WO2025062024A1 - Control of cell production in a bioreactor - Google Patents

Abstract

The invention relates to a method, implemented by a control system, for controlling cell production in a bioreactor. The bioreactor comprises a cell culture vessel containing a culture medium and microcarriers, on the surface of which the cells grow. The method comprises dynamic adjustment of parameters during production. The dynamically-adjusted parameters include at least a temperature of the culture medium, a pH of the culture medium, a concentration of oxygen dissolved in the culture medium, and a quantity of microcarriers present in the vessel. The method improves the quality and quantity of the biomass produced in the bioreactor.

Description

Description

Title of the invention: Control of cell production in a bioreactor

Technical field

[0001] The present disclosure relates to a method implemented by a control system for controlling cell production in a bioreactor, a computer program for implementing such a method, a control system for such a bioreactor and a bioreactor incorporating such a control system.

Technical background

[0002] One of the challenges of the major national bioproduction challenge defined in 2021 by the government is the control of the quantity and quality of the biomass produced. However, the initial cultivation conditions between production batches are never identical. They can cause significant variability in output and rejection of the products obtained whose characteristics and quality are no longer within the expected specifications.

[0003] Everyone now agrees that the Quality-by-Testing approach is unsuitable for the rapid and efficient development of innovative pharmaceutical products. The emergence in the mid-2000s of good pharmaceutical QbD (Quality by Design) practice (ICH Q8-Q12 international guidelines) made it possible to introduce predictive risk analysis as early as possible in the development chain (references Bastogne, 2017, Bastogne, 2022, Bastogne et al., 2022 and Bevers et al., 2022 from the list provided below). Other technological and digital advances, grouped under the term Process Analytical Technology, have brought the means to the field to monitor certain critical variables in real time during product manufacturing. In addition, there are control strategies, also defined in the ICH Q8(R2) guideline. Based on the measurements provided by analytical technologies, they allow action on the process to ensure that the product complies with the expected specifications. There are at least three levels of sophistication for these control strategies. The most advanced level allows for automatic process control and fits naturally into the broader framework of the Industry 4.0 paradigm. For the automation community, there is nothing really new on the horizon except for data (spectra, images and/or genomics) that are increasingly available in real time (online), which allows now the implementation of more sophisticated regulators such as those based on predictive control in order to better control quality and productivity objectives and to correct certain variables during cultivation if necessary. The first studies, published under the name "Quality by Control", were applied to bioproduction between 2015 and 2020. However, these existing solutions for production control in a bioreactor are not sufficient.

[0004] List of references:

- Bastogne, T., 2022. iQbD: a TRL-indexed Quality-by-Design Paradigm for Medical Device Engineering. Transactions of the ASME, Journal of Medical Devices 16.

- Bastogne, T., 2017. Quality-by-design of nanopharmaceuticals-a state of the art. Nanomedicine: Nanotechnology, Biology and Medicine 13, 2151-2157.

- Bastogne, T., 1997. Identification of multivariable systems by subspace methods. Application to a belt drive system (Doctoral thesis). Henri Poincaré University, Nancy 1.

- Bastogne, T., Caputo, F., Prina-Mello, A., Borgos, S., Barberi-Heyob, M., 2022. A State of the Art in Analytical Quality-by-Design and Perspectives in Characterization of Nano-enabled Medicinal Products. In revision in Journal of Pharmaceutical and Biomedical Analysis.

- Bevers, S., Kooijmans, S.A.A., Van de Velde, E., Evers, M.J.W., Seghers, S., Gitz-Francois, J.J.J.M., van Kronenburg, N.C.H., Fens, M.H.A.M., Mastro-battista, E., Hassler, L., Sork, H., Ahmed, K.E., Andaloussi, S.E., Breckpot, K., Bastogne, T., Schiffelers, R.M., De Koker, S., 2022. LNPs tuned for systemic immunization induce strong antitumor immunity by engaging splenic immunity. Molecular Therapy.

- Camacho, E.F., Bordons, C., 2007. Nonlinear model predictive control: An introductory review. Assessment and future directions of nonlinear model predictive control 1-16.

- Ferrari, C., Balandras, F., Guedon, E., Olmos, E., Chevalot, I., Marc, A., 2012. Limiting cell aggregation during mesenchymal stem cell expansion on microcarriers. Biotechnology progress 28, 780-787.

- Guideline, I.H.T., others, 2005. Validation of analytical procedures: text and methodology. Q2 (Rl) 1.05.

- ICH Management Committee, 2018. ICH Q14: Analytical Procedure Development and Revision of Q2(R1) Analytical Validation. (Final Concept Paper). International Council for Harmonization of Technical Requirements for Pharmaceuticals for Human Use., ICH Secretariat, Switzerland.

- Lindberg, Y., 2014. A comparison between MPC and PID controllers for education and steam reformers. Master's thesis.

- Maciejowski, J.M., Huzmezan, M., 2007. Predictive control, in: Robust Flight Control: A Design Challenge. Springer, pp. 125-134.

- Nekanti, U., Mohanty, L., Venugopal, P., Balasubramanian, S., Totey, S., Ta, M., 2010. Optimization and scale-up of Wharton’s jelly-derived mesenchymal stem cells for clinical applications. Stem cell research 5, 244-254.

- Rafiq, Q.A., Ruck, S., Hanga, M.P., Heathman, T.R., Coopman, K., Nienow, A.W., Williams, D.J., Hewitt, C.J., 2018. Qualitative and quantitative demonstration of bead-to-bead transfer with bone marrow-derived human mesenchymal stem cells on microcarriers: Utilizing the phenomenon to improve culture performance. Biochemical engineering journal 135, 11-21.

- Van Overschee, P., De Moor, B., 1996. Subspace identification for linear systems - Theory, implementation, applications. Kluwer Academic Publishers.

- Voisin, C., Cauchois, G., Bensoussan, D., Huselstein, C., 2021. Effects of Cell Confluence on the Immunological and Migration Receptors of Wharton Jelly’s Mesenchymal Stem Cells.

[0005] In this context, there is still a need to improve production in a bioreactor.

Summary

[0006] For this purpose, a method implemented by a control system is proposed for controlling cell production in a bioreactor. The bioreactor comprises a cell culture tank containing a culture medium and microcarriers on the surface of which the cells grow. The method comprises, during production, a dynamic adjustment of parameters. The dynamically adjusted parameters include at least a temperature of the culture medium, a pH of the culture medium, a concentration of dissolved oxygen in the culture medium and a quantity of microcarriers present in the tank.

[0007] The dynamic adjustment of the quantity of microcarriers may comprise a measurement of the cell confluence rate at the surface of the microcarriers in the tank, and an addition of microcarriers in the tank so that the measured cell confluence rate follows a predetermined setpoint kinetics.

[0008] The addition of microcarriers into the tank can be carried out by a pump.

[0009] The measurement of the cell confluence rate can be carried out by an online sensor.

[0010] The production may take place during a period divided into successive time intervals. The addition of microcarriers to the tank may comprise, for each time interval, a prediction of an evolution of the cell confluence rate in the tank. The addition of microcarriers to the tank may comprise, for each interval of time, three steps. The first step can be a comparison of the predicted evolution with the predetermined setpoint kinetics. The second step can be a determination of a quantity of microcarriers to be added for the following time interval depending on the result of the comparison. The third step can be an addition of the determined quantity of microcarriers into the tank.

[0011] The prediction may be based on a dynamic model comprising a state matrix, an input matrix and an output matrix. The contents of the matrices may be determined from training data.

[0012] Determining the quantity of microcarriers to be added may include applying an optimization algorithm based on an objective criterion. The objective criterion may preferably have the following formula:

^[0013] minJ( AM

[0014] in which:

[0015] - A u is the increment of the control signal for adding microcarriers in the tank minimizing a cost function J, taking into account the various constraints;

[0016] - y ( t + ) denotes the prediction of the cell confluence rate at time t + J based on the measurements up to time ^f ;

[0017] - r ( t + j ) designates a reference signal at time t + j of the predetermined setpoint kinetics to be continued;

[0018] - denotes the start of a prediction horizon;

[0019] - N ₂ denotes the end of the prediction horizon;

[0020] - N _u denotes a calculation horizon of the control signal; and

[0021] - is a predetermined weighting term.

[0022] The microcarriers may be beads and/or discs. The microcarriers may have a diameter between 50 micrometers and 1000 micrometers.

[0023] The bioreactor may comprise a culture medium agitator and at least one sensor for measuring the quantity of particles suspended in the tank. The method may further comprise dynamic adjustment of the culture medium agitator as a function of the quantity of particles suspended measured by at least one sensor.

[0024] Also provided is a computer program comprising instructions which, when the program is executed by a bioreactor control system, cause the latter to implement the method.

[0025] A control system for a bioreactor is also provided. The control system may be, for example, a SCADA system. The system comprises a processor coupled to a memory. The memory has recorded the computer program. The processor is configured to execute the instructions of the computer program to control the bioreactor according to the method. [0026] A bioreactor integrating the control system is also provided.

Brief description of the figures

[0027] Non-limiting examples will be described with reference to the following figures: [0028] [Fig.1] shows a flowchart of an example implementation of the method.

[0029] [Fig.2] illustrates an example of a bioreactor control system.

[0030] [Fig.3] shows an example of monitoring the predetermined setpoint kinetics.

[0031] [Fig.4] illustrates an example of a control system including agitation regulation.

[0032] [Fig.5] shows examples of acceptable value ranges in accordance with international ICH Q8(R2) guidelines.

[0033] [Fig.6] shows an example block diagram of determining the quantity of microcarriers to be added for each time interval.

[0034] [Fig.7] shows an example of the dynamic model.

[0035] [Fig.8] shows an example of minimization by the objective criterion.

[0036] [Fig.9] shows an example of a signal for stimulating the flow rate of microcarriers to the pump.

[0037] [Fig.10] shows an example of calibration of confluence rate measurements.

[0038] [Fig.11] illustrates an example of an MPC regulator.

Detailed description

[0039] With reference to the flowchart of [Fig.1], a method implemented by a control system is proposed for controlling cell production in a bioreactor. The bioreactor comprises a cell culture tank containing a culture medium and microcarriers on the surface of which the cells grow. The method comprises, during production, a dynamic adjustment S 10 of parameters. The dynamically adjusted parameters include at least a temperature of the culture medium, a pH of the culture medium, a concentration of dissolved oxygen in the culture medium and an amount of microcarriers present in the tank.

[0040] The process improves the quality and quantity of the biomass produced in the bioreactor.

[0041] Indeed, the dynamic adjustment of the quantity of microcarriers makes it possible to control the confluence rate on microcarriers, which is a critical point for improving the quality and production yield. In particular, microcarriers are culture supports for cells and the dynamic adjustment of their quantity in the bioreactor makes it possible to reduce cell degradation when the cell confluence on the microcarriers is too high. Cell confluence indeed affects the behavior and growth of cells. When the cell confluence on the microcarriers becomes too high, the quality of production decreases and the quantity no longer increases (the surface area of all microcarriers in the tank being colonized). It has been shown that it may be one of the main mechanisms that can inhibit their expansion (Nekanti et al., 2010). In Voisin et al., 2021, it is also shown that confluence has an impact on the immunomodulatory phenotype and migration. In addition, once the entire surface area has been recovered by the cells, a phenomenon called contact inhibition occurs and the cells stop growing. Cell confluence on microcarriers also leads to aggregations of cells and microcarriers. Furthermore, it has been shown in Ferrari et al., 2012 that some cells can migrate from microcarriers to other microcarriers during confluence. Rafiq et al., 2018 demonstrated this particular property of bead-to-bead transfer with human bone marrow mesenchymal stem cells, using different types of microcarriers in the same culture. Therefore, automatically controlling the addition of new microcarriers allows to reduce the percentage of confluence and therefore the cell damage associated with this phenomenon. Furthermore, by improving the parameter setting, the process also reduces the risk of initial culture conditions between different production batches being different.

[0042] In particular, the process improves quality and production yields, particularly in the context of the development of innovative therapy drugs. This is the main motivation behind national calls for tenders such as the major biomedicine challenge in 2020 and, more recently, that linked to innovations in biotherapies and bioproduction.

[0043] The method is implemented by the control system. This means that the steps (or substantially all of the steps) of the method are executed by the control system, or any other similar system. Thus, the steps of the method are executed by the control system, possibly in a fully automatic or semi-automatic manner. In some examples, the triggering of at least some of the steps of the method may be carried out by an interaction between the user and the control system.

[0044] A typical example of implementing the method is to execute the method using a control system adapted for this purpose. The control system may comprise a processor coupled to a memory and a graphical user interface (GUI). The memory may have recorded a computer program comprising instructions for executing the method. The memory may also store a database. The memory is a hardware adapted for this type of storage, possibly comprising several distinct physical parts (for example, one for the program, and possibly one for the database).

[0045] Controlling cell production in the bioreactor means mastering the production of cells in the bioreactor by directly adjusting, i.e. dynamically, certain critical variables (i.e. parameters) of the process. By dynamic adjustment, it is understood that the process is carried out during the production of cells in the bioreactor, and that the process adjusts the parameters during the production (i.e. several times, at several different times during the production). For example, the production may take place during a period divided into successive time intervals, and, at each time interval, the process may comprise an adjustment of the parameters. The method may comprise, after each elapse of a time interval, a determination of the value of the parameters (for the elapsed time interval), then, an adjustment of the parameters (i.e. the value of the parameters) for a following time interval (after the elapsed one). The following time interval may be the time interval immediately following the elapsed one, or the second (or any subsequent interval after the elapsed one). The method may comprise a recording of the determined value of each parameter (for example in the database). The dynamically adjusted parameters are critical parameters of the production.

[0046] The production period and/or the division into time intervals may be predetermined. For example, the length of the production period and/or the time intervals considered may be defined by the user before the execution of the method (for example by means of user actions on the control system). The production period may be several hours, for example to several days. The division into time intervals may be a division into intervals of one or more seconds, for example one or more tens of minutes.

[0047] For one or more parameters (e.g. for temperature, pH and/or oxygen concentration), the determination of the value of the parameters may comprise a measurement (e.g. by a sensor) of the value of the parameter in the tank during the time interval. For example, the determination of the temperature may comprise a measurement of the temperature in the tank by a temperature sensor. Similarly, the pH may be measured by a pH measuring sensor and the oxygen concentration may be measured by an oxygen quantity sensor. The determined value of the parameter may then correspond to an average of the measured value of the parameter over the time interval (e.g. when the measurement comprises several measurement points over the interval). Alternatively, the determined value of the parameter may be a value taken during the time interval (e.g. at the beginning or in the middle of the interval). For the quantity of microcarriers, the determination of the value of this parameter may be calculated. For example, determining the value may include calculating the amount of microcarriers in the tank based on the amount of microcarriers determined for an interval of previous time (e.g. initially present in the tank for the first time interval) and the quantity of microcarriers added since the elapse of this previous time interval.

[0048] The cells produced in the bioreactor may be adherent cells that grow on the surface of microcarriers. For example, the cells may be HEK293 cells, mesenchymal stem cells, Vero cells, or CHO cells. The cells may colonize the surface of the microcarriers that are present in the cell culture vessel (i.e., the microcarriers initially present in the vessel and those that are added during the execution of the process). The cells may develop (i.e., grow and multiply) by feeding on the culture medium present in the vessel.

[0049] The method comprises a dynamic adjustment S10 of at least the parameters temperature, pH, dissolved oxygen concentration and quantity of microcarriers in the tank. For example, the method may comprise, for each of these parameters, the execution of commands to increase or decrease the value of the parameter (these commands being executed at different times during production). In examples, the method may comprise, in parallel, a dynamic adjustment of one or more other production parameters (other than the temperature, the pH, the dissolved oxygen concentration and the quantity of microcarriers in the tank).

[0050] In examples, the method can repeat the dynamic adjustment S 10 for different production batches. For each production batch, the method can carry out, during production, the dynamic adjustment S10 of at least the parameters temperature, pH, dissolved oxygen concentration and quantity of microcarriers in the tank in the same manner. The method thus makes it possible to compensate for the effects due to uncontrollable differences in initial culture conditions between the different production batches, which improves the reproducibility of production in the bioreactor.

[0051] The dynamic adjustment S 10 of temperature, pH and dissolved oxygen concentration is now discussed.

[0052] The method can perform the dynamic adjustment S 10 of the temperature, pH and dissolved oxygen concentration in any manner. For example, the method can dynamically adjust these parameters to follow a predetermined setpoint (e.g. a curve indicating a desired evolution for each parameter as a function of time) and/or so that the value of these parameters remains within predetermined value intervals (e.g. by standards or norms). The adjustment of these parameters can be done by adding substances (e.g. air for the oxygen concentration or carbon dioxide gas for the pH) or energy to the tank (e.g. heat for the temperature). The dynamic adjustment S 10 can, at each time interval, and for each in- time interval, including a measurement of the parameter (temperature, pH and dissolved oxygen concentration), then, for each parameter, the addition of a quantity of the respective substance to the parameter so as to follow the predetermined set value of the parameter.

[0053] The adjustment S 10 of the quantity of microcarriers is now discussed in more detail.

[0054] The adjustment S10 of the quantity of microcarriers may comprise a measurement S20 of the cell confluence rate at the surface of the microcarriers in the tank. The measurement S20 of the cell confluence rate may be carried out by an online sensor (or probe). The online sensor may be a multi-angle sensor, which is configured to find the coverage rate on the microcarriers. The online sensor may be a near-infrared spatially resolved spectroscopic probe. For example, the online sensor may be the SAM-Spec sensor manufactured by Indatech (Chauvin-Arnoux). Alternatively, the online sensor may be a probe for measuring permittivity in a liquid medium (such as the Incyte Arc sensor manufactured by Hamilton). Alternatively, the measurement may be an offline measurement. For example, the measurement may comprise taking a sample from the tank (for example automatically by a sample taking device located in the tank, or manually by an operator). After collection, the measurement may include processing the sample collected to deduce the cell confluence rate. The processing may include determining the absorbance and/or turbidity of the sample collected, followed by calculating the cell confluence rate in the tank based on the determined absorbance and/or turbidity.

[0055] As previously described, the production may take place during a period divided into successive time intervals. The addition S30 of microcarriers into the tank may comprise, for each time interval, a prediction S31 of an evolution of the cell confluence rate in the tank. The prediction may be a prediction in future instants of the evolution of the cell confluence rate in the tank. The time interval over which the prediction is made may be a future time interval. For example, the prediction S31 may take place during a time interval (or time step) /, and the prediction S31 may predict the evolution of the cell confluence rate for the time interval (or time step) t+1, i.e. the one following the time interval currently elapsing while the method is performing the prediction S31 for the interval t. The prediction may be based on the evolution of the cell confluence rate before the time interval. For example, the prediction can be based on the cell confluence rate measured for previous time intervals. The prediction can also be based on measured values of the other parameters (temperature, pH and/or dissolved oxygen concentration), e.g. example the values measured for these other parameters during their dynamic adjustment.

[0056] In examples, the prediction may be based on a dynamic model comprising a state matrix, an input matrix and an output matrix. The content of the matrices may be determined from training data. The dynamic model may be a dynamic model of the bioreactor describing it by a state representation. The content of the matrices may be predetermined. The content of the matrices may have been determined before the execution of the method. The dynamic model may be a model belonging to the family of black box models, i.e. an empirical model whose content of the matrices is determined from training data. The training data may have been collected during experiments carried out on the bioreactor to be controlled. The dynamic model makes it possible to establish predictions on the cellular confluence rate of the microcarriers in the tank during the time interval.

[0057] The prediction S31 of the evolution of the cell confluence rate may be a prediction over the time interval. For example, the prediction may comprise several predicted values of the cell confluence rate at different times in the time interval. Alternatively, the prediction may comprise a single predicted value of the cell confluence rate, this value being able to be an average for the interval or at a time in the interval (for example in the middle or at the end of the interval). Alternatively, the prediction may comprise a predicted variation compared to a measured and/or predicted value for the previous time interval (for example an increase or a decrease in the cell confluence rate).

[0058] After the prediction, the addition S30 of microcarriers may comprise a comparison S32 of the predicted evolution with the predetermined setpoint kinetics. The predetermined setpoint kinetics may comprise an evolution of the setpoint cell confluence rate as a function of time. This evolution may be optimal in terms of quality and quantity of biomass produced. The predetermined setpoint kinetics may be represented by a curve representing the evolution of the setpoint cell confluence rate as a function of time. The setpoint kinetics may have been predetermined on prior tests. The curve representing it may have the shape of a sigmoid, for example with as parameters: K a static gain and T a time constant. The comparison S32 may comprise a calculation of a difference between the predicted evolution for the time interval and the cell confluence rate given by the setpoint evolution for this time interval, and for example also a calculation of a difference between the predicted evolution for one or more previous time intervals and, for each, the respective cell confluence rate given by the setpoint evolution. [0059] The addition S30 may comprise a determination S33 of a quantity of microcarriers to be added for the following time interval based on the result of the comparison. The result of the comparison may comprise all of the calculated differences. The determination S33 may comprise an application of an optimization algorithm. The optimization algorithm may be based on an objective criterion. The objective criterion may take as input a control signal for the addition of microcarriers to the tank. The optimization algorithm may comprise a minimization of this objective criterion. The quantity of microcarriers to be added determined may minimize the result of the objective criterion. The minimization may comprise a determination of a control signal for the addition of microcarriers to the tank minimizing the result of the objective criterion. The objective criterion may preferably have the following formula:

[0061] dans laquelle : [° ⁰⁶⁰ ] minJ(

[0061] in which:

- A u is the increment of the control signal for the addition of microcarriers in the tank minimizing a criterion J, taking into account the various constraints;

- y ( t + ) denotes the prediction of the cell confluence rate at time t + J ^; based on measurements up to time ^f ;

- r ( t + j ) denotes a reference signal at time f + . of the predetermined setpoint kinetics;

- N j denotes the start of a prediction horizon;

- N denotes the end of the prediction horizon;

- N _u denotes a calculation horizon of the control signal; and

- is a predetermined weighting term

[0062] The prediction horizon: N = N ₂ - N ₁ may be greater than or equal to 1 and/or less than or equal to 50. For example, N may be equal to 10. The weighting term may be predetermined by prior comparison tests. The weighting term may be greater than or equal to 0 and/or less than or equal to 10. For example, the weighting term may be equal to 1.

[0063] After the determination S33, the addition S30 may comprise the addition S34 of the determined quantity of microcarriers in the tank. The method may execute the steps S31 to S33 at the time interval t (using the notations previously introduced). The method may execute the addition S34 of the determined quantity of microcarriers during the time interval /+ 1 (the time interval following the time interval t during which the prediction is executed). The addition S34 of the determined quantity of microcarriers in the tank may comprise an application of the control signal for the addition of microcarriers in the tank determined (i.e. minimizing the result of the objective function). The addition of microcarriers to the tank can be achieved by a pump. The control signal can be a control signal for this pump. For example, the pump can include a valve, and the control signal can define an opening value for this valve (the pump valve being configured so that the opening value modifies a flow rate of microcarriers added to the tank). The opening value can be applied during the time interval t+1.

[0064] The microcarriers may be of any shape. The microcarriers may be beads and/or discs. For example, all of the microcarriers in the vessel (initially present or added) may be beads. Alternatively, all of the microcarriers in the vessel may be discs. Alternatively, the vessel may comprise a mixture of beads and discs. The microcarriers may have a diameter of between 50 micrometers and 1000 micrometers.

[0065] In examples, the bioreactor may include an agitator. The agitator may be any device capable of agitating the contents of the tank. For example, the agitation may be a propeller. Alternatively, the agitator may be a flow agitator or an external device such as wave bioreactors. The bioreactor may also include at least one sensor for measuring the amount of particles suspended in the tank. The at least one measurement sensor may be configured to measure the absorbance and/or turbidity of the container of the tank. For example, the bioreactor may include two in-line sensors: a sensor for the absorbance of the container of the tank and a sensor for the turbidity of the container of the tank. The method may include measuring the amount of particles (e.g., in real time, at each interval t for example). The measurement may include a calculation of the cell confluence rate based on the result of at least one sensor present in the tank (for example based on the measured absorbance and/or turbidity).

[0066] When the bioreactor comprises an agitator and at least one sensor for measuring the quantity of particles suspended in the tank, the method may further comprise dynamic adjustment of the agitator as a function of the quantity of particles suspended measured by the at least one sensor. The dynamic adjustment of the agitator may comprise, for each time interval, a measurement of the quantity of particles suspended at a time interval (or time step) t, then, a determination of a control signal for the agitator to be applied for the following time interval t+1. After this, the dynamic adjustment may comprise an application of the determined control signal at the time interval t+1 (and also at this time a repetition of the measurement and determination of the signal for the following time interval t+2). The determined control signal may be such that the quantity of particles suspended remains within a given range of values. For example, the determined control signal may increase the agitation when the quantity of suspended particles is less than the values in the given range and decrease agitation when the quantity of suspended particles is greater than the values in the given range.

[0067] Dynamic adjustment of agitation contributes to improving the quality and quantity of biomass produced in the reactor. Indeed, the quantity of suspended particles is one of the parameters influencing the quality and quantity of biomass produced because it makes it possible to increase the culture surface. Thus, it increases the yield of the bioreactor while avoiding excessively high cell confluence leading to cell degradation. Adjustments to the quantity of microcarriers and the agitation speed are interdependent. Indeed, when the number of microcarriers increases, the number of cells also increases and it becomes necessary, due to this greater mass, to increase the agitation speed to maintain homogenization of the whole above a critical value. If the agitation is too weak, all the microcarriers and cells gradually fall to the bottom of the bioreactor. In particular, dynamic adjustment of agitation makes it possible (in addition to adjusting the quantity of microcarriers) to reduce cell damage. Dynamic agitation control allows the microcarriers to remain in suspension, while preventing the cells from being broken (in the event of excessive agitation).

[0068] In particular, the hydrodynamic effect, the agitation parameters and the agitation speed influence cell damage. In particular, the hydrodynamic death mechanisms are due to collisions of a microcarrier covered with cells with other beads, collisions with parts of the reactor (mainly the propeller) and interaction with turbulent vortices. Cell death increases with the agitation power and decreases with the viscosity of the fluid. Automatically adjusting the agitation in the bioreactor therefore makes it possible to reduce damage due to collisions and to reproduce more faithfully a setting between different batches produced in the bioreactor.

[0069] With reference to Figures 2 to 11, examples of implementation of the method will now be described.

[0070] [Fig. 2] illustrates an example of a control system. The control system drives a pump 102 to automatically adjust, over the course of the culture time, the quantity of microcarriers to be added to a bioreactor which comprises a cell culture tank 101 containing cells, microcarriers and a culture medium, but also a cell confluence rate sensor 103 which returns the measurement data to the control system 104. This sensor can be a near-infrared spatial resolution spectroscopic probe, such as SAM-Spec manufactured by Indatech (Chauvin-Amoux) or a capacitance probe such as the Incyte-Arc d' Hamilton. From the information from these sensors, the control unit 104 calculates the microcarrier supply rate so that the cell confluence measured on the surface of the microcarriers follows as closely as possible a predefined setpoint kinetics and does not exceed an upper limit that is also predefined.

[0071] [Fig.3] shows an example of monitoring the predetermined setpoint kinetics. [Fig.3] shows the measured cell confluence 110 at the surface of the microcarriers and the predefined setpoint kinetics 111. The method determines at each time step the quantity of microcarriers to be added to the tank so that the cell confluence in the tank follows the predefined setpoint kinetics 111.

[0072] To carry out the operation of controlling the time trajectory of the confluence rate of the microcarriers, the central control unit can comprise software, which, depending on the differences between the confluence rate measurement signal and the desired reference trajectory of this same quantity, calculates the adjustments (variations) to be made to the bare microcarrier supply flow rate, to reduce this difference as much as possible.

[0073] [Fig.4] illustrates an example of a control system including agitation regulation. The multivariable bioreactor control system comprises a biomass culture tank 201, a temperature sensor 207, a pH probe 209, a dissolved oxygen sensor 208, a biomass analysis probe 213, an optical density sensor 214, a turbidity sensor 215, a soda (NaOH) feed pump 206A, mass flow regulators 206B, 205A, 205B for the air, nitrogen and CO2 supplies, a heating resistor 204, and a culture medium and microcarrier feed pump 203. The contents of the bioreactor are continuously mixed by a stirrer 202. The temperature 210, dissolved oxygen 211 and pH 212 are controlled by three independent control loops acting on the actuators 203, 204, 205, 206. These five critical process parameters: temperature of the medium, pH level, dissolved O2 concentration, stirring power and quantity of microcarriers, have combined effects on the quantity and quality of the biomass produced. A central control unit 214 simultaneously calculates the modifications to be made to the set values of the local regulators managing the temperature, the pH level and the dissolved oxygen concentration (dO2), as well as on the microcarrier feed rate and the stirrer speed so that:

[0074] - the quantity of biomass measured 220 by a probe inside the bioreactor always follows the same reference trajectory 221 during the culture time. The culture trajectory is imposed by the user responsible for the bioreactor;

[0075] - all of the microcarriers are kept in suspension without unnecessarily increasing the stirring power which would result in shocks and deterioration of the state of the cells on the microcarriers;

[0076] - the cellular confluence on the surface of the microcarriers follows predefined kinetics and that it does not exceed an upper limit which is also predefined;

[0077] - the values of the variables of temperature, dissolved oxygen concentration, agitator rotation speed and pH level are always maintained within a range of acceptable values, called Control Operating Region 232, a sub-region of the “Design Space” 231 in compliance with international directives in Quality by Design (ICH Q8 to Q12) (see the ranges of acceptable values in compliance with international directives shown in [Fig.5]).

[0078] The central control unit comprises software which, depending on the deviations between the measurement signals and the desired reference trajectories, simultaneously calculates the adjustments (variations) to be made to the instructions of the local regulators, to reduce these deviations while maintaining the values of the control variables within a working region.

[0079] The control software may include a predictive control algorithm. The software may include instructions for determining the quantity of microcarriers to be added at each time interval. [Fig.6] shows an example of a block diagram of the predictive control principle.

[0080] Compared to a conventional closed-loop control scheme, the setpoint ^r of the predictive control contains a quantity at different times in the future: t + 1, ■ ■ ■ , t + N ₂ , in order to form a horizon corresponding to the reference signal. This horizon of N ₂ future setpoint (or reference) values, noted r(t + i), is compared at each time ^f (iteration) to the N ₂ future predicted values v(/ + z) of the output variable which here corresponds to the average cellular confluence rate of the microcarriers.

[0081] The difference between the desired values r(t + i) and the predicted values y(t + z) is then used by the optimization algorithm to generate a horizon of N _u future values of the control signal, noted ut + z), where only the first value corresponding to the instant t + 1 is applied to the actuator, here the microcarrier supply pump. This control horizon is then recalculated at each instant ^z in order to avoid disparities between the predictions and the observed reality. N _u is a parameter of the control unit which can be adjusted by an automation engineer during the implementation and testing phases of the control unit on site.

[0082] An example of a minimization algorithm is now discussed. The bioreactor may include an MPC (Model (Based) Predictive Control) controller that is constructed from 3 main elements: a prediction model, an objective function, and a control law. [0083] The prediction model can be a dynamic model of the bioreactor described by a state representation. [Fig.7] shows an example of the dynamic model. This dynamic model can be made up of three matrices: state matrix (A), input matrix (B), output matrix (C). It can be a model belonging to the family of black box models, i.e. an empirical model whose matrix content is determined from training data, themselves collected during experiments carried out on the system to be controlled. This model makes it possible to establish predictions y on the cellular confluence rate of the microcarriers, predictions used in the MPC regulator.

[0084] The objective criterion can be of the form:

^[0085] minJ( Au

[0086] which describes a mathematical optimization problem where the objective is to find the best increment of the control signal A u allowing to minimize the cost function J, taking into account the various constraints. This function is composed of the following elements:

[0087] - y ( t + ) denotes the prediction of the output (cell confluence rate) at time t + j knowing all the measurements up to time ^f .

[0088] - r ( t + j ) denotes the reference signal at time t + j.

[0089] - denotes the start of the prediction horizon.

[0090] - N ₂ denotes the end of the prediction horizon.

[0091] - N _u denotes the calculation horizon of the control signal.

[0092] - is a weighting term used to choose between tracking performance and energy cost spent on the input signal. Like N _u , it can be a parameter to be adjusted by the automation engineer during the test phases.

[0093] This objective function is composed of 2 terms:

[0094] - the first having the role of minimizing the quadratic error between the output y ( t ) and the reference r(f ), as illustrated in the diagram in [Fig.8], for better trajectory tracking precision;

[0095] - the second having the role of minimizing the energy cost on the control signal and makes it possible to obtain a physically achievable control signal.

[0096] The control law can correspond to the optimal control value: ^^), solution of the objective criterion, determined by the resolution of the optimization algorithm, allowing the calculation of the following value of the control signal, noted

[0097] Table 1 below shows the contents of an example MPC controller optimization algorithm that minimizes an objective function while taking takes into account a set of constraints.

dite active au point ^xk si g = 0 ^cl inactive au point ^xk si g^J > Q. L’ensemble

actif W_k au point ^xk est l'ensemble des contraintes actives au point courant. Pour

chaque point ^xk et son ensemble actif W_k associé, l’optimisation est réalisée en utilisant la méthode de résolution par Lagrangien. La solution optimale est trouvée lorsque tous les multiplicateurs de Lagrange sont positifs. Dans le cas contraire, cela indique que la fonction d’objectif peut encore être optimisée, en rendant inactives les contraintes correspondant aux multiplicateurs de Lagrange négatifs. L’algorithme réalise alors un pas ^akP_k, avec ^ak le paramètre de longueur de pas, correspondant à la plus grande valeur de la plage [0,1] pour laquelle toutes les contraintes ^{sont sa}"

tisfaites, et Pk le paramètre de direction de pas. Les contraintes pour lesquelles le

minimum est atteint sont appelées contraintes bloquantes : si «< = I et qu'aucune nouvelle contrainte n'est active à ^xt+i alors il n’y a pas de contrainte bloquante à cette itération. Si -k ^< 1, cela signifie que le pas le long de Pk a été bloqué par une contrainte extérieure à W_k. Dans ce cas, un nouvel ensemble actif

est formé en ajoutant une des contraintes bloquantes à W_k. L'algorithme est ensuite répété pour chaque itération, en ajoutant des contraintes à l'ensemble actif W_k jusqu’à ce que la solution optimale, qui minimise la fonction d'objectif sur son ensemble actif W_k soit trouvée. [0099] Any optimization problem such as this is defined using an objective function but also a set of constraints that define the set of admissible solutions, which corresponds to the set of ^x k to be tested in order to find the optimal solution to the problem. Let ^x k be an admissible point, a constraint j is

said to be active at point ^x k if g = 0 ^cl inactive at point ^x k if g^J > Q. The set

active W _k at point ^x k is the set of constraints active at the current point. For

each point ^x k and its associated active set W _k , the optimization is performed using the Lagrangian resolution method. The optimal solution is found when all Lagrange multipliers are positive. Otherwise, this indicates that the objective function can be further optimized, by making inactive the constraints corresponding to negative Lagrange multipliers. The algorithm then performs a step ^a kP _k , with ^a k the step length parameter, corresponding to the largest value in the range [0,1] for which all constraints ^{are sa} "

satisfies, and Pk the step direction parameter. The constraints for which the

minimum is reached are called blocking constraints: if «< = I and no new constraint is active at ^x t+i then there is no blocking constraint at this iteration. If -k ^< 1, this means that the step along Pk has been blocked by a constraint outside W _k . In this case, a new active set

is formed by adding one of the blocking constraints to W _k . The algorithm is then repeated for each iteration, adding constraints to the active set W _k until the optimal solution, which minimizes the objective function over its active set W _{k ,} is found.

[0100] The dynamic model used to estimate the predictions y of the output variable (cell confluence rate) may be a black box state model determined from a supervised machine learning method based on subspace decomposition (for example with the subspace decomposition algorithms described in Van Overschee and De Moor, 1996 and/or Bastogne, 1997). Commercial software tools exist to implement these decomposition methods and they make it possible to determine a state model directly from the training data, in particular for example in the System Identification toolbox of Mathworks' Matlab software.

[0101] The determining method may be executed prior to execution of the method. Alternatively, the determining method may be included in the method, which may comprise, prior to production, an execution of the determining method to determine the state model that will be used, during production, to make the predictions.

[0102] The determination method can be based on three steps.

[0103] The first step may include carrying out experiments dedicated to the acquisition of learning data. Carrying out the experiments may be based on stimulations of the input signal w(f) (the microcarrier feed rate) from a pulse signal of variable heights and widths, such as that shown in [Fig.9]. These parameters may be adjusted by a person skilled in the art, an automation engineer, assisted by a bioprocess engineer. During the experiments, the measurement signals m(t) from the confluence sensor may be recorded and samples may be taken to measure in the laboratory, with a reference technique, an average number of cells per microcarrier: y ( t ) •

[0104] The second step may include a calibration of the measurement signals. This step may include a determination of a calibration function /(.) making it possible to reconstruct an estimate y = f nï) of the cell confluence rate. The good practices defined in the ICH Q14 guidelines (ICH Management Committee, 2018) and Q2 (Guideline and others, 2005) can be used for this purpose. [Fig.10] shows an example of such a calibration of confluence rate measurements.

[0105] The third step may comprise an identification of the state model. The third step may consist of applying subspace decomposition algorithms (for example such as those described in Van Overschee and De Moor, 1996 and Bastogne, 1997) on the signals (ut), y (t)) to estimate the matrices (A, B, C) of the following state model:

[0107] where x(t) is a state vector of dimension ⁿ corresponding to the order of the model. This order is also estimated by the identification algorithm. v(t), wt) are centered, independent Gaussian random variables with constant variances, which represent the measurement and state noises. These random variables make it possible to describe the uncertainty on the predictions of the confluence rate.

[0108] The dynamic model used to estimate the predictions of the output variable (cell confluence rate) can also be a white-box model, i.e. a model using biological laws of cell culture to describe the behavior of the bioreactor, such as that defined in the equations below, which considers a first non-linear dynamic equation representing the concentration of cells X(t) and a second non-linear differential equation representing the concentration of substrate S(t).

[0109]

[0110] in which:

[0111] L _s is the substrate efficiency coefficient,

[0112] S _in is the substrate concentration of the input stream,

[0113] F _s is the flow rate of the input stream,

[0114] V(t) is the volume of the bioreactor, and

[0115] is the cell growth rate. This parameter is itself a function of the process parameters ((t)) such as temperature (T), pH, dissolved oxygen (DO2) in the bioreactor, but also of certain key nutrients in the culture medium such as the amount of glucose and glutamine.

[0116] The process has several advantages and consequences on the quality and yield of bioproduction. Firstly, it allows to control cell colonization at the surface of the microcarriers and more specifically the phenomenon of cellular confluence which can be the cause of rapid degradation of the biological state of the cells if it exceeds a certain level. Secondly, it allows the bioreactor to follow a predefined kinetics of the average confluence during the culture. Thirdly, it makes it possible to improve the reproducibility of production between each batch.

[0117] An example of adjusting the production parameters in the bioreactor (temperature, pH level, dissolved oxygen concentration, substrate quantity, nutrient concentration, stirring speed and microcarrier quantity) is now discussed. The adjustment of the parameters is based on the prediction of the evolution over time of critical quality attributes (output variables to be controlled) such as cell confluence, the quantity of living cells and the level of population doubling. This prediction is based on a dynamic model of the bioreactor such as the state models previously described.

[0118] The bioreactor control can incorporate a Model Predictive Control (MPC) algorithm, which adjusts in real time the values of critical process parameters to follow predefined trajectories of critical quality attributes. [Fig. 11] illustrates an example of such an MPC controller. This MPC optimization algorithm calculates the best fit of the values for the process parameters from the prediction of the previously described modeling of the bioreactor in order to minimize the error between the predicted kinetics, such as Cc(t), X(t) and PDL(t), and the desired trajectory of the same critical quality attributes: Cc*(t), X*(t) and PDL*(t). This controller also minimizes the cost of energy and materials.

[0119] The MPC regulator helps improve biomass production in the bioreactor since it allows predefined ideal trajectories to be followed for certain critical quality attributes such as cell confluence (Cc(t)), the number of living cells (X(t)) and the population doubling level (PDL(t)). This regulator also allows constraints on value ranges on the input variables (critical process parameters) to be taken into account.

Claims

Claims [Claim 1] A method implemented by a control system (104) for controlling a production of cells in a bioreactor, the bioreactor comprising a cell culture vessel (101) containing a culture medium and microcarriers on the surface of which the cells grow, the method comprising, during the production, a dynamic adjustment (S 10) of parameters including at least: - a temperature of the culture medium; - a pH of the culture medium; - a concentration of dissolved oxygen in the culture medium; and - a quantity of microcarriers present in the tank. [Claim 2] The method of claim 1, wherein the dynamic adjustment (S 10) of the quantity of microcarriers comprises: - a measurement (S20) of the cell confluence rate (110) at the surface of the microcarriers in the tank; and - an addition (S30) of microcarriers in the tank so that the measured cell confluence rate (110) follows a predetermined setpoint kinetics (111). [Claim 3] Method according to claim 1 or 2, in which the addition (S30) of microcarriers in the tank is carried out by a pump (102). [Claim 4] A method according to any one of claims 1 to 3, wherein the measurement of the cell confluence rate is carried out by an in-line probe (103). [Claim 5] Method according to any one of claims 2 to 4, in which the production takes place during a period divided into successive time intervals, the addition (S30) of microcarriers in the tank comprising, for each time interval: - a prediction (S31) of an evolution of the cell confluence rate in the tank; - a comparison (S32) of the predicted evolution with the predetermined setpoint kinetics; - a determination (S33) of a quantity of microcarriers to be added for the following time interval as a function of the result of the comparison; and an addition (S34) of the quantity of microcarriers determined in the tank. [Claim 6] A method according to claim 5, wherein the prediction is based on: - a dynamic model comprising a state matrix, an input matrix and an output matrix, the contents of the matrices being determined from training data, or - a nonlinear state model of the transparent box type using biological laws of cell culture to describe the behavior of the bioreactor. [Claim 7] A method according to claim 5 or 6, wherein determining the quantity of microcarriers to be added comprises applying an optimization algorithm based on an objective criterion, the objective criterion preferably having the following formula:

in which: - A u is the increment of the control signal for adding microcarriers in the tank minimizing a cost function J, taking into account the various constraints; - y ( t + ) denotes the prediction of the cell confluence rate at time t + j based on measurements up to time ^z ; - r ( t + j ) designates a reference signal at time t + j of the predetermined setpoint kinetics to be continued; - N j denotes the start of a prediction horizon; - N 2 denotes the end of the prediction horizon; - N _u denotes a calculation horizon of the control signal; and - is a predetermined weighting term. [Claim 8] A method according to any one of claims 1 to 7, wherein the microcarriers are beads and/or discs, the microcarriers having a diameter between 50 micrometers and 1000 micrometers. [Claim 9] Method according to any one of claims 1 to 8, in which the bioreactor comprises an agitator of the culture medium and at least one sensor for measuring the quantity of particles suspended in the tank, the method further comprising: - dynamic adjustment of the culture medium agitator according to the quantity of suspended particles measured by at least one sensor. [Claim 10] A computer program comprising instructions which, when the program is executed by a bioreactor control system, cause the latter to implement the method according to any one of claims 1 to 9. [Claim 11] A control system for a bioreactor such as a SCADA system, the system comprising a processor coupled to a memory, the memory having recorded the computer program of claim 10, the processor being configured to execute the instructions of the computer program to control the bioreactor according to the method of any one of claims 1 to 9. [Claim 12] Bioreactor incorporating a control system according to claim 11.