Bioprocess and Biosystems Engineering

, Volume 26, Issue 1, pp 1–10

Improvement of a mammalian cell culture process by adaptive, model-based dialysis fed-batch cultivation and suppression of apoptosis

Original Paper

DOI: 10.1007/s00449-003-0335-z

Cite this article as:
Frahm, B., Lane, P., Märkl, H. et al. Bioprocess Biosyst Eng (2003) 26: 1. doi:10.1007/s00449-003-0335-z

Abstract

Both conventional and genetic engineering techniques can significantly improve the performance of animal cell cultures for the large-scale production of pharmaceutical products. In this paper, the effect of such techniques on cell yield and antibody production of two NS0 cell lines is presented. On the one hand, the effect of fed-batch cultivation using dialysis is compared to cultivation without dialysis. Maximum cell density could be increased by a factor of ~5–7 by dialysis fed-batch cultivation. On the other hand, suppression of apoptosis in the NS0 cell line 6A1 bcl-2 resulted in a prolonged growth phase and a higher viability and maximum cell density in fed-batch cultivation in contrast to the control cell line 6A1 (100)3. These factors resulted in more product formation (by a factor ~2). Finally, the adaptive model-based OLFO controller, developed as a general tool for cell culture fed-batch processes, was able to control the fed-batch and dialysis fed-batch cultivations of both cell lines.

Keywords

Process control Fed-batch Apoptosis bcl-2 overexpression NS0 cells 

Abbreviations

A

membrane area (dm2)

cGlc,F

glucose concentration in nutrient feed (mmol L−1)

cGlc,FD

glucose concentration in dialysis feed (mmol L−1)

cGlc,i

glucose concentration in inner reactor chamber (mmol L−1)

cGlc,o

glucose concentration in outer reactor chamber (dialysis chamber) (mmol L−1)

cLac,FD

lactate concentration in dialysis feed (mmol L−1)

cLac,i

lactate concentration in inner reactor chamber (mmol L−1)

cLac,o

lactate concentration in outer reactor chamber (dialysis chamber) (mmol L−1)

cLS,FD

limiting substrate concentration in dialysis feed (mmol L−1)

cLS,i

limiting substrate concentration in inner reactor chamber (mmol L−1)

cLS,o

limiting substrate concentration in outer reactor chamber (dialysis chamber) (mmol L−1)

cMab

monoclonal antibody concentration (mg L−1)

FD

feed rate of dialysis feed (L h−1)

FGlc

feed rate of nutrient concentrate feed (L h−1)

Kd

maximum death constant (h−1)

kd,LS

death rate constant for limiting substrate (mmol L−1)

kGlc

monod kinetic constant for glucose uptake (mmol L−1)

kLac

monod kinetic constant for lactate uptake (mmol L−1)

kLS

monod kinetic constant for limiting substrate uptake (mmol L−1)

KLys

cell lysis constant (h−1)

KS,Glc

monod kinetic constant for glucose (mmol L−1)

KS,LS

monod kinetic constant for limiting substrate (mmol L−1)

µ

cell-specific growth rate (h−1)

µ d

cell-specific death rate (h−1)

µ d,min

minimum cell-specific death rate (h−1)

µ max

maximum cell-specific growth rate (h−1)

PGlc

membrane permeation coefficient for glucose (dm h−1)

PLac

membrane permeation coefficient for lactate (dm h−1)

PLS

membrane permeation coefficient for limiting substrate (dm h−1)

qGlc

cell-specific glucose uptake rate (mmol cell−1 h−1)

qGlc,max

maximum cell-specific glucose uptake rate (mmol cell−1 h−1)

qLac

cell-specific lactate uptake/production rate (mmol cell−1 h−1)

qLac,max

maximum cell-specific lactate uptake rate (mmol cell−1 h−1)

qLS

cell-specific limiting substrate uptake rate (mmol cell−1 h−1)

qLS,max

maximum cell-specific limiting substrate uptake rate (mmol cell −1 h−1)

qMab

cell-specific antibody production rate (mg cell−1 h−1)

qMAb,max

maximum cell-specific antibody production rate (mg cell−1 h−1)

t

time (h)

Vi

volume of inner reactor chamber (culture chamber) (L)

Vo

volume of outer reactor chamber (dialysis chamber) (L)

Xt

total cell concentration (cells L−1)

X

viable cell concentration (cells L−1)

YLac/Glc

kinetic production constant (stoichiometric ratio of lactate production and glucose uptake) (−)

1 Introduction

Animal cell culture has become an important field in the pharmaceutical industry for the production of diagnostics and therapeutics. With products such as Erythropoietin (EPO), recombinant Factor VIII and t-PA, the production of monoclonal antibodies and recombinant proteins is an important field of biotechnology. Nevertheless, worldwide conditions for companies are becoming more and more difficult. The health system's demands for low prices of pharmaceuticals, the pressure of competition and the globalizing market are emphasizing the economic side of animal cell culture and are asking for more efficiency. But potential savings do not only originate from short times to market, advanced downstream processing, high manufacturing flexibility, low research and development costs or good sales networks. The process itself and its potential for improvement should not be underestimated. Biological techniques for the optimization of cell lines and engineering techniques for the enhancement of the process and its control can lead to higher product yields.

As industrial large-scale production systems, batch and fed-batch suspension cultures are still dominant [1, 2, 3]. Research on the characterization and improvement of these types of process has been carried out for many years [4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15], but significant improvements are still possible [16, 17, 18].

In the following, three techniques applied to the cultivation of a NS0 cell line producing a monoclonal antibody are presented: the use of dialysis, adaptive model-based control and the overexpression of bcl-2 for suppression of apoptosis.

1.1 Dialysis cultivation

Metabolites produced from animal cells during the cultivation can inhibit cell growth. These metabolites can accumulate to toxic levels. The use of dialysis membranes for the removal of low molecular weight metabolites can significantly enhance the performance of the process. The metabolites diffuse across the membrane into a dialyzing fluid (dialysate) according to the concentration gradient. Large molecules such as monoclonal antibodies are simultaneously retained with the cells leading to an enrichment of the product. Medium is often used as dialysate to remove the metabolites and to provide the cells with fresh nutrients. A scheme of a dialysis fed-batch process is given in Fig. 1 [19].
Fig. 1.

Scheme of a dialysis fed-batch process with "nutrient-split"-feeding strategy [20]

1.2 Adaptive model-based fed-batch control

Existing control concepts in fed-batch cell culture are still not sufficiently developed. The transfer of available control strategies established in bioprocess engineering is restricted due to changes in cell metabolism during cultivation, between similar cultivations, between different cell lines and the complexity of the cell metabolism itself. Up to now, there is no sufficient knowledge on the metabolism of animal cells available, so that appropriate mathematical models can be set up. Nevertheless, the performance of existing models can be increased by the adaptation of the model parameters. In addition, the use of analytical equipment for on-line measurements of relevant variables is problematic because of high complexity, insufficient accuracy and the risk of contamination. Results of off-line measurements are available for control purposes only after some delay. A controller, which must be designed in order to achieve optimized process control, must account for these difficulties [21, 22]. They result in requirements towards adaptability and flexibility of the applied process control strategy. In addition, such a control strategy must show reproducibility and robustness for usage in approved processes.

The combination of a fed-batch-process and an open-loop-feedback-optimal (OLFO) control provides a new approach for cell culture process control, which couples an efficient cultivation concept to a capable process control strategy [23]. An advantage is the universal character, because this concept is not bound to a certain cell line or a certain type of reactor. The OLFO controller is an adaptive, model based, long-term predictive controller. Major elements of the OLFO controller are a process model [24, 25, 26], a model parameter identification and an optimization part. They permit adaptation of model parameters and prediction of the future process course as well as its optimization. Its basic structure is shown in Fig. 2. Various theoretical as well as experimental publications show the potential of this strategy [27, 28, 29].
Fig. 2.

Scheme of the elements of the adaptive, model-based process control strategy for animal cell fed-batch cultivation using the OLFO controller

1.3 Suppression of apoptosis by bcl-2 overexpression

One of the characteristics of animal cells that can limit productivity is the genetic program that induces cell death called apoptosis [30, 31]. Suboptimal conditions, especially at higher cell densities, lead to apoptosis and prevent further product formation of the cells. Therefore it is of commercial interest to investigate and suppress apoptosis [32, 33]. One way is the transfection of cell lines with vectors producing various proteins that regulate apoptosis. Bcl-2, the product of an oncogene, has been demonstrated to suppress apoptosis in cell lines [34]. The bcl-2 gene has also been inserted into the GS-NS0 cell line 6A1 (100)3 and renamed into 6A1 bcl-2 [35]. Cultivations of both cell lines and the effect of bcl-2 overexpression are presented in this paper.

2 Materials and methods

2.1 Cells and media

The NS0 6A1 (100)3 and NS0 6A1 bcl-2 cell lines, producing chimeric cB72.3 IgG monoclonal antibodies (MAb) for cancer therapy, were used in the experiments. The antibody is directed against an epitope on a tumour-associated antigen TAG-72 found on various human neoplasms. As a special characteristic created by transfecting the NS0 host with a vector that encodes the gene for glutamine synthetase, no glutamine is required in the culture medium. This simplifies the cultivation because monitoring and control of glutamine concentration is not necessary. The two cell lines differ by the overexpression of the bcl-2 gene in the NS0 6A1 bcl-2 cell line for suppression of programmed cell death (apoptosis). The two cell lines were made available through Prof. Dr. M. Al-Rubeai, Animal Cell Technology Group, University of Birmingham, UK.

The cells were routinely maintained in T25 flasks in a humidified incubator at 37°C in an atmosphere of 5% CO2. For reactor experiments they were passaged to T75 flasks, T175 flasks and roller bottles to reach a preculture volume of ~200 mL at ~2×106 viable cells mL−1 for reactor inoculation at ~3.5–5×105 viable cells mL−1.

Cultivations were carried out in serum-free Pro CHO 4 CDM medium (#12-029 Q, Biowhittaker, Belgium) containing 23 mmol L−1 glucose supplemented with 25 µM L-Methionine sulfoximine (Sigma, Germany). For cultivation of NS0 6A1 (100)3 cell line 1% penicillin/streptomycin (PAA Laboratories, Austria) was supplemented and for NS0 6A1 bcl-2 in flask and roller bottle cultures also 1 g L−1 G418 (Geneticin, Sigma, Germany).

For fed-batch mode, a feed containing 100 mM glucose in an amino acid/vitamin concentrate (referred to as 'nutrient concentrate' feed) was employed. The amino acid/vitamin concentrate matches a tenfold concentrated DMEM/Ham's F12 (1:1) medium without salts. The osmolality of the concentrate was 356 mOsmol kg−1.

2.2 Analytical methods

During cultivation, 3 mL samples were taken twice a day and analyzed for viable and total cell concentration, glucose, lactate, ammonia, antibody concentration, DNA content and osmolality. Viable and total cell concentrations were measured by haemocytometer counts using trypan blue exclusion. Glucose and lactate concentrations were determined by YSI-Analyzer 2700 (Yellow Springs Instruments, USA). Ammonia was measured using enzymatic test kits (Boehringer Mannheim, No. 1112732, Germany).

Antibody concentration was determined using an ELISA that measures assembled human IgG. This involved capture of assembled antibody in samples and standard onto a 96-well plate coated with goat anti-human IgG Fc. Bound antibody was revealed with goat anti-human kappa light chain linked to horseradish peroxidase and the chromogenic substrate tetramethylbenzidine. Color development was proportional to the concentration of antibody present in the sample when compared to a reference standard prepared from an antibody stock solution. All antibody samples were measured at the same time in one set of 6 ELISAs. DNA degradation was estimated using DNA material from cell culture supernatants and analyzed by capillary electrophoresis with an Agilent 2100 Bioanalyser [36]. Osmolality measurements were performed in Osmomat 030, Cryoscopic Osmometer (Gonotec, Germany).

2.3 Bioreactor cultivation

Fed-batch experiments were carried out in a 2-L polyamide-foil stirred bioreactor (Visual Safety Fermenter VSF 2000, Bioengineering, Switzerland, working volume at inoculation 1 L) and a membrane dialysis reactor (Bioengineering, Switzerland). The membrane dialysis reactor consists of an inner culture chamber (2 L, working volume at inoculation 1 L) integrated into an outer dialysis chamber (8 L, working volume 4 L), separated by a cylindrical Cuprophane dialysis membrane (cut-off: 10 kDa, Akzo Nobel, Germany). This enables low molecular metabolites to pass through the membrane according to the specific permeability and the concentration gradient between both chambers [19]. The set-up for the membrane dialysis reactor with process control and periphery for gas mixing and substrate-, DO- and pH-control is shown in Fig. 3. The set-up for the 2-L bioreactor is the same without periphery for the dialysis chamber and bubble aeration.
Fig. 3.

Set-up for the membrane dialysis reactor with process control and periphery for gas mixing and substrate-, DO- and pH-control

Process temperature was kept at 37°C. Culture volumes started at 1 L and reached final volumes up to 1.7 L (maximum filling volume). For the dialysis fed-batch experiments the culture medium was also used as dialysate at a volume of 4 L and was continuously perfused at a rate of 400 mL day−1.

Dissolved oxygen was maintained at 50% air saturation by surface aeration using a self-developed PID-controller implemented in the process control system (Sect 2.4). The reactor inlet gas was a mixture of oxygen, nitrogen and, when required, carbon dioxide for lowering pH, mixed in appropriate proportions by mass flow controllers. The volume-specific oxygen transfer coefficient ( k L a O2) in the 2-L stirred bioreactor was 1.5 h−1 at a volume of 0.9 L, determined by gas exchange measurement (N2 to O2). When oxygen content in the reactor inlet gas increased towards 100%, stirring speed was stepwise increased in order to improve surface aeration. Therefore, the cell suspension was agitated at 80–225 rpm (2 L bioreactor) and 100–200 rpm (membrane dialysis bioreactor) using rushton turbines. In the dialysis fed-batch experiments, the same gas mixture as for the culture chamber was used for bubble aeration in the dialysis chamber (stirring speed 250 rpm). This enables oxygen supply via the dialysis membrane. Nevertheless, the capacity of the provided oxygen transfer was reached at higher cell densities. Further increase of stirring speed to rates exceeding 200 rpm was avoided because of the risk of cell damage by shear stress. Additional bubble aeration of pure oxygen directly to the culture chamber now provided additional oxygen transfer. An on/off controller turned on bubble aeration for 20–40 s every minute if required to keep dissolved oxygen at 30% air saturation.

Culture pH was controlled to a lower limit of 7 by the addition of NaHCO3 (75 g L−1, osmolality 1408 mOsmol kg−1, Merck, Germany) directly to the cell suspension. The usage of 1-N NaOH (Merck, Germany) caused cell damage (for reasons not investigated). Supply of CO2 to the reactor inlet gas avoided an increase of the culture pH above ~7.5.

2.4 Process control system

The bioreactor control system and the fed-batch control program package for calculation of the feed profiles (including OLFO controller, process model, routines for model parameter identification and process optimization, Sect. 2.5.3) were installed on separate computers for safety reasons. Data exchange (e.g. process data) and transfer of calculated feed profiles was provided via TCP/IP (transmission control protocol/Internet protocol).

The bioreactor control system PMCS (PCS, Switzerland) is based on field bus technology (Interbus-S, Phoenix Contact, Germany) running under the Unix-compatible real time environment LynxOS 2.1 (Real-Time Systems, USA). Routines for controlling dissolved oxygen and running the peristaltic feed pump (SP-GLV, Meredos GmbH, Germany) were implemented as separate programs that exchange data with the process control system via shared memory. The pump was digitally controlled (on/off) at a fixed flow rate of 26 mL h−1 to manage the small flow rates. The feeding routine guarantees a minimum on-time of the pump of 4 s. Shorter on-times lead to decreased pump flow accuracy. Feedback of the pumped volume is given by a balance (providing an accuracy of 1 g, Meredos GmbH, Germany).

The control platform and data management work independent for safety reasons. Remote control is provided via TCP/IP protocol to separate computers.

2.5 Control of fed-batch and dialysis fed-batch cultivations by the adaptive, model-based open-loop feedback-optimal (OLFO) controller

A detailed description of the OLFO controller and its application to cell culture fed-batch control is presented in [23]. An overview of the important elements is given in the following.

2.5.1 The model

The model consists of 16 equations (Table 1) and of 14 variable model parameters (Table 2). The model equations can be separated into ten coupled differential equations (description of viable and total cell concentration and all major substrate and metabolite concentrations) and six mostly Monod-type equations (description of substrate uptake and metabolite production kinetics). Furthermore, the model includes a limiting substrate (LS), which is used to fit the death phase.
Table 1.

Model equations

Balances biophase

\( \frac{{{\text{d}}X_{v} }} {{{\text{d}}t}} = {\left( {\mu - \mu _{{\text{d}}} - \frac{{F_{{{\text{Glc}}}} }} {{V_{{\text{i}}} }}} \right)} \cdot X_{v} \)

\( \frac{{{\text{d}}X_{t} }} {{{\text{d}}t}} = \mu \cdot X_{v} - \frac{{F_{{{\text{Glc}}}} }} {{V_{{\text{i}}} }} \cdot X_{t} - K_{{{\text{Lys}}}} \cdot {\left( {X_{t} - X_{v} } \right)} \)

Balances liquid phase inner reactor chamber

\( \frac{{{\text{d}}c_{{{\text{Glc,i}}}} }} {{{\text{d}}t}} = P_{{{\text{Glc}}}} \cdot \frac{A} {{V_{{\text{i}}} }} \cdot {\left( {c_{{{\text{Glc,o}}}} - c_{{{\text{Glc,i}}}} } \right)} + \frac{{F_{{{\text{Glc}}}} }} {{V_{{\text{i}}} }}{\left( {c_{{{\text{Glc,F}}}} - c_{{{\text{Glc,i}}}} } \right)} - q_{{{\text{Glc}}}} \cdot X_{v} \)

\( \frac{{{\text{d}}c_{{{\text{Lac,i}}}} }} {{{\text{d}}t}} = P_{{{\text{Lac}}}} \cdot \frac{A} {{V_{{\text{i}}} }} \cdot {\left( {c_{{{\text{Lac,o}}}} - c_{{{\text{Lac,i}}}} } \right)} - \frac{{F_{{{\text{Glc}}}} }} {{V_{{\text{i}}} }} \cdot c_{{{\text{Lac,i}}}} + q_{{{\text{Lac}}}} \cdot X_{v} \)

\( \frac{{{\text{d}}c_{{{\text{Mab}}}} }} {{{\text{d}}t}} = - \frac{{F_{{{\text{Glc}}}} }} {{V_{{\text{i}}} }} \cdot c_{{{\text{Mab}}}} + q_{{{\text{Mab}}}} \cdot X_{v} \)

\( \frac{{{\text{d}}c_{{{\text{LS,i}}}} }} {{{\text{d}}t}} = P_{{{\text{LS}}}} \cdot \frac{A} {{V_{{\text{i}}} }} \cdot {\left( {c_{{{\text{LS,o}}}} - c_{{{\text{LS,i}}}} } \right)} - \frac{{F_{{{\text{Glc}}}} }} {{V_{{\text{i}}} }} \cdot c_{{{\text{LS,i}}}} - q_{{{\text{LS}}}} \cdot X_{v} \)

\( \frac{{{\text{d}}V_{{\text{i}}} }} {{{\text{d}}t}} = F_{{{\text{Glc}}}} \)

Balances liquid phase outer reactor chamber

\( \frac{{{\text{d}}c_{{{\text{Glc,o}}}} }} {{{\text{d}}t}} = - P_{{{\text{Glc}}}} \cdot \frac{A} {{V_{{\text{o}}} }} \cdot {\left( {c_{{{\text{Glc,o}}}} - c_{{{\text{Glc,i}}}} } \right)} + \frac{{F_{{\text{D}}} }} {{V_{{\text{o}}} }} \cdot {\left( {c_{{{\text{Glc,FD}}}} - c_{{{\text{Glc,o}}}} } \right)} \)

\( \frac{{{\text{d}}c_{{{\text{Lac,o}}}} }} {{{\text{d}}t}} = - P_{{{\text{Lac}}}} \cdot \frac{A} {{V_{{\text{o}}} }} \cdot {\left( {c_{{{\text{Lac,o}}}} - c_{{{\text{Lac,i}}}} } \right)} + \frac{{F_{{\text{D}}} }} {{V_{{\text{o}}} }} \cdot {\left( {c_{{{\text{Lac,FD}}}} - c_{{{\text{Lac,o}}}} } \right)} \)

\( \frac{{{\text{d}}c_{{{\text{LS,o}}}} }} {{{\text{d}}t}} = - P_{{{\text{LS}}}} \cdot \frac{A} {{V_{{\text{o}}} }} \cdot {\left( {c_{{{\text{LS,o}}}} - c_{{{\text{LS,i}}}} } \right)} + \frac{{F_{{\text{D}}} }} {{V_{{\text{o}}} }} \cdot {\left( {c_{{{\text{LS,FD}}}} - c_{{{\text{LS,o}}}} } \right)} \)

Growth kinetics

\( \mu = \mu _{{\max }} \cdot \frac{{c_{{{\text{Glc,i}}}} }} {{c_{{{\text{Glc,i}}}} + K_{{{\text{S,Glc}}}} }} \cdot \frac{{c_{{{\text{LS,i}}}} }} {{c_{{{\text{LS,i}}}} + K_{{{\text{S,LS}}}} }} \)

Death kinetics

\( \mu _{{\text{d}}} = \mu _{{{\text{d,min}}}} + K_{{\text{d}}} \cdot \frac{{k_{{{\text{d,LS}}}} }} {{c_{{{\text{LS,i}}}} + k_{{{\text{d,LS}}}} }} \)

Substrate uptake-/metabolite production kinetics

\( q_{{{\text{Glc}}}} = q_{{{\text{Glc,max}}}} \cdot \frac{{c_{{{\text{Glc,i}}}} }} {{c_{{{\text{Glc,i}}}} + k_{{{\text{Glc}}}} }} \cdot \frac{{c_{{{\text{LS,i}}}} }} {{c_{{{\text{LS,i}}}} + k_{{{\text{LS}}}} }} \)

\( q_{{{\text{Lac}}}} = Y_{{{\text{Lac/Glc}}}} \cdot q_{{{\text{Glc}}}} - q_{{{\text{Lac,max}}}} \cdot \frac{{c_{{{\text{Lac,i}}}} }} {{c_{{{\text{Lac,i}}}} + k_{{{\text{Lac}}}} }} \cdot \frac{\mu } {{\mu _{{\max }} }} \)

\( q_{{{\text{LS}}}} = q_{{{\text{LS,max}}}} \cdot \frac{{c_{{{\text{LS,i}}}} }} {{c_{{{\text{LS,i}}}} + k_{{{\text{LS}}}} }} \)

\( q_{{{\text{Mab}}}} = q_{{{\text{Mab,max}}}} \)

Table 2.

Overview of model parameters. During parameter identification they are either allowed to be identified based on the initial values (3rd column) (plus/minus bounds are given for each parameter (not listed) to prevent a too drastic variation of the parameters) or set as constant (4th column)

Parameter

Units

Initial value before parameter identification

Constant value

µ d,min

h−1

0.003

µ max

h−1

0.028

K d

h−1

0.045

k d,LS

mmol L−1

0.008

k Glc

mmol L−1

0.4

k Lac

mmol L−1

1.4

k LS

mmol L−1

0.15

K Lys

h−1

0.028

K S,Glc

mmol L−1

0.021

K S,LS

mmol L−1

0.018

q Glc, max

10−9×mmol cell−1 h−1

0.22

q Lac, max

10−9×mmol cell−1 h−1

0.052

q LS, max

10−9×mmol cell−1 h−1

0.029

q MAb, max

10−9×mg cell−1 h−1

1.25

Y Lac/Glc

1.67

Additional parameters for dialysis cultivation:

   A

dm2

5.655

   c Glc,FD

mmol L−1

22.7

   c Lac,FD

mmol L−1

0.156

   c LS,FD

mmol L−1

8

   F D

L h−1

0.0167

   P Glc

dm h−1

0.09

   P Lac

dm h−1

0.1176

   P LS

dm h−1

0.12

   V o

L

4

If no dialysis is applied, the terms describing mass transfer over the dialysis membrane and perfusion of the dialysis chamber can be neglected. The underlying hypotheses of the model are described in [23]. The model can be applied to different cell lines, e.g. a hybridoma cell line as presented in [23]. Equations describing glutamine/ammonia metabolism are then analogous to the ones describing glucose/lactate metabolism [37].

Based on this model, the courses of viable cell, total cell, glucose, lactate and antibody concentration as well as the culture volume can be described. The courses of glucose and lactate concentration can also be calculated for the dialysis chamber. This allows the determination of a suitable perfusion rate for the dialysate in order to prevent lactate inhibition.

2.5.2 Process control loop

The OLFO controller includes routines for model parameter identification, optimization of the future process course and solving of nonlinear differential equations. A user interface has been developed, which guides the user through the control loop shown in Fig. 2. Files for the input of process measurements (viable cell, total cell, glucose and lactate concentration) are automatically opened. An initial measurement or estimation of antibody concentration is only entered at the beginning of the experiment. Using the process model and the set of parameter starting values, the course of important process states from the end of the lag phase to the actual process time is calculated. Subsequently, an automated parameter identification takes place, based on the available off-line measured process states. Then, the optimization part predicts the future course of process states using the model and its adapted parameters. It calculates a feed profile based on an optimization criterion (e.g. controlling glucose concentration at a certain level, maximizing time–space–yield). Finally, the calculated feed profile is transferred to the process control system. The control cycle is repeated within a chosen time interval after having taken the next sample.

2.5.3 Model parameter identification and feed optimization algorithms

The differential model equations are solved by the fourth-order Runge–Kutta algorithm [38] with variable step size. The model parameter identification part and feed optimization part is done by the Nelder–Mead algorithm [38]. The Nelder–Mead algorithm minimizes a performance function that includes weighted least squares [23]. To improve the fit of certain data sets, their data can be additionally weighted using a weighting factor [23]. In the experiment presented, all measurement data has been weighted by the factor 1 except for viable cell and glucose concentration. In order to obtain an especially good fit for these data, a weighting factor of 100 has been chosen. In addition, the latest measurements of viable cell and glucose concentration have been weighted by a factor of 200. A deviation between the fit of viable cell and glucose concentration and the latest measurements would influence the prediction of the future process course and could decrease the quality of future process control.

During the model parameter identification part, input to the Nelder–Mead algorithm are experimental data, the model (Table 1), the set of model parameters (Table 2) and the identification time interval. Output is an identified set of model parameters that gives an optimized model fit of the experimental data. During the feed optimization part, input to the Nelder–Mead algorithm are artificially preset data for optimization instead of experimental data, the model, the set of previously identified parameters, a feed flow starting value [23] and the optimization time interval. Output is an optimized feed profile that best fulfills the optimization criterion.

3 Results and discussion

Two fed-batch and two dialysis fed-batch cultivations of the NS0 cell lines 6A1 (100)3 (control cell line) and 6A1 bcl-2 (overexpressing bcl-2) are presented. They have all been controlled by the adaptive model-based open-loop-feedback-optimal (OLFO) control. Therefore, the performance of the control strategy is illustrated at first on the basis of one cultivation. Second, the effects of culture mode (fed-batch and dialysis fed-batch) and bcl-2 overexpression on cell yield and antibody production are discussed separately. All cultivations were controlled at a glucose target concentration of 16 mmol L−1. The aim was to investigate the range of cell-specific glucose uptake rate at this glucose level and the shift of cell-specific glucose uptake during cultivation. More sophisticated optimization criteria such as optimizing the time–space yield were not applied.

3.1 Performance of the adaptive model-based OLFO controller

After inoculation, the processes were run in batch mode during the lag phase, until measurements of viable cell concentration indicated the beginning of the exponential growth phase (Fig. 4). At this point, the OLFO controller was started, using the two previous measurement points of viable cell, total cell, glucose and lactate concentration. During the OLFO-controlled course of the experiment (from 48 h to finish), the model parameters were adjusted once or twice a day after having measured viable cell, total cell, glucose and lactate concentrations off-line. Since the model was not designed to describe the lag phase, the model parameters were identified from the complete data available during the cultivation excluding the lag phase. Using the criteria of keeping glucose concentration at 16 mmol L−1, the controller allowed the glucose concentration to drop to 16 mmol L−1 from its starting concentration during the first part of the cultivation. From hour 100 on, the controller calculated a feed rate of nutrient concentrate feed in order to keep the glucose concentration at 16 mmol L−1. The feed rate was calculated until the point of time at which the next sample should be taken. Within this prediction horizon, the predicted courses of all model variables were also calculated.

Presented in Fig. 4 are the predicted and measured courses of viable cell, glucose and lactate concentration in the inner reactor chamber, as well as the feed profile of the dialysis fed-batch experiment of the NS0 6A1 bcl-2 cell line over cultivation time. Each predicted concentration results from the model parameter optimization and feed optimization, part, which was carried out at the time of the previous measurement.
Fig. 4.

Dialysis fed-batch cultivation of NS0 cell line 6A1 bcl-2 via OLFO controller. Comparison of measured and predicted viable cell, glucose and lactate concentrations in inner reactor chamber over cultivation time as well as nutrient concentrate feed rate

The OLFO controller predicted the viable cell concentration with an average accuracy of ~8% during the growth phase. The accuracy was calculated according to the following equation:
$$ {\text{accuracy}} = \frac{{{\left| {{\text{measurement}} - {\text{prediction}}} \right|}}} {{{\text{measurement}}}} \cdot 100\% $$
The one value of the death phase showed a deviation of ~200%. The reason is the used model, which can only predict the death phase on the basis of parameters (index LS) obtained from a previous experiment. The accuracy of this prediction depends on the similarity of the cultivations. Nevertheless, the model can fit the death phase as soon as it started. Therefore one deviated value in viable cell concentration is needed in order to adapt the model to the occurring cell death.

The controller predicted the glucose level with an average accuracy of ~5%. During the phase of the cultivation, where the controller not only predicted glucose concentration but also calculated the feed rate (from hour 100 on), the average accuracy was ~9% and showed ~25% overfeeding at the beginning of the death phase. Lactate concentration was predicted with an average accuracy of ~35%. Accuracy of lactate is lower than that of glucose because glucose measurement data was weighted higher during parameter identification (Sect. 2.5), resulting in a better prediction and control of glucose concentration to the disadvantage of lactate concentration. State variables like total cell and lactate concentration could also be predicted as well as states of the dialysis chamber (data not included).

3.2 Effect of culture mode (fed-batch and dialysis fed-batch) and bcl-2 overexpression

The courses of viable cell density and viability over cultivation time of all four experiments are shown in Fig. 5.
Fig. 5.

Comparison of viable cell concentration and viability in fed-batch and dialysis fed-batch cultivations of NS0 cell lines 6A1 (100)3 and 6A1 bcl-2

The 6A1 bcl-2 cell line shows a prolonged growth phase and a higher maximum cell density. In addition, cell death during the stationary/apoptotic phase does not occur as quickly as for the control cell line 6A1 (100)3. The 6A1 (100)3 cell line reached a maximum cell density of 1.5×106 cells mL−1 (fed-batch) and 1.1×107 cells mL−1 (dialysis fed-batch) in comparison to the bcl-2 cell line with a maximum cell density of 3×106 cells mL−1 and 1.6×107 cells mL−1 respectively.

This means a higher maximum viable cell density of 6A1 bcl-2 cells in comparison to 6A1 (100)3 cells in fed-batch cultivation (factor 2), but cell densities of the same magnitude in dialysis fed-batch cultivation. Dialysis fed-batch cultivation increases maximum cell density by a factor of ~5–7, in comparison to fed-batch cultivation. This could indicate that one or more substances that limit or inhibit cell growth in fed-batch cultivation are supplied/removed by the dialysis. Lactate concentration did not exceed 17 mmol L−1 in all experiments and did not cause any inhibition.

Nevertheless, a reason for cell death could not be detected in both cultivation systems. Ammonia measurements (not presented) indicated low concentrations (<0.5 mmol L−1), not causing any inhibition. Osmolality (not presented) stayed in the usual range around 330 mOsmol kg−1, which is recommended for cell proliferation.

Viability of the cell line 6A1 bcl-2 ranges at 90–95% during growth phase and at 80–85% for the 6A1 (100)3 cell line. Similar results were obtained in batch cultures in T-flasks. This is due to a higher death rate of the 6A1 (100)3 cells in comparison to the 6A1 bcl-2 cells. In addition, the effect of dialysis can be seen again in maintaining the viability for a prolonged time.

An interesting matter is whether the previously described differences between the cell lines could be due to bcl-2 overexpression. Figure 6 shows the DNA concentration of the dialysis fed-batch experiment of each cell line in supernatant samples except for the last sample.
Fig. 6.

DNA concentration profiles in supernatant samples in dialysis fed-batch cultivations of NS0 cell lines 6A1 (100)3 and 6A1 bcl-2

DNA concentration in the medium is related to cell death because DNA is released into the medium by cell lysis. For the 6A1 (100)3 cell line, DNA concentration is continuously increasing during cultivation, especially the 180 base-pair oligomer DNA fragments characteristic of apoptosis. During the death phase of the cultivation, the DNA laddering typical for apoptosis is also visible at this resolution. For the bcl-2 cell line, DNA concentration is not visible during the whole experiment except for some single bands. Even during the death phase no bands appear, showing no apparent apoptosis. This indicates that the membrane integrity of the cells stays intact. DNA results from the fed-batch experiments are similar (not presented). Therefore bcl-2 overexpression in the bcl-2 cell line could be the reason for the positive differences between the cell lines. Besides, bcl-2 overexpression leads to a cleaner product or easier downstream processing, respectively.

Courses of antibody concentration in the medium over cultivation time are shown in Fig. 7 for all experiments. The antibodies produced are intact, as shown by electrophoresis using protein chip analysis (data not presented).
Fig. 7.

Comparison of antibody concentration in fed-batch and dialysis fed-batch cultivations of NS0 cell lines 6A1 (100)3 and 6A1 bcl-2

Antibody concentration increases continuously during the first ~80 h. In combination with similar cell densities during this phase (Fig. 5), cell-specific antibody production of the two cell lines is within the same range of 1.2 10−9 mg cell−1 h−1 (results not presented). Differences in antibody concentration arise at the end of the experiments. The 6A1 (100)3 cell line reached a maximum MAb concentration of 170 mg L−1 (fed-batch) and 1000 mg L−1 (dialysis fed-batch) in comparison to the bcl-2 cell line with a maximum MAb concentration of 340 mg L−1 and 1030 mg L−1, respectively.

In fed-batch experiments, the prolonged growth phase and the higher maximum cell densities of the bcl-2 cells yield higher product formation. Product formation still takes place during the stationary phase in spite of missing cell growth. At the high cell densities during the stationary phase, the bcl-2 cell line still produces antibody, whereas the cells of the 6A1 (100)3 cell line already undergo apoptosis.

In dialysis fed-batch experiments, both cell lines reach similar antibody concentrations. The increase of maximum viable cell density and the duration of the experiment of the bcl-2 cell line in comparison to the 6A1 (100)3 cell line seems to be not high enough to cause significant differences in antibody concentration.

The advantage of cell cultivation using the adaptive model-based control can be seen in comparison to results obtained by Perani et al. (at Lonza Biologics, Slough, UK) with the same cell lines [39]. Fed-batch cultivation in a 10-L airlift bioreactor under optimized conditions using an optimized feed yielded in a maximum viable cell concentration of 1×107 cells mL−1 (cell line 6A1 bcl-2) and 6×106 cells mL−1 (cell line 6A1 (100)3). In case of antibody concentration, Perani et al. obtained 500 mg L−1 (6A1 bcl-2) and almost 600 mg L−1 (6A1 (100)3). The adaptive model-based control achieved higher concentrations in dialysis fed-batch mode and lower concentrations in fed-batch mode, but without any optimization of medium, feed and culture conditions. This shows that the control is very suitable for obtaining good results with new cell lines in a short period of time.

4 Conclusion

Experimental results show the improvement of maximum cell density and antibody production of the two NS0 cell lines by culture mode (fed-batch and dialysis fed-batch) and suppression of apoptosis. On the one hand, the fed-batch cultivation using dialysis increases maximum cell density and product accumulation by a factor of 3–7 in comparison to cultivation without dialysis. On the other hand, the effect of the successful suppression of apoptosis by bcl-2 overexpression in the one NS0 cell line also gives positive results in contrast to the control cell line. In fed-batch cultivation, the NS0 6A1 bcl-2 cells show a prolonged growth phase and a higher viability and maximum cell density. These factors yield in higher product formation. In both fed-batch and dialysis fed-batch cultivations, bcl-2 overexpression leads to a cleaner product or easier downstream processing, respectively. Finally, the adaptive model-based OLFO controller, developed as a general tool for cell culture fed-batch processes, could control the fed-batch and dialysis fed-batch cultivations of both cell lines.

Acknowledgements

This project "Suppression of programmed cell death in industrial scale biological production systems", contract number QLK3-CT-2000-00076, is kindly financed by the European Union (framework V).

Copyright information

© Springer-Verlag 2003

Authors and Affiliations

  1. 1.Bioprozess- und BioverfahrenstechnikTechnische Universität Hamburg-HarburgHamburgGermany
  2. 2.Institut für Technologie und BiosystemtechnikBundesforschungsanstalt für LandwirtschaftBraunschweigGermany

Personalised recommendations