coli YYC202 ldhA::Kan strain. The experimental data were used from the recently described batch and fed-batch experiments [ Zelić B, Study of the process development for Escherichia coli-based pyruvate production. PhD Thesis, University of Zagreb, Faculty of Chemical Engineering and Technology, Zagreb, Croatia, July 2003. (In English); Zelić et al. Bioproc Biosyst Eng 26:249–258 (2004); Zelić et al. Eng Life Sci 3:299–305 (2003); Zelić et al Biotechnol Bioeng 85:638–646 (2004)]. The neural networks were built out of the experimental data obtained in the fed-batch pyruvate production experiments with the constant glucose feed rate. The model validation was performed using the experimental results obtained from the batch and fed-batch pyruvate production experiments with the constant acetate feed rate. Dynamics of the substrate and product concentration changes was estimated using two neural network-based models for biomass and pyruvate. It was shown that neural networks could be used for the modeling of complex microbial fermentation processes, even in conditions in which mechanistic unstructured models cannot be applied.     

Fed-batch culture automated by uses of continuously measured cell concentration and culture volume

An automated control method of fed-batch culture in which the nutrient feed rate was determined from continuously measured cell concentration and culture broth volume was developed. Theoretical background was elucidated, from which it was found that the method is unique in that it controls specific substrate consumption rate of the microorganism. The method was experimentally applied to the fed-batch cultures of recombinant Escherichia coli HB101. It was observed that the specific substrate feed rate affects not only the specific growth rate but also the growth yield. If some conditions are satisfied, this type of automated fedbatch culture can be applied widely to any microbial systems and seems especially useful when the culture medium is composed of natural complex nutrient(s) because their concentrations are very difficult to detect and control.     

Demonstration of the Crabtree effect in Phaffia rhodozyma during continuous and fed-batch cultivation

By monitoring cell yield and fermentation products during fed-batch and continuous growth, Pfaffia rhodozyma was shown to exhibit the Crabtree effect. In fed-batch culture at feed concentrations of 27 and 55 g glucose/l there was good agreement between the observed biomass formation and that predicted by a mass balance model. At 125 g glucose/l in the feed, biomass formation was less than predicted and fermentation products such as ethanol and acetic acid accumulated in the culture medium. In continuous culture with a feed concentration of 10 g glucose/l, the Crabtree effect became apparent at a dilution rate of 0.1 h -1 . Aerobic fermentation did not occur provided the sugar substrate was maintained at a concentration of less than 0.5 g/l. Although the cell yield coefficient was reduced from 0.5 g/g to 0.16 g/g during aerobic fermentation, the carotenoid content of the cells was unaffected.     

Dynamic optimization of hybridoma growth in a fed-batch bioreactor

This study addressed the problem of maximizing cell mass and monoclonal antibody production from a fed-batch hybridoma cell culture. We hypothesized that inaccuracies in the process model limited the mathematical optimization. On the basis of shaker flask data, we established a simple phenomenological model with cell mass and lactate production as the controlled variables. We then formulated an optimal control algorithm, which calculated the process-model mismatch at each sampling time, updated the model parameters, and re-optimized the substrate concentrations dynamically throughout the time course of the batch. Manipulated variables were feed rates of glucose and glutamine. Dynamic parameter adjustment was done using a fuzzy logic technique, while a heuristic random optimizer (HRO) optimized the feed rates. The parameters selected for updating were specific growth rate and the yield coefficient of lactate from glucose. These were chosen by a sensitivity analysis. The cell mass produced using dynamic optimization was compared to the cell mass produced for an unoptimized case, and for a one-time optimization at the beginning of the batch. Substantial improvements in reactor productivity resulted from dynamic re-optimization and parameter adjustment. We demonstrated first that a single offline optimization of substrate concentration at the start of the batch significantly increased the yield of cell mass by 27% over an unoptimized fermentation. Periodic optimization online increased yield of cell mass per batch by 44% over the single offline optimization. Concomitantly, the yield of monoclonal antibody increased by 31% over the off-line optimization case. For batch and fed-batch processes, this appears to be a suitable arrangement to account for inaccuracies in process models. This suggests that implementation of advanced yet inexpensive techniques can improve performance of fed-batch reactors employed in hybridoma cell culture.     

Maximum production in a bakers' yeast fed-batch culture by a tubing method

A feedback control system of the glucose feed rate in a bakers' yeast fed-batch culture was developed by keeping the ethanol concentration constant. A PID controller and on–off controller were applied and discussed with the aid of the porous Teflon tubing method. Experimental results showed the effectiveness of the control system for avoiding the glucose effect and glucose starvation. It was shown that the feedback control system developed hare could achieve a maximum specific growth rate of 0.3 h^?1 or a maximum cell yield of 0.5 g cell/g glucose in the fedhyphen;batch culture.     

Fed-batch culture with a modified DO-stat method

To support a high growth rate of microorganism in fed-batch culture with high cell density, a modified DO-stat method was developed. In this method, an exponential substrate feed was coupled with the usual DO-stat method, i.e., a fixed amount of substrate per DO signal was exponentially fed to the culture based on the estimation of the substrate consumption rate and thereafter the feed was stopped in order to prevent the oversupply of substrate until an abrupt increase in the concentration of dissolved oxygen (DO) in the broth appeared. After that, the feed was started again and this cyclic operation was repeated throughout the cultivation. This method was applied to the fed-batch cultivation of ethanol utilizing yeast, Candida brassicae. At high cell densities (> 10 g/l), this modified method was more effective than the usual one in keeping a higher growth rate.     

A predictive and feedback control algorithm maintains a constant glucose concentration in fed-batch fermentations.

A combined predictive and feedback control algorithm based on measurements of the concentration of glucose on-line has been developed to control fed-batch fermentations of Escherichia coli. The predictive control algorithm was based on the on-line calculation of glucose demand by the culture and plotting a linear regression to the next datum point to obtain a predicted glucose demand. This provided a predictive "coarse" control for the glucose-based nutrient feed. A direct feedback control using a proportional controller, based on glucose measurements every 2 min, fine-tuned the feed rate. These combined control schemes were used to maintain glucose concentrations in fed-batch fermentations as tight as 0.49 +/- 0.04 g/liter during growth of E. coli to high cell densities.     

A predictive and feedback control algorithm maintains a constant glucose concentration in fed-batch fermentations

A combined predictive and feedback control algorithm based on measurements of the concentration of glucose on-line has been developed to control fed-batch fermentations of Escherichia coli. The predictive control algorithm was based on the on-line calculation of glucose demand by the culture and plotting a linear regression to the next datum point to obtain a predicted glucose demand. This provided a predictive "coarse" control for the glucose-based nutrient feed. A direct feedback control using a proportional controller, based on glucose measurements every 2 min, fine-tuned the feed rate. These combined control schemes were used to maintain glucose concentrations in fed-batch fermentations as tight as 0.49 +/- 0.04 g/liter during growth of E. coli to high cell densities.     

The control of glucose concentration during yeast fed-batch cultivation using a fast measurement complemented by an extended Kalman filter

In this contribution results are presented from the control of glucose during a yeast fed-batch cultivation. For glucose measurements a special flow injection analysis (FIA) system was employed, which uses a glucose oxidase solution instead of immobilized enzymes. To avoid the large delay time caused by probing systems samples containing cells, i.e., samples containing the ordinary culture broth, are injected into the FIA system. Based on a special evaluation method the glucose concentration can be measured with a delay time of about 60 s. Employing an extended Kalman filter, the biomass, the glucose concentration as well as the w_max (Monod model) are estimated. Based on the estimation a feed forward and a PI-control with a set point of 0.5 g/l was carried out. The mean deviation of the set point and the estimated value as well as the set point and the measured value were 0.05 and 0.11 g/l respectively for a control period of 8 h producing a cell dry mass of more than 6 g/l.     

基于差分进化算法的酿酒酵母分批补料培养在线自适应控制

酿酒酵母分批补料培养中,葡萄糖添加过量会导致乙醇大量积累,破坏细胞结构及功能,降低葡萄糖利用效率;葡萄糖添加不足会限制细胞生长。为解决这一矛盾,提出了一种基于差分进化算法的在线自适应控制策略,并利用计算机仿真方法对该策略、传统的间歇流加、分段恒速流加及PID控制策略的控制性能进行了研究和比较。结果表明,在该控制策略下,发酵液中的乙醇浓度能够被稳定地控制在1g/L的低水平,而细胞浓度却达到34.45g/L的高水平,比采用间歇流加、分段恒速流加及PID控制策略的批次分别提高了243%、18%和29%。由此可知,该自适应控制策略能够将葡萄糖流加速率控制在适宜水平,避免乙醇过量积累的同时保证细胞的快速增殖。     

Protein-free fed-batch culture of non-GS NS0 cell lines for production of recombinant antibodies

Presented is a novel antibody production platform based on the fed-batch culture of recombinant, NS0-derived cell lines. A standardized fed-batch cell culture process was developed for five non-GS NS0 cell lines using enriched and optimized protein-free, cholesterol-free, and chemically defined basal and feed media. The process performed reproducibly and scaled faithfully from the 2-L to the 100-L bioreactor scale achieving a volumetric productivity of > 120 mg/L per day. Fed-batch cultures for all five cell lines exhibited significant lactate consumption when the cells entered the stationary or death phase. Peak and final lactate concentrations were low relative to a previously developed fed-batch process (FBP). Such low lactate production and high lactate consumption rates were unanticipated considering the fed-batch culture basal medium has an unconventionally high initial glucose concentration of 15 g/L, and an overall glucose consumption in excess of 17 g/L. The potential of this process platform was further demonstrated through additional media optimization, which has resulted in a final antibody concentration of 2.64 +/- 0.19 g/L and volumetric productivity of > 200 mg/L per day in a 13-day FBP for one of the five production cell lines. Use of this standardized protein-free, cholesterol-free NS0 FBP platform enables consistency in development time and cost effectiveness for manufacturing of therapeutic antibodies.     

Run-to-run fed-batch optimization for protein production using recombinant Escherichia coli

A two-phase design approach is introduced to determine the optimal feed rate, fed glucose concentration and fermentation time to maximize protein productivity using recombinant Escherichia coli BL21 (pBAW2) strain. The first phase is applied to determine a primary S-system kinetic model using batch time-series data. Two runs were carried out in the second phase to achieve the maximum protein productivity for the fed-batch fermentation process. The computational results using the S-system kinetic model obtained from the second run are in better agreement with the experiments than those using the kinetic model obtained from batch time-series data. For cross-validation, two extra fed-batch experiments with different feed strategies were carried out for comparison with the optimal fed-batch result. From the experimental results, this approach could improve productivity by at least 3%.     

Optimising fed-batch production of recombinant proteins using the baculovirus expression vector system

Fed-batch culture can offer significant improvement in recombinant protein production compared to batch culture in the baculovirus expression vector system (BEVS), as shown by Nguyen et al. (1993) and Bedard et al. (1994) among others. However, a thorough analysis of fed-batch culture to determine its limits in improving recombinant protein production over batch culture has yet to be performed. In this work, this issue is addressed by the optimisation of single-addition fed-batch culture. This type of fed-batch culture involves the manual addition of a multi-component nutrient feed to batch culture before infection with the baculovirus. The nutrient feed consists of yeastolate ultrafiltrate, lipids, amino acids, vitamins, trace elements, and glucose, which were added to batch cultures of Spodoptera frugiperda (Sf9) cells before infection with a recombinant Autographa californica nuclear polyhedrosis virus (AcNPV) expressing beta-galactosidase (beta-Gal). The fed-batch production of beta-Gal was optimised using response surface methods (RSM). The optimisation was performed in two stages, starting with a screening procedure to determine the most important variables and ending with a central-composite experiment to obtain a response surface model of volumetric beta-Gal production. The predicted optimum volumetric yield of beta-Gal in fed-batch culture was 2.4-fold that of the best yields in batch culture. This result was confirmed by a statistical analysis of the best fed-batch and batch data (with average beta-Gal yields of 1.2 and 0.5 g/L, respectively) obtained from this laboratory. The response surface model generated can be used to design a more economical fed-batch operation, in which nutrient feed volumes are minimised while maintaining acceptable improvements in beta-Gal yield.     

Dissolved-oxygen-stat controlling two variables for methanol induction of rGuamerin in Pichia pastoris and its application to repeated fed-batch

A DO-stat control strategy for two variables was introduced to the rGuamerin production process in Pichia pastoris and applied to repeated fed-batch culture. Two interrelated variables, namely the ratio of partial pressure of pure O2 in the inlet air-stream and the methanol feed rate, were controlled simultaneously. By using this control strategy, methanol feeding for induction could be controlled automatically while efficiently controlling the dissolved oxygen level. As a result, the cell concentration reached more than 140 g l(-1) and rGuamerin expression level 450 iu l(-1). rGuamerin was secreted into the culture medium and reached a level that was 40% higher than achieved in a fed-batch process using manual control of the methanol feeding rate. Repeated rGuamerin induction was achieved by repeating the methanol feeding and withdrawing the culture broth during extended production. During more than 250 h of culture, expression of rGuamerin was maintained at an average of about 430 iu l(-1 )(473 mg l(-1)), without causing the cell density to decrease. In addition to the rGuamerin production process, the proposed control system might be applied to cultivation of other methylotrophic yeasts in the production of therapeutic proteins.     

Studies on the growth, modelling and pigment production by the yeast Phaffia rhodozyma during fed-batch cultivation

Summary As Phaffia rhodozyma is a Crabtree positive yeast, its cell yield and pigment production are reduced at high sugar concentrations. A method for maintaining low growth medium sugar concentrations is fed-batch culture. Using a mass balance approach and Monod growth kinetics a model is presented which describes the fed-batch culture of Phaffia rhodozyma and enables the calculation of a feed regime to obtain the maximum yield of cells and pigment. Although developed on a glucose medium, the model was also applied successfully to a molasses-based medium.     

Effect of feeding strategy on Zymomonas mobilis CP4 fed-batch fermentations and mathematical modeling of the system

In this work, the effect of the feeding strategy in Zymomonas mobilis CP4 fed-batch fermentations on the final biomass and ethanol concentrations was studied. Highest glucose yields to biomass (0.018 g/g) and to ethanol (0.188 g/g) were obtained in fed-batch fermentations carried out using different feeding rates with a glucose concentration in the feed equal to 100 g/l. Lower values (0.0102 g biomass/g glucose and 0.085 g ethanol/g glucose) were obtained when glucose accumulated to levels higher than 60 g/l. On the other hand, the highest biomass (5 g/l) and ethanol (39 g/l) concentrations were obtained using a glucose concentration in the feed equal to 220 g/l and exponentially varied feeding rates. Experimental data were used to validate the mathematical model of the system. The prediction errors of the model are 0.39, 14.36 and 3.24 g/l for the biomass, glucose and ethanol concentrations, respectively. Due to the complex relationship for describing the specific growth rate, a fed-batch culture in which glucose concentration is constant would not optimize the process. Received: 30 November 1999 / Received revision: 24 March 2000 / Accepted: 7 April 2000