Appl Microbiol Biotechnol (2012) 93:575584 DOI 10.

1007/s00253-011-3507-9

BIOTECHNOLOGICAL PRODUCTS AND PROCESS ENGINEERING

Investigation of the potential of biocalorimetry as a process analytical technology (PAT) tool for monitoring and control of Crabtree-negative yeast cultures
Moira Monika Schuler Senthilkumar Sivaprakasam Brian Freeland Adel Hama Katie-Marie Hughes Ian W. Marison

Received: 2 March 2011 / Revised: 28 June 2011 / Accepted: 20 July 2011 / Published online: 16 August 2011 Springer-Verlag 2011

Abstract Biological reaction calorimetry, also known as biocalorimetry, has led to extensive applications in monitoring and control of different bioprocesses. A simple real-time estimator for biomass and growth rate was formulated, based on in-line measured metabolic heat flow values. The performance of the estimator was tested in a unique bench-scale calorimeter (BioRC1), improved to a sensitivity range of 8 mW l1 in order to facilitate the monitoring of even weakly exothermic biochemical reactions. A proportionalintegral feedback control strategy based on these estimators was designed and implemented to control the growth rate of Candida utilis, Kluyveromyces marxianus and Pichia pastoris by regulating an exponential substrate feed. Maintaining a particular specific growth rate throughout a culture is essential for reproducible product quality in industrial bioprocesses and therefore a key sequence for the step from quality by analysis to quality by design. The potential of biocalorimetry as a reliable biomass monitoring tool and as a key part of a robust control strategy for aerobic fed-batch cultures of Crabtree-negative yeast cells in defined growth medium was investigated. Presenting controller errors of less than 4% in the best cases, the approach paves the way for the development of a generally applicable process analytical technology platform for monitoring and control of microbial fed-batch cultures.

Keywords Biocalorimetry On-line biomass estimation Specific growth rate Fed-batch Candida utilis Kluyveromyces marxianus Pichia pastoris

Introduction The specific growth rate (), describing the variation of the cell concentration over time in relation to the actual cell concentration, is an important process parameter. Its correlation to the metabolic state of the cells and by inference to other cell-related process variables, such as product quality attributes, is undeniable. Studies have highlighted the effect of changes in the specific growth rate on the glycosylation pattern (Schenk et al. 2008) or on the secretory expression (Puertas et al. 2010) of recombinant proteins and described the importance of maintaining threshold values of in order to avoid overflow metabolite production (Nahku et al. 2010). Moreover, arbitrary induction of a given cell state in order to produce a specific compound, such as prodigiosins (Chang et al. 2011), is of particular relevance to biotechnology and pharmaceutical industries. A particular process parameter can be defined as critical as soon as it is related to a critical product quality attribute (ICH 2008). The control of critical process parameters is a prerequisite in the step from quality by analysis (QbA) to quality by design (QbD) as highlighted by the Food and Drug Administration (FDA) initiative for industrial process analytical technology (PAT) (CDER 2004). Research is often restricted to continuous cultures, where the biological variable is regulated by the dilution rate ( D), a physical parameter, in order to study

M. M. Schuler S. Sivaprakasam B. Freeland A. Hama K.-M. Hughes I. W. Marison (B) Laboratory of Integrated Bioprocessing (LiB), Dublin City University (DCU), Dublin 9- Ireland e-mail: ian.marison@dcu.ie

576

Appl Microbiol Biotechnol (2012) 93:575584

growth rate-related phenomena and effects (Hellmuth et al. 1994; Tomson et al. 2006; Kontoravdi et al. 2007). However, fed-batch cultures are gaining importance since they allow maintaining at a given setpoint by the choice of an appropriate control strategy. Various types of control strategies for fed-batch cultures have been proposed by different research groups over the past two decades, as shown by (Lee et al. 1999). The application of complex soft models, such as artificial neural network or fuzzy control, gives interesting results in terms of controller stability and adaptivity. Simulations with such soft models perform well in certain cases, yet there is a potential risk of loosing biologically relevant information. Stoichiometric or mechanistic models provide more information about the cellular state, especially if there is a switch from a pure black box model to a parametric-based one. It should, however, be kept in mind that simplicity is a keyword when the developed control strategy is intended for industrial application. Indeed, some studies on complex model-predictive control showed interesting results at simulation level but proved to be difficult to implement in practice (Arndt et al. 2005). In this study, there was a quest for a simple control strategy, in order to ensure a facilitated applicability to actual fed-batch cultures. A slight variation of a simple, exponential feeding strategy, described in previous work (Dabros et al. 2010), is attempted here. The chosen control strategy enables to maintain a constant specific growth rate without extensive softmodel-based adaption calculations. However, a control strategy is only as good as the measurements used to estimate the control variable. Therefore, all strategies to control rely on the availability of reliable biomass measurements or estimations. Direct measurements of biomass can be achieved through capacitance readings (Claes and Van Impe 1999; Rhiel et al. 2002) or by means of near-infrared spectroscopy (McLeod et al. 2009). Off-gas analysis or metabolite detection by Fourier transform infrared spectroscopy measurements for instance enables indirect biomass estimation through the calculation of different rates, such as oxygen uptake rate, carbon dioxide evolution rate or C- or N-source consumption rates. Heat evolution rate, inherent to every process involving living organisms, has been investigated as a powerful monitoring tool over the past 30 years (Ishikawa and Shoda 1983; Marison et al. 1985; von Stockar and Marison 1989; Larsson et al. 1991; Guan et al. 1998; Maskow et al. 2004; Ma et al. 2007; Schubert et al. 2007; Senthilkumar et al. 2008; Maskow et al. 2010). Biocalorimetry enables a deeper insight into the metabolic state of cells and brings a greater process understanding in general. Even so, attempts to use the

metabolic heat flow as sole measurement to estimate and control the specific growth rate throughout microbial fed-batch cultures are scarce. Indeed, heat measurements were used either in combination with other parameters, such as carbon dioxide evolution rate, or in a derived form, such as the oxycalorific equivalent (Randolph et al. 1990; von Stockar et al. 1997), to control the substrate feed rate into the system.The presented work is an investigation of the potential of biocalorimetry as a tool to provide reliable measurements to create a simple control strategy for . The present study is centered around three different Crabtree-negative yeast strains, Candida utilis, Kluyveromyces marxianus and Pichia pastoris, grown in a modified, high sensitivity biocalorimeter. The organisms were chosen due to their simple growth requirements. They do not suffer from any known metabolic bottleneck, enabling the direct correlation between heat generation rate and specific growth rate in its simplest way (von Stockar and van der Wielen 1997). A simple proportionalintegral feedback strategy, where the specific growth rate, estimated from metabolic heat flow measurements, as well as the error term of the controller are embedded in the exponential term of an equation described in previous work (Dabros et al. 2010), is presented. The development of a simple strategy to control the specific growth rate of cells in fedbatch processes has its importance in the realm of the PAT framework in terms of process understanding and process control. Moreover, two independent research groups (Voisard et al. 2002; Tuerker 2004) have shown the feasibility of large-scale calorimetry bringing the proposed control strategy a step closer to the development of a general applicable, heat flow measurementbased platform for industrial control of the specific growth rate of microbial cells in fed-batch cultures.

Material and methods Cell strains and culture conditions Cell strains Two wild-type yeast strains, K. marxianus (DSM 5422, DSMZ, Braunschweig, Germany) and C. utilis ( DSM 2361, DSMZ, Braunschweig, Germany), were chosen. For the assessment of the general applicability of the developed control strategy, another yeast strain (P. pastoris, hSA producing strain, Invitrogen, Carlsbad, CA, USA) was cultured. All strains were preserved as 1.8 ml aliquots in a 20 g l1 glycerol solution at 80 C.

Appl Microbiol Biotechnol (2012) 93:575584

577 Table 1 Composition of the batch and feed media for culturing K. marxianus and C. utilis Component (g l1 ) Glucose (NH4 )2 SO4 KH2 PO4 MgSO4 7H2 O (ml l1 ) Trace elements and vitamins solution Polypropylene glycol 2000 (antifoam) 5 0.5 15 2 10 5 3 0.5 300 50 35 3 Batch medium Feed medium

Chemicals All chemicals were purchased from Sigma-Aldrich (Sigma-Aldrich, St. Louis, MO. USA). Inoculum preparation The preculture was obtained by suspending cells from a 1.8-ml aliquot in a 1-l, baffled Erlenmeyer flask containing 100 ml of preheated, sterile, complex preculture medium (at pH 4 for C. utilis and at pH 5 for K. marxianus and P. pastoris, respectively) containing 20 g l1 of glucose for K. marxianus and C. utilis and 20 g l1 of glycerol for P. pastoris, 10 g l1 of yeast extract and 10 g l1 of peptone and incubating the flask in an orbital shaker incubator (SHEL LAB S19, Sheldon Manufacturing, Cornelius, NC, USA) for 24 h at 30 C and at 150 rpm. Culture broth (90 ml) was then centrifuged at 3,000 rpm for 10 min, the supernatant discarded and the cell pellet resuspended in 10 ml of a sterile saline solution (9 g l1 of NaCl). Reactor description, media composition and culture conditions The preculture was used to inoculate a 2-l BioRC1 (modified RC1, Mettler Toledo, Greifensee, Switzerland) with a working volume of 1.3 l, equipped with a six-blade Rushton-type agitator, baffles, a pH probe and controller, gas inlet and outlet ports, a base inlet port, a port for a capacitance probe (Biomass Monitor, ABER Instruments, Aberysthwyth, UK) and a sampling valve. The aeration rate was set to 2.5 l min1 using a thermal massflow controller (5850E, Brooks, The Netherlands), and the air was sterilised by passage through a 0.22 m filter before entering the biocalorimeter. Outlet gas was passed through a Wolff bottle followed by a 0.22 m filter before entering a gas analyser (Duet, Applied BioSystems Ltd, UK). The measured values of O2 and CO2 were corrected for water vapour according to Duboc and von Stockar (Duboc and von Stockar 1998) and used to evaluate in real time the oxygen uptake rate, the carbon dioxide evolution rate as well as the respiratory quotient. A solution of 4 M NaOH was used to maintain the pH at the required value. Acid control was not necessary due to the composition of the culture medium and the cell metabolism. The operating principle of the BioRC1 has been reported previously in literature (Duboc and von Stockar 1995). All experiments were performed with a initial working volume of 1.3 l, and the reaction temperature was maintained constant at 30 C. Agitation rate was maintained at 800 rpm throughout all

cultures. A detailed account of liquid volumes entering (base and medium feed) and leaving the reactor (samples) was made gravimetrically (Analytical Balances, Mettler Toledo, Greifensee, Switzerland) and acquired through a LabVIEW (LabVIEW 8.2, National Instruments, Austin, TX, USA) programme in order to have a continuous inventory of the reactor volume throughout the experiment. Prior to inoculation, the bioreactor was sterilised in situ at 120 C for 20 min with deionized water using an automated WINRC sterilisation programme (WINRC Software, Mettler Toledo, Greifensee, Switzerland), cooled down to room temperature, drained and filled with sterile batch medium. Cells were grown in batch mode until depletion of the carbon and energy source. For the assessment of the growth rate estimator, fedbatch cultures were carried out, where an exponential feed mode was started after the end of the batch phase. Tables 1 and 2 outline the composition of the defined batch and feed media. The media were prepared with deionized water, sterilised by filtration (Steritop, 0.22 m pore size, Millipore, Billerica, MA,
Table 2 Composition of the batch and feed media for culturing P. pastoris Component (g l 1 ) 10 0.59 9.1 7.45 2.06 9 500 90 Batch medium Feed medium

Glycerol CaSO4 2H2 O K2 SO4 MgSO4 7H2 O KOH NH4 Cl (ml l1 ) H3 PO4 85% Trace elements and vitamins solutions Polypropylene glycol 2000 (antifoam)

13.35 4.35 1

578

Appl Microbiol Biotechnol (2012) 93:575584

USA) and supplemented with sterile antifoam solution, trace elements and vitamins as prescribed by (Verduyn et al. 1992) and (Cannizzaro et al. 2004) for K. marxianus and C. utilis and suggested by Invitrogen for P. pastoris. Monitoring, data acquisition and control strategy Of f-line analysis Samples (10 ml) were taken at regular intervals during the cultures in order to determine off-line the biomass concentration. Biomass dry cell weight was determined by filtering a known volume of the culture broth through a pre-weighed 0.22 m pore size filter (GSWP 0.22 m nitrocellulose membrane filters, Millipore, Billerica, MA, USA), drying the filter to constant weight and subsequently reweighing. The off-line dry cell weight measurements were used to calculate the off-line values for the specific growth rate. Glucose and glycerol concentrations of the samples were quantified by HPLC (Agilent Instruments 1200, Agilent Technologies Ltd., Cork, Ireland) with a Supelcogel C610H column (Sigma-Aldrich, St. Louis, MO, USA) equipped with a guard column and using a 0.027% (v/v ) H2 SO4 mobile phase for isocratic elution. Calibration curves for each component were prepared for each run using five synthetic standards of appropriate concentration range. An internal standard (30 g l1 of isopropanol) was run with each sample and each standard. Data acquisition and process control Data acquisition of the process parameters from the different probes and sensors was carried out through a FieldPoint (ModelFP 2000, National Instruments, USA) and an interfacing hardware (NB-MID-32X, National Instruments) with a PC. A LabVIEW programme (LabVIEW 8.2, National Instruments, Austin, TX, USA) was developed in-house for data acquisition, storage and display as well as for calibration and process control. The acquired raw values, saved in a separate file, were continuously averaged over 50 points and used for the real-time calculations of the baseline heat flow, the metabolic heat flow rate, the biomass concentration, the specific growth rate, the nutrient feed rate and the error term for the feedback control loop. Biomass and specif ic growth rate estimator Specific growth rate estimators are commonly based on models derived from Monods equation for cellu-

lar growth, requiring reliable direct biomass measurements such as capacitance measurements or indirect estimations based on oxygen uptake rate (Lubenova 1996). However, a heat yield based approach may be employed as stated by (von Stockar et al. 1997). The detailed mathematical development for the heat-based biomass estimator and the specific growth rate estimator for fed-batch cultures, described in Eqs. 1 and 2, has been described in previous work (Sivaprakasam et al. 2011). xt Vt = x0 V0 + Qt Q0 Y Q/ X (1)

est,t =

qm x0 V0 Y Q/ X + ( Qt Q0 )

(2)

Equation 1 relates the biomass concentration at a given time (xt in grams per litre) and the actual culture broth volume (Vt in litres) to the biomass concentration at the end of the batch phase (x0 in grams per litre) and the culture volume at this very same moment (V0 in litres), the cumulative metabolic heat production at the end of the batch ( Q0 in kilojoules), the cumulative metabolic heat production at a given time ( Qt in kilojoules) and the heat yield coefficient (Y Q/ X in kilojoules per gram). The specific growth rate at a given time (est,t ) is related to the same physical quantities and the instantaneous heat flow rate (qm ) due to the metabolic cell activity as given in Eq. 2. The biomass and the specific growth rate estimators are relying on a good estimation of the biomass yield coefficient. This parameter is depending on environmental conditions (Nagai 1979) and was determined for each run on the experimental data during the batch phase. The biomass yield coefficients were very similar from batch to batch for each strain. It could therefore be thought of taking an average value, so that the biomass yield coefficient does not need to be determined each time. Control strategy design Independently of the control strategy, each controlled quantity or controlled variable is associated with a manipulated variable. Furthermore, a sensed variable, which is the measured value of the controlled variable, needs to be taken in account (Murrill 1991). In this study, the controlled variable is , the specific growth rate, and the associated manipulated variable is the flow rate of a C-source-limited substrate feed. The sensed variable is the metabolic heat flow rate. Measurements of the heat flow rate, acquired through the BioRC1, are used to estimate the actual state of the controlled variable as indicated in Eq. 2. The difference

Appl Microbiol Biotechnol (2012) 93:575584

579

between the setpoint for the specific growth rate sp and its actual value est is evaluated and used to define the tracking error as defined in Eq. 3. The tracking error is included into the feedback loop as usually in a closed-loop system. (t) = sp est (t) (3)

However, oscillatory behaviour may be expected since this type of controller is more prone to instability. FPI (t) = F0 e
est + K P (t)+ K I
t 0

(t)dt t

(6)

Here, a variation of the classical PI controller was designed, based on previous work (Dabros et al. 2010) where proportional and integral terms were included in the exponential term of the equation for feed rate. The equation for the feed rate was obtained by combining substrate and biomass balances as suggested by other authors (Lee et al. 1999; Dabros et al. 2010). The feedforward component of the closed-loop control strategy is described in Eq. 4. FFF (t) = F0 e(est t) (4)

The control block diagram developed during this study and implemented in a LabVIEW programme is shown in Fig. 1. The control action, FPI (t) is executed by a remote-controlled peristaltic pump.

Results The present section is divided into five parts, highlighting first the potential of biocalorimetry as a biomass monitoring tool Validity of the biomass and specific growth rate estimator before emphasizing the importance and limits of data pretreatment Importance and limits of data pretreatment and controller constant tuning. The differences in controlled and uncontrolled situations are shown in Differences in controlled and uncontrolled situation, followed by a discussion about the performance of the controller in Experimental robustness and long-term stability of the controller and finally the applicability of the developed strategy to another yeast strain is assessed in General applicability. Validity of the biomass and specific growth rate estimator The applicability of biocalorimetry as a monitoring tool for biomass in microbial cultures has been assessed in previous work (Sivaprakasam et al. 2011) for the three yeast strains under study and is illustrated in Fig. 2 for a batch culture of C. utilis. Glucose concentration is decreasing and biomass, predicted by the biomass estimator based on calorimetric measurements, increases as the culture evolves. The estimated specific growth rate shows abnormally high values initially, due to heat signal disturbances inherent to the inoculation procedure. However, after the first hour of culture, the specific growth rate starts to follow an expected pattern, increasing until reaching a constant, slightly oscillating value, max during the exponential growth phase. Importance and limits of data pretreatment and controller constant tuning Raw estimations might consist of noise which will have an effect on the controller action. Simple mathematical data pretreatment, such as a moving-point

The initial feed rate F0 , as shown in Eq. 5 is a function of the biomass concentration at the previous timestep (xt1 ) for a given setpoint, the reactor volume (Vt1 ) at the same moment, the biomass yield coefficient (Y Q/ X ) and the substrate concentration ( S F ) in the feed solution. F0 = xt1 Vt1 sp Y X/ S S F (5)

A proportionalintegral feedback control, described in Eq. 6, is chosen, since a proportional controller, except in the presence of a pure capacity process, invariably leads to a non-zero steady-state offset (Ogunnaike and Ray 1994). A simple way of removing the offset effect of the proportional controller is to add an integral term. The action of this controller is based on the summing of the instantaneous error over time and therefore responding to accumulated errors from the past, enabling to remove the residual steady-state error.

Fig. 1 Control block diagram of the feedback control strategy based on heat flow measurements (dotted line)

580

Appl Microbiol Biotechnol (2012) 93:575584

Fig. 2 Batch culture of C. utilis, illustrating the glucose and biomass concentrations as well as the specific growth rate over time. Triangles represent glucose values, circles represent cell concentration values estimated from the heat flow values and black line represents the specific growth rate

in a decrease in the peak-to-peak amplitude. A 100points moving average corresponds, in the case of a data point sampling interval of 10 s, to a total average time of 16.7 min. Considering the doubling time of yeast cells at maximal growth rate and moving-point average values reported elsewhere (Wechselberger et al. 2010; Dabros et al. 2010), a 16.7-min windows was regarded as acceptable. An important point of a feedback control strategy is the choice of the appropriate values for K P and K I . Based on previous experiments (Dabros et al. 2010), several values for K P (Fig. 4) and later for K I were tested and the response of the controller qualitatively construed. Finally, K P was set to 0.1 and K I to 0.15 for all experiments that followed. Differences in controlled and uncontrolled situation The specific growth rate (Fig. 5) results in less oscillations during the feed phase where a given setpoint value is imposed to the system than in the batch phase, where the cells are free to grow at their maximal specific growth rate. Other studies (Xiong et al. 2008; Dabros et al. 2010) have shown that oscillations in the values of the specific growth rate, even at the end of the batch phases, are present. A shift from a highly oscillating system response to a smoother signal occurs during the transition from batch to fed-batch mode (Fig. 5). Despite remaining oscillations, the controller has actually a stabilising action on the system when compared to the uncontrolled situation during the batch phase.

average or a high/low pass filter, can smoothen the signal. However, if the data pre-processing is too pronounced, a potential loss of essential biological information may occur. Therefore, it is important to find an appropriate trade-off. Several different moving-point averages as well as some built-in (LabVIEW) high/low pass filters were tested during a fed-batch and the most suitable one was chosen and applied to all cultures that followed. The response of the system is affected by the applied data processing technique (Fig. 3). Indeed, an increase in the number of points averaged results

Fig. 3 Study of the influence of different filtering techniques on the specific growth rate during a culture of K. marxianus. Moving-point averages with values ranging from 50 (from t = 7.5 to t = 8.5 h), over 80 (from t = 8.5 to t = 11.5 h) to 100 (from t = 11.5 to t = 14 h) were chosen. The dotted lines represent the specific growth rate values calculated off-line from dry cell weight measurements

Fig. 4 K P tuning during a fed-batch culture of K. marxianus. The K P value were set to 1 (region A), 0.5 (region B) and 0.1 (region C). A K P value of 0.1 leads to the least marked oscillations

Appl Microbiol Biotechnol (2012) 93:575584

581 Table 3 Summary of the different average values for the specific growth rate for different setpoints with different controllers and the overall controller error Controller Feedforward Feedforward P P PI PI PI Setpoint (h1 ) 0.2 0.15 0.2 0.15 0.2 0.15 0.2 Average (h1 ) 0.27 0.23 0.24 0.17 0.25 0.16 0.21 Controller error (%) 37.16 54.17 18.14 10.73 24.26 3.73 3.48

Fig. 5 Illustration of differences in the oscillating behaviour of the specific growth rate in batch (region A) and feed (region B) phases during a culture of K. marxianus

Experimental robustness and long-term stability of the controller To assess the performance of the developed control strategy, a series of fed-batches were carried out with K. marxianus, controlling the specific growth rate at a desired setpoint. In a very first attempt, only the feedforward part of the controller was used; in second step, the proportional feedback part was brought into play and finally, the proportionalintegral feedback controller, as described in Eq. 6, was tested. During the first cultures, two different setpoints were imposed (sp1 = 0.2 h1 and sp2 = 0.15 h1 ), where only one setpoint was assessed during an additional culture in order to observe the long-term stability of the control strategy. For each setpoint, the average specific growth rate over the time of the setpoint was calculated and the overall controller error estimated in order to assess the robustness of the specific growth estimator under the given culture conditions. The results of these K. marxianus cultures are summarized in Table 3. The feedforward control strategy (Eq. 4) implemented in this part of the study showed a satisfactory ability to control the specific growth rate at a constant value. However, as expected for a control strategy without any feedback term, the setpoint is not reached. Feed-forward control lacks the ability to compensate for overshooting within a reasonable frame of time, resulting in high controller errors for each setpoint (37.16% and 54.17%). A proportional feedback controller invariably leads to a off-set in the response, except in the case of a pure

capacity process (Ogunnaike and Ray 1994). Despite the risk of ending up with an unstable controller by adding an integral term, the proportionalintegral feedback term was implemented. The resulting PI controller showed a surprisingly high stability and resulted in a decreased controller error for all setpoints (Table 3) compared to the feedforward controller results. The overall controller error is much higher in the first setpoint than in the second one. This more pronounced controller error could be related to the importance of the moment of the feed starting at the end of the batch phase. Indeed, if too much time elapses before the controller is switched on, cells would enter stationary phase. As a result, a delayed response of the system might result in an accumulation of substrate and in a later stage to an overshooting of the system, as illustrated in Figs. 8 and 5. The end of the batch phase should indeed be easy to detect since it is accompanied by a sudden drop in the heat signal as reported in the past by several authors (Larsson et al. 1991; Biener et al. 2010). Therefore, a robust, systematic and automated way of controlling the onset of the first setpoint should be implemented. The robustness of the developed -estimator implemented in the present study is highlighted in Table 3. The average tracking error was not higher than 0.08 h1 . Furthermore, the response is smooth (Fig. 6) giving evidence for a high potential as a robust control platform for microbial fed-batch cultures. General applicability In the last step, the developed feedback control strategy was applied to cultures of a different yeast strain (P. pastoris) under different culture conditions in terms of medium composition. During these batch and fedbatch cultures, glycerol was used as a C-source, as described in Table 2. Despite this difference, no changes

582

Appl Microbiol Biotechnol (2012) 93:575584

Fig. 6 Feed phase of a fed-batch culture of K. marxianus where the specific growth rate was set to 0.2 h1 using the developed PI feedback control strategy. The dotted and the black lines represent the biomass concentration and the specific growth rate, respectively, both estimated based on heat flow values. The squares represent the cell concentration based on off-line measurements of the dry cell weight

Fig. 8 Fed-batch culture of P. pastoris in open-loop control where the specific growth rate (black line) was set to 0.25 h1 . The triangle represent the glycerol concentration values, the black dots the dry cell weight measurements and the white dots the biomass concentration estimated from the heat flow measurements

were applied to the experimental setup. The appropriated values for Y X/ S and S F needed to be entered into the LabVIEW programme prior to the experiment. The potential of biocalorimetry as a tool for monitoring the specific growth rate, as well as biomass, is illustrated in Fig. 7. The values for biomass concentration, predicted in real-time through the estimator described in Eq. 1, match the off-line measurements obtained through dry cell weight analysis. After initial perturbations in the specific growth rate signal, a constant max , typical of the exponential growth phase, is observed.

During the fed-batch culture of P. pastoris which was carried out with the feedforward part of the control strategy only, a concordance between off-line measurements and predicted biomass values can once more be observed (Fig. 8). The specific growth rate (window in Fig. 8), set to 0.25 h1 , overshoots the setpoint in a first time, decreases slightly and seems to stabilise above sp . HPLC measurements of the culture revealed a dramatic increase in the glycerol concentration, due to an inappropriate reaction of the controller at the beginning of the feed phase. The feedforward part of the control strategy alone is not able to deal with an overshooting in the specific growth rate. The inability of the open-loop control, more commonly used, to cope with such situations is strong evidence for the need for a reliable feedback control strategy, as proposed by this study. Feedback control for the specific growth rate, associated with a close control of the substrate concentration in the medium, for instance based on spectroscopic measurements, would allow the application of the proposed strategy to Crabtree-positive organisms.

Discussion
Fig. 7 Batch culture of P. pastoris, illustrating the biomass concentration (dots for heat flow based measurements and squares for dry cell weight values) and the specific growth rate over time (black line)

The present work confirmed the reliability of biocalorimetry as a biomass monitoring device (Sivaprakasam et al. 2011) and established the potential of the

Appl Microbiol Biotechnol (2012) 93:575584

583 Biener R, Steinkmper A, Hofmann J (2010) Calorimetric control for high cell density cultivation of a recombinant Escherichia coli strain. J Biotechnol 146(12):4553 Cannizzaro C, Valentinotti S, von Stockar U (2004) Control of yeast fed-batch process through regulation of extracellular ethanol concentration. Bioprocess Biosyst Eng 26:377383. doi:10.1007/s00449-004-0384-y CDER (2004) Guidance for industry pata framework for innovative pharmaceutical development, manufacturing and quality assurance. FDA, Silver Spring Chang C-C, Chen W-C, Ho T-F, Wu H-S, Wei Y-H (2011) Development of natural anti-tumor drugs by microorganisms. J Biosci Bioeng 111:501511 Claes JE, Van Impe JF (1999) On-line estimation of the specific growth rate based on viable biomass measurements: experimental validation. Bioprocess Biosyst Eng 21:389395. doi:10.1007/s004490050692 Dabros M, Schuler M, Marison I (2010) Simple control of specific growth rate in biotechnological fed-batch processes based on enhanced online measurements of biomass. Bioprocess Biosyst Eng 33:11091118 doi:10.1007/s00449-010-0438-2 Duboc P, von Stockar U (1995) Energetic investigation of Saccharomyces cerevisiae during transitions. Part 1. Mass balances. Thermochim Acta 251:119130. Calorimetric and Thermodynamic Studies in Biology Duboc P, von Stockar U (1998) Systematic errors in data evaluation due to ethanol stripping and water vaporization. Biotechnol Bioeng 58(4):428439 Guan Y, Evans PM, Kemp RB (1998) Specific heat flow rate: an on-line monitor and potential control variable of specific metabolic rate in animal cell culture that combines microcalorimetry with dielectric spectroscopy. Biotechnol Bioeng 58(5):464477 Hellmuth K, Korz D, Sanders E, Deckwer W-D (1994) Effect of growth rate on stability and gene expression of recombinant plasmids during continuous and high cell density cultivation of Escherichia coli tg1. J Biotechnol 32(3):289298 Hocalar A, Tuerker M (2010) Model based control of minimal overflow metabolite in technical scale fed-batch yeast fermentation. Biochem Eng J 51(1-2):64 71 ICH (2008) Pharmaceutical development qr2. ICH, Geneva Ishikawa Y, Shoda M, (1983) Calorimetric analysis of Escherichia coli in continuous culture. Biotechnol Bioeng 25(7):1817 1827 Kontoravdi C, Asprey SP, Pistikopoulos EN, Mantalaris A (2007) Development of a dynamic model of monoclonal antibody production and glycosylation for product quality monitoring. Comput Chem Eng 31(56):392400. In: ESCAPE-15 selected papers from the 15th European symposium on computer aided process engineering held in Barcelona, Spain, May 29June 1, 2005 Larsson C, Lidn, G, Niklasson C, Gustafsson L (1991) Calorimetric control of fed-batch cultures of Saccharomyces cerev isiae. Bioprocess Biosyst Eng 7:151155 doi:10.1007/BF00387410. Lee J, Lee SY, Park S, Middelberg APJ (1999) Control of fedbatch fermentations. Biotechnol Adv 17(1):2948 Lubenova V (1996) On-line estimation of biomass concentration and non stationary parameters for aerobic bioprocesses. J Biotechnol 46(3):197207 Ma J, Qi W, Yang L, Yu W, Xie Y, Wang W, Ma X, Xu F, Sun L (2007) Microcalorimetric study on the growth and metabolism of microencapsulated microbial cell culture. J Microbiol Methods 68(1):172177 Marison I, Birou B, von Stockar U (1985) Use of a novel heat-flux calorimeter for the measurement of heat generated during growth of on lactose. Thermochim Acta 85:497500

technique to predict the specific growth rate in real time. Furthermore, substantial evidence is shown that open-loop control in the form of an exponential feedforward supply of substrate is not able to cope with a setpoint overshoot, requiring the development of a robust feedback strategy to control the specific growth rate. The experiments discussed through this study pave the way to the establishment of biocalorimetry as a potential PAT tool and as a basis for a general applicable strategy to control the specific growth rate in microbial fed-batch cultures. Admittedly, further research needs to be carried out, especially in terms of controller tuning and data pre-processing, before the proposed feedback strategy can become a robust and flexible PAT platform for microbial culture. In its present form, the control strategy can be applied to closely control the specific growth rate of Crabtree-negative yeast cells in defined medium cultures under aerobic conditions.Thought needs to be given to an appropriate secondary measurement, as well as to a monitoring enhancement technique, such as data reconciliation, if the developed controller is applied to Crabtree-positive microorganisms. The sensitivity of the biocalorimeter may require improvements if the device is to be applied to animal cell cultures. Despite these shortcomings, biocalorimetry has considerable potential, even for industrial application (Hocalar and Tuerker 2010). This monitoring technique provides both real-time information about the biomass and specific growth rate evolution throughout a culture, as well as a considerable insight into the metabolic activity of cells and changes in process conditions, thereby leading to a greater process understanding and a deeper knowledge about the underlying phenomena. Process understanding is a key step in the shift from QbA to QbD as recommended by the FDA initiative about industrial PAT and biocalorimetry. Providing a realtime insight into the ongoing process, biocalorimetry can therefore be considered a potential PAT tool.
Acknowledgements The Science Foundation of Ireland (SFI Research Programme Number 08/IN.1/B1948) and the Irish Council for Scientific and Engineering Technology (IRCSET Post Doctoral Fellowship D/2008/44) are greatly acknowledged for financial support of this work.

References
Arndt M, Kleist S, Miksch G, Friehs K, Flaschel E, Trierweiler J, Hitzmann B (2005) A feedforward-feedback substrate controller based on a Kalman filter for a fed-batch cultivation of Escherichia coli producing phytase. Comput Chem Eng 29(5):11131120

584 Maskow T, Kiesel B, Schubert T, Yong Z, Harms H, Yao J (2010) Calorimetric real time monitoring of lambda prophage induction. J Virol Methods 168(12):126132 Maskow T, Olomolaiye D, Breuer U, Kemp R (2004) Flow calorimetry and dielectric spectroscopy to control the bacterial conversion of toxic substrates into polyhydroxyalcanoates. Biotechnol Bioeng 85(5):547552 McLeod G, Clelland K, Tapp H, Kemsley EK, Wilson RH, Poulter G, Coombs D, Hewitt CJ (2009) A comparison of variate pre- selection methods for use in partial least squares regression: a case study on NIR spectroscopy applied to monitoring beer fermentation. J Food Eng 90(2): 300307 Murrill PW (1991) Fundamentals of process control theory. Instrument Society of America, Research Triangle Park Nagai S (1979) Mass and energy balances for microbial growth kinetics. In: Advances in biochemical engineering, volume 11, of advances in biochemical engineering/biotechnology, vol 11. Springer, Berlin, pp 4983 Nahku R, Valgepea K, Lahtvee P-J, Erm S, Abner K, Adamberg K, Vilu R (2010) Specific growth rate dependent transcriptome profiling of Escherichia coli k12 mg1655 in accelerostat cultures. J Biotechnol 145(1):6065 Ogunnaike BA, Ray HW (1994) Process dynamics modeling and control. Oxford University Press, Oxford Puertas J-M, Ruiz J, de la Vega MR, Lorenzo J, Caminal G, Gonzlez G (2010) Influence of specific growth rate over the secretory expression of recombinant potato carboxypeptidase inhibitor in fed-batch cultures of Escherichia coli. Process Biochem 45(8):13341341 Randolph T, Marison I, Martens D, von Stockar U (1990) Calorimetric control of fed-batch fermentations. Biotechnol Bioeng 36(7):678684 Rhiel M, Ducommun P, Bolzonella I, Marison I, von Stockar U (2002) Real-time in situ monitoring of freely suspended and immobilized cell cultures based on mid-infrared spectroscopic measurements. Biotechnol Bioeng 77(2):174185 Schenk J, Balazs K, Jungo C, Urfer J, Wegmann C, Zocchi A, Marison IW, von Stockar U (2008) Influence of specific growth rate on specific productivity and glycosylation of a recombinant avidin produced by a Pichia pastoris mut(+) strain. Biotechnol Bioeng 99(2):368377 Schubert T, Breuer U, Harms H, Maskow T (2007) Calorimetric bioprocess monitoring by small modifications to a standard bench-scale bioreactor. J Biotechnol 130(1):2431

Appl Microbiol Biotechnol (2012) 93:575584 Senthilkumar S, Surianarayanan M, Swaminathan G (2008) Biocalorimetric and respirometric studies on biological treatment of tannery saline wastewater. Appl Microbiol Biotechnol 78:249255. doi:10.1007/s00253-007-1309-x Sivaprakasam S, Schuler M, Hama A, Hughes K-M, Marison I (2011) Biocalorimetry as a process analytical technology process analyser; robust in-line monitoring and control of aerobic fed-batch cultures of Crabtree-negative yeast cells. J Therm Anal Calorim 111 doi:10.1007/s10973-010-1259-x Tomson K, Barber J, Vanatalu K (2006) Adaptastata new method for optimising of bacterial growth conditions in continuous culture: interactive substrate limitation based on dissolved oxygen measurement. J Microbiol Methods 64(3):380390 Tuerker M (2004) Development of biocalorimetry as a technique for process monitoring and control in technical scale fermentations. Thermochim Acta 419(12):7381 Verduyn C, Postma E, Scheffers WA, Van Dijken JP (1992) Effect of benzoic acid on metabolic fluxes in yeasts: a continuous-culture study on the regulation of respiration and alcoholic fermentation. Yeast 8(7):501517 Voisard D, Pugeaud P, Kumar AR, Jenny K, Jayaraman K, Marison IW, von Stockar U (2002) Development of a largescale biocalorimeter to monitor and control bioprocesses. Biotechnol Bioeng 80(2):125138 von Stockar U, Duboc P, Menoud L, Marison IW (1997) On-line calorimetry as a technique for process monitoring and control in biotechnology. Thermochim Acta 300(12):225236. A Collection of Invited Papers in Celebration of Volume 300 von Stockar U, Marison I (1989) The use of calorimetry in biotechnology. In: Bioprocesses and engineering, vol 40 of advances in biochemical engineering/biotechnology. Springer, Berlin, pp 93136. doi:10.1007/BFb0009829 von Stockar U, van der Wielen LAM (1997) Thermodynamics in biochemical engineering. J Biotechnol 59(12):2537 Wechselberger P, Seifert A, Herwig C (2010) Pat method to gather bioprocess parameters in real-time using simple input variables and first principle relationships. Chem Eng Sci 65(21):57345746. Pharmaceutical Engineering ScienceA Key for Tomorrows Drugs Xiong Z-Q, Guo M-J, Guo Y-X, Chu J, Zhuang Y-P, Zhang S-L (2008) Real-time viable-cell mass monitoring in high-cell-density fed-batch glutathione fermentation by Saccharomyces cerev isiae t65 in industrial complex medium. J Biosci Bioeng 105(4):409413