Contribution to the development of methods and systems for the automatization during the early stages of bioprocess development

Departament d'Enginyeria de Sistemes, Automàtica i Informàtica Industrial Contribution to the development of methods and systems for the automatization during the early stages of Bioprocess development. Thesis presented for the qualification of Ph.D. Program: Biomedical Engineering Author: Andreu Francesc Fontova i Sosa Director: Ramon Bragós i Bardia. Barcelona 2015 Agraïments. El conjunt dels reptes i les dificultat afrontades durant el desenvolupament d’aquest treball no es restringeixen només al context de la tesi doctoral, sinó que han format part del dia a dia d’una aventura empresarial anomenada HEXASCREEN CULTURE TECHNOLOGIES s. l. amb l’ambiciós objectiu de desenvolupar i portar al mercat un producte tecnològicament avançat i capaç de donar solucions a un dels sectors industrials més exigents del món, la Biotecnologia mèdica i farmacèutica. Malgrat que el projecte empresarial no ha tingut l’èxit esperat, cal posar en valor tant la feina com l’actitud amb la que els treballadors i alguns socis de la companyia van recolzar el projecte i que va permetre no només fer realitat el producte sinó també aconseguir èxits remarcables com ara la col·laboració amb el Biotechnology Process Engineering Center del Massachusets Institute of Technology i la reconversió del producte pel cultiu de cèl·lules mare embrionàries humanes per a la University of Calgary. Agrair també l’ajuda sempre incondicional per part dels grups de recerca d’Enginyeria Cel·lular i Tissular de la Universitat Autònoma de Barcelona i de Instrumentació i Enginyeria Biomèdica de la Universitat Politècnica de Catalunya. Així com en especial al meu director de tesi, el Dr. Ramon Bragós qui a més de proporcionar un assessorament crític i didàctic ha estat sempre un exemple a seguir per la seva talla professional i humana. Finalment adreço el meu agraïment més profund a la meva germana Laura i el meus pares Núria i Andreu, els quals sempre m’han recolzat, tant en els moments d’èxit com els més difícils. 1 Contents: 1. ABSTRACT 3 2. INTRODUCTION 7 2.1. METABOLISM AND KINETICS OF MICROBIAL GROWTH 9 2.2. STIRRED TANK BIOREACTORS 11 2.3. SHEAR STRESS 16 2.4. MASS TRANFER 19 2.5. OBJECTIVES 24 3. STATE OF THE ART 27 3.1. MINIBIOREACTOR SYSTEMS 28 3.2. DISSOLVED OXYGEN MEASUREMENT BY MEANS OF FLUORESCENCE QUENCHING 37 3.2.1.TIME DOMAIN METHODS (Lifetime) 39 3.2.2.FREQUENCY DOMAIN METHODS (Phase shift) 42 3.2.3.RATIOMETRIC METHODS 46 3.3. O.U.R. ESTIMATION METHODS 47 3.3.1.GLOBAL MASS BALANCE METHOD 47 3.3.2.STATIONARY LIQUID PHASE MASS BALANCE METHOD 48 3.3.3.DYNAMIC METHOD 50 4. DEVELOPMENT OF SINGLE USE BIOREACTORS 58 4.1. THE HEXASCREEN BIOREACTOR 59 4.1.1.SYSTEM OVERVIEW 59 4.1.2.HEXASCREEN ARCHITECTURE 63 4.1.3.HEXASCREEN OPTICAL LAYOUT 69 4.1.3.1. OPTICAL ABSORBANCE SPECTROSCOPY MEASUREMENT (pH & OD) 69 4.1.3.2. FLUORESCENCE MEASUREMENT (DO) 78 4.1.4.MINIBIOREACTOR STIRRING, AEREATION & MASS TRANSFER 79 4.1.5.EXPERIMENTATION WORKFLOW & RESULTS 83 4.2. THE MONOSCREEN FED-BATCH SYSTEM 89 4.2.1.SYSTEM OVERVIEW 90 4.2.2.MONOSCREEN FED-BATCH ARCHITECTURE 94 4.2.3.MONOSCREEN FED-BATCH OPTICAL LAYOUT 103 4.2.3.1. NIR LASER TURBIDIMETRY (Cell density) 107 4.2.3.2. FLUORESCENCE MEASUREMENTS (DO & pH) 108 4.2.4.MINIBIOREACTOR LIQUID HANDLING (Medium addition & sampling) 110 4.2.5.MINIBIOREACTOR STIRRING, AEREATION & MASS TRANSFER 114 4.2.6.EXPERIMENTATION WORKFLOW & RESULTS 116 2 4.3. THE BIOSTAT B-PLUS BENCH-SCALE BIOREACTOR & AERATION TEST SETUP 120 4.3.1.DESCRIPTION OF THE TEST SETUP 120 4.3.2.BIOREACTOR AEREATION CONTROL & MASS TRANFER 123 5. SIMPLIFIED IMPLEMENTATION OF THE OUR STATIONARY LIQUID MASS BALANCE METHOD 128 5.1. METHOD DESCRIPTION AND MODELIZATION 129 5.2. SIMULATIONS 135 5.3. EXPERIMENTAL RESULTS 141 6. CONCLUSIONS & WORK IN PROGRESS 149 6.1. CONCLUSIONS 149 6.2. WORK IN PROGRESS 151 7. PAPERS, PATENTS AND CONFERENCE CONTRIBUTIONS 155 Abstract 1 3 Abstract. The present dissertation deals about some of the tools and methods used in Biomedicine and Biotechnology to discover new innovative therapeutic agents. Whose development and production still involve an important amount of manual labour. Currently, such fields are considered as a zone of frontier in science [1]. Therefore, it exists an urgent need for more efficient and reproducible manners to carry out experiments and get results. In this direction bioreactors and fermenters are being used not just for production but for testing the potentiality of some cell species to produce such therapeutic agents. The most widespread concept of a bioreactor is made of a vessel connected to a control unit by means of an ensemble of probes and actuators with the aim of monitoring the growth of some cell specie (the biocatalyst), its metabolic activity, and controlling some important physical and chemical conditions, for instance: temperature, pH, pO2… In order to make such systems useful for the early bioprocess development stages (Cell screening, Clone optimisation, and Process optimisation) engineers and biotechnologists have the challenge of designing the processes keeping in mind the following requirements: Abstract 1 4 - Capability to perform an statistically significant number of experiments (High Throughput Screening), - Capability to produce results being representative of the future production stages (Scalability). - Capability to produce reliable results at low cost. That’s why during the last decade an important scientific and commercial interest on miniaturised disposable cell culture systems has risen. Different approaches have been developed [2] [3] [4]. Some of them focused on increasing the experimental throughput by reducing the culture volume and others on offering bigger volumes but including stunning automation and monitorization features. Probably it doesn’t still exist an optimum compromise between the experimental throughput and how bio-process significant the results are for every application. Therefore, the right measurement and control methods shall be chosen depending on the type of cells cultured and the type and size of the bioreactor used. Figure 1 Trade-off between the data throughput and how bio-process significant the data are. From D. Doig S., I. Betts J. et al. [5] [6] The work here presented aims to be a contribution to the state of the art on cell culture monitoring and control techniques applied to the bioprocess development taking into account the requirements mentioned above. Specifically about the design and construction of instrumentation for the estimation of the Oxygen Uptake Rate (OUR) on miniaturized disposable bioreactors. OUR has been described as a key variable for tracking and monitoring the metabolic activity in animal cell culture [7][8]. This work proposal fits within this framework and tries to demonstrate the feasibility of a new method for the continuous estimation of the OUR. The proposed method is based on the accurate control of the oxygen concentration by means of Pulse Width Modulated (PWM) valves and the use of internal control loop variables to estimate the OUR. The method claims for a cheaper, continuous and accurate estimation, as well as being free of cell stress due to strong changes in the medium’s oxygen concentration, as happens with the Dynamic Method, being this the most common technique. Abstract 1 5 To that end, after the study of the background and the system modelling and simulation three different testing platforms were built and every technical mean required to carry out the experimentation was developed. The first and the second testing platforms are experimental prototypes named as HexaScreen® Hexa-Batch and MonoScreen® Fed-Batch; they are disposable Minibioreactors (MBR’s) manufactured by HEXASCREEN CULTURE TECHNOLOGIES S.L., the third platform is based on a Biostat® Bplus Bench-Scale bioreactor by SARTORIOUS AG. Such developments required some additional effort to design instruments and algorithms not just for measuring and control the Dissolved Oxygen (DO), but the pH, Temperature, Cell Concentration, and a wide set of means to allow the cell growth (such design will be specifically introduced in chapter 4. Therefore, although the main topics under focus are the OUR estimation and the Dissolved Oxygen (DO) measurement, due to the fact that the cultivation of cell species requires a number of considerations besides the oxygen consumption, the general scope of the present dissertation ranges from the state of the art in Minibioreactor design to different techniques applied for measure and control, as well as for monitoring the metabolic activity of the cell species. A number of original hardware and instrumental contributions will be introduced. Below some of the main achievements: - Development and construction of two state-of-the-art Minibioreactor platforms useful to essay the OUR estimation and DO measurement methods under study. - Development and construction of a low cost, state-of-art. DO measurement instrument to perform a highly accurate control of the medium’s DO concentration. - Demonstration of the feasibility of a new method for the continuous estimation of the OUR, based on the accurate control of the DO concentration and especially suited for animal cell culture. Therefore, the following chapters introduce a theoretical approach to a new method for the estimation of the OUR, which has been demonstrated by means of the experiments carried out with three different types of bioreactors, being two of them, as will be shown, an innovative embodiment within the field of disposable bioreactors. References: [1] M. S. F. M. S. G. Audrey R. Chapman, “Stem Cell Research and Applications Monitoring the Frontiers of Biomedical Research,” American Association for the Advancement of Science & Institute for Civil Society, 1999. [2] V. Glasser, “Disposable Bioreactors gaining favor,” Genetic Engineering and Biotechnology News, vol. 26, no. 12, 2006. [3] V. Glasser, “Bioreactor and Fermentor market trends,” Genetic Engineering and Abstract 1 6 Biotechnology News, vol. 29, no. 14, 2004. [4] V. Glaser, “Bioreactor and Fermentor Market trends,” Genetic Engineering & Biotechnology News, vol. 29, no. 19, 2009. [5] B. F. J. L. G. D. Doig S, “High-Throughput screening and process optimization,” in Basic Biotechnology, Cambridge, 2006, pp. 289-305. [6] B. F. I Betts J., «Miniature Bioreactors: Current practices and future opportunities,» Microbial Cell Factories, p. 5:21, 2006. [7] E. G. V. E. S. J. C. M. Felix Garcia-Ochoa, “Oxygen uptake rate in microbial processes: An overview,” Biochemical Engineering Journal, vol. 49, pp. 289-307, 2010. [8] U. v. S. I. W. M. Pierre-Alain Ruffieux, “Measurement of volumetric (OUR) and determination of specific (qO2) oxygen uptake rates in animal cell cultures,” Journal of biotechnology, vol. 63, no. 2, pp. 85-95, 1998. Introduction 2 7 Introduction. The fundamentals of the in vitro tissue and animal cell culture techniques were developed in the late nineteenth and early twentieth centuries and their basis are still valid (W. Roux 1885, Arnold 1887, Ljunggren and L. Loeb 1897, R. Harrison 1907, Carrel A. Burrows 1911 and 1912) [1].These essentially consist in placing the cells inside some type of container (Petri dishes , test tubes, Carrel bottles, T-flasks ...) along with an appropriate culture medium, typically a buffered solution of amino acids, vitamins, and other nutrients, all incubated in sterile conditions in an atmosphere of 5 % CO2and warmed at 37 °C. This ensures that the cell line will grow continuously, even being able to express its phenotype. In some cases the surface of the container allows the cells to attach forming a monolayer growth. Using this procedure, Alexis Carrel was able to keep a strain of fibroblasts obtained from the heart of chicken embryos in active multiplication over twenty-five years [2]. The evolution of these techniques led to the concept of Bioreactor, where the most widespread embodiment is the stirred tank, which is made of a sterile vessel featuring some sort of turbine for mixing, a gas supply and the culture medium. Once the Bioreactor is seeded with a certain initial concentration of cells, they will keep growing attached to the vessel walls or in suspension Introduction 2 14 parameter affecting the flow topology. Rotating speed, Density and viscosity of the medium, and bubbles coalescence are additional variables affecting the flow’s behavior. This is usually represented through the Power number Npand the Reynolds number Reifor a given geometry, medium’s rheology and stirring conditions. Figure 4 A) Unbaffled vessel with paddle/bar for Embryonic Stem Cell, B) Baffled vessel with marine propeller or pitch blade turbine for productive animal cells like CHO, C) Baffled vessel with Rushton turbine for microbial culture. Adapted from [13] [14] [15] The Reynolds number Reiis used to predict the Laminar or Turbulent behaviour within some fluid. For instance, a non aereated Newtonian medium being stirred within some of the bioreactor configurations represented above will typically show a Laminar flow for Rei< 10, and a Turbulent flow for Rei> 10,000. Anything in between is considered as a Transition regime and both types of flow can be found within the volume. =∙ ∙ Where: •r[kq/m3] : Fluid’s density •[kg/m·s]: Fluid’s viscosity •Di[m]: Impeller diameter •Ni[rps]: Stirrer speed The power number Npregards only to the mechanical energy applied to the medium within the bioreactor; A difference must be made between the actual energy applied to the fluid and the total electrical power consumed by the stirrer motor, since the bioreactor’s Introduction 2 15 power input Pis not taking into account the motor efficiency nor the loss of energy due the mechanical parts (gearbox, seal rings, etc…). =∙ ∙ =2∙ ∙ ∙ ∙ ∙ Where: •P[W/m3]: Power input •M[N·m]: Induced torque due to the impeller’s friction and the drag of the fluid. •[kq/m3] : Fluid’s density •Di[m]: Impeller diameter •Ni[rps]: Stirrer speed The Np-Reiratio it has been determined experimentally for different types of impellers and bioreactor configurations. The figure below displays such relationship for the bioreactor configurations shown in figure 3 and non-aereated Newtonian fluids. It can be observed how the power number Nptends to become constant within the turbulent regime and lower than under laminar flow conditions. Next section will explain how this is related to the shear stress. Figure 5 Np-Reiexperimental correlations for: Rushton turbines (1); downward pumping Pitch-blade turbines (2); and Marine propellers (3) in baffled bioreactors and non-aereated Newtonian fluids. From [13] Despite that the previous expressions have been defined just for non-aerated Newtonian fluids, they can be used for a rough design of the bioreactor mixing system. However, aeration is usually a key point for most of purposes in biotechnology since the use of non-aereated systems is commonly restricted to research applications with very shear sensitive cell species. Therefore, the calculation of the power input Pneeds to take into account the reduction of the fluid’s density due to the presence of small bubbles; this is made through empiric correlations obtained for every bioreactor configuration depending on the gas volume flowing through the bioreactor, the following equation is defined for Rushton turbines in baffled vessels [6] [16]: =0.783∙ ∙ ∙ .. Introduction 2 16 Where: •P0[W/m3]: Power input under unaerated conditions •Q[m3/s]: Volumetric aeration rate •Di[m]: Impeller diameter •Ni[rps]: Stirrer speed For a more accurate assessment on the physical phenomena affecting the mixing capability in aereated systems, as well for non-Newtonian fluids, a further detailed and deeper explanation is provided by references [6] [7] [13] [16]. 2.3 Shear Stress After the previous introduction on the basis of stirred tank bioreactors design, in this section the most common methodology for estimating the shear stress in stirred tank bioreactors will be presented. The shear stress term is used as an ambiguous concept regarding any hydrodynamic conditions leading to the cell death or a slowing down of the cell growth or product synthesis. This may happens due to different reasons when cells are brought under a gradient of forces produced by spatial differences in the fluid’s velocity; such situation is not significantly harmful when all forces are applied in the same direction (laminar flow). In this case, cells are trapped within the flow in a rotational movement induced by the differences in the fluid’s velocity. However, when forces are distributed in different directions (shear forces) the possibility of cell damaging is dramatically increased, this happens for high stirring and gas sparging rates (Turbulent flow). In stirred tank bioreactors, the Turbulent regime is featured by the chaotic formation of eddies around the impeller and baffles, the length of the fluid eddies depends on the impeller’s geometry, its size and the rotational speed. The faster the stirring is the shorter eddies are. Figure 6 Freely suspended particles: (A) Under laminar flow conditions; (B) Under turbulent flow conditions A straightforward method for estimating the shear stress is by comparison of the mean length of the fluid eddies and the diameter of the biocatalyst [6] [13]. The bigger the biocatalyst is the more harmful eddies are, that’s why small organism like bacteria and yeast, besides of their protective cell wall, are so insensitive to the high turbulences produced by fermenters, since they become easily trapped by the eddies without experiencing a significant Introduction 2 17 shear stress. On the other hand, bigger cell species like animal, plant and fungi cells are definitely more sensitive to the distribution of the shear forces within the fluid. Those lines which are not adapted to grow in suspension demand special care since they will grow forming relatively big cell flocks attached on microcarrier beads or embryonic bodies which can be easily damaged by bead to bead collisions and small eddies. Figure 7 (A) Cell flocks on microcarriers [17]; (B) Embryonic stem cells [18]. The mean length of the fluid eddies can be calculated through the Kolmogorov scale if the type of flow can be considered as isotropically turbulent: = Where: •[m2/s]: Kinematic viscosity of the fluid. •E[m2/s3]: Turbulent energy dissipated. In order to avoid an underestimation of the dissipated energy it’s important to decide whether the turbulent regime is isotropically distributed along the whole volume or just around the impeller, according to the size of the bioreactor: Turbulence is dissipated isotropically along the whole volume: =∙ Turbulence is concentred around the impeller: =∙ Where: •P[W/m3]: Aereated or unaerated power input calculated as explained in the previous section. •[kq/m3] : Fluid’s density. •Di[m]: Impeller diameter. •Vl[m3]: Total volume of the fluid. In Stirred tank bioreactors the shear rate can be calculated through the following equation [6]: Introduction 2 18 = ∙ 4∙ 3∙ +1 ∙ Where: •n[]: Flow index of the fluid, equals 1.0 for all Newtonian liquids which are any sort of aqueous solutions and most of culture media. •Ni[rps]: Stirrer speed •ki[m]: Parameter depending on the impeller type, Typical values are: o11-13 for six-bladed disk turbines o10-13 for paddle impellers o~ 10 for propellers o~ 30 for helical ribbon impellers The shear rate can be converted to a parameter known as the shear stress through the following expression: = ∙ Where: •[kg/m·s]: Fluid’s viscosity Other cell damaging phenomena different than turbulence are related to the dynamics of the aeration through gas sparging, bubbles formation, coalescence and rupture can dissipate energies producing high shear forces around the event, this is particularly important for the bursting bubbles reaching the bioreactor headspace. As the bubble rises some cells attach to its surface, when the bubble approaches the surface the liquid is pushed into a thin film hemispherical shape until tension forces become too weak to stand the bubbles buoyancy, the rupture of the film is produced and high acceleration forces are released so the cells attached around the bubble’s surface are brought under high pressure conditions, afterwards upwards and downwards jets are produced. The shear forces produced during this sequence use to be higher enough to destroy the wall of most animal cells. Figure 8 Bubble burst at the liquid-gas interface [19]. Introduction 2 19 For both cases, turbulent eddies and bursting bubbles, some protective products may be added to the medium. On one hand, the use of thickening additives to increase the viscosity gives as a result longer and less dangerous eddies. On the other hand, surfactants minimize the magnitude of the bubble bursting as well as avoid the cell attachment to the bubbles surface. 2.4 Mass transfer Oxygen is usually considered as the most important nutrient in aerobic processes due to the fact that its depletion implies an immediate slowdown of the cell growth or metabolite production. That’s why it is imperative to ensure a proper oxygen supply to the cells. However, its reduced solubility in water makes it a pretty difficult task, which is strongly influenced by an important number of variables related to the bioreactor’s geometry, the aeration system, the stirring system and the medium’s rheology. In most types of bioreactors, the aeration is performed by means of some sort of submerged gas sparger or microdiffusor used to produce bubbles that will transfer the oxygen and other gasses from the gas phase to the liquid phase through its surface. Flick’s law states that the gas flux Jthrough the walls of a steady bubble is proportional to the oxygen concentration gradient in the direction of the transport (between both phases): =− ∙ Where Dis the diffusivity, dC/dx is the transfer coefficient and dC is the increment of oxygen concentration or driving force. Such definition was used by Whitman (1923) for the development of the first and simplest theory on gas-liquid mass transfer, known as the twofilm model. Figure 9 Whitman’s two-film model. Adapted from [7] [16] [20] Introduction 2 20 Two attached films are described to explain the properties of the medium between both phases. Within the gas film, oxygen is diffused towards the interface driven by an increment of the partial pressure and then is decreasingly transferred into the liquid film. The overall oxygen flux JOknown as oxygen transfer rate (OTR) is hence limited by the film resistances at both sides of the phase boundary. This can be written as follows: = ∙(−)= ∙(−) Nevertheless, the interfacial values CLi and pGi are not directly measurable. That’s why mass transfer rate is normally written in function of the oxygen saturation concentration in the bulk liquid CL *and the global transfer coefficient K: = ∙(∗−) It can be demonstrated that the global transfer coefficient KLmay be written as the combination of the gas and liquid mass transfer coefficients kG,kL: 1=1∙+1 Where: •H[l·atm/mol]: Is the Henry’s constant under certain medium and temperature conditions. The reduced solubility of oxygen in water makes commonly accepted that the global mass transfer coefficient approximately equals the liquid mass transfer coefficient. That means that the main resistance happens within the liquid film. The volumetric OTR is then obtained multiplying the overall oxygen flux JOby the gas-liquid interfacial area per unit of volume of the bioreactor a.OTR= · ·(∗−) The kL·a product is an essential parameter used to express how fast oxygen is transported from the gas phase to the liquid where is consumed by the cells. As well as to estimate the potential of a bioreactor to support a certain cell concentration. The kL·a is usually referred as the most important criterion for bioprocess scale up. However, the complexity to separately measure both parameters kLand a, makes necessary its use as a single parameter. Due to said importance, different methods and numerous empirical correlations have been developed to measure or estimate its value. The measurement methods of the kL·a are classified depending on whether they are performed with or without oxygen consumption. A number of chemical methods have been described, however they are not currently very well accepted due to their influence in the properties of the culture broth that could lead to overestimated values of the kL·a. Physical methods are by far more accurate and reliable, they are usually based on the global mass balance equations like: =OTR−OUR= · ·(∗−)−OUR Introduction 2 21 Where: •dCL/dt [mol/s]: Oxygen accumulation rate. • OUR[mol/l·s]: Oxygen uptake rate or cell’s volumetric oxygen consumption. A difference can be made between the available measurement methods depending on whether oxygen consumption exists or not. The most widely used procedures are the Dynamic method and the Gas phase global balance method. The simplest version of the Dynamic method is performed without cells (OUR = 0) and is based on the measurement of the rise and decay times of the dissolved oxygen concentration during the absorption and desorption of oxygen. The first step consists on the aeration of the bioreactor with air or pure oxygen until the dissolved concentration in equilibrium with the gas phase is reached, that’s the saturation concentration. The whole process needs to be realized using some medium, stirring rate and bioreactor configuration relevant of the culture conditions. The second step consists on replacing the air flow with nitrogen to force a total desorption of oxygen (CL= 0). Thus, the mass balance equation within the liquid state is modified as follows: =− · · The general solution is: =∗··· |=0→ = ∗ Thus, if logarithms are taken on both sides of the general solution, kL·a may be found as the slope of the linear regression of the logarithm of the oxygen extinction profile divided by the saturation concentration: ∗=− · · This formula may be found in several books and reviews on bioprocess engineering and assumes that the kLfactor is the same for oxygen and nitrogen. Nevertheless, this is not an accurate approach since the nitrogen’s solubility in water is approximately half of oxygen’s solubility [21]. Such difference lead to a slower desorption of oxygen than absorption. Thus, it makes sense to distinguish between two different mass transfer coefficients depending on desorption or absorption: ∗=− · During the third step, reoxygenation of the bioreactor is produced by aeration with air or pure oxygen at the culture conditions flow rate. This time, just the consumption term is removed of the mass balance equation: = · ·(∗−) Introduction 2 22 The general solution is:=∗· 1− · · |=0→ =0 Therefore, the kL·a can be solved the same after taking logarithms on both sides of the equation: 1− ∗=− · · Figure 10 Profile of the dissolved oxygen after application of the Dynamic method The Dynamic method is widely accepted to be used for well mixed bench scale bioreactors due to its accuracy and simplicity. Limited only by the transient response of the oxygen probe. Fortunately, nowadays this is not a common situation, since the transient response of the currently available optical dissolved oxygen sensors uses to be fast enough for almost any application (a few seconds). The Gas phase global balance method can be applied under cell growth conditions (OUR ≠ 0). It takes into account the global balance at steady state, which implies the existence of some sort of control loop to keep the dissolved oxygen concentration constant. Figure 11 Schematic representation of a generalized aeration system showing phase boundaries. From I. Betts J. & Baganz F. [22]. Introduction 2 23 Figure 11 represents a generalized model of the phase boundaries within a bioreactor. An inlet gas flow Gin with a certain molar fraction of oxygen yin enters the gas phase of the bioreactor VGwhere some of the oxygen diffuses towards the liquid phase VLthrough the interfacial area adriven by the concentration gradient. No matters if diffusion happens just across the headspace surface or the bubble surface, the mentioned control loop modifies the Gin·yin product to keep the dissolved oxygen concentration CLconstant so the transfer rate equals the consumption rate (OTR = OUR) and the concentration yout of the outlet flow Gout = Gin is decreased. This is written as: ·(−)=OUR· Where from the global mass balance equation: OUR=OTR= · ·(∗−) Then the kL·a product can be solved: · = ·(−) ·(∗−) The Gas phase global balance method shows two major advantages. It is well suited for large scale bioreactors and OUR estimation is implicit. However, the need for a well modelled gas mixing system and highly sensitive mass spectrometers makes the procedure a complex and expensive option strongly dependent on the accuracy of the inlet and outlet gas composition measurements, since tiny calibration errors can lead to significant errors for low level consumption rates. Additionally to the methods available for the measurement of the kL·a product, numerous empirical correlations have been defined for different bioreactor topologies, mediums and stirring systems. However, such correlations cannot be directly extrapolated unless no significant changes are made in comparison with the reference conditions. For instance, the addition of surfactants or other chemicals produces considerable discrepancies between the theoretical and actual kL·a values. Nevertheless, depending on the combination of the correlated variables, stirring rate N, superficial gas velocity Vs, effective viscosity 0or several dimensionless parameters like the Reynolds number, certain empirical correlations can still be useful to help on the design of bioreactors and their scale up. A very popular correlation for non-viscous Newtonian fluids in stirred tank bioreactors is the one proposed by Van’t Riet (1979), stating that no influence exists of stirrer geometry on mass transfer, this is a controversial issue that has been investigated by numerous authors and a number of correlations have been proposed under different experimentation conditions. The equation below shows the applicable form to such correlations, changing the value of the coefficients: · = · · · Where: State of the art. 3 30 The greatest obvious advantage of the SimCell™ system is a stunning number of concurrent possible experiments; this makes it currently the most powerful tool for DoE and hence for process development. However, despite that a good reproducibility with bench-top bioreactors has been reported, a proper scale-up assessment cannot be done due to the difficulty to compare other scale-up criteria different than the kL·a such as the power number Npor the stirring rate. Besides, its complexity of use and an unaffordable price makes the SimCell™ system a non-realistic solution for most companies. -24™ [13] was initially developed and introduced by MicroReactor Technologies in 2007 and is currently distributed by the companies Pall and Applikon (Price ≈ 80,000 €). It shows some similarities with the BioLector™, on one hand the cultivation unit also consists of a disposable micro-well plate with embedded pH and DO fluorescence sensors that must be attached to a shaking tray inside an incubation chamber. However, on the other hand the vessels of the -24’s plate may hold a slightly bigger volume, the temperature of each vessel can be individually controlled and the aeration is performed through a gas permeable membrane placed on the bottom of the vessel, which is capable to supply oxygen enough even for some microbial applications (kL·a = 33 to 57 h-1). The BioLector was designed just for monitoring, on the contrary the -24 is meant for the pH and DO control through the individual mixing of different gas supplies. Again, a system that shows some interesting features for screening applications but lacks of some other essential requirements useful for development of bioprocesses, e.g., an on-line, non-invasive cell density related measurement instrument, and some mechanism for the liquids handling, at least for culture sampling and media addition. Figure 3 m-24™ micro-bioreactor system by MicroReactor Technologies. The AMBR15™ & AMBR250™ systems were developed by The Automation Partnership, and nowadays they may be considered as the ones that better fulfil the market requirements within the Minibioreactor niche. The success of the AMBR15™ introduced in 2011 is due to the combination of a pseudo stirred tank topology (tic-tac box), the use of non-invasive pH and DO sensing technologies, and a robotized approach that makes possible to take care of the liquid handling for a significant number of vessels during the experiment, as well as the previous medium preparation. State of the art. 3 31 (Price ranging from 166,000 € to 319,000 €). Additionally, since the disposable Minibioreactors feature a miniaturized sparger (microdiffusor is optional for the AMBR250™) both cell culture and microbial applications are possible (kL·a = 8,5 to 40 h-1). The AMBR15™ [14] consists of a table holding a number of vessels for mediums and chemicals, pipetting supplies and up to four cultivation units in charged to provide thermoregulation, gas supply, stirring and pH and DO measurements, each cultivation unit has capacity to hold up to 12 individual Minibioreactors. During the experiment the robot sequentially handles every Minibioreactor taking samples, adding feed medium, or acid and alkali solutions for pH control. Despite that the AMBR15™ system still lacks of some on-line non-invasive cell density measurement, its features make it a good of option for a lot of tasks where DoE and QbD are must. Figure 4 Tic-Tac box Minibioreactor & AMBR15™ system by The Automation Partnership. Up to now no appreciation has been made regarding sterility, that’s because most of the previously presented systems are supposed to be operated within a sterile area like a cabinet hood or laminar flow. This becomes an important restriction as the culture volume gets bigger. To evaluate the increase in size and complexity observe the following figure showing the AMBR250™ [15] in a four Minibioreactors set-up. The system is built as a cabinet hood big enough to contain a sophisticated robotic arm, a number of vessels for the different mediums and chemicals, pipetting supplies, and of course all the individual instrumentation and hardware per each Minibireactor. Taking into account that the system can be expanded to conduct up to 24 simultaneous experiments the total size of the system becomes significantly big. From the operation point of view the system works mostly the same as the AMBR15™. A remarkable difference is the increase of the pH measurement range in comparison with the optical fluorescence technology thanks to the use of an embedded pH electrode within the disposable vessel. The AMBR250™ introduced in 2014 is the newest in the list (price ≈ 639,000 €), and an important number of good attributes can be pointed; a standardized stirred tank design (scalability); robotized liquid handling; fully individual control per Minibioreactor; disposable sensing technologies; and all the benefits of the single use philosophy. However, the size, price and complexity of the system make it an inappropriate solution for most biotechnologists and just a few big bio-pharma companies are currently taking advantage of its features. State of the art. 3 32 Figure 5 AMBR250™ multiple bioreactor system by The Automation Partnership. DASbox™ presented in 2012 can be explained as a multiple miniaturized conventional stirred tank bioreactor system that uses either disposable MiniBLU© vessels from Epperdorf or reusable glass vessels. The whole system supports up to 24 simultaneous experiments, four per each cultivation unit. The product is appropriate for most cell lines and any type of cultivation strategies (batch, fed-batch, perfusion and continuous) and useful for DoE and QbD. Nevertheless, despite that the size and price could be relatively affordable for many users, the need for numerous manual operations during the set-up process, and the fact that the instrumentation for pH and DO measurement relies on conventional electrochemical probes, strongly limits the final potential uses of the system, just because of the need for probes sterilization and labour required. Figure 6 DASbox multiple bioreactor system by The DASGip Despite that Mini Bio™ is no more than a single miniaturized conventional stirred tank, with little innovation in comparison with bigger bench-top bioreactors (Price ≈ 25,000 €/unit). The product is becoming quite popular due to comprehensive control software based on an embedded web server and its compact design which lets the user to add as many units as wanted whenever is needed. The commercial success of this product since 2010 demonstrates that as far the required functionalities are properly met; a cheap and small enough product supported by appropriate software can be as State of the art. 3 33 powerful and useful as some of the more expensive previously introduced systems. The obvious disadvantages are the lack of some sort of on-line cell density measurement; the labour required to set-up every new experiment and the need for a relatively high volume of culture medium, leading to increased economic cost per experiment. Figure 7 MiniBio bioreactor by Applikon Despite that many other products could be found in the market, the ones mentioned above make a good picture of the state of the art in the development of Minibioreactor systems during the last five years. The author of this thesis has also developed several Minibioreactor systems, where some of them will be described in chapter 4. Besides the achievements of the industry, remarkable academic contributions to the state of the art have also been made; the following table appoints some relevant contributions. Authors Type Volume [ml] Stirring/Aeration Instrumentation Number of Channels Peter Harms, Yordan Kostov, Joseph A. French, Mohammed Soliman, M. Anjanappa, Arun Ram, Govig Rao [16] Stirred tank 1 Turbine / Sparger Non-invasive DO, pH OD, GFP probes. 24 Yordan Kostov, Peter Harms, Lisa Randers-Eichhorn, Govind Rao [17] Doig SD, Ortiz-Ochoa K., Ward JM, Baganz F. [18] Bubble column 2 Shaker / Sparger Non-invasive DO and pH probes 12 Doig SD, Diep A. Baganz F. [19] S. R. Lamping, H. Zhang, B. Allen, P. Ayazi Shamlou [20] Stirred tank 6 Turbine / Sparger Non-invasive DO, pH and OD probe. 1 Robert Puskeiler, Kaufmamm K., Dirk Weuster-Botz [21] Stirred tank 8-12 Customized Turbine / Sparger Non-invasive DO, pH and OD probes. 8-48 Robert Puskeiler, Andreas Kusterer, Gernot T. John, Dirk Weuster-Botz [22] Emmanuel Franchon, Vincent Bondet, Hélène Munier Lehmann, Jacques Bellalou [23] Bubble column 80 Sparger only Electrochemical DO and pH electrodes plus a non-invasive OD probe. 8 Antonio de León, Héctor Mayani & Octavio T. Ramirez [24] Stirred tank 75 to 250 Magnetic stir bar / Headspace Electrochemical DO, pH and Redox electrodes. 1 Table 2 Remarkable academic contributions in Minibioreactors design. One of the first integrations of a Minibioreactor system was reported by A. de León et al. in 1998 [24]. Such development consisted of a glass vessel placed on a magnetic stirrer-heater plate and connected to a number of commercial devices for control and monitorization of the culture conditions. A set of temperature and electrochemical State of the art. 3 34 conventional probes (pH, DO and Redox potential) were placed on the cap of the vessel and connected to a computer through an A/D - D/A interface module. Aeration was supplied through headspace and three independent mass flow controllers for O2, N2and CO2were used to obtain a proper pH and DO control. This set-up allowed online determination of the OUR from the mass balance in the liquid phase (This method will be further explained in the third section of this chapter). The system was validated for the expansion of human hematopoietic cells from umbilical cord blood. The stirring method, a hanging stir bar, demonstrated its feasibility in terms of shear stress for such very specific application. Nevertheless, the low mass transfer coefficient achieved (kL·a = 1 to 3,6 h-1), would make it insufficient for high cell density applications with microbes or some productive animal cell lines. In spite of the fact that the use of OD, pH, and DO non-invasive sensors in bioreactors had already been discussed by R. P. Cox [25] and B. H. Weigl et al. [26] in 1984 and 1994 respectively, the department of Chemical and Biochemical Engineering of the University of Maryland deserves recognition due to the appliance of such non-invasive measurement techniques for the monitorization of a Minibioreactor system with multi-parallelization potentiality. The first version of the system was described by Y. Kostov et al. in 2000 [16] and G. Rao in 2001 [27]. The system was based on a 4 ml cuvette, featuring pH and DO fluorescence sensors, as well as an optical path to measure the cell culture’s optical density. Despite that the set-up was validated for microbial applications using an Escherichia Coli strain, the initially obtained mass transfer coefficient was not very high (kL·a = 20 h-1). This happened due to the insufficient mixing power of stirring/aeration method consisting of a tiny magnetic stir bar lying on the bottom of the cuvette and a small sparger. Two new improved approaches were reported by P. Harms in 2005 [16], both consisting of 24, 1 ml stirred tanks. The first prototype, consisted of a standard 24-well plate with pH and DO fluorescence sensors attached on the well’s bottom surface. The micro-well plate was held by a measurement board including all the signal conditioning and acquisition stages for pH, DO, OD and GFP, and a sterilizable lid featuring miniaturized spargers and stirrers. The system shown a significant high volumetric mass transfer coefficient (kL·a > 75 h-1) for stirring rates > 500 rpm what it makes it suitable for most microbial applications. However, the large number of tubes and wiring harnesses made it impractical for its use with a pipetting robot (pH control, fed-batch and sampling). Hence, a second prototype was developed; made of an array of 24 independent 5 ml flat bottom glass vials, each of them featuring pH, DO fluorescence sensors and a sterilizable cap with individual sparger and stirrer. Such embodiment became to be more appropriate for a future implementation of a robotized liquid handling system. Fluorometrix, a spin-off company of the University of Maryland claimed for the development of such system named as CellStation™ consisting of a carrousel of 12 independent 35 ml disposable glass Minibioreactors, featuring aeration by means of a sparger and a stirring turbine. The parameters (pH, DO and OD) were sequentially monitored and controlled by a robot but no liquid handling abilities were included. Nevertheless, either its detailed features or validation of such system has never been properly reported. State of the art. 3 35 In 2003 S.R. Lamping et al. [20] reported a nice embodiment of a miniature bioreactor, with a footprint equivalent to a standard 24 micro-well plate. The system included miniaturized baffles, sparger and turbines, as well as fiber optic based probes for the measurement of the pH, DO and OD. Regardless of the high mass transfer rate obtained (kL·a = 100 to 400 h-1) the aim of the study was not to obtain a high biomass density within a miniaturized bioreactor but to provide comprehensive data of its performance in comparison with a 20 l conventional fermenter. The most important conclusion of the study was that in spite of the lower kL·a of the Minibioreactor than the one obtained for the 20 l fermenter under the same power input conditions, it was found that the linear fit for both cases could easily be predicted within a 40 % deviation from the Van Riet’s equation, showing the same slope. That, makes possible to stablish the volumetric power input scale-up ratio to keep the kL·a constant in microbial bioprocess development. Figure 8 Main components of the miniature bioreactor. Adapted from [20] The minibioractor was also operated under shaken like and bubble column conditions, that is without aeration and stirring respectively, giving the following conclusions: since in shaken micro-well systems oxygen transfer is achieved by headspace surface aeration only, the potential risk of oxygen depletion is highly related to the culture volume and shaking speed; However, bubble column bioreactors have the potentiality to provide a mass transfer comparable to typical values obtained at low stirring rates but still bigger than shaken systems. Doig et al. [18] [19] described in 2005 the construction and characterization of a glass made bubble column Minibioreactor in the scales from 2 to 100 ml, equipped with sintered gas diffusors for aeration, fluorescent patches for pH and DO monitoring, and temperature-control by means of heat-exchange liquid circuits. It was demonstrated that the mass transfer (kL·a ≈ 220 h-1) was comparable with values usually obtained for stirred tanks. Regardless of the fact that such development was intended for microbial applications only (tested for E. Coli and Bacillus Subtilis). Doig et al. made a remarkable contribution regarding the correlations of the kL·a with the diameter of the bubble, the superficial gas velocity and the volumetric power consumption. State of the art. 3 36 The chair for Bioprocess Engineering of the Faculty of Mechanical Engineering in the Technische Universität München, made a great contribution to the aeration techniques in stirred tanks for microbial fermentation by introducing a new gas induction turbine for maximizing the mass transfer. R. Puskeiler et al. [21] [22] used such turbine to describe in 2005 a system for Batch and Fed-Batch culture in the scale of 8 to 12 ml, with capability up to 48 simultaneous experiments. A stunning value of the kL·a (kL·a > 1440 h-1) ranks the system as the most efficient in terms of mass transfer ever reported, becoming ideal for the production of ultra-high biomass densities. On the other hand, the high level of turbulence produced makes it absolutely inappropriate for any sort of shear sensitive species such as mycellial fungi or animal and insect cell lines. Additionally, the system was complemented with an automated liquid handling robot (Tecan, RSP 150) that once or twice per hour sequentially sampled each Minibioreactor and delivered the sample on a standard multi-well plate enabled for pH and DO fluorescence measurements as well as for OD (Fluostar Galaxy). The Fed-Batch operations as well the addition of NaOH for the pH regulation were also performed by means of the liquid handling robot. In spite of the great achievement related to the design of the mentioned above novel gas inducing turbine capable of providing an unmatched mass transfer rate. A number of drawbacks are obvious, the consumption of culture media for the monitorization, lack of DO control, as well as a very slow pH control action for microbial species (just once or twice per hour) for sure will require a further development. In 2006 E. Franchon et al. [23] introduced a multiple mini-fermenter battery based on a bubble column topology analogous to the one propose by Doig et al. nevertheless a bigger kL·a value was reported, probably due to the different volume and dimensions (kL·a = 750 h-1). The most relevant contribution of this work regards to the subsystem for the measurement of an expanded range of OD (0.05 to 100) which consists of a near infrared LED and a photodetector placed at both sides of the vessel and a control loop that modulates the power applied to the LED in order to keep the photodetected signal constant regardless of the biomass density. The measurement dynamic range is maximized by the fact that a minimum amount of light is always measured by the photodetector and by the use of a single near infrared wavelength to diminish the effect of scattering. Finally, the OD is related to the modulating signal through an empirically obtained calibration curve. HexaScreen Culture Technologies introduced the HexaScreen system in 2012 [28]. Nevertheless, given that such device was developed by the author of this thesis, its description and features will be deeply introduced in chapter 4. The references provided above demonstrate the existence of a well stablished niche of Minibioreactor systems for bioprocess development, as well as a significant dispersion of results especially from the academic point of view. That’s why there’s still need for products with well-defined correlations easy to scaleup and affordable enough to be used for extensive screening studies. State of the art. 3 37 3.2 Dissolved oxygen measurement by means of fluorescent quenching. It has been demonstrated that the DO control or even just monitorization is a key point in most cell culture and fermentation processes and a good indicator of the metabolical activity [28] [29]. A number of well-known techniques for the detection of oxygen are available: gas chromatography, mass spectroscopy, paramagnetic resonance, and the methods based on electrochemical electrodes and fluorescence. Nevertheless, just the last two options are appropriate to be used within the liquid phase. Regarding the first technique, polarographic probes or Clark electrodes which were developed since 1940, are made of a pair of electrodes Pt / AgCl submerged in an electrolytic solution and isolated from the medium by means of a porous membrane permeable to the O2molecules. Electrodes are connected to a potentiostat circuit in order to produce the oxidation-reduction reaction that generates the electrical current which is proportional to the oxygen partial pressure. Then, such electrical current can be measured by means of an electrometric amplifier. Despite that this technique is widely used and still valid, it shows some inconveniences: Slow response time (depends on the analyte’s diffusion through the membrane), reduced signal to noise ratio, electrolyte consumption, analyte consumption (inappropriate for monitoring low volumes), drift due to aging and temperature, and finally the need for sterilization since the electrode must be in touch with the culture medium. On the other hand, the measurement techniques based on the fluorescence emitted by certain chemical compounds or elements depending on the dissolved oxygen concentration, show some advantages: Fast response, good signal to noise ratio, as well as no need for maintenance. In addition, the possibility to build patch like sensors that can be glued on the inner wall of any transparent vessel and measured from outside makes this technology an ideal choice for the development of sterile disposable products. That’s why the use of DO fluorescence probes as well as for other parameters (pH and CO2) is currently becoming a golden standard in Bioprocess engineering. Instruments based on fluorescence usually consist of three stages; a porous polymeric matrix permeable to the analyte that is wanted to be measured, such matrix is used to immobilize the fluorescent element or compound; an optical front-end made of lenses and/or fiber optics for irradiating the matrix and collecting the emitted light; and finally a photodetector and conditioning circuit to transduce the electrical signals into the optical domain and vice versa [30] [31] [32] [33] [34]. When the immobilized fluorophore is irradiated by a beam of light at a given wavelength the absorbed photons push electrons to a higher quantum level for a short period of time (lifetime). After that, when electrons fall down to their original fundamental quantum level, a new wave of light at a longer wavelength is produced. Hence the emitted wavelength depends on the energy difference between such fundamental and excitation quantum levels: 01 SS hc e    State of the art. 3 38 Where: •  e[m]: Emitted wavelength. •c[m·s-1]: Speed of light in vacuum (299 792 458 m·s-1). •h[m2·kg·s-1]: Planc constant (6.626 069 57 × 10-34 m2·kg·s-1). •S0[kg·m2·s-2]: Energy of the fundamental level. •S1[kg·m2·s-2]: Energy of the excitation level. Figure 9 Jablonsky energy diagram of fluorescence The mean time before a single photon is emitted due to fluorescence uses to be quite short (10 ns) and may be modulated by the existence of certain analytes. That’s known as fluorescence quenching and is expressed through the Stern-Volmer equation which relates the intensity and duration of the response depending on the analyte’s or quencher concentration: Figure 10 The Stern-Volmer diagram suggests how for low quencher concentrations Linear fitting may be applied for calibration State of the art. 3 39     QkQk I I qSV  0 00 11    Where: •I: Luminescent intensity in presence of the analyte. •I0: Luminescent intensity in absence of the analyte. •t[s]: Lifetime in presence of the analyte. •t0[s]: Lifetime in absence of the analyte. • [Q][mol·m-3]: Analyte’s or quencher concentration. •kSV[m3·mol-1]: Stern-Volmer constant. •kq[m3·mol-1·s-1]: Quenching rate coefficient. It may be referred also as the bimolecular extinction constant and calculated from the Stokes-Einstein relation:  ·3 8TR kg   Where: •R[J·K-1·mol-1]: Universal constant of ideal gases (8.314 472 J·K-1mol-1). •T[K]: Temperature. •[kg·m-1·s-1]: Lifetime in presence of the analyte. The immediate consequence of the Stern-Volmer equation is the measurement method itself. Analyte concentration can be determined either from the intensity ratio or the fluorescence delay ratio .               1 1 1 100   SVSV kI I k Q Nevertheless, a number of several physical issues have been demonstrated to have a great impact on the fluorescence response; the method for the fluorophore immobilization within the matrix; the aging of the fluorophore compound known as photobleaching; Temperature; as well as the front-end’s optical topology [35] [36] [37]. It also has been found that the determination of [Q] through the fluorescence delay ratio becomes more robust than through the intensity ratio, since the lifetime cannot be affected by the attenuation or other mechanical artefacts. Additionally, the ratio offers two measurement options either in the time or frequency domains. 3.2.1 Time domain method (Lifetime) The time domain absolute method consists on the excitation of a fluorescent sensor patch by a burst of pulsed light where the pulse width is significantly shorter than the fluorescence lifetime. Based on the previous description and the given electrical engineering State of the art. 3 46 It is important to mention that despite the frequency domain methods are virtually insensitive to the magnitude uncertainty; this is only true if the optical cross-talk is kept below the sensitivity threshold of the system. Otherwise, the phase variation range as well the signal to noise ratio will be reduced and the uncertainty will be increased proportionally to the magnitude of the excitation light reaching the photodiode. Therefore, the feasibility of such fluorescence detection systems is not only a matter of electronic design but optical and mechanical issues must be also carefully taken into account. 3.2.3 Ratiometric methods Some variations of the previously explained methods include ratiometric techniques [33] [38], developed to overcome the limitations imposed by the lack of stability of the fluorescence intensity due to not only mechanical reasons, i.e. temperature or photodecomposition may have a great impact on the calibration’s accuracy as well as on the general feasibility of the sensor. Ratiometric techniques are based on the combination of two individual fluorophores within the same matrix. One of them is meant to be used a reference and the other to produce a signal dependent on the analyte concentration. Fluorescence ratiometric techniques can also be referred as Dual Lifetime Referenced (DLR) or Dual Emission Ratiometric (DER) techniques. They are already very well stablished for the measurement of some relevant analytes (pH, Ca2+, Mg2+ and Zn2+). The fact that the reference fluorophore is insensitive to the analyte concentration and features a longer lifetime than the signal fluorophore, leads to several possible measurement strategies, where the lifetime and phase delay techniques can still be used. a) Dual Excitation-Single Detection: The sensor is alternatively irradiated by two different wavelengths and the fluorescence response of both fluorophores is measured at once by a single photodiode within a wavelength region where the individual responses are partially overlapped or as a whole by means of a longpass filter. Since each fluorophore shows different absorption maxima and analyte sensitivity, the ratio of both measured intensities can be related to the analyte concentration [34]. b) Single Excitation-Dual Detection: The sensor is irradiated at a single wavelength and each fluorophore individual response (signal and reference) is measured individually by means of a separate filter and photodiode. The analyte concentration is then related to the ratio of both measured intensities [39]. c) Single Excitation-Single Detection: In this case, since only one light source and photodetector are used, some imaginative method is needed to produce a differentiated fluorescence response dependent on the analyte concentration. This is achieved by operating the system in the frequency domain. The excitation light source is modulated by a sinusoid at two different frequencies. When the modulation frequency approaches the inverse lifetime of the signal fluorophore the measured response equals the total emission of both fluorophores. However, when the frequency gets closer to inverse lifetime of the reference fluorophore State of the art. 3 47 the response can be approached just to the reference fluorophore response. Then the ratio can be solved [40] 3.3 OUR estimation methods. Given the importance of oxygen as a substrate for the cell’s metabolism, the Oxygen Uptake Rate (OUR) has been pointed as a key indicator of the metabolical activity. Nevertheless, the OUR estimation cannot be directly measured. Hence, its determination may show some difficulties especially for animal cell lines [41]. Their low specific consumption rates make the minimum measurement threshold to become an important merit factor; the cell’s sensitivity to the sudden changes of the oxygen concentration; and finally the difficulty to satisfy an appropriate oxygen transfer rate without damaging the cells because of the shear stress produced by the aeration and stirring methods. qO2x10-12 mol O2·cell-1·h-1 Cell line 0.15 – 0.36 KS1/4 (Hybridoma) 0.21 – 0.25 NB1 (Hybridoma) 0.223 – 0.248 Cla (Hybridoma) 0.05 FS-4 (Human diploid cell) 0.19 – 0.4 AB2 – 143.2 (Hybridoma) 0.023 – 0.087 167.4G5.3 (Hybridoma) qO2x10-12 mol O2·kg X-1·h-1 Cell line 2 – 15 Xanthomomas Campestris NRRL 1775 (Bacteria) 0.9 – 23 Escherichia Coli K 12a (Bacteria) 31 – 31.2 Bacillus Acidocaldarius NRRC-207 F (Bacteria) 2–3 Trigonopsis Variabilis CBS 4095 (Yiest) 0.3 – 1 Candida Bombicola NRRL Y-17069 (Yiest) 0.8 Hansenula or Pichia Anomala CBS6759 (Yiest) Table 3 Specific consumption rates qO2for some cell species commonly used in productive biotechnology. Adapted from Ruffieux P-A. et al. [41] and Felix Ochoa et. Al. [43] Most of the available methods for OUR estimation are based on the mass balance equation [42]; some cases are considering the whole bioreactor or just the liquid phase where the oxygen concentration can be kept constant or not. For a proper understanding of the different methods the generalized model of the phase boundaries within a bioreactor must be considered [Figure 11, Introduction]. 3.3.1 Global mass balance method This method is based on the comparison of the gas compositions measured on the gas Inlet and outlet taking into account the whole bioreactor. Hence, the mass balance equation is expressed as follows: State of the art. 3 48 LoutoutininG G L LVyGyGV dt dC V dt dC  OUR Where: •VL[l]: Liquid phase volume. •VG[l]: Gas phase volume. •Gin[lpm]: Inlet gas flow. •Gout[lpm]: Outlet gas flow. •yin[mol/l]: Oxygen inlet molar fraction. •yout[mol/l]: Oxygen outlet molar fraction. If no oxygen accumulation happens within the bioreactor (the medium’s DO concentration is kept constant) then the equation can be rewritten as: L outoutinin V yGyG  OUR This is in fact equivalent to the Gas phase global balance method for the kL·a determination explained in section 2.4. As mentioned, this method has not historically been very widespread due to the need for very sophisticated and expensive instrumentation like mass spectrometers and very accurate DO control systems (A very tight DO drift below ±0.05 % is required [41] [43]). Nevertheless, the current capabilities of the latest DO probes and mass flow controllers, as well as the improvements on the accuracy of the gas phase oxygen sensor technologies (fluorescence quenching and ZrO2[44] [45]) will probably make in the near future a more common usage of the Global mass balance method. Aehle et al. [46] compared the OUR data of a CHO cell line fermentation, when inlet and outlet gas compositions were measured by means of a mass spectrometer and an oxygen/CO2combined sensor (ZrO2/IR, BCpreFerm, Bluesens [47]). Results showed a much lower transient response for the combined sensor than the mass spectrometer but still fast enough for the application since an almost perfect correlation of the data provided by both instruments was demonstrated. 3.3.2 Stationary liquid phase mass balance method This method consists on determining the required oxygen concentration on the gas phase to set a constant DO value in the liquid phase. Therefore, no accumulation happens and the mass transfer equals the consumption. On the contrary to the Global mass balance, only the gas to liquid mass exchange is considered:   OUR * LLl LCCak dt dC   OTROUR * LLl CCak Where: State of the art. 3 49 •CL[mol/l]: DO concentration in the liquid. •CL *[mol/l]: DO concentration in the liquid in equilibrium with the gas phase. Therefore to solve the OUR, previous knowledge of the mass transfer coefficient and the dissolved oxygen concentration in equilibrium with the gas phase are necessary. On one hand, the kL·a value can be obtained through the application of any of the procedures introduced in section 2.4 while reproducing some physical and chemical conditions representative of the experiment. On the other hand CL *can be related to the gas phase oxygen’s partial pressure and molar fraction through the Henry’s law. Hy H yP CL Gtot L  ** Where: •Ptot [atm]: Total gas supply pressure. •yG[ ]: Oxygen’s molar fraction in the gas phase. •yL *[ ]: Oxygen’s molar fraction in the liquid phase. •H[l·atm·mol-1]: Henry’s constant. Which it can be approximated for a certain temperature in a aqueous solution through the Van’t Hoff equation: ()=( )∙∙ Where: •T0[K]: Reference temperature (298 K) •H(T0)[l·atm·mol-1]: Oxygen’s dilution constant in water at the reference temperature (769.23 l·atm·mol-1). •K[K]: Scale factor (1700 K) As for the previous method, the fact that the DO concentration is kept constant throughout the culture is an important advantage from the biological point of view, since the cells are not subjected to stressful changes of the oxygen tension. Despite that the Stationary liquid mass balance method offers minimum cell stress and great estimation accuracy. It has not become of extended usage due to the need of expensive mass flow controllers and means for the determination of the inlet gas composition. Ducommun et al. [48] integrated a set-up consisting of 2l stirred tank aereated by means of a PTFE made porous pipe connected to three mass flow controllers for DO an pH control (O2, N2, CO2). The DO was measured by means of a standard polarographic probe and its signal acquired by a computer running custom control software used to operate the mass flow controllers. The inlet oxygen’s molar fraction was calculated as a function of the set points of each mass flow controller. The set-up was tested under cell culture conditions for a CHO cell line showing that it was possible to obtain an online accurate OUR estimation within ±5 %. Other realisations chasing a higher accuracy, may also include additional instrumentation for a more precise determination of the gas phase composition (infra-red, paramagnetic analysers or mass spectrometers) [41]. State of the art. 3 50 3.3.3 Dynamic method. The Dynamic method is currently the most widely used for the OUR determination. The method consists on replacing the oxygen in the gas phase with some inert gas to force the dissolved oxygen desorption, producing an extinction profile that fits to a negative exponential curve featured by a slope proportional to the oxygen consumption and a time constant inversely proportional to the mass transfer coefficient. This task is commonly done by means of a 3-way electrovalve that periodically switches the gas supply from air to nitrogen always maintaining the oxygen concentration between two predefined levels (typically 50 to 30 %) where the duration of both states are named tOn and tOff respectively. That’s why, on the contrary to the previously explained methods the mass balance equation needs to be written in two parts:                 OffLdes OnLLl L tttCk tttCCak dt tdC OUR OUR * Figure 20 DO response for the dynamic estimation of the OUR. Adapted from F García-Ochoa et al. [42] To solve the mass balance equations, two assumptions need to be made. The first is that the removal of the oxygen in the gas phase of the bioreactor happens fast enough to consider it as an instantaneous event, the second assumption is that the consumption variation is really slow when compared to the tOn and tOff periods, so it can be considered as a constant. Obviously, such assumption implies inconsistency when t→∞. However, the solution will be valid if the aeration period tOn and the oxygen removal period tOff are kept small when compared to the growth rate of the cell species but still bigger than the absorption and desorption constants. ∙ , ≪ , ≪ State of the art. 3 51 Hence, the solution of the mass balance equation can be written as:               Off tk des On tk des L L tte k tte k C tC des des   OUR OUR * The OUR instantaneous value can be found through the integration of CL(t) along the tOff period:       OffOffOff tt Ldes tLdtdttCk dt tdC 000 OUR       Off t Ldes t l t dttCktC Off Off   0 0· OUR Iterated repetition of this process allows reliable on-line monitoring of the oxygen consumption. However, this technique shows some drawbacks. Previous knowledge of kl·a is required; Accuracy is dependent on the oxygen consumption; Need for an initial minimum inoculum to set the lower estimation threshold; and finally an obvious compromise between the “measurement” period and the potential damage suffered by the growing cells because of the necessary variations of the DO concentration. In animal cell culture, such compromise comes out with a very poor temporal resolution of 5 to 10 samples a day making difficult a proper on-line interpretation to define new cultivation strategies. Nevertheless, such compromise is not that dramatic for the monitorization of microbial processes due to their robustness and higher oxygen demand. This has been reported by numerous authors specially for monitoring activated sludge systems [49] [50] [51]. A simplified version of the Dynamic method was used by Anderlei T. et.al. to develop the RAMOS system (Respiration Activity Monitoring System) [52] [53] [54]. It consists of several shake flasks using a customized cap featured by two inlet and outlet aeration ports. The system operates in two stages, during the rising stage the vessels are put down a constant air flow to force DO equilibrate with the gas phase, during the measurement stage two electrovalves close both aeration ports producing and certain increment of the oxygen partial pressure in the headspace due to respiration of the microorganisms. Then OUR can be calculated from such increment: L G VTRt VpO    2 OUR Where: •T[K]: Temperature •R[atm·l·K-1·mol-1]: Universal constant of ideal gases (0.08205746 atm·l·K-1·mol-1). •VL[l]: Liquid phase volume. •VG[l]: Gas phase volume. Between all the explained methods only the dynamic method can still be considered as a standard. This is because a good trade-off between accuracy and the economical investment State of the art. 3 52 for its implementation. Some other methods have also been reported but they are just suitable for very specific applications like the Yield coefficient method [55] or may have physical implementation concerns especially for low-volume bioreactors like the OUR probe developed by K. Konstantinov et.al. [56] [57]. References: [1] A. d. Palma, «Microbioreactors Carve Out Growing Niche,» Genetic Engineering & Biotechnology News, vol. 30, nº 3, 2010. [2] F. B. Jonathan I Betts, “Miniature bioreactors: Current practices and future opportunities,” Microbial Cell Factories, vol. 5, no. 21, 2006. [3] C. W. E. H. Sathish Kumar, “Minibioreactors,” Biotechnology letters, no. 26, pp. 1-10, 2004. [4] V. Glasser, «Finding a Bioreactor that's right for you,» Genetic Engineering and Biotechnology News, vol. 31, nº 14, 2011. [5] V. Glasser, “Bioreactor and Fermentor market trends,” Genetic Engineering and Biotechnology News, vol. 29, no. 14, 2004. [6] V. Glasser, “Disposable Bioreactors gaining favor,” Genetic Engineering and Biotechnology News, vol. 26, no. 12, 2006. [7] M.-N. G. B. J. Samorski M, «Quasi-continous combined scattered light and fluorescence measurements: a novel measurement technique for shake microtiter plates.,» Biotechnology and Bioengineering, vol. 92, nº 1, pp. 61-68, 2005. [8] E. Z. C. F. R.-K. T. J. B. Frank Kensy, “Validation of a high-throughput fermentation system based on online monitoring of biomass and fluorescence in continuously shaken microtiter plates,” Microbial Cell factories, vol. 8, no. 31, 2009. [9] C. E. J. B. Frank Kensy, “Scale-up from microtiter plate to laboratory fermenter: evaluation by online monitoring techniques of growth and protein expression in Escherichia coli and Hansenula polymorpha fermentations,” Microbial Cell Factories, vol. 8, no. 68, 2009. [10] J. M. O. M. H. Z. A. B. S. A. H. P. R. Ashraf Amanullah, “Novel Microbioreactor High Throughput Technology for cell culture development: Reproducibility, and Scalability assessment in Fed-Batch CHO Cultures,” Biotechnology and Bioengineering, vol. 106, no. 1, pp. 57-67, 2010 . [11] G. S. Z. R. L. S. Ache Stokelman, «High Throughput Cell Culture Experimentation with BioProcessors SimCell™ Platform,» de CPAC Sattellite Meeting, Rome, 2006. State of the art. 3 53 [12] G. Wilson, “Evaluation of a Novel Micro-Bioreactor System For Cell Culture Optimisation,” in 9th ESACT Meeting, Harrogate, UK, June 5-8, 2005. [13] R. C. D. C. A. A. Aaron Chen, “Twenty-four well plate Miniature Bioreactor System as a Scale-Down Model fer Cell Culture Process Development,” Biotechnology and Bioengineering, vol. 102, no. 1, pp. 148-160, 2008. [14] N. B. R. O. K. W. D. P. Rachel Bareither, “Automated disposable small scale reactor for high throughput bioprocess development: A proof of concept study,” Biotechnology and Bioengineering, vol. 110, no. 12, p. 3126–3138, 2013. [15] T. D. C. Z. U. H. G. T. A. C. F. Davy De Wilde, «Superior Scalability of Single-Use Bioreactors,» BioProcess international, vol. 12, nº 8s, 2014. [16] Y. K. J. A. F. M. S. M. A. A. R. G. R. Peter Harms, “Design and performace of a 24-Station High Throughput Microbioreactor,” Biotechnology and Bioengineering, vol. 93, no. 1, pp. 6-13, 2006. [17] P. H. L. R.-E. G. R. Yordan Kostov, “Low-Cost Microbioreactors for High-Throughput Bioprocess,” Biotechnology and Bioengineering, vol. 72, no. 3, pp. 346-352, 2001. [18] O.-O. K. W. J. B. F. Doig SD, “Characterization of Oxygen Transfer in Miniature and LabScale bubble column bioreactors and comparison of microbial growth performance based on constant kla,” Biotechnology progress, no. 21, pp. 1175-1182, 2005. [19] D. A. B. F. Doig SD, “Characterization of a novel miniaturized bubble column bioreactor for high throughput cell cultivation bioreactor system for mammalian cell culture,” Biochemical Engineering, no. 23, pp. 97-105, 2005. [20] H. Z. B. A. P. A. S. S. R. Lamping, “Design of a prototype miniature bioreactor for high throughput automated bioprocessing,” Chemical Engineering Science, no. 58, pp. 747-758, 2003. [21] K. K. D. W.-B. Robert Puskeiler, “Development parallelization and automation of a gasinducing millilitre-scale bioreactor for high-throughput bioprocess design,” Biotechnology and Bioengineering, no. 89, pp. 512-523, 2005. [22] A. K. G. T. J. D. W.-B. Robert Puskeiler, “Miniature bioreactors for automated highthroughput bioprocess design (HTBD): reproducibility of parallel fed-batchcultibations with Escherichia coli,” Biotechnology and Applied Biochemistry, no. 42, pp. 227-235, 2005. [23] V. B. H. M. L. J. B. Emmanuel Franchon, “Multiple Microfermentor Battery: a Versatile Tool for Use with Automated Parallel Cultures of Microorganisms Producing Recombinat Proteins and for Optimization of Cultivation Protocols,” Applied and Environmental Microbiology, vol. 72, no. 8, pp. 5225-5231, 2006. State of the art. 3 54 [24] H. M. O. T. R. Antonio de León, “Design, characterization and application of a minibioreactor for the culture of human hematopoietic stem cells.,” Cytotechnology, no. 28, pp. 127-138, 1998. [25] M. M. J. B. N. M. N. J. K. T. Raimond P. Cox, “Continous turbidometric measurements of microbial cell density in bioreactors using a light emitting diode and a photodiode,” Journal of Microbiological Methods, no. 10, pp. 25-31, 1989. [26] A. H. W. T. I. K. H. K. P. O. O. S. W. B. H. Weigl, «Optical Triple Sensor for Measuring pH, Oxygen and Carbon Dioxide,» Journal of Biotechnology, nº 32, pp. 127-138, 1994. [27] G. Rao, “Bioreactor and Bioprocessing Technique”. USA Patent US 6,673,532 B2, 4 January 2004. [28] N. H., “Mass transfer,” in Basic Biotechnology, Cambridge University Press, 2006, pp. 201217. [29] M. H. H. S. Y. K. K. A. B. D. D. F. A. R. M. G. R. Xudong Ge, “Validation of an optical sensorbased high-throughput bioreactor system for mammalian cell culture,” Journal of Biotechnology, vol. 122, no. 3, pp. 293-306, 2006. [30] I. Bergman, “Rapid-response Atmospheric Oxygen Monitor based on Fluorescence Quenching,” Nature , 396 (27 April 1968); doi:10.1038/218396a0, vol. 218, no. 396, 1968. [31] R. C. M. J. Palatine and S. M. L. Hoffman, “Optical sensor for monitoring the partial pressure of oxygen”. USA Patent 4,725,115, 21 June 1988. [32] T. E. P. J. A. M. R. B. G. C. B. B. Arthur E. Colvin Jr, “A novel Soli-State Oxygen Sensor,” Johns Hopkins apl Technical Digest, vol. 17, no. 4, pp. 377-384, 1996. [33] O. Wolfbeis, “Fiber-Optic Chemical Sensors and Biosensors,” Analytical Chemistry, vol. 80, p. 4269–4283, 2008. [34] Y. K. H. Lam, “Optical Instrumentation for Bioprocess Monitoring,” in Optical Sensor Sytems in Biotechnology, Springer, 2009, pp. 14-23. [35] N. W. P. J. Q. F. H. K. M. C. R. R. A. B. G. C. L. R. J. L. M. Thanassis Papaioannou, “Performance evaluation of fiber optic probes for tissue lifetime fluorescence spectroscopy,” in Proceedings of SPIE Vol. 4958, San Jose CA, 2003. [36] M. A. M. J. R. D. B. L. I. V. A. G. M. V. P. G. L. C. D. Patrick O'Neal, “Oxygen sensor based on the fluorescence quenching of a Ruthenium complex immobilized in biocompatible Poly(Ethylene Glycol) hydrogel,” IEEE Sensors Journal, vol. 4, no. 6, pp. 728-734, 2004. [37] L. C. K. C. C. Samantha M. Grist, “Optical Oxygen sensors for applications in microfluidic cell culture,” Sensors, vol. 10, pp. 9286-9316, 2010. State of the art. 3 55 [38] S.-B. I. C.-F. J. S.-M. A. Hochreiner H., “Dual emission probe for luminescence oxygen sensing: a critical comparison between intensity, lifetime and ratiometric measurements,” Talanta, vol. 66, no. 3, pp. 611-618, 2005. [39] R. L. M. W. S. K. P. J. S. John F. McCarthy, “A Fiber-optic System for Measuring Single Excitation-Dual Emission Fluorescence Ratios in Real Time,” Biotechnology progress, vol. 8, no. 4, pp. 360-368, 2008. [40] P. H. R. S. P. G. R. Yordan kostov, “Ratiometric oxygen sensing: detection of dual-emission ratio through a single emissio filter.,” The Analyst, vol. 125, pp. 1175-1178, 2000. [41] U. v. S. I. W. M. Pierre-Alain Ruffieux, “Measurement of volumetric (OUR) and determination of specific (qO2) oxygen uptake rates in animal cell cultures,” Journal of biotechnology, vol. 63, no. 2, pp. 85-95, 1998. [42] E. G. V. E. S. J. C. M. Felix Garcia-Ochoa, “Oxygen uptake rate in microbial processes: An overview,” Biochemical Engineering Journal, vol. 49, pp. 289-307, 2010. [43] D. C. D. T. J. E. C. Dorresteijn R. C., “A method for simultaneous determination of solubility and mass transfer coefficient of oxygen in aqueous media using off-gas mass spectrometry,” Biotechnology and Bioengineering, vol. 43, pp. 149-154, 1994. [44] SensorTechnics, “Oxygen sensors XYO Series,” [Online]. Available: http://www.sensortechnics.com/cms/upload/datasheets/DS_StandardXYO_E_11818.pdf. [Accessed 4 12 2014]. [45] SensorTechnics, “Oxygen sensors XYA Series,” [Online]. Available: http://www.sensortechnics.com/cms/upload/datasheets/DS_StandardXYA_E_11637.pdf. [Accessed 4 12 2014]. [46] A. K. S. S. R. S. A. L. Mathias Aehle, “Simplified off-gas analyses in animal cell cultures for process monitoring and control purposes,” Biotechnology Letters, vol. 33, no. 11, pp. 2103-2110, 2011. [47] BlueSens, “Brochure BCpreFerm,” [Online]. Available: http://www.bluesens.com/fileadmin/user_upload/grafik/pdf/Brochure_BCpreFerm.pdf. [Accessed 4 12 2014]. [48] P.-A. R. M.-P. F. I. M. U. v. S. Paul Ducommun, “A new method for on-line measurement of the volumetric oxygen uptake rate in membrane aerated animal cell cultures,” Journal of Biotechnoloy, vol. 78, pp. 139-147, 2000. [49] P. R. B. J. S. d. R. N. A. C. v. H. Valnyr V. Lira, “Estimation of Dissolced Oxygen Dynamics for Sequencing Batch Aerobic Reactors,” in IMTC 2004 - Instrumentation and Measurement Bioreactors 4 62 Communication is point to point, between the host PC and the workstation, and just requires a RJ45 crossover cable. TCP/IP protocol is used to support the data exchange, such a capability allows setting up a 6-device array in order to increase the screening capability or even to provide remote access to the on-line data .That requires the use of a simple Ethernet hub or switch as for any Local Area Network. Figure 5 6-device array Regarding the HexaScreen control & acquisition software, it basically acts as a wizard, designed to assist the user through the process. Experiments are executed in three stages, configuration, pre-culture calibration and data acquisition. Main features of the software are: - Creation or edition of experiment recipes. - Creation of customized graphs combining up to 6 variables. - Creation of reports, including: oThe results of the experiment. oName of users involved on the experiment (design, execution, …) oAcquisition dates (Start time, End time and Time stamp) oWorkstation ID number’s (s/n, IP address, port, …) oExperiment set-up parameters. oWorkstation’s calibration parameters. oList of unexpected events (Errors happened) - Exportation of reports to EXCEL and HTML files. - 21CFR-11 administration oData files encryption oUsers administration (Supervisor, Operator, User) oSystem audit trail Bioreactors 4 63 4.1.2HexaScreen® architecture The HexaScreen system is made of a significant number of parts of different kind: electronic circuits, fiber optic bundles, lenses, machined parts in a wide sort of materials, valves, TEC devices, etc… The cross-sectional view below shows a glimpse of the system’s internal configuration: where the lower box contains most of the electronics and actuators as well as the opening mechanism, the upper box contains the gas circuit, fiber optic probes, the central stirrer, and heaters for thermoregulation, the walls are thermally insulated in order to keep the temperature some degrees over the culture’s temperature to avoid condensation phenomena within any part of the gas circuit, and finally the insulation chamber provides visual inspection as well as thermoregulation, stirring, light excitation, and secures gas connections. Figure 6 HexaScreen’s workstation cross-sectional view and actual implementation The control architecture is based on an RCM2200 Rabbit core; a C-programmable module with Ethernet connectivity especially suited for controlling embedded systems, two additional ADuC218 microcontrollers are also used to carry out especially critical functions. The ADuC812 MicroConverter® is also C-programmable and a fully integrated 12-bit data acquisition system-on-a-chip, featuring precision A/D & D/A converters. As shown in Figure 7 the system is made of four printed circuit boards: Control, Dissolved oxygen, Drivers and Temperature. The modules are connected each other in a PC104 like stackable way where all the logic and math functions are performed by the control board and the most of the signal conditioning and power driving functions are supported by specific circuits. Bioreactors 4 64 Figure 7 HexaScreen Block Diagram A specific microcontroller (ADuC812) is used to synthesize the PWM stirring command signals where an 8 bit resolution sampled sinusoid is used for modulation, two pointers are constantly reading the signal profile keeping a constant π/4 phase delay, the returned values are used to load the microcontroller’s internal timers that generate the command signals (further details will be explained in the 4.1.4 section, Minibioreactor aeration, stirring & mass transfer). A second microcontroller (ADuC812) is also used for acquiring the multiplexed DO signals as well as the temperatures inside the lower and the upper boxes and in the Minibioreactor plate. The second ADuC812 is also in charge of solving the PID control signals for the upper box and Minibioreactors plate thermoregulation. Such control signals are provided to the Temperature board by means of two built-in digital to analog converters. In order to allow optical absorbance spectroscopy measurements the control board holds an UV/VIS spectrometer module (Steag Microparts). The spectrometer is based on a hollow cavity waveguide design with no moving parts which is attached to a silicon-photodiode detector array (Hamamatsu S8378-256N24). Light is coupled into the spectrometer through a 300/330µm silica fibre and an entrance slit. A focusing flat field Echelette grating and a camera mirror lead the light inside the spectrometer cavity. These elements are arranged in the Rowland design, which guarantees mechanical, thermal and optical stability. Virtually, there is no thermal drift of the wavelength calibration due to the fixed geometrical position of the optical components. Hence, wavelength to pixel calibration function is stable over the device lifetime and no recalibration is required [2]. This is a very convenient feature for the monitorization of long term phenomena as cell growth. Bioreactors 4 65 Figure 8 Steag Microparts  spectrometer features The RCM2200 module is responsible for the remaining tasks of the control board: TCP/IP communications, Firmware downloading, LCD interface and control of the multiple peripheral devices (Reference LED’s for optical absorbance spectroscopy, electrovalves for MBR’s aeration, ventilation fans and DC motor for the water-bath central stirring). To perform such tasks the RCM2200 module executes a state machine observing the following ten conditions related to the system’s different stages of operation: Figure 9 HexaScreen Bioreactor states diagram •Initialisation: Default configuration is loaded from flash memory and all the working variables are initialised. If the loaded data is found to be corrupted or inexistent this circumstance is treated as a fatal error and the system moves to the “Final Error” state. Bioreactors 4 66 •Autotest: Automated check of the internal communications and sequenced actuation of the peripheral devices for fast inspection. If the internal communication check fails the system will move to the “Final Error” state. Next state depends on the value of the “Last State” parameter recorded on the flash memory, any value different than “Idle” points to an unexpected lack of power or analogue circumstance, if the recorded value was “Acq & Control” it means that something happened during the execution of the previous experiment so the last experiment configuration is restored and the system moves to the “Quiescent Cell Culture” state. •Idle: Once the “Autotest” process has successfully been performed, the system starts “listening” the TCP port waiting for any service request coming from the host. This state can also be activated after any operation supported by the “Test & Set-Up” state, the “Instrument Calibration” state, the “Load Configuration” state, the “Preculture calibration” state, and the “Acquisition & control” state. •Test & Set-Up: This state is intended for service tasks, on one hand provides access to the configuration data recorded on the flash memory, and on the other hand allows executing commands for testing single functions which are useful during the assembly procedure. •Instrument Calibration: This state is analogue to the “Test & Set-Up” state but focused on the calibration procedure, it supports commands useful to obtain the thermometers calibration parameters, as well as to test the DO and the light spectra measurements. •Load Configuration: It supports experiment configuration upload/download commands to/ from the host computer. •Preculture Calibration: Automated procedure previous to the cell culture process itself. Supported commands are mostly related to thermoregulation, as well as to DO and OD calibrations. •Acquisition & Control: Automated procedure that keeps the physical conditions constant all along the experiment (temperature, stirring, and aeration) and acquires real-time data on cell culture evolution (DO, pH & OD). If during the experiment the flux of data would be interrupted due to any reason, the system would try to reestablish the communication, if that would not be possible such circumstance would then be notified through the LCD and the system would move to the “Quiescent Cell Culture” state. •Quiescent Cell Culture: The aim of this state is to preserve the contents of the Minibioreactors once a non-solved communication or unexpected power-up problem has been detected. In such circumstance the system restores the culture parameters according to the values established by the last valid experiment’s configuration. Besides, the system stays “listening” the TCP port waiting for a service request from the host to re-establish the connection and move back to the normal “Acquisition & Control” state. •Final Error: If some internal error happen a wait and retry procedure will take place in order to solve the problem, basically the internal errors that can be solved are the ones related to the communications between the RCM2200 module and the ADuC812 Bioreactors 4 67 microcontrollers, if finally the wait and retry procedure times out the system moves to the “Final Error” state and notifies the corresponding error code through the LCD. Since the treatment of other error types as the detection of aberrant and out of range data may require a certain interaction with the user. Then, high level error management is under control of the host computer software. The Dissolved Oxygen board is in charge of measuring the fluorescence light emitted by the disposable oxygen sensors attached inside each MBR. The HexaScreen system uses the previously explained XOR phase detector (3.2.2 section) implemented by means of an AD8302 RF/IF gain and phase detector typically used for RF applications. However, it has been also documented for low frequency applications [3]. An oscillator block provides a 7…8 kHz sinusoid signal which is used to modulate the excitation LED’s. Obviously the modulation signal is biased according to the LED’s forward voltage. In order to reduce the cost of the DO measurement system a time multiplex technique was applied so it was possible to use only one photodetector to measure the fluorescence from every MBR. An optical long pass filter was directly mounted on the photodiode’s window to reject the excitation wavelength light or other noise sources. Note that the hereby introduced module is a cost effective alternative to the proposed solution by the oxygen sensor manufacturer consisting of six multiplexed independent OEM modules [4]. Due to the relatively small magnitude of the light emitted by the oxygen sensor, about 3 % of the excitation peak (-15 dB), the long-pass filter must be carefully selected in order to maximize the detection sensitivity. Therefore, in order to consider that the excitation light compound reaching the photodetector is negligible in comparison with the emitted light, it is important to ensure a minimum crosstalk ratio between the of excitation and the detection channels in the order of -30 dB. Observe how for the selected filter (Deep orange - Wratten 22), the percentage of transmittance within the wavelength excitation region is < 0.01 % which is equivalent to a crosstalk ratio around -40 dB. Figure 10 Backscattered light spectra emitted by a Presens SP-PSt3 oxygen sensor under two situations, 0 % and 100 % oxygen-saturation when irradiated at 505 nm. Bioreactors 4 68 Figure 11 Deep Orange Wratten 22 assembly and Transmittance/Density curve [5] Due to the significant number of actuators and the long required operation time per experiment the system was designed to avoid excessive power consumption. Therefore, in order to drive power consuming devices (TEC, heaters and electrovalves as well as any other actuator that could be necessary in the future), the Temperature and Drivers boards were based on PWM power management IC’s like Texas Instrument’s DRV593 and DRV103. Figure 12 HexaScreen’s Acquisition & Control hardware, inspired on the PC104 standard Bioreactors 4 69 4.1.3HexaScreen® optical layout. In order to simplify manipulation and reducing the contamination chances, most of the measurements were designed to be performed using optical techniques where light beams are applied through the Minibioreactor’s transparent walls and guided from the corresponding ports to the measurement electronics by means of a variety of optical parts. Two different optical set-ups were developed, one for absorbance spectroscopy measurements (used for pH and OD), and another for fluorescence measurements (used for DO). The following sections explain each set-up implementation as well as their theory of operation. 4.1.3.1 Optical absorbance spectroscopy measurement (pH & OD): As mentioned above, absorbance spectroscopy was chosen to perform pH and OD measurements. One of the most common configurations used on optical density spectroscopy is based on the dual beam principle; on Figure 13 a possible optical layout for such configuration is shown. It is a complex instrument that allows solving spectral absorbance, compensating any change due to thermal drift or part’s aging. It requires the use of some wide spectrum light source including optics to provide a parallel light beam, a beam splitter to produce a reference beam, two mirrors, one beam combiner, some diffraction system and one shutter or chopper device to alternatively select between the reference and the sample beam. Obviously, such approach implies a significant size and prize related to the mechanics used to hold the optics and to avoid unwanted movements. Figure 13 Double beam-Offner spectrometer. Bioreactors 4 70 The HexaScreen concept did not allow taking benefit of the dual beam configuration due to the reduced size of the Minibioreactor’s plate; imposing the need of looking for some alternative realisation able to match the following requirements: •Small size. It had to be integrated with the heating and the stirring subsystems. •Multiplexing. It had to use a single detector device for the 6 Minibioreactors. •Robustness and repeatability. The probes had to be connected and disconnected to the Minibioreactor’s plate before and after every experiment. The finally chosen optical layout it is shown on Figure 14 It is composed by a 6 legs fiber bundle (probes), one fiber optic taper, two inline SMA-SMA connectors, and 6 GRIN lenses working as collimators, the aim of all these parts is to collect the light that goes through the culture medium and to guide it to the sensing spectrometer. The simulation in Figure 16 shows the light distribution into the medium and how is collected by the GRIN lens. GRIN lenses were primarily designed as components for fiber to fiber and laser to fiber connections [6]. However, they have been found to be useful for sensing purposes in multiple applications from analytical sensing to endoscopy [7] [8] [9] [10] [11] [12]. Due to the important trade-off between size and functionality the chosen light source was a wide spectrum white LED 1 which was embedded onto the heat pump’s thermal plate in order to minimize any change of intensity produced by thermal drift. When the LED is switched on, the light travels through a 1 mm long and 0,5 mm diameter pinch hole which is filled with optical glue of known refractive index. The simulation (A) shows how the light beam is spread due to the pinch hole’s numerical aperture and the refractive indexes of the different materials. It’s important to mention that for such simulation seawater was used to mimic the culture broth. Nevertheless, seawater behaves as a homogeneous medium which is not true for actual cultures, where the presence of cells distorts the light beam producing a very wellknown effect known as light scattering. So it was predictable that the final sensitivity shall be affected by this phenomenon. Simulation (B) shows the maximum numerical aperture of the GRIN probe able to focus the received light into a 300 m fiber optic. It means that any ray out of the acceptance angle here represented will not reach the fiber entrance implying that only a small fraction of the emitted light is finally collected by the fiber. On the other side, it makes the probe more robust to the distorting effects of the scattered light and Fresnel reflections [13]. That is more obviously represented on simulation (C) where the excitation beam and the probe’s numerical aperture were overlapped in order to show that only the transmitted and the scattered light within the intersected volume are contributing to the absorbance measurement. Experimental measurements proved that this is a good, and cheap, approach to a parallel beam compared to the ones produced by standard spectrophotometers. During normal operation, each LED shall be switched on and off sequentially and the respective beams of light shall go through the culture medium with a relatively high numerical aperture. Hence, the light collected by each probe shall be affected not just by the Beer’s law but the interaction of light through the interfaces (Fresnel reflections) and the suspended biomass. 1 Luxeon Warm White Emitter LXHL-BW03 Bioreactors 4 71 Figure 14 HexaScreen’s Optical absorbance setup Figure 15 Post-equalization spectra profile per channel provided by the excitation LED’s. Bioreactors 4 78 •A: Sigmoid parameter. •B: Sigmoid parameter. •C: Sigmoid parameter, Residual pH value. •D: Sigmoid parameter, Residual absorbance value for zero pH. •  pH [nm]: pH measurement wavelength depends on the dye chosen. •Abs []: Measured absorbance at the pH measurement wavelength (  pH). •Abso[]: Initial absorbance at the pH measurement wavelength (  pH) corresponding to the medium’s pH initial value which it can be solved by: = +1+ 4.1.3.2 Fluorescence measurement (DO): The dissolved oxygen concentration is obtained by means of the light response produced by a fluorescence patch sensor attached inside each Minibioreactor using the Phase shift method. The fluorescence kit is composed by a unique part, a 6 to 6+1 leg fiber bundle which guides the excitation light from the excitation LED’s to the fluorescence ports, and delivers the collected response to a sensing photodiode. Figure 21 HexaScreen’s DO fluorescence setup Bioreactors 4 79 The excitation fibers are hexagonally arranged around the collecting fiber in the probe’s tip, such configuration was built based on the simulations performed by Papaioannou [24] to ensure a proper irradiation of the sensor spot. During operation, each individual excitation LED switches on and off sequentially, so the sensor spot’s emitted light is synchronously read by the photodetector placed in the oxygen module. The principle of measurement was introduced in the 3.2.2 section of State of the art chapter. Figure 22 Phase shift acquired by means of the HexaScreen’s DO measurement module and a Presens DO sensor. 4.1.4 Minibioreactor aeration, stirring & mass transfer. System’s aeration is performed by means of a set of seven electrovalves. Each Minibioreactor is connected to an On-Off electrovalve which opens periodically for a certain period of time in order to allow individual culture aeration. The system features two gas inlets that can be connected to air+CO2and N2gas supplies depending on the user’s needs. An extra switching electrovalve allows selecting between both gas supplies which is useful to perform OUR estimation by removing the oxygen of the headspace volume and measuring the oxygen extinction profile into the liquid phase (Dynamic method). Additionally, both gas inlets include means for humidification in order to compensate the loss of volume in the Minibioreactors due to evaporation. The following illustration shows the pneumatic model of the gas circuit, consisting of six solenoid valves one per each Minibioreactor connected to a common line served by the seventh solenoid valve which acts as the gas supply switcher. Figure 23 Aeration circuit model Bioreactors 4 80 Once the gas supplies are connected and both relevant pressures and flows are adjusted, air and nitrogen flow through the humidifier columns and water traps. However, given the fact that initially all valves are closed; gases will stop flowing once pressure equilibrium is reached. Under normal operating conditions, electrovalves will operate sequentially, starting again at the first electrovalve once after the sixth Minibioreactor has been aereated. This ensures that the flow rate is approximately the same for each Minibioreactor during an individually programmed operating cycle which enables the user to apply different aeration regimes. Figure 24 HexaScreen’s Aeration electrovalves timing chart The aeration method is a key point aspect in bioreactor design. However, is not the only parameter to take into account to ensure a sufficient mass transfer. Stirring shall also be considered if high cell concentration is wanted to be achieved. In stirred tank bioreactors proper mixing is required to avoid gradients of oxygen, nutrients and cell density [25], for bench scale bioreactors and bigger volumes the most widely used types of impellers are Marine, Rushton and Pitch-Blade, the selection of a certain type of impeller or even a combination of them mostly depends on the productivity target for a specific bioreactor, as well as the cell specie that is going to be cultivated. For bioreactor’s hydrodynamics characterization a number of parameters have been defined: Flow regime (turbulent, laminar), Mass transfer, Hold up time, Mixing time, Shear stress… However, when the application involves single use miniaturized bioreactors some other constraints arise, on one hand the impeller must be cheap and small enough to be included on a disposable product and fit into a small volume, on the other hand it must be able to support the required culture conditions. During the early stages of the development a simple stirrer was implemented by means of a stir barr located inside the Minibioreactor [1] driven externally by an electronically controlled electromagnet. However, it was found that the mechanical forces produced by the stir barr (friction and shear) were dangerous for growing mammalian cells. Hence, the finally Bioreactors 4 81 chosen stirrer for the construction of the HexaScreen’s Minibioreactor plate consisted of a flexible pendulum made of biocompatible silicone 4 with an enclosed Neodymium magnet. Figure 25 Flow topology simulation based on a flexible pendulum stirrer 5 . Experiments were carried out in order to test its functionality using Hybridoma as mammalian cell model; due to the vessel’s geometry it was found the existence of dead volumes between the walls of the Minibioreactor and the optical ports, leading limitations in terms of power transfer and stirring smoothness. Nonetheless, the pendulum was finally selected since it shown to be useful for animal cells being able to keep a Hybridoma culture alive for several days avoiding sedimentation. Figure 26 Left: Percentage of cell sedimentation. Right: kLavalues Both graphs are expressed as function of the stirring rate. [23] Regarding the driver, the concept of the classical magnetic stirrer based on AC asynchronous motors was rejected due to the presence of perishable moving parts. A different approach with no moving parts was developed (Figure below): 4 Rhodorsil® MF 970 USP (Bluestar Silicones) 5 This simulation was performed in the frame of collaboration with the Dept. of Chemical Engineering - Federal University of Rio de Janeiro. Bioreactors 4 82 Figure 27 Left.- Electromagnet’s connections and current senses. Right.- Magnetic field distribution. A four pole electromagnet electronically controlled by means of PWM signals was implemented. Such electromagnet was made of four solenoids arranged orthogonally and a ferromagnetic core geometrically designed to maximize the magnetic field within the Minibioreactor’s culture volume. In this way a driving magnetic field was obtained by the composition of the fields produced by the solenoids A, B, C, D which were connected in pairs A-D, B-C in such a manner that currents flow clockwise and counter clockwise producing complementary fields coming in and out on each pole. With some slight differences such concept of electromagnet has been used by the German company VARIOMAG to develop a range of magnetic stirring products [26]. However, the fact that they are not able to produce low rotation speeds, which is required for some sensitive animal cells i.e. Stem Cells, suggests that the solenoids are activated sequentially. Therefore, if a too long sequence period would be applied the stir bar would rotate in a discontinuous and unexpected manner, producing unwanted shear forces over the cells. On the contrary, the solenoid’s driving method used by the HexaScreen system allows a soft rotation from zero up to several hundred rpm’s, which is the optimal range for most animal cell applications. Figure 28 Electromagnet’s driving signals generation The block’s diagram above shows the driving signals generation scheme, the front block consists of two H-bridges and logic control modules that convert the PWM signals into currents flowing through the solenoids, this part of the circuit may be implemented by any Stepper Motor IC. However, attention must be paid when choosing the right IC since the H- Bioreactors 4 83 Bridge must include fly-back diode protections in order to avoid voltage spikes due to the solenoids reactance. What it makes possible to obtain a soft stirring motion is the type of signal used for modulation, a sinusoidal oscillator was used to feed both modulators with a 90° phase difference, finally regardless of the stirrer’s physical limitations like mass, geometry, or medium’s viscosity, the obtained stirring rate was directly proportional to the oscillation frequency. This was proofed in the HexaScreen system for aqueous solutions between 0 to 400 rpm’s. Even though many different solutions were possible for the PWM signals synthesis. The fact that the required oscillation frequency to achieve a high stirring rate is still very low (167 Hz per 1.000 rpm’s) made possible to use of an inexpensive microcontroller to generate the PWM signals by Direct Digital Synthesis. 4.1.5 Experimentation workflow & results. In this section a brief explanation of the HexaScreen system experimentation workflow is developed, as well as the obtained results for some possible applications of the system using productive animal cell lines are explained. Further results including OUR estimation by means of the dynamic method will be discussed on section 5.3 Steps of the experimentation workflow: •Experiment definition: It consists on the definition of the experiment’s conditions and parameters for a certain application: Cell type, inoculum’s concentration, stirring profile if needed (i.e. for adherent cells culture on micro-carriers), aeration rates per each MBR, temperature, composition of the culture medium, etcetera. Once all variables have been defined an experiment file is to be created and then being loaded by the acquisition software just before starting the cell culture. •Workstation preparation: Some previous verification shall also be performed before starting the experiment. The pressure of the gasses supply and the humidification unit’s water level shall be checked to ensure a proper mass transfer as well as to avoid medium evaporation along the culture time. •Minibioreactor plate inoculation: It consists on the procedure of preparing the Minibioreactor plate to place it into the workstation. Filling the external water bath and each Minibioreactor with fresh medium and inoculate them with the selected cells. Before the inoculation step is advisable to preheat both water and culture medium in order to minimize the cell stress due to temperature changes. Moreover, due to the fact that most of the animal cell culture mediums are carbonated it becomes necessary to pre-set its pH to the cell’s physiological value by placing it inside a CO2incubator for a certain period. Once the plate is ready for use and just before starting the experiment, the pH of the basal medium shall be measured using a benchtop pH-meter in order to provide the initial pH conditions required by the acquisition software. Bioreactors 4 84 Figure 29 Minibioreactors plate inoculation under sterile conditions •PreCulture stage: Once the Minibioreactor plate is located inside the insulation chamber the experiment can be started. To do so, some previously defined experiment file may be loaded by the software and the initial pH values or other variables may be entered depending on the actual conditions. After that, the experiment configuration is uploaded to the workstation and the Preculture wizard is launched. This is a wizard like procedure to ensure the right temperature and aeration conditions before any reference measurement is acquired, once the temperature reaches the culture setpoint both fluorescence and light spectra references are acquired per each channel and related to the 100% DO and the initial conditions of pH and OD respectively. Figure 30 Placing the Minibioreactor’s plate •Acquisition and control stage (Results evaluation): Once the PreCulture stage finishes the system begins to display on-line data on Temperature, DO, pH, OD and Cell concentration. The following culture examples demonstrate the HexaScreen system as a useful tool for screening proposes in animal cell biotechnology. Bioreactors 4 85 Example I Medium optimization – Fetal Calf Serum (FCS) efficiency evaluation Experiment description The aim of the experiment was to evaluate the efficiency of two different FCS supplements (batches A and B) in comparison to a commercially FCS (batch C) when culturing NS1 mouse myeloma cells. Prior to the experiment, cells were thaw out and cultured for a week in DMEM + 10% standard FCS (batch C). During inoculation, cells were rinsed twice to eliminate the FCS used during cell expansion, and every Minibioreactors was inoculated from this common inoculum at 2·105cells/ml with a total volume of 12 ml per Minibioreactor. The three FCS batches were added as follows: FCS batches and concentration used per each Minibioreactor Minibireactor FCS batch FCS concentration MBR1 A 10 % MBR2 A 5 % MBR3 B 10 % MBR4 B 5 % MBR5 C 10 % MBR6 C 5 % Culture Conditions Medium: DMEM + 3 different FCS batches (“A”, “B” and “C” as a control) at 5% and 10% Cell line: NS1 mouse myeloma. Volume: 12 ml Initial cell concentration: 2·105cells/ml Stirring rate: 200 rpm Bioreactor temperature: 37 °C Aeration rate: 0.2 slpm (Duty cycle: 2%) Results The graphs below show the on-line profiles of optical absorbance, cell concentration, pH and dissolved oxygen obtained along 6 days of culture. Optical absorbance and Cell concentration may be considered proportionally equivalent for OD values below 0.8 Bioreactors 4 86 pH and DO evolution Analyzing the different profiles it can be observed that the cell growth is quite similar for every one of the FCS batches tested, with almost the same profile during the tropophase and reaching the same cell concentration after stabilization 23·105to 25·105cells/ml. The slightly lower cell concentration reached by the MBR5 compared to the other MBR’s it can be related to different error sources. For instance when the Neubauer chamber is used for the cell counting procedure previous to the inoculation typically yields an estimation error between 20 to 30 %. Another common source of error may be a drift of the optical absorbance measurements due to the existence of small bubbles interfering with the optical path during the PreCulture stage, such bubbles can easily appear if the water bath or the culture medium are not pre-heated or the Minibioreactor’s plate preparation takes too long before placing it inside the workstation and starting the experiment. The pH and dissolved oxygen graphs also shown similar profiles indicating almost the same acetic acid production and oxygen consumption rates independently of the FCS present in the medium. Only a slightly faster acidification of medium can be observed with controls, although it is barely significant. A deeper analysis yields the following data. Cell growth parameters Minibireactor /batch Growth rate [h-1] Lag phase duration [h] MBR1/A 26.4 11 MBR2/A 22.8 17 MBR3/B 20.4 2 MBR4/B 19.2 3 MBR5/C 16.8 5.5 MBR6/C 19.2 1.5 Conclusions The reproducibility and time resolution of the measurements led not just to a proper determination of the culture yield and the tropophase duration, but also a good estimation of the cell growth parameters. In this case it was found that the efficiency of the two FCS batches tested was very similar compared to a commercial FCS batch, taking the same time to achieve the same maximum cell concentration. However, batch A lasted longer during the lag phase. Bioreactors 4 87 Example II Clone selection – KB26.5 vs. BHRF1 Experiment description Three parallel growths for each Hybridoma clone (KB 26.5 and BHRF1) for the same culture conditions. The aim is to establish which clone offers the best tradeoff between culture time and cell concentration reached. Culture Conditions Medium: DMEM + 10% (Fetal Calf Serum) Cell line: Hybridoma KB26.5 and BHRF1 Volume: 12 ml Initial cell concentration: 2·105cells/ml Stirring rate: 200 rpm Bioreactor temperature: 37 °C Aeration rate: 0.2 slpm (Duty cycle: 2%) Results Comparison of a Hybridoma culture growth (optical cell concentration) and its metabolic (pH and dissolved oxygen) curves obtained with HexaScreen® for the two different clones: KB 26.5 (black lines) and BHRF1 (green lines) after three days of culture. Cell concentration obtained from the OD measured at 650 nm. pH and DO evolution for both clones. For a better comparison between both hybridoma clones, cell growth rate can also be Bioreactors 4 94 4.2.2 MonoScreen® Fed-Batch architecture As its predecessor HexaScreen, the MonoScreen Fed-Batch system also includes customized electronics, mechanisms and some optical parts. The pictures below show a snapshot of its internal construction and front panel. The box is divided in two sections thermally insulated to ensure an almost constant working temperature of the gas circuit, to avoid the possibility of condensation within the gassing pipes or any unwanted thermal drift of the electronics. The non-insulated part holds the power supply unit and the liquid cooling circuit heat sink. Figure 34 MonoScreen Fed-Batch. Bioreactors 4 95 The following diagram shows the main elements of the DO and pH control loops. Due to the fact that their operation is connected to each other and to the sampling mechanism, which is another important functionality of the system, such elements become especially sensitive parts of the workstation. That is why the design of the system’s global architecture was not faced until the right control strategies were chosen for each variable and key parts selected. Figure 35 DO & pH control model. Essentially, the DO & pH control loops are acting over the O2and CO2partial pressures within the gas phase of the Minibioreactor by properly regulating the Air, N2and CO2flows, which are mixed in the humidification unit. Each individual gas flow is regulated by means of a proportional electrovalve controlled by a Pulse Width Modulated (PWM) signal. In order to keep the total gas flow constant regardless of its composition, a condition must be observed. The addition of the duty cycles of every PWM signal must be constant. On one hand, the ratio between the Air flow and the other gases states the pO2and therefore DO, on the other hand increasing the ratio between CO2and N2increases medium’s acidity. On the contrary, to reach a high pH value no CO2should flow through the Minibioreactor’s gas phase and a certain volume of an alkali solution like NaOH should be delivered by means of the syringe pump. The architecture was based on the RCM4300 Rabbit core; a more powerful Cprogrammable module than the one used for the HexaScreen system. It features a multichannel 12 bit A/D converter, Ethernet connectivity and mass storage capabilities by means of an on board 1 GB SD card. Most of the communication with the different peripheral devices was based on an I2C bus. The workstation is made of the following modules: Control, Power drivers, Syringe pump drivers and the Signal acquisition and conditioning module. Bioreactors 4 96 The Control module holds the system’s intelligence and is in charge of the user interface functions (Touch panel, web & ftp server), data acquisition and logging and culture control and system’s supervision. The control module counts with gas pressure and flow sensors as well as proportional valves for the gas mixing control, multiplexers for digital I/O and an on board thermometer for internal monitoring. Figure 37 MonoScreen Fed-Batch Control Module block diagram Bioreactors 4 97 The Power driver’s module holds the parts related to some of the main functions like the Minibioreactor’s stirring and temperature control, and secondary functions like the Humidifier’s temperature control, as well as the needed elements for the automated calibration of the Humidifier’s thermometer. The Power driver’s module is connected to the Control module by means of an I2C bus which allows multiplexed access to different peripheral IC’s. One additional feature is the possibility of dual operation; the module may be remotely controlled (by the RCM4300 microcomputer) or manually operated in order to simplify service tasks. Figure 38 MonoScreen Fed-Batch Power driver’s module block diagram Bioreactors 4 98 The MonoScreen Fed-Batch system features two identical Syringe pump driver modules consisting of a 10 ml syringe pump and its corresponding electronics, one module is meant for the administration of pH control solutions, the other is intended for feeding the Minibioreactor. The second features the additional capability to keep the syringe’s temperature within a certain range. Therefore, the Feeding module’s printed circuit board holds some extra elements not just for driving the mechanics but for temperature control and automated thermometer calibration. Syringe pump driver modules are directly controlled by the RCM4300 microcomputer by means of general purpose I/O bits in order to optimize control over the steppers motors and hence over the liquid addition resolution. Like the Power driver’s module they also can be operated manually for service purposes. Figure 39 MonoScreen Fed-Batch Medium addition module block diagram. The Signal acquisition and conditioning module is responsible for the conditioning and acquisition of the signals provided by the sensors directly applied to the Minibioreactor. The different analog signals are collected via a multichannel, 12-bit ADC, and transmitted to the control board via an I2C bus. The module includes an auto-calibrated thermometer, two fluorescence detectors (DO – pH), and one NIR laser Turbidimeter made of a low power 850 nm laser and three logarithmic photodetectors, based on a LOG112 precision logarithmic amplifier to increase the Turbidimeter’s measurement range. Turbidimeter’s operation is represented by the figure below where the gain of the three photodetectors (Reference, Transmitted & Scattered) must be matched kRef ≈kFwr ≈kSctt. Bioreactors 4 99 = ∙log Where: •v0[V]: Signal response. •IPht i [A]: Photocurrent proportional to the light intensity for every phodetector (Reference, Transmitted & Scattered) •Ii[A]: Phodetector offset current control. •ki[ ]: Phodetector sensitivity. Figure 40 Turbidimeter’s Photodetector module block diagram. Figure 41 Signal acquisition and conditioning module block diagram. Bioreactors 4 100 The DO measurement is based exactly in the same instrumentation technique used by the HexaScreen system, XOR phase detection (3.2.2). Regarding the pH measurement, this is based on Dual Lifetime Referencing (DLR). Where pH is related to the fluorescent response of two independent fluorophores one of them used as the reference due to its lack of sensitivity to the pH and the other to record the pH evolution as its lifetime changes. The pH measurement circuit used is analogous to the DO measurement scheme but including a second phase detector for the reference signal. Figure 42 shows the pH sensor fluorescent response when irradiated at 470 nm. Two clearly differentiated emission peaks were detected at 525 nm (measurement wavelength) and 610 nm (reference wavelength) consequently two specific optic filters were chosen for rejection of the ambient light and the backscattered excitation. Figure 44(d) shows the optic filters attached to the photodiodes for both DO & pH measurement circuits. Figure 42 Fluorescent emitted spectra by a Presens sensor (PS-HP5-D5-US pH) When irradiated at 470 nm, for different acidity conditions from 1.8 to 10.8 pH Figure 43 Schott OG570 & Semrock FF01-525/15-25 long-pass and band-pass filters Transmittance curves. pH reference and measurement wavelengths. [29] [30] Bioreactors 4 101 Figure 44 MonoScreen Fed-Batch Hardware: A)Control module, B) Medium addition module, C) Power driver’s module, D) Measurement module, E) Logarithmic photodetector. A big improvement on the MonoScreen’s firmware was made in comparison with the HexaScreen system. The development of the built-in www and ftp servers made unnecessary the need for having a customized client software resident on the host computer. This way any computer or “smart” device running a standard browser or ftp client could easily gain access to each individual workstation as far they were connected to the same LAN. Additionally, a FAT file system was implemented for recording different type of data (setup parameters, calibration values, experiment results and culture recipes). Culture recipes are scripts describing all the conditions and events to be carried during the experiment i.e. Administration of a certain feed volume or the execution of a specific stirring profile. The firmware code is mostly based on Digi’s Dynamic C real time functions and consists of a number of concurrent processes to perform very specific tasks (acquisition and control, data logging, etc…) such processes are enabled and disabled depending on the conditions of a simple state machine. Bioreactors 4 102 Figure 45 MonoScreen Fed-Batch states diagram and list of concurrent task •Initialisation: All working variables are initialised with the configuration data recorded in the flash memory and different configuration files in the FAT file system. If the loaded data would be found to be corrupted or inexistent this circumstance would be treated as a fatal error and the system would move to the “Final Error” state. After successful initialization of the working variables an automated verification of the internal I2C bus as well as on the proper operation of most actuators is performed. Failure of any peripheral device would be reported. However, the system would not immediately move to the “Final Error” state since the system could be still functional depending on the error importance. Next state shall be “Instrument Calibration”; However, depending on the value of the “Last State” parameter recorded on the flash memory, any value different than “Idle” points to an abnormal power up situation due to an unexpected lack of power or analogue circumstance and then the user is prompted to restart the system. •Instrument Calibration: After initialisation, the user is prompted to calibrate the system at least once previously to any experiment. The calibration does not necessarily need to be executed immediately after the initialization. At this point the calibration step can be skipped and come back to it later from the “Idle” status. The Calibration process takes care of the following tasks: Determination of the syringe pumps resolution. Determination of the off-set parameters for pressure gauges a flowmeters. Automated calibration of the internal thermometers. Testing the proper operation of the thermoregulation and stirring control loops. Determination of the DO calibration parameters. Determination of the NIR turbidimeter base-line value. Whilst the “Instrument Calibration” stage the execution of any other concurrent tasks is interrupted. •Operation: Once the Minibioreactor is placed and connected to the dock bay, the system enters the “Operation” status and the following concurrent tasks are activated: Bioreactors 4 103 “Data Logging”, “Alarms Control” “Acquisition & Control”. When the workstation enters this stage the active recipe is automatically executed. That means that the different control loops are started and set points are applied following the profile previously programmed by the user. Monitorization of the experiment must be done by means of a www client that retrieves the real time data from the instrument through the network. Nevertheless, critical data regarding the current culture conditions (Temperature, DO, pH, stirring, etc…) can also be locally checked via the touch panel. The operation status also allows the user to perform other tasks such as on-line modification of the active recipe, aliquot sampling and the introduction of offline data that will later be included in the experiment’s report file. •Idle: After the “Initialisation-Calibration” process, the system enters the “Idle” state and the tasks “Calendar”, “www Server” and “ftp Server” are started. They are the ones providing the user with access to the FAT file system to edit or upload recipes and download experiments results. The idle state can also be entered after finishing an experiment or a calibration process. •Final Error: If some fatal error happens, the RCM4300 will try solving it by means of a wait and retry procedure. Errors that can be solved are the ones related to the I2C communications bus, access to memory and the lack of some feedback signals i.e. The syringe pumps optical encoders or the power supply OK signal. If finally the wait and retry procedure would time out the system would move to the “Final Error” state and notify the corresponding error code through the touch panel. 4.2.3 MonoScreen® Fed-Batch Optical Layout. Due to the more demanding requirements related to the maximum cell concentration for the MonoScreen Fed-Batch system and the fact that diluted pH indicators like Phenol red may interfere with some spectrophotometric and fluorescent essays, or even interact with some cell types [31] [32]. A different approach to the optical layout was planned for the MonoSreen Fed-Batch. Figure 46 Preliminary optical layout initially considered. Bioreactors 4 110 Figure 54 MonoScreen Fed-Batch DO an pH optical setup 4.2.4 Minibioreactor liquid handling (Medium addition & sampling) The ability to take samples and adding mediums in an automated manner is a very important feature of any bioreactor; especially for fed-batch, perfusion or continuous operation. For the MonoScreen Fed-Batch system, the feeding mechanism was based on syringe pumps which are coherent with the single use product concept and provide an Bioreactors 4 111 unmatched addition resolution in comparison to the peristaltic pumps, something which is pretty important when coming to a bioreactor within the millilitres scale. Therefore, the reservoirs were made of two 10 ml syringes connected to the vessel by means of silicone tubbing and 30G x1/4” dispensing needles used to provide a typical addition resolution around 5 to 8 l/drop. In some cases the feed medium may contain temperature sensitive components. That is why the second syringe pump was additionally equipped with a TEC device based cooling system capable for to keep the reservoir’s temperature below 4 °C all along the experiment. Figure 55 Syringe pump’s testing setup. Observe the reading of the multimeter, 1.988 °C An innovative sterile sampling device for single use bioreactors was also integrated within the system and constituted a patent request derived from the thesis evolution [27]. Its performance is based on the physical principles of fluid mechanics (Law of connected vessels, Tate’s Law and Hydrostatic pressure) and the use of a non-sealed vessel pneumatically driven through gas filters. The cross-sectional view of the device is shown in the Figure 57. The sampling procedure comprises three stages: A) Normal regime. The pressure gradient through the suction cannula is null or too small to force the fluid flow towards the accumulation chamber. B) Accumulation stage: The sampling port cap is replaced by an Eppendorf tube and the pressure gradient between the vessel’s headspace and the accumulation chamber is increased to fill the accumulation chamber. Once filled, the sample volume remains there due to the fluid’s surface tension experienced around the dispensing nozzle. C) Sample dispensation: The pressure gradient is reversed and the sample volume is delivered in drops avoiding the presence of any sort of continuous stream between the accumulation chamber and the Eppendorf tube. Therefore, the possibility of external contamination is also minimized. Bioreactors 4 112 . Figure 56 Parts of the sterile sampling device Figure 57 Stages of the sampling procedures Still, there are certain dimensional aspects to keep in mind for a proper operation of the device. The first key point is the ratio between the nozzle’s diameter and the height of the accumulation chamber. In order to avoid uncontrolled dripping of the sample volume, the weight of the mass accumulated over the nozzle’s section must be smaller than the fluid’s surface tension around the nozzle perimeter. This is an empiric ratio where the maximum height of the accumulated fluid is inversely proportional to the nozzle’s radius and the drop’s contraction and surface tension parameters may be determined through the Tate’s Law, which states the minimum mass of a drop to break the surface tension around the nozzle where is hanging from and fall. Bioreactors 4 113 ∙ = ∙2∙ ∙ ∙ Figure 58 Representation of the Tate’s Law. · ∙ · > ∙2∙ → · > ∙2∙ ∙ Where: •k[ ]: Experimentally found contraction factor for a given nozzle. •[N/m2]: Surface tension (72,75x10-3 N/m per water at 20 ºC). •[kg/m3]: Fluids’s density. •h[m]: Fluid’s column height. •ro[m]:Nozzle’s radius. i.e. For a nozzle’s radius of 0.25 mm and water at 20 °C we get the following empiric data: mg= 11mg ∙ =68,8×10 Where: •g[m/s2]: Gravity’s acceleration. •mg[kg]: The mass of a drop. Then, it is possible to state the maximum height of the water column: ∙2∙ ∙=1,4×10 → > ∙2∙ ∙ ∙ >56mm Obviously, these are very rough calculations, where other phenomena and fluid properties such as capillarity, viscosity or geometrical imperfections are not taken into account. Therefore, it is advisable to be conservative when designing the height of the accumulation chamber and to apply a correction factor no bigger than 0.5. This will make possible to extent its use to other application conditions different than the reference ones. Bioreactors 4 114 Additionally, the diameter of the sampling port must be wide enough to avoid drops touching its inner walls. Once the mass of a drop is known, if we approach its shape as a sphere, it will be possible to calculate its radius and therefore its ratio versus the nozzle’s radius: ∙ = → = 3∙∙4∙ =3∙ ∙ ∙ ∙2∙ This gives for the previous reference conditions a drop’s radius of 1.38 mm. A sampling port with a minimum inner diameter bigger than twice the drop’s radius should be enough to ensure a clean and dry wall which is the second key point to properly operate the sampling device. Of course, the fluid’s characteristics, the geometrical design and other imperfections like verticality or the length of the sampling port could also modify the actual result and should obviously be taken into account to preserve the inner wall of the sampling port to get wet. Figure 59 Simplified representation of the sampling port. 4.2.5 Minibioreactor aeration, stirring & mass transfer. The MonoScreen’s aeration strategy is completely different in comparison to the HexaScreen’s system. Figure 35 shows the gas circuit with all the necessary valves and sensors for having an accurate control of the gas flow, as well as the humidification vessel used to compensate evaporation and for proper gas mixing. Due to the need for a constant ratio between the gas flow through the bioreactor with respect to the culture volume, the aeration circuit was conceived to provide a continuous gas flow even though continuous aeration uses to be an economical issue for bench-scale bioreactors. Fortunately, due to the reduced size of the MonoScreen Fed-Batch vessel the total gas consumption is not significant, typically from 15 to 30 smlpm for animal cell culture and from 30 to 60 smlpm for microbial. Additionally, the use of Rushton and Pitch-Blade turbines as well as the standardized dimensional design of the Bioreactors 4 115 vessel, leads to an optimized oxygen transfer that increases the Minibioreactor’s capability to hold a higher cell concentration. The flow topology and turbulence were investigated by means of Computational Fluid Dynamics (CFD) simulation. Such study gave three important results: a) Laminar flow happens approximately below 500 rpm and Turbulent flow starts over 750 rpm (Maximum kl·a will be obtained over 750 rpm when using baffles. This is a very aggressive agitation condition typical of microbial fermentation). b) Dead volumes behind the baffle areas are almost negligible over 500 rpm (The whole culture volume will contribute to the cell growth). c) On one hand, mammalian cell culture will be possible even for a quite high stirring rate with Pitch Blade turbines. On the other hand, optimization of microbial growth could be achieved by the use of baffles and Ruston turbines over 750 rpm. Figure 60 Up: Transition from laminar flow to turbulent regime. Down: Evolution of the Flow topology 9 . Some trials were carried out to evaluate the actual kl·a values for different stirring and aeration rates. The results shown that for simple sparging (by means of a bent needle submerged into the medium with no baffles), the kl·a values were high enough to ensure proper scalability with most Bench-Scale bioreactors. Despite that, if an even higher growth is wanted, it would be necessary to mimic the production conditions by using a microdifusor and/or baffles to reduce the bubble’s size and increasing both, the gas to liquid interfacial area and the bubble’s hold up time. 9 This simulation was performed in the frame of collaboration with the Heat and Mass Transfer Technological centre – Polytechnic University of Catalonia Bioreactors 4 116 Figure 61 Left: kLavalues for Rushton & Pitch-Blade turbines for different aeration rates. Right: kLavalues for Rushton & Pitch-Blade turbines for different stirring rates. It was found that the effect on the kl·a depending on the aeration rate shows a similar behavior for both turbines if no baffles were used, being slightly bigger for the Pitch-Blade. However, when coming to the effect of the stirring the Rushton turbine shown a bigger slope. These results were coherent with the previous simulations and reinforce the idea that the Pitch-Blade turbine is a good choice for cell species with moderate oxygen consumption [28]. On the contrary, the Rushton turbine is the one to be used, always with baffles, for very prolific cell lines. Nonetheless, the most important conclusion was that the achieved MonoScreen’s mass transfer behavior is comparable to most common benchtop stirred tank bioreactors [40]. 4.2.6 Experimentation workflow & results. The MonoScreen Fed-Batch experimentation workflow is quite similar to the one for the HexaScreen system. However, MonoScreen is mostly intended for controlling the physical and chemical culture conditions which are connected to the algorithm for the OUR estimation and not that much for monitoring. A deeper discussion on the more relevant results will be conducted in section 5.3.4 Steps of the experimentation workflow: •Workstation preparation: The system needs to be calibrated at least once before starting any experiment. It consists of several steps for obtaining the calibration parameters of the internal instrumentation (Temperature, pH, DO & OD) and to ensure a proper operation of the gas mixing system and servos (Temperature control loops, stirring and syringe pumps). Such process needs to be performed using the same Minibioreactor that will later be used to perform the experiment. Both, the Minibioreactor and the syringe reservoirs are required to be prefilled with the culture and feed mediums. Bioreactors 4 117 •Minibioreactor inoculation: After the calibration, the Minibioreactor and the workstation are ready to be used. However, it is advisable to pre-set the medium’s pH to the cell’s physiological value by placing the Minibioreactor together with a sample of the basal medium inside a CO2incubator for a certain period before being seeded. Then, the basal medium pH shall be measured using a bench-top pH-meter in order to introduce the initial pH conditions into the culture recipe. Figure 62 Minibioreactor inoculation under sterile conditions Figure 63 Minibioreactor attached to the workstation’s dock bay Bioreactors 4 118 •Experiment definition: Before starting the experiment a culture recipe needs to be created; this can be done by editing a template recipe. Where besides of some information about the cell’s type, the mediums used and the experiment’s initial conditions, a set-points table for the operation of the different control loops needs to be defined as well as any user defined variable that could be wanted to be introduced during the experiment after some off-line analysis. •Acquisition and control stage (Results evaluation): The experiment is started through the workstation’s touch panel. Once the MBR is connected, the active recipe is executed automatically and data begin to be displayed and recorded. Interaction with the experiment in terms of modifying or adding new set-points as well as introducing Off-line data will need to be done remotely via the built-in web server. Example I Characterization of the NIR Turbidimetric response under high cell density conditions. Experiment Description A highly concentrated solution of fresh baker’s yeast was used to study the accuracy of the NIR Turbidimeter. A Minibioreactor was filled with 20 ml of the mentioned solution and the absorption of light was measured. Afterwards, the solution was diluted 50 % and the absorption measured again. The operation was repeated until the detection limit was reached and a crystal clear solution was obtained. In order to ensure a homogeneous solution, stirring and temperature were kept constant. Finally the Turbidimetric Absorption (TA) characteristic was compared with its logarithmic form (OD). Culture Conditions Medium: Physiological saline (NaCl, 9.0 g per liter). Also used for the successive dilutions. Cell line: Saccharomyces Cerevisiae Volume: 20 ml Initial cell concentration: 100 g per liter Stirring rate: 500 rpm Bioreactor temperature: 37 °C Results The comparison of both biomass representations can be observed in the figure below. On one hand, the OD shows the well-known saturation behaviour as the biomass density increases. On the other hand, it can be seen how up to an approximate concentration of 20 g/l TA displays a clearly linear trend, which is more than two orders of magnitude over the OD’s “linear” range. Despite that TA becomes non-linear for the higher values of biomass density; it is obvious that offers a more appropriate expression to display the on-line evolution of the biomass. On the other side, the non-linear stretch of the TA characteristic can easily be calibrated by means of an error function of the form: TA∗= ·TA+(+1)·TA+ Bioreactors 4 119 Nevertheless, the calibration parameters a, b, c will not be valid for every cell type. Hence, in order to provide an accurate estimation a previous characterization of the NIR Turbidimeter will always be necessary. Comparison between the TA and OD within a wide range of biomass density. This example can be considered as a worst case, and demonstrates the viability of NIR Turbidimetry for monitoring a wide range of cell species under high cell densities conditions. Example II Evaluation of the DO control capabilities. Experiment Description As widely explained throughout the introduction DO is one of the most important scaleup parameters in Bioprocess engineering. Accuracy of the OUR estimation is highly dependent on the stationary error of the DO control loop. Therefore, the optimization of the DO control Loop parameters, as well as the determination of the system’s control accuracy within the specified control range is a must. 5 different DO set-points from 10 to 90 % were programmed every 2.5 hours at a constant temperature and stirring rate. Culture Conditions Medium: Distilled water Volume: 25 ml Stirring rate: 500 rpm Bioreactor temperature: 37 ⁰C Aeration rate: An approximately constant gas mixture of 1.88 vvm was applied. Bioreactors 4 126 [24] N. P. Q. F. H. K. M. C. R. R. A. B. G. C. L. J. L. M. Thanassis Papaioannou, “Performance evaluation of fiber optic probes for tissue lifetime fluorescence spectroscopy,” in SPIE - Advanced Biomedical and Clinical Diagnostic Systems, San José, CA, 2003. [25] P. M. Doran, “Fluid Flow and Mixing,” in Bioprocess Engineering Principles, Sydney, Elsevier Science & Technology Books, 1995, pp. 141-153. [26] 2-MagUSA, “2-MagUSA magnetic e-motion,” [Online]. Available: http://www.2magusa.com/. [Accessed 17 10 2012]. [27] A. Fontova and E. Sarró, “Sample Taking Device”. ES-EU Patent P2144EP00, 9 March 2012. [28] R. Mirro and K. Voll, “Which impeller is Right for your Cell Line? A guide to Impeller Selection for Stirred-Tank Bioreactors,” BioProcess International, pp. 52-57, 2009. [29] UQG Optics, [Online]. Available: http://www.uqgoptics.com/materials_filters_schott_orange_OG570.aspx. [Accessed 26 5 2014]. [30] Semrock, [Online]. Available: http://www.semrock.com/FilterDetails.aspx?id=FF01525/15-25. [Último acceso: 26 May 2014]. [31] Y. Berthois, J. A. Katzenellenbogen y B. S. Katzenellenbogen, «Phenol red in tissue culture medium is a weak estrogen: Implications concerning the study of estrogen-responsive cells in culture,» Proceedings of the National Academy of Sciences of the United States of America, vol. 83, nº 8, pp. 2496-2500, 1985. [32] A. Bukovsky, M. Svetlikova y M. R. Caudle, «Oogenesis in Cultures Derived from Adult Human Ovaries,» Reproductive Biology and Endocrinology, vol. 3, nº 5, p. 17, 2005. [33] M. P. D. &. E. T. Davies, Non-Invasive Biomass Monitor with Wide Linear Range, BugLab LLC (www.bugbab.com), 2012. [34] J. McFarland, «The Nephelometer:An instrument for estimating the number of bacteria in suspension used for calculating the opsonic index and for vaccines.,» Journal of American Medical Association, vol. 49, nº 14, pp. 1176-1178, 1907. [35] T. R. L. K. K. N. T. A. M. Paw Dalgaard, “Estimation of bacterial growth rates from turbidimetric and viable count data,” Food Microbiology 23 (1994), vol. 23, pp. 391-404, 1994. [36] EPA, «EPA Guidance Manual, Turbidity Provisions,» nº 11, pp. 1-13, April 1999. [37] P. W. K. B. H. Asa Kolmert, “A fast and simple turbidimetric method for the determination of sulfate in sulfate-reducing bacterial cultures,” Journal of Microbiological Methods, vol. Bioreactors 4 127 41, p. 179–184, 2000. [38] M. Z. B. M. Ahmad Fairuz Bin Omar, «Turbidimeter Design and analysis: A review on Optical fiber Sensors for Measurement of Water Turbidity,» Sensors, vol. 9, pp. 83118335, 2009. [39] M. Sadar, Turbidity Measurement: A simple Effective Indicator of Water Quality Change. [40] V. Glaser, «Bioreactor and Fermentor Market trends,» Genetic Engineering & Biotechnology News, vol. 29, nº 19, 2009. [41] Parker, “VSO-Low Flow Miniature Proportional Valve,” August 2013. [Online]. Available: http://ph.parker.com/us/en/vso-low-flow-miniature-proportional-valve. [Accessed January 2015]. [42] X. G. Montoya, “Estudi d'estratègies de cultiu per cèl·lules animals basades en eines de instrumentació i control,” Memòria per optar al grau de doctor per la UAB, Progama de doctorat en Biotecnologia, Bellaterra, Novembre 2000. [43] B. M. Weidgans, New Fluorescent Optical pH sensors with Minimal Effects of ionic stregth - Dissertation zur Erlangung des Doktorgrades der Naturwissenschaften, Regensburg: Naturwissenschaftlichen Fakultät IV – Chemie und Pharmazie der Universität Regensburg, 2004. [44] Presens, pH-10 Mini Instruction Manual, Regenburg: Precision Sensing GmbH, 2010. Simplified implementation of the OUR stationary liquid mass balance method. 5 128 Simplified implementation of the OUR stationary liquid mass balance method. The OUR stationary liquid mass balance estimation method offers advantages in terms of estimation accuracy and cells stress due to the constant DO concentration. However, the need for sophisticated instrumentation like mass flow controllers and sometimes gas analyzers, has historically limited its use. In this chapter, a new simplified implementation for the continuous estimate of the OUR based on inexpensive electrovalves is introduced. It will be demonstrated to be not just a cheap but a reliable alternative to monitor the metabolic activity in many biotechnological processes where the lack of knowledge on the culture conditions could become a limiting factor. Simplified implementation of the OUR stationary liquid mass balance method. 5 129 5.1 Method description and modelization. The proposed method consists on controlling the gas phase oxygen molar fraction by means of regulating the flow provided by two independent gas supplies (Oxygen and Nitrogen). Such regulation will be performed by two PWM driven electrovalves leading to the accurate control of the DO concentration. Additionally, the control loop internal signals will be used for solving the OUR without generating a glimpse of cellular distress. For a better understanding of the control mechanism, it will be useful to look at the model of the simplest bioreactor and consider some physical analogy. An electrical circuit in this case. The model below shows a gas supply Ps, connected to a bioreactor through an ideal pipe (no resistance against the flow or accumulation will happen) and an inlet filter represented by pneumatic resistance Rinlet, a certain volume of gas flows from the supply to the Bioreactor Gin and finally goes out through the outlet filter Routlet. Then, as stated by the Henry’s Law diffusion will happen between the gas phase and the liquid phase until the equilibrium situation is reached. Notice that this is no more than a simple serial circuit where if we consider the filters to be identical, a voltage divider analogy can easily be applied and the oxygen’s partial pressure in the gas phase calculated as follows:   100100 2 22 2 O atmfiltre O atm atms y PRG y P PP pO          Figure 1 Simplest aeration bioreactor model Hence, for a continuous air flow the DO concentration in equilibrium with the gas phase is: H y P PP H pO C O atm atms L 100 2 2 2 *          Where: Simplified implementation of the OUR stationary liquid mass balance method. 5 130 •Patm [atm]: Atmospheric pressure. •yO2 [%]: Oxygen’s molar fraction in the gas phase. •H[l·atm·mol-1]: Henry’s constant. Let’s now take into account the same model adding a switching valve that follows a certain duty cycle and periodically interrupts the inlet flow. In this case, the mean value of the oxygen partial pressure can be suspected to be modulated by such duty cycle: Figure 2 Simplest aeration bioreactor model Where: •T[s]: Switching period. •a[%/100]: Duty cycle (0…1) This would be true for an ideal model free of any "dead" volume conferring memory properties to the pneumatic circuit, i.e., to assert that there is a direct relationship between the oxygen’s partial pressure and the duty cycle operating the valve some element must be incorporated in order to get rid of the inherent "capacity" introduced by the gas phase. The strategy will consist on the rapid replacement of the air volume existent in the gas phase by means of another gas containing no oxygen (typically nitrogen).This implies the following two approaches: 1) The replacement speed must be virtually instantaneous so the global flux can be considered continuous. 2) No existence of dead volumes. Therefore, the pneumatic circuit admits a linear model without memory elements. In order to keep ahead with the development the following agreement is also taken: All the pressure expressions are related to the atmospheric pressure. The direct relationship between the mean value of the oxygen partial pressure and the duty cycle can be demonstrated through the Mean value theorem for integrals, which is visually represented by the following figure: Simplified implementation of the OUR stationary liquid mass balance method. 5 131 Figure 3 Gas switching makes the DO concentration proportional to the mean value of the oxygen partial pressure in the gas phase.   100 2 100 22 2 O atm atms O atmfilter y P PP y PRGpO           Once the linear relationship between the mean value of the oxygen partial pressure and the duty cycle is proofed, is equally possible to calculate the DO concentration in equilibrium with the gas phase for a discontinuous air supply * L C : H y P PP H pO C O atm atms L 100 2 2 2 *            Obviously, for a discontinuous but periodic air flow the value of * l C will tend to be a times the value obtained for a constant flow. This suggests the possibility to implement a DO control loop by means of the iterated switching between both gas supplies. Figure 4 shows the basic diagram of such control loop. Nevertheless, two key conditions regarding the switching and measurement periods need to be taken into account in order to keep the DO concentration constant: ≤0.1·; <10 i.e., Imagine a bioreactor intended for low concentration microbial fermentation, which features a mean mass transfer coefficient of kL·a =20 h-1. That implies a switching Simplified implementation of the OUR stationary liquid mass balance method. 5 132 period smaller than 18 s and a maximum measurement period of 1.8 s. This is technically feasible. However, the use of the switching valve topology could become problematic for bioreactors with higher kL·a values. Figure 4 Gas phase replacement control loop topology. Hence, the mass balance equation may be written as a function of the duty cycle applied to the switching valve:     OUR·· )( * tCCak dt tdC LLL L  Now, considering how the signals around the error detector are related and their differential form for a given constant set point. It is obvious to find an expression of the OUR in function of the control loop parameters a(t) and e(t): dt dC dt de dt dC dt dC dt de L ctnC L sp L sp L       dt de tCCak LLL  * ··OUR  Notice that for the PI and PID controllers the stationary error will tend to disappear. Hence, as far the previously mentioned conditions are respected, a time domain defined expression for the OUR estimation is proposed, proportional to the duty cycle and the mass transfer coefficient. In order to overcome the limitation related to the ratio between the measurement period and the mass transfer coefficient a new approach based on proportional valves is considered. The new scheme considers a continuous proper gas mixing before Simplified implementation of the OUR stationary liquid mass balance method. 5 133 reaching the bioreactor’s gas phase by means of said proportional valves which are equally operated through PWM signals. In this case the duty cycle of the control signal ais not related to the replacement of the gas phase. Hence, its period is independent of the control loop operation. Section 4.3.1 provides an additional description on the operation of the proportional valves. Figure 5 Continuous flow control loop topology. In order to obtain the oxygen partial pressure in the bioreactor’s gas phase, it is possible to do it by redrawing the gas circuit of the diagram above using an electrical analogy and applying basic electrical circuit theory: Figure 6 Gas circuit electrical analogy. Simplified implementation of the OUR stationary liquid mass balance method. 5 134 Where Rvrepresents the pneumatic resistance featured by the proportional valves when they are completely open and some gas is flowing through. So then, if the condition Rv >>> 2·Rinlet is met, it possible to state that: [ ] ≈ · ·100 →[ ] ≈· · 100 2 This is equivalent to the mean partial pressure obtained for the discontinuous model. Therefore, it has been considered that both control topologies under the appropriate conditions of switching period and mass transfer behave as a causal linear and time invariant system, which can be modelled by a transfer function in the Laplace domain. This can be useful for a better understanding of the system dynamics as well for the optimization of the control loop parameters. As previously stated, the mass balance equation considering the control signal ais:     OUR·· )( * tCCak dt tdC LLL L  This matches the form of a first order differential equation that can be solved and expressed by means of the Laplace transformation:   OUR· )( * LLLL LCaktCak dt tdC                  t ttak l d o tak o tak LL dlll ee ak t xqO eCeCtC 2ln · 2 · * · 2ln 1                          d l o l o l L LL t saks xqO aks C akss C tCsC 2ln · ·· 2 *  L Where: •Co[mol/l]: Initial value of the DO concentration •qO2[mol/l·h]: Oxygen specific consumption. •xo[cell/ml]: Initial cell concetration. •td[h]: Duplication period The expression of the mass balance frequency behaviour can now be used for drawing the control loop block diagram, next figure. Where it can be seen how, despite that the open loop transfer function is only dependent on the bioreactor’s mass transfer capability, the dynamics is also clearly affected by the biological evolution. Since the aim of the current approach is to use control loop internal variables to estimate the oxygen consumption, it will be necessary an adequate design of the controller to have a good compromise between the control loop performance and the sensitivity of the control signal  (t) respect to the biological Simplified implementation of the OUR stationary liquid mass balance method. 5 135 “disturbance”. As well as keeping in mind that any artifact or stationary drift on the control loop variables will be propagated to the OUR estimate. Figure 7 Control loop block diagram in the frequency domain. 5.2 Simulations. As part of the feasibility assessment, several numerical simulations were performed in order to guess the operational ranges as well as identifying possible design compromises (All the simulations were carried out assuming an exponential growth model with no existence of other limiting factors than the lack of oxygen). The goal of the first simulation was to investigate the convergence conditions between both control topologies: Continuous flow by means of proportional valves or through gas phase replacement by means of a switching valve. This is shown below: Figure 8 DO control profiles for both topologies: Continuous flow & Gas phase replacement. Simplified implementation of the OUR stationary liquid mass balance method. 5 142 DO, pH and OUR evolution. Simplified implementation of the OUR stationary liquid mass balance method. 5 143 Conclusions The HexaScreen system was successfully able to provide the culture conditions and monitor the metabolical activity of a Hybridoma culture. The span of the OD measurement was small enough to provide a good estimation of the cell concentration. Another evidence of the success of the experiment is the obvious correlation of the graphs that matches the expected behavior of the cells. Regarding the performance of the Dynamic method for the OUR estimate, despite it demonstrated to be a robust technique, it clearly shown two important drawbacks: The lack of temporal resolution, only 7 measurements along the whole culture is not enough to take some decisions, i.e.: An accurate reading of the limitation time, and the periodic disturbance on the pH and the DO concentration produced by the sudden replacement of the gas phase. This implies a serious danger for many cell species. That’s why the Dynamic method could only be addressed as an appropriate method for research applications and experimental essays were non GLP or GMP rules need to be applied. Experiment II Validation of the DO controller and the OUR estimation method for microbial applications Experiment description The extremely fast duplication time of bacteria in comparison with animal cells makes microbial applications more demanding in terms of oxygen transfer and robustness of the DO control loop. In this experiment the MonoScreen Fed-Batch system was evaluated under such worst case condition. After calibrating the system using the same disposable Minibioreactor and culture medium to be used for the experiment, the Minibioreactor was inoculated and a Batch culture was carried out along 17 hours. Regarding the OUR estimate, since the aeration system of the MonoScreen Fed-Batch is based on the continuous mixing of several gas flows (Air, N2, CO2), the Simplified stationary liquid mass balance method was applied. Culture Conditions Medium: LB Cell line: Escherichia Coli Volume: 25 ml Initial cell concentration: 0.2 gdw/l Stirring rate: 500 rpm Bioreactor temperature: 37 °C Aeration rate: 53 smlpm (≈2 VVM) Results The overall behaviour of the DO controller worked as expected. After a transient response of approximately 30 minutes, the DO concentration reached the set-point of 20 %. Besides of some artifacts due to an excessive gain of the integral term, the DO concentration was kept constant around the set-point during the following 2.5 hours. Then, the sudden Simplified implementation of the OUR stationary liquid mass balance method. 5 144 extinction of the dissolved oxygen happened just as predicted in the previous section. This event is correlated with the increase of the biomass density and the evolution of the oxygen demand. When the signal that controls the gas flow mixing reached the maximum, it became impossible for the system to keep on transferring the required oxygen to maintain the primary metabolism of the cells. Hence, during the following hours the bacteria adapted to the new environment and kept on growing at a slowing down rate. After 14 hours of culture the cell death happened. This was clearly pointed by two signs: A slight decrease of the Turbidimetric absorbance and the sudden increase of the DO concentration due to the lack of oxygen consumption and also by the OUR estimator. DO, TA, Duty cycle and OUR evolution. Conclusions The MonoScreen Fed-Batch system was also successful on providing the culture conditions and monitoring the metabolical activity of a typical microbial application. The NIR turbidimeter demonstrated to be a feasible tool for monitoring the biomass concentration in microbial applications. And the Simplified OUR stationary liquid mass balance estimation method provided realistic data on the evolution of the oxygen demand. Nevertheless, the mass transfer coefficient was found still too low to support high biomass densities. This will require in the future some additional effort to improve the system’s mass transfer capability, by means of including a miniaturized sparger or microdiffusor. Simplified implementation of the OUR stationary liquid mass balance method. 5 145 Experiment III Effect of the culture temperature on the growth dynamics of the E. Coli Experiment description In order to evaluate the DO controller of the MonoScreen Fed-Batch system for a longer period, a new experiment was designed to compare the evolution of the same E. Coli strain at three different culture temperatures, based on the fact that a reduction of the temperature will produce and increment of the duplication time. Therefore, after a slight correction of the PI gains, three consecutive batch cultures were carried following the same procedure explained for the previous experiment. Culture Conditions Medium: LB Cell line: Escherichia Coli Volume: 25 ml Initial cell concentration: 0.2 gdw/l Stirring rate: 500 rpm Bioreactor temperature: 37, 30, 23 °C Aeration rate: 53 smlpm (≈2 VVM) Results After the initial transient response, the DO concentration was kept constant around the set-point (20 %) until the saturation of the control signal. This occurred in increasing times, consistent with the temperature increase. Obviously, the Turbidimetric absorbance and the OUR estimate also displayed great differences in the biomass evolution of every batch. DO, TA and OUR evolution. Simplified implementation of the OUR stationary liquid mass balance method. 5 146 Conclusions After some fine tuning of the DO controller, the stability of the DO concentration around the set-point was increased for the three cases without showing any drift before the saturation of the gas mixing system. Additionally, a reduction of the noise on the OUR estimations was also observed. The three graphs above clearly demonstrated the effect of culture temperature on the growth dynamics. It was possible to notice a curious effect. The decrease of the temperature produced not just a slower growth rate but a bigger biomass concentration. This is especially visible in the comparison of the OUR graphs. Experiment IV Validation of the DO controller and the OUR estimation method for animal cell applications Experiment description The use of the simplified stationary liquid mass balance OUR estimation method has already been demonstrated for Microbial cultures by means of the MonoScreen Fed-Batch system. In this new experiment the method is applied for animal cell culture using the customized BIOSTAT B-plus bench scale bioreactor. The bioreactor’s vessel was inoculated and a Batch culture was carried out for 10 days at a fixed DO concentration 30 %. The gas supply consisted initially of a mixture of Air, N2plus a minimum constant flow of CO2for adjusting the physiological pH. After four days the Air supply was replaced by pure O2in order to avoid the saturation of the gas mixing system. Additionally, a number of parameters were also analyzed through different techniques after daily manual sampling (Lactate, Glucose, Cell concentration and Viability). Culture Conditions Medium: SFM4TransFx + 5 % FBS + 10 % CB5 + 4 mM GlutaMAX + 0.2 % Pluronic F69 + 50 ppm Antifoam C Cell line: HEK293 Volume: 1500 ml Initial cell concentration: 0.25·106cells/ml Stirring rate: 100 rpm Bioreactor temperature: 37 °C Aeration rate: 0.35 slpm Results The DO set-point was reached after an initial transient period of 6 hours and the concentration was kept constant along the culture time. A short transient occurred after the substitution of the air supply by pure O2. Then, the control signal was biased and compressed due to the increment of the oxygen partial pressure in the gas phase. Such increment had to be taken into account in order to correct the signal before the calculation of the OUR estimate. Simplified implementation of the OUR stationary liquid mass balance method. 5 147 DO, Duty cycle and OUR evolution. The evolution of the OUR shown the typical behavior of any cell culture. During the first 6 days the cell concentration grown exponentially consuming glucose and producing lactate. When the remaining glucose was not enough to feed the current cell concentration a secondary metabolic path was activated and the lactate previously produced started being consumed by the cells. Once that both analytes became too low the oxygen consumption started to fall down dramatically and the number of viable cells decreased. This is shown in the following graph. DO, pH and OUR evolution. Simplified implementation of the OUR stationary liquid mass balance method. 5 148 Conclusions Additionally to the obvious demonstration of the method’s viability for cell culture processes and bigger volumes (Bench Scale). The off-line data available in this experiment permitted a crucial observation: the fact that the turning point in the OUR graph allowed anticipating the time of maximum viable cell concentration (75 hours before). This is a very important advantage that leads to multiple possibilities regarding the culture strategy. i.e.: When the administration of a bolus may be required to avoid cell death, to maximize the cell concentration or to define some feeding profile. This fact has been confirmed in several (not shown) experiments. Conclusions & work in progress. 6 149 Conclusions & work in progress. As for the explanation of the state-of-the-art, the conclusions will regard not just to the simplified implementation of the OUR stationary liquid mass balance estimation method but also to other aspects like the development of the several Minibioreactor systems and their complementary instrumentation techniques. 6.1 Conclusions The fact that the all the work in this thesis was strongly imbricated with the business project of the start-up company HEXASCREEN CULTURE TECHCNOLOGIES made that any aspect had to be faced from a product development point of view in order to match the market’s requirements. Four different Minibioreactor systems were developed for testing and evaluating the most appropriate control architectures, instrumentation and operational ranges. It was found that tiny microcomputers such as the ones selected from Rabbit Semiconductor were a cost effective option for developing even complex and technologically demanding instruments. However, the development time required for such devices used to be much higher than for other platforms specifically designed for the rapid prototyping of Conclusions & work in progress. 6 150 embedded systems. On the other hand, the initial Master-Slave communication scheme where a host computer was in charge of the monitorization and control of the different instruments was finally replaced by a web server based user interface. This offered a more robust solution since only one piece of software had to be developed and the need for a custom Master-Slave communication protocol became unnecessary. Hence, despite that the last version of the system is still a prototype; it must be seen as a preindustrial Minibioreactor system including all the state-of-the-art features. Figure 1 Evolution of the different Minibioreactor systems developed One of the most important parameters to measure in any cell culture is obviously the cell concentration. To that end, several non-invasive optical probes and configurations were investigated and finally the differences between Optical density and NIR Turbidimetric absorbance were pointed. It was confirmed that Optical Density, which is a well stablished method for off-line measurement in microbial applications, was only useful for on-line monitoring when low cell concentration values were achieved. This happened due to the effect of the light scattering phenomena within the whole visible spectra. On the other side, NIR Turbidimetric absorbance demonstrated a wider linear range of measurement feasible for almost any application with animal cells. This achievement was possible thanks to the use of a NIR laser source which offered a highly collimated light beam that combined with its narrow spectral pattern minimized the effect of the light scattering. The second key parameter was pH. Optical absorbance spectroscopy was initially chosen as the appropriate technique. Nevertheless, the need for some diluted dye as well as the fact that the optical absorbance at the dye’s characteristics wavelengths is interfered by the Optical Density makes this option not useful for many cases, even with mammalian cells where in spite of the relatively small drift of the pH, sometimes the use of indicator dyes can be unacceptable. Therefore, the development has evolved towards the use of disposable pH fluorescence sensors and the DLR (Dual Lifetime Referencing) detection method, whose response is independent of the cell concentration and the presence of any diluted pH indicator. Unfortunately, the DLR method has not been yet tested enough. It is in fact the last development milestone to provide the MonoScreen Fed-Batch system with complete functionality. Conclusions & work in progress. 6 151 The third basic parameter from the metabolical point of view is DO (Dissolved Oxygen). During the development and testing of the first prototypes, several approaches were carried out for the implementation of a non-invasive polarographic probe (Clark’s electrode). However, the difficulty to avoid mechanical defects or positioning mismatches of the electrode with respect to the permeable membrane led to the generation of quite noisy signals. This problem was finally overcome by means of the use of disposable DO fluorescence sensors. Despite that OEM readers for such sensors were commercially available; a custom low cost coherent XOR detector was designed and validated for the application. Such complementary instrumentation permitted to successfully face the challenge of proofing the feasibility of the previously mentioned Minibioreactor systems. The high degree of innovation associated to the development of these products becomes obvious when observing that the development rate of the commercial business ran in parallel with the evolution of the different systems developed by HEXASCREEN CULTURE TECHNOLOGIES. With respect to the major original contribution of this thesis, the simplified implementation of the OUR stationary liquid mass balance estimation method, the feasibility of the procedure was demonstrated for both animal and microbial cells; it also was demonstrated for two different culture scales. In comparison with the Dynamic method the proposed method shows obvious advantages in terms of time resolution and DO stability (lack of cell stress). Regarding the method’s accuracy, unfortunately any standard method to set a known value of oxygen consumption has been reported yet. Hence, it is not possible to provide a measured value on the estimation error. Nevertheless, the accuracy is expected to be comparable to the non-simplified stationary liquid mass balance method due to the fact that both procedures are dependent on the previous knowledge of the mass transfer coefficient and the DO control’s stationary error. That is, a precision of 1% on the DO control allows measurement of the consumption of 104cells·ml-1 assuming a typical consumption rate of 0.2x10-12 mol·cell·h-1 (Ruffieux et al. 1998) 6.2 Work in progress In order to promote the use of the method, another experiment is currently being conducted out in cooperation with members of the Chemical Engineering Department of the Universitat Autònoma de Barcelona. The goal is to compare the performance of the simplified method versus the global de mass balance through the use of an in-line Bluesens O2and CO2 analyzer. The following figure shows the diagram of the experiment set-up. The four valves around the Bluesens analyzer are periodically switched on an off in order to sequentially measure the oxygen molar fraction in the bioreactor’s inlet and outlet. This lets to solve the global mass balance.