I would like to express my sincere gratitude to my external mentor from IIT Delhi Prof. I would also like to thank the Central Instrumentation Center, the Department of Biosciences and Bioengineering, the Center for the Environment and the Department of Chemical Engineering for their support. - analytical objects of art.
ABSTRACT
The final phase of the study focuses on the development of various optimization case studies to achieve increased productivity. The objective of the case study (1) focused on maximizing the total biomass in the reactor at a minimum broth volume.
NOMENCLATURE
146 6.3 Modeling and Validation of Fermenter Batch and Feed Process 148 6.4 Optimization Studies to Maximize Biomass Production and Harvest Time Prediction 149 .
Experimental and Cole-Cole model predicted capacity profile for four different time intervals of fermentation namely a (batch), b (fed batch), c (induction) and d (harvest). The two marked areas (gray and gold quadrangle) represent two important areas for operating the reactor at the desired volume. Figure 5.12 Pareto front for the two targets f(1) and f(2) obtained from Pareto search algorithm for reactor volume of 10 L..142.
LIST OF TABLES
Introduction
- Overview of Biotherapeutics
Process optimization would be especially crucial in the therapeutic manufacturing process, as millions of lives depend on the timely delivery of quality products. Autonomous bioprocessing using artificial intelligence could effectively address the bottlenecks to ensure product quality and speed up the manufacturing process, and the biopharmaceutical industry is moving towards the same in the coming years.
1.1.2 Production methods: Continuous bioprocessing
- Significance of Process Analytical Technology implementation
The application of various PAT tools can provide us with insight into the physical, chemical and biological attributes of the process. Important steps involved in implementing PAT include: (i) Identify: design experiments to identify CQAs and relevant CPPs that regulate the same, (ii) Monitor: deploy appropriate analytical tools to monitor the process, ( iii) Analysis: statistical analysis of monitored CQA and to understand its relationship with product efficiency, and (iv) Control: initiation of appropriate control schemes to ensure process quality (Streefland et al., 2013 ).
1.1.3.1 Model development in bioprocesses
- Optimization and advanced process control
For any biotherapeutic manufacturing application, the effectiveness of the process is determined by the quality of the product. In addition, the advanced controller could effectively manage the interaction between multiple process inputs and outputs (MIMO).
- Scope and motivation of the research
In addition, the incorporation of the real-time process measurements into the developed process model to enable subsequent implementation of real-time optimization and control strategies is essential. Product concentration and productivity of a biological substance are important for commercialization, and thus optimization of the total biomass in a reactor plays a significant role (Kaiser et al., 2008; Rathore, 2016).
1.3 Overview of optimization and advanced process control strategy
- Objectives
Online optimization and advanced process control strategies are implemented as a monitoring layer that provides the set points for the regulatory layer (Proportional-Integral-Derivative (PID) control), which further keeps the underlying process in control by providing appropriate manipulating variables (MVs). ). Overview of implementation of advanced process control strategies such as online optimization and supervisory control (model predictive control (MPC)) in bioprocesses.
1. Application of dielectric spectroscopy for real-time monitoring of biotherapeutic
- Organization of the thesis
Furthermore, this chapter presents the implementation of a robust methodology using the Cole-Cole model for the real-time estimation of biomass physiological properties. Thus, this chapter summarizes the implementation of the various case studies for optimization, which improves the production of the therapeutic protein considered in this study.
Chapter 6 presents the overall summary and highlights the conclusions from each of
Review of literature
2.2 Dielectric spectroscopy as a PAT tool for real-time monitoring
- Dielectric spectroscopy: Principle and applications
In addition, it could also be estimated using mechanical models or soft sensors such as exhaust gas analysis (Reichelt et al., 2016; Wechselberger et al., 2013). Thus, the total capacitance of the suspension is measured to estimate the total viable biomass concentration (Markx and Davey, 1999; Yardley et al., 2000).
Schematic representation of capacitance-based monitoring in bioprocesses The response of the material subjected to the applied electric field can be described by conductance (G) and capacitance (C), which are converted to conductivity [σ (S m-1) ] and relative permittivity [ε (dimensionless)], respectively, using the probe constant. With the increase in the frequency of the electric field, the permittivity of the cell suspension decreases while the conductivity increases in a step-like shift called dispersions, which occurs due to the loss of the cells' polarization capabilities and increasing permeability of the plasma membrane (Davey et al., 1992).
- Models for estimation of physiological properties from capacitance measurements
In the two-frequency mode, Δε is measured at two different frequencies, i.e. one at a lower frequency plateau and the other at a higher dispersion frequency β. In addition to dual-frequency scanning, another method of measuring capacitance is frequency scanning, where the capacitance is measured over a range of electrical frequencies, typically in the 0.1–10 MHz range, where β dispersion occurs.
- Cole-Cole modeling
The parameters in the Cole-Cole model can be used to describe the morphological characteristics of the cells. The ion content of the media (which affects the conductivity of the suspension medium σe) and the internal cytoplasmic conductivity (σi) are considered to be influenced by the measurement frequency as described in equation (2.6).
Table 2.1. Application of dielectric spectroscopy for process monitoring and control
Kaiser et al. (2008) used a dual-frequency method to measure live cell density in recombinant protein processes by growing cells at high density E. Cole et al. (2015) investigated the combined use of PAT, DS, and biocalorimetry tools to monitor immobilized Chinese hamster ovary (CHO) cells.
Combining online measurement with data reconciliation enables accurate estimation of specific growth rate and implementation of a simple robust control strategy. Predicted viable biomass concentration for two phases (growth and decline) was used to develop control strategies that adapt to changing biomass yields.
2.2.4 Gaps and challenges
- Model development in bioprocesses
- Modeling in upstream processes
The success of the PAT tool depends on the accuracy of the process model that can interpret the measured variables. Model development from a bioprocess perspective is a method to develop and validate a mathematical representation of a process.
- Data pre-processing
- Sensitivity analysis and model validation
Some of the commonly used pre-processing techniques include the moving average (MA) filter applied for the pre-processing of DS data (Horta et al., 2012). Soft sensors enable real-time estimation of the process variables for online optimization and implementation of process control strategies, thus contributing to bioprocess automation (Sonnleitner, 2012).
2.3.3 Gaps and challenges
- Optimization studies and control strategy development for overall process optimization
- Optimization studies for different applications
Since model-based optimization studies are built on the valid process model, the success of the method relies on the reliable representation of the system (von Stosch et al., 2016b). This step is followed by optimization of the underlying process once the process model is validated (Proß and Bachmann, 2012).
In case of multiple objectives that might also be conflicting, multiobjective optimization
- Conventional control strategies for fermentation
- Open-loop control
Lee et al. (1999) reviewed the use of various control strategies for batch feeding over the past decade. Ehgartner et al. (2017) investigated the application of closed-loop control to filamentous fungi using a specific growth rate based on an estimate of viable biomass from DS, where the feed rate F was determined based on the PID output.
- Advanced process control strategies
- Adaptive control
- Model predictive control
According to Jenzsch et al., (2006b), Artificial Neural Networks (ANN) can estimate biomass concentration (X) when several data records are available. Jenzsch et al., (2006d) developed a general control model where the process dynamics were described by a simple mechanical model and a nonlinear state estimator (EKF), and it proved to be an important approach.
Table 2.5. Application of various control strategies
- DO setpoint
- Gaps and challenges
- Summary of the state of the art
Some of the challenges in implementing advanced control strategies can be briefly described below. The use of advanced control strategies such as MPC can handle the dynamic changes and interactions between different input and output variables of the fermenter.
Application of dielectric spectroscopy for real-time
Problem statement
Real-time monitoring and estimation of the physiological variable, biomass concentration, is significant for establishing control strategies based on this variable. Therefore, establishing reliable monitoring and obtaining real-time estimation of this physiological variable is essential.
- Materials and Methods
- Reactor configuration and experimental setup
- Strain and reactor operating conditions
The cloning host (E.coli DH5α) and expression host (E. coli BL21 DE3) were obtained from Promega corporation (Madison, WI, USA). The schematic view of the 2:2 clone of ranibizumab is presented in Figure 3.1 (adapted from (Priyanka and Rathore, 2021)).
The seed culture grown overnight was prepared using the working cell bank (glycerol
- Offline biomass measurements
- Substrate concentration measurement (glucose and glycerol)
- Therapeutic protein measurement
Furthermore, the speed of the stirrer was chosen accordingly so that the interference of the electric field was minimized. Quantification of the recombinant protein (Ranibizumab) was performed using reverse phase gradient HPLC (RP-HPLC) with a Zorbax 300 SB C8 column.
- Data pre-processing and noise removal
- Cell size and viability studies
- Cell size studies: Particle size analysis (quantitative) and Field Emission Scanning Electron Microscope (FESEM) (qualitative)
For the validation of the preprocessing methods, RMSE was applied as one of the statistical measures and was considered sufficient for the comparison of the methods. Cell size and viability studies were performed to validate the physiological properties estimated by the Cole-Cole model in this study.
- Cell viability studies: Flow cytometry (quantitative) and Fluorescence microscopy (qualitative)
- Model development for real-time biomass estimation .1 Linear modeling
- Cole-Cole modeling
The stained samples were centrifuged and washed with saline after the incubation before analysis. The samples were analyzed using a CytoFLEX flow cytometer (Beckman coulter, USA) using CytoFLEX solution as the sheath fluid.
- Results and discussion
- Data pre-processing and noise removal
Noise in the raw capacitance data was reduced by applying various pre-processing algorithms.
SS were applied, and the comparison of the aforementioned methods is presented in Table
- Capacitance measurements from recombinant E. coli
- Real-time biomass estimation from dual-frequency capacitance measurements
The preprocessed capacitance profile shown in Figure 3.6 indicates physiological changes between the different phases, including batch, fed-batch and induction (Ehgartner et al., 2015). As shown in Figure 3.7 (B), the linear relationship between the pretreated capacitance and DCW is observed.
3.5.3.2 Capacitance measurements using frequency scanning
- Offline analysis
The offline analysis for the biomass (DCW), substrate (batch phase substrate - Sb, fed- batch phase substrate - Sf) and product concentration (P) is presented in Figure 3.9. Offline analysis for biomass (DCW, filled circle), substrate (Sb, empty circle, Sf, filled downward pointing triangle) and product (P, empty downward pointing triangle) concentration.
3.5.5 Cell size and viability studies
- Cell size analysis
- Cell viability analysis
Cell viability significantly decreased after induction, as reflected in the viability values of samples c and d in Figure 3.11 (A, B). A decrease in viability was also observed by fluorescence microscopy with PI staining from Figure 3.11 (C).
Table 3.2 Linear correlation model for capacitance and offline biomass concentration
- Cole-Cole modelling
- Comparison of the deterministic (LMA) and stochastic (GA) method for solving the non–linear equations of the Cole-Cole model
- Cole–Cole model predictions
- Real-time estimation of physiological properties using Cole-Cole model The overall procedure for real-time estimation of physiological properties using the Cole-
Obtaining reliable parameter values from the nonlinear least-squares fitting of the Cole-Cole model is essential for estimating the physiological properties of the cell. Comparison of experimental and model-predicted values of physiological properties of recombinant E.
3.7 Summary
Modelling and validation of
Problem background
In this chapter, a mechanistic model for the representation of the fed-batch fermentation process of a biotherapeutic protein production from recombinant E. Furthermore, the biomass concentration estimated from the capacitance measurements was integrated with the developed model for the model calibration and validation.
- Methodology
- Experimental studies
- Model development for biotherapeutic production from E. coli
The substrate mass balance for Sb was active during the batch phase (0 - tb), and the mass balance for Sf was active during the fed-batch phase (tb - tf), where Sb and Sf. The mass balance for biomass was defined based on the assumption that the substrate was used only for biomass growth and maintenance.
- Model calibration and validation .1 Parameter sensitivity analysis
- Parameter estimation and model validation
- Results and discussion
- Model calibration and validation .1 Parameter sensitivity analysis
- Summary
The data flow between the model calibration tool and model equations and the overall procedure for parameter estimation is presented in Figure 4.2. The model predicted the experimental values of the validation data set with an average error value of 14.97%.
Optimization studies for
Problem statement
The implementation of optimization studies aimed at the desired objectives can help increase productivity and thereby improve process performance. The results of optimization studies can result in improved biomass growth and efficient metabolism by directing the feed collection process to work.
- Process model and experimental data
- Optimization studies
The model equations were developed separately for the batch (0 – tb) and batch (tb – tf) phases as discussed in Section 4.3.2. Optimization studies were thus demonstrated for a validated process model obtained from experimental studies of batch cultivation of recombinant E.
- Case study (1): Optimizing the total biomass at a minimum broth volume In order to achieve maximum biomass with a minimum broth volume, the validated
- Simulation studies
- Formulation of the objective function
The total biomass (XV, g) was obtained by multiplying the biomass concentration X by the corresponding slurry volume V (L) at time tf. Decision variable bounds and constraints were chosen based on the sensitivity of the decision variable to the objective function through simulation studies and experimental constraints.
The objectives f(1) and f(2) were observed to be of conflicting nature; that is, when the
The GA is expected to give a global optimal solution for nonlinear problems, but SQP can converge to a local minimum in the case of multiple minima. However, GA is expected to consume more computational time and may not be suitable for real-time optimization applications.
- Case studies for testing the effect of fault in actuators
- Formulation of the objective function
Similar to case study (1), we solved the MOO using the epsilon method for different values of λ from 0-1. A time interval of 10 minutes was taken forward for case study (2) based on the best interval obtained from the simulation studies of case study (1).
- Case study (1): Optimizing total biomass at a minimum broth volume .1 Simulation studies
The result of simulation studies with different constant substrate flow is shown in Figure 5.4. Therefore, it was concluded that a multiobjective optimization was necessary to address the conflicting nature of the two objectives and simultaneously achieve maximum total biomass and minimum broth volume.
Among equal time interval cases, minimum f was achieved in cases starting from 10 min
- Multiobjective optimization for enhanced production of therapeutic proteins in E. coli
Furthermore, based on the outputs of the downstream processes, the operator can choose any point in the asymptotic region. Consequently, any of the algorithms can be applied for solving the MOO developed in this study.
As shown in Figure 5.6 (B), the obtained optimal feeding profile initially has a spike at
- Case studies for testing the effect of fault in actuators
- Sample procedure to select optimal point for given cost factors
As previously mentioned, the two objective functions, total biomass [f(1)] and broth volume [f(2)], can be converted to a single objective function by including relevant cost factors, and this approach is explored in this section with the following assumptions. The cost of the therapeutic protein product (C-product) is in $/g product, where it is assumed that the final yield (g) of therapeutic protein is directly proportional to the total biomass with ϕ as the proportionality factor.
For different values of broth volume, respective total biomass (XV) and subsequent
- Multiobjective optimization with different fed-batch harvest time
- Pareto Front for two objectives f(1) and f(2)
However, it was observed that at a trend value of 30 hours, the Pareto points were less, indicating that the objective function may not improve significantly after this harvest time. The two regions correspond to the asymptotic value of the Pareto functions observed at 0.995 and 0.999 for different trend values.
- Summary
The total biomass obtained from the optimal substrate feeding profile was 20.6% higher than the total biomass from the experimental constant substrate feeding profile implemented in this study. It can be deduced from the case study (1) that the maximum total biomass can be obtained at a desired minimal broth volume by choosing the appropriate substrate nutrition profile based on the Pareto front.
Conclusions and Future recommendations
Application of dielectric spectroscopy for real-time monitoring of biotherapeutic protein production
Real-time estimation of the physiological properties of the organism was accomplished by correlating the DS measurements using a non-linear Cole-Cole model. A robust methodology for real-time estimation of the physiological properties was proposed as the outcome of this study.
Therefore, it could be inferred that MA filtering eliminates the signal noise
- Modelling and validation of batch and fed-batch process of the fermenter The second objective explored in this thesis focused on the development of a mechanistic
A mechanistic model representing the two stages of the fermentation process was developed based on the mass balance for the state variables. Thus, the importance of applying the Monod-type dual substrate kinetic model for the description of the biomass profile was observed by the sensitivity analysis.
4. The model predicted the experimental values of the validation dataset with an
- Optimization studies for maximizing biomass production and predicting harvest time
- Recommendations for the future
Likewise, the validation studies to harvest the fed-batch crop at a suitable optimal harvest time can be performed to verify the practical feasibility of the optimal results achieved. The proposed methodology for real-time estimation of biomass can be used to measure the critical process variable, and.
Soft sensor control of metabolic fluxes in a recombinant Escherichia coli fed-batch culture producing green fluorescent protein. Model-based estimation and optimal control of fed-batch fermentation processes for the production of antibiotics.
