<<

66 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 2, NO. 2, JUNE 2008 A Real-Time Multi-Channel Monitoring System for Stem Culture Process Xicai Yue, Emmanuel M. Drakakis, Mayasari Lim, Anna Radomska, Hua Ye, Athanasios Mantalaris, Nicki Panoskaltsis, and Anthony Cass

Abstract—A novel, up to 128 channels, multi-parametric physio- Clinical treatment with stem cells may consist of autologous logical measurement system suitable for monitoring hematopoietic or allogeneic transplantation by means of systematic infusion culture processes and cell cultures in general is presented or local injection, a fact that raises unprecedented opportunities in this paper. The system aims to measure in real-time the most im- portant physical and chemical culture parameters of hematopoi- for the treatment of diseases and trauma. The transplantation of etic stem cells, including physicochemical parameters, nutrients, hematopoietic stem cells in particular, which can be obtained and metabolites, in a long-term culture process. The overarching from marrow, peripheral blood, and umbilical , scope of this research effort is to control and optimize the whole has been used for many years in the treatment of leukemia, other bioprocess by means of the acquisition of real-time quantitative types of and autoimmunity and even in broader range of physiological information from the culture. The system is designed in a modular manner. Each hardware module can operate as an such as , ischemia and cirrhosis. independent gain programmable, level shift adjustable, 16 channel For example, transplant has been used for the re- data acquisition system specific to a sensor type. Up to eight such construction of a patient’s hematopoietic system after under- data acquisition modules can be combined and connected to the going or radiotherapy to treat and disease host PC to realize the whole system hardware. The control of data such as aplastic anemia, thalassemia, Gaucher’s disease. The re- acquisition and the subsequent management of data is performed by the system’s software which is coded in LabVIEW. Preliminary alization and development of technologies based on these cells experimental results presented here show that the system not only require a readily available source of stem cells and/or their dif- has the ability to interface to various types of sensors allowing the ferentiated derivatives outside a living body. Unfortunately, the monitoring of different types of culture parameters. Moreover, it application of stem cell is still clinically limited owing can capture dynamic variations of culture parameters by means of to the demands associated with highly specialized cell cultures real-time multi-channel measurements thus providing additional information on both temporal and spatial profiles of these param- [2]. eters within a . The system is by no means constrained In contrast to many traditional processes which use a cell’s in the hematopoietic stem field only. It is suitable for capacity to produce a product or virus, this new tech- monitoring applications in general. nology aims to generate the cells themselves as the products. Index Terms—Bioprocess, cell culture, data acquisition, physio- The output of a cell culture process depends on specific chemical monitoring, stem cell. physicochemical conditions together with optimal nutrient, metabolite, and cytokine concentrations that are unique to specific cell types. These parameters keep changing during the I. INTRODUCTION culture process as the cultivated cells differentiate and expand in number. Even slight deviations in the culture parameters TEM cells with their properties of self-renew and multilin- can affect the type and amount of final cell output [3]–[6]. S eage differentiation have the capability of developing unre- The dynamic variation of the culture parameters during cell lated cell and tissue types, such as bone, , neural cells, growth means that the optimal parameters for culturing the pneumocytes, muscle, , endothelial, epithelial cells and hep- cells also vary. Therefore, it is necessary to monitor the culture atcytes used in , cellular therapies and drug parameters in real-time to capture these dynamic changes. screening [1], [2]. Although ingenious microsystems have been introduced to cell culture [7], the most common set-up for on-line, in-situ measurement for cell culture is the incorporation of flow injec- Manuscript received October 24, 2007; revised February 26, 2008, and April tion analysis (FIA) where a sample is taken from the bioreactor 25, 2008. Current version published September 10, 2008. This work was sup- ported by U.K. BBSRC and EPSRC under Project BBS/B/17298 offered to the and the sensor is not in direct contact with the culture medium Intelligent Stem Cell Culture Systems (ISCCS). [8]. This method produces readings of a delayed response since X. Yue and E. M. Drakakis are with the Department of Bioengineering, Impe- the physiological data are measured at the outlet of the sample rial College London, London SW7 2AZ, U.K. (e-mail: [email protected]). M. Lim, H. Ye and A. Mantalaris are with the Department of Chemical En- port and therefore it averages the data in time and space. gineering and Chemical Technology, Imperial College London, London SW7 The practical monitoring of cell culture processes is related to 2AZ, U.K. the adopted sensor and bioreactor technology. New cell biore- A. Radomska and A. Cass are with the Institute of , Imperial College London, London SW7 2AZ, U.K. actor technologies [9] make it possible to monitor and control N. Panoskaltsis is with the Department of Hematology, Northwick Park the physical and chemical environment of the cell culture. New Campus, Imperial College London, London HA1 3UJ, U.K. sensor technologies [10]–[12] make it possible to monitor the Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. long-term culture process. These facts when combined with a Digital Object Identifier 10.1109/TBCAS.2008.925639 physiological monitoring system, such as the one presented in

1932-4545/$25.00 © 2008 IEEE YUE et al.: REAL-TIME MULTI-CHANNEL MONITORING SYSTEM FOR STEM CELL CULTURE PROCESS 67

TABLE I It is a requirement that all quantities will be measured si- REQUIREMENTS OF THE PRACTICAL MONITORING SYSTEM multaneously in real-time; each real-time interfacing channel would be associated with a sensor positioned at a specific lo- cation within the bioreactor. In this way the acquisition of both the spatial and temporal physicochemical profiling of the cell culture becomes feasible. Most operational specifications shown in Table I could be sat- isfied by means of commercially available, albeit single-channel this paper suggest, in future, the ability to supply cells with dy- equipment. However if a multi-parametric measurement and, namic profiles of nutrients, oxygen and growth factors in an op- most importantly, a multi-channel-interfacing system were to be timal manner. built out of single-channel equipment, then factors such as size, Pioneers have attempted to monitor the parameters of cell cul- cost, and operational inflexibility would render it totally imprac- ture processes. Culture parameters such as pH, glucose, lactate tical. It is for this reason that the newly developed system opted and dissolved oxygen have been monitored as growth indicators for the application-specific development of a compact, portable, [13]. On-line detection of the changes in dissolved oxygen has multi-channel and multi-sensor measurement system tailored to been reported in high bacterial/cell density cultures [14]. How- the real-time monitoring needs of cul- ever, currently there is no compact measurement system that tures. records a variety of physicochemical parameters (such as pH, dissolved oxygen, tension, nutrient and metabolite concen- B. Electrical Specifications trations) simultaneously and in real-time. Hence, the scope of The measurement requirements shown in Table I can be the newly developed system is to provide a credible technolog- translated to electrical signal range and accuracy according ical answer to the emerging need for the on-line and in-situ mon- to the specifics of the sensor type and sensor manufacturing itoring of the stem cell bioprocess, by means of real-time mea- technology [15], [16]. surement of physicochemical parameters using our multi-site, Several types of sensors have been used for the monitoring of multi-channel monitoring system. cell culture processes and can be classified into three groups ac- This paper introduces the hardware and software of the cording to the type of the electrical output signal of the sensor: proposed monitoring system including the system specifica- potentiometric, amperometric and ohmic sensors. Bearing in tions, the hardware module design (used for multi-channel and mind Table I, the pH sensor is a potentiometric sensor, whereas multi-parametric measurements), and the software modules the and glucose sensors are amperometric sensors. The (used for measurement control and data management). The temperature sensor is an ohmic sensor. hardware modules are benchmarked against commercially Potentiometric sensor—The typical potentiometric sensor available instruments. Preliminary experimental results of stem is that of pH. A pH sensor can be considered as a mV-level cell culture process monitoring collected by means of the newly voltage source with a series source resistance dependent upon developed system are also reported in this paper. However, the the electrode’s composition and configuration. The typical glass authors would like to stress that though these results suggest bulb pH sensor resistance varies between and . the usefulness of the reported system in future, in no way At 25 C, the output changes by 59 mV/pH unit and therefore, can they substantiate claims related to the control of stem cell the resolution of 0.1 pH units listed in Table I translates to a differentiation paths at this stage. voltage resolution of 5.9 mV. Amperometric sensor—Amperometric sensors can be II. SYSTEM OVERVIEW modeled as high impedance nA-level current sources. The typical amperometric sensor is the glucose sensor, which has A. Measurement Requirements three electrodes: reference electrode, counter electrode and the Generally speaking, the most important physicochemical pa- working electrode. Generally speaking, to measure the current rameters are pH, oxygen tension, carbon dioxide tension and signal sourced from the working electrode, an “excitation” temperature. These affect the cell expansion rate and cell pop- voltage signal is applied between the reference electrode and ulation. Glucose, glutamine, lactate and ammonia are nutrient the working electrode. The output current is in the orders of and metabolite parameters which determine cell growth, differ- nAs or pAs. When the glucose concentration varies in the range entiation and cell death. of 0–35 mM, the output current range for the glucose sensors Table I lists the range of values and the corresponding ac- we intend to use changes by a factor of 100. For a current range curacy for each one of the parameters targeted for monitoring of 100 pA to 10 nA, the measurement resolution of 0.5 mM by the practical system. As it will become clear in the rest of (see Table I) translates to 140 pA current resolution. the paper, the list of Table I is non-exhaustive. The type of Ohmic sensor—The temperature sensor is a typical ohmic the sensed parameters depends on the availability of the cor- sensor. There are three basic types of temperature sensors: the responding sensor technology and the accuracy offered by the thermocouple, the thermistor and the resistance temperature de- potentiometric and amperometric data acquisition channels of tector (RTD). Of them, the RTD is the most stable and accurate our system. Hence, parameters such as stem cell factor (SCF), device [17]. The resistance of PT100 RTD [18] we used changes ammonia and lactate can also be sensed as long as the relevant almost linearly from 100 at 0 C to 138.4 at 100 C. From sensors are available. 25 Cto39 C, its resistance changes by about 6 . The required 68 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 2, NO. 2, JUNE 2008

Fig. 1. Basic operational architecture of the multi-channel and multi-parametric measurement system for stem cell culture monitoring.

Fig. 2. General block diagram of each data acquisition module. measurement resolution of 0.1 C (see Table I) translates to a For different types of sensors, the signal conditioning blocks change of 0.038 . are designed exclusively to account for their characteristics. Based on the interpretation of the system requirements from Therefore there are three types of data acquisition modules biomedical to electrical figures, the nominal resolutions for the depending on the sensor type and the conditioning circuit asso- monitoring system become: ciated with it : the potentiometric module, the amperometric, Current: 140 pA (e.g., glucose) and the ohmic one. Voltage: 5.9 mV (e.g., pH) 1) Signal-Conditioning: Sensor interfacing and conditioning Ohmic: 0.038 (e.g., temperature) circuits ensure the appropriate conversion of the sensor output The bandwidth of the above signals is much less than 1 KHz signal values to a proper range and form for further processing. (probably 1 Hz). Potentiometric signal conditioning—This signal con- ditioning circuit is designed for pH measurements. High C. Hardware Overview impedance voltage signals from pH sensors are directly ap- plied to an ultra-low bias operational amplifier of high input The PC-based, on-line, real-time measurement system is de- impedance , low offset voltage (maximum 2 mV), signed based on established multi-channel data acquisition tech- low drift, low input bias current and low input offset current niques [19], [20]. The operational system architecture is shown (in the orders of fAs). The potentiometric signal conditioning in Fig. 1. block is a non-inverting amplifier with a gain of 11. This gain Analog signals from different types of sensors are input to is produced by means of two 0.01% precision resistors. The different data acquisition modules where they are conditioned designed circuits can also be used to measure from other types and subsequently converted to digital form. The digital signal is of high impedance potentiometric sensors such as ammonia then transferred to the host PC. The setting of the measurement sensors. Their measurement range lies from 200 mV to 200 configuration parameters is achieved through a graphical user mV. As illustrated in the general block of Fig. 2, a stage of interface by means of the LabVIEW (National Instrumentation, programmable gain amplifier (which follows the non-inverting Austin, TX, USA) [21] application program. signal conditioning amplifier) can provide an additional gain varying from 1 to 4096 and an adjustable DC level. Hence, D. Data Acquisition Modules the potentiometric module is flexible enough to interface with The generic block diagram of the eight data acquisition mod- potentiometric sensors of a wide range of input signals. ules is shown in Fig. 2. Sixteen sensor output signals are con- Ohmic signal conditioning—To measure temperature, a ditioned and subsequently multiplexed, band limited by a low- small constant current is applied to the RTD sensor and the pass filter, amplified and level adjusted by a gain programmable voltage drop across the RTD is measured. The schematic dia- amplifier to fit the 0–5 V input range of the analog to digital gram of the ohmic sensor signal conditioning is shown in Fig. 3. converter (ADC). A 16-bit ADC is used in each data acquisition A four-wire RTD (two wires for carrying the “sense” current module to convert analog signals to digital form with a resolu- and two for measuring the voltage across the element) rather tion of 0.076 mV [22]. than a three-wire Wheatstone bridge is adopted to measure the YUE et al.: REAL-TIME MULTI-CHANNEL MONITORING SYSTEM FOR STEM CELL CULTURE PROCESS 69

Fig. 4. Amperometric signal interface and conditioning.

Fig. 3. RTD sensor (Ohmic) signal conditioning/interfacing diagram. voltage set by the DAC is re-measured at the reference electrode (R) of the amperometric sensor (See Fig. 4). value of the RTD resistance. The signal conditioning circuits The amperometric measurement module is designed for the are composed of two parts: a current source and a signal condi- glucose sensor, but it is generic enough for other amperometric tioning amplifier. A 2.5 V 1 mV precision voltage reference sensors as well. With the gain programmable amplifier, the am- and a 0.04% tolerance ultra precise 2.5 resistor form a 1 perometric measurement module covers a current measurement mA current feeding through the RTD sensor. A low leakage range from 0.1 nA to 1 mA and the excitation voltage for each current (in the orders of pAs) amplifier is used for the current to channel can be set from 2.5 V to 2.5 V with a 13-bit accu- voltage conversion. The potential difference across pins 2 and 3 racy. of the 4-wire RTD is amplified by means of an instrumentation 2) Measurement Control: The data acquisition modules are amplifier (see Fig. 3) and can be further amplified by the sub- controlled by the host PC via one of its USB ports [23]. A com- sequent programmable gain amplifier as shown in the generic mercially available 24 digital inputs/outputs (DIO)-USB inter- block diagram of Fig. 2. face card is adopted. Some of the ports are used as output ports to The resistor value can be calculated combining the measured transmit measurement commands such as “channel select data” voltage signal and the known current source value. The temper- and the gain programmable amplifier’s setting data from the ature is finally calculated by the relation: host PC to the data acquisition modules. Other ports are used to input the measured data from the data acquisition modules C to the host PC. The basic structure of measurement control is shown in Fig. 5. where denotes the measured resistance value (in Ohms) at The measurement control procedure is organized as follows: temperature C, is a known resistance value whereas and to measure from a target sensor, the module address is sent are known constants specific to the RTD sensors. For the used to all data acquisition modules. The ADC in the data acquisi- PT100, , , . tion module whose address setting matches the module address Since the leakage currents of both amplifiers are in the order signal is enabled. The channel selection data and gain of ampli- of pAs, the wire resistors of the 4-pin RTD (especially those of fier data are sent to the selected module and the measured data wires #2 and #3) have no significant effect on the accuracy of the are transferred from the enabled ADC to the host PC. test result even though their value is relatively high compared to Apart from the above basic control functions, additional the required measurement resolution of 0.038 . The specific controls are needed for amperometric modules. The excitation ADC used exhibits a worst-case error less than 1 mV. Clearly voltage for the amperometric sensor is set by the DACs that the combined gain offered by the instrumentation amplifier and are programmable via the host PC. This control is not very the PGA of the flexible architecture shown in Fig. 2, ensures that different from the basic control and therefore it is not shown in the targeted measurement precision can be met. detail in Fig. 5. Amperometric signal conditioning—As shown in Fig. 4, The control structure equips the system with flexible config- the low current signal from the working electrode (W) of the uration. Its modular design and structure allows for it to be con- sensor is measured by means of a low leakage current (in the figured for the measurement of the same type of sensor with order of pAs), low offset voltage and high input impedance main up to 128 channels or for up to eight different types of sensors amplifier. A 2 0.01% precision resistance is used in the with 16 channels allocated to each one. Any other combination feedback branch of the main amplifier. As the sensor’s output of sensor types (up to 8) and 16-channel data acquisition mod- currents are tiny (in general), special effort is undertaken to ules (up to 128 channels) is possible. In practice, the system can avoid interference. At printed circuit board (PCB) level, guard support more types of sensors as sensors with the same type of rings are placed near the input pins of amplifiers to minimize output signal may share the 16 channels within a module. interference caused by spurious undesired signals. One ground E. Software Overview plane is placed beneath the surface mounted amplifier to reduce the leakage current from other layers of the PCB. A 13-bit DAC The main function of the software is to control the data ac- is used to set the sensor excitation voltage levels. Low offset quisition process and to manage the acquired data. The National voltage amplifiers are used to feed the excitation signal. To en- Instrumentation LabVIEW 7.1 is adopted as the programming sure the system’s high measurement accuracy, the excitation language. The software of the monitoring system is designed 70 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 2, NO. 2, JUNE 2008

Fig. 5. Measurement control structure.

played and all acquired data are written to data files. More de- tails of the four software modules are provided below: Configuration module—The configuration module is de- signed to set and update the parameters of the system such as the number of modules and the number of channels in each module, the type of the module (e.g., amperometric, potentio- metric), the active or inactive status of each module and each channel, the measurement interval of each module and the alarm level of each module. For the amperometric data acqui- sition module, the excitation voltage of each channel can also be set via this module. The user interface of the configuration module is illustrated in Fig. 6. Measurement control module—The measurement control module is linked with the hardware in a manner similar to that of hardware drivers which isolate other software modules from the hardware. This module is designed to send the “address control” data, the “channel select” data and the “amplifier control” data from the host PC to the data acquisition modules Fig. 6. Pop-up sub-panel enabling system configuration. and read the measured data from the data acquisition modules. These control data are sent automatically in accordance with the system configuration. The basic structure of a typical control with four functional modules allowing the user to reconfigure process is shown in Fig. 7. the system, display raw data and waveforms, save data file and Sequential structures and timers are employed in the mea- replay measured data files. surement control module. Sequential structures are used to send Four concurrently running software modules facilitate the use the “module-select” data, the “channel-select” data and the “am- of the system and allow its configuration according to specific plifier-gain” and “offset-control” data from the host PC to the practical needs. Both the measurement interval and the alarm data acquisition modules and read the measured data from the level can be set for each module. After the system is configured data acquisition modules in chronological order, while a timer by the above parameters, the whole data acquisition system runs is used to determine when the measurement should be taken ac- automatically. The acquired data from selected channels are dis- cording to a pre-recorded configuration file. Sample averaging YUE et al.: REAL-TIME MULTI-CHANNEL MONITORING SYSTEM FOR STEM CELL CULTURE PROCESS 71

Fig. 7. Basic operational structure of the measurement control software module. Fig. 9. Test results of the potentiometric module.

stem cell culture process can take several weeks, the data pro- duced in this period must be kept at a reasonable size. The con- figuration parameters of measurement interval are used to con- trol the data writing process for this purpose.

F. System Tests The electrical performance of the whole system has been evaluated. Our potentiometric module was compared against a single channel pH/mV/ C meter (Model: MP 220, Met- tler Toledo Inc., Columbus, OH, USA). A commercial pH sensor was used (WTW Sen-Tix 42 pH electrode, WTW Wissenschaftlich-Technische, Werkstätten GmbH, Germany). Voltages produced by the pH sensor were measured by both our system and the pH/mV/ C meter. Fig. 9 reports indicative test Fig. 8. Front panel of the graphical user interface. results for the potentiometric module. The “ideal response” line in Fig. 9 corresponds to (ideal) module readings that equal the pH meter reading against which our potentiometric module is techniques which have been successfully used to extract very compared. From Fig. 9 it is clear that the measurement errors of weak electroencephalography (EEG) [24] signals from back- potentiometric module are within the targeted (see Section II-B) ground noise are adopted to improve the signal to noise ratio 5.9 mV range. In fact, the measured measurement errors are (SNR). These techniques are particularly useful for the reduc- below 3 mV. tion of white noise which shares bandwidth with the desired The amperometric module has been evaluated by measuring signal and can not be easily filtered out by conventional filters pre-set currents generated by a commercial ultra-high-preci- (N times averaging can increases the SNR times ). sion current source (Model: 6220, Keithley Instruments Inc., User interface module-—A virtual instrumentation user in- Cleveland, OH). The known input current varied from 1 to terface module is designed for the display of raw data and wave- 10 nA in steps of 1 nA and was measured by our amperometric forms as shown in Fig. 8. This module can display at the same module. For each current setting five module readings have time the 16 raw data originating from all 16 channels within a been recorded. The errors of the measurement (input cur- data acquisition module. A channel can be selected to show its rent-measured current) are shown in Fig. 10. It can be verified trace/waveform marked with the alarm level. The alarm infor- that the measurement errors are contained within the required mation along with the module number, the channel number, and 140 pA accuracy range. the “start” and “stop” times of the alarm are also displayed via The ohmic module has been evaluated by comparing the this interface. voltage readings of the PT100 RTD when a current of 1 mA is Data log module—The data log module is designed to record applied to the sensor. The current is fed through pins 1 and 4 measured data and alarm information into data files for further of the RTD (see Fig. 3). Subsequently, the voltage drop across use (e.g., “data replay” and “report generation”). As the whole pins 2 and 3 is measured by both the newly developed module 72 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 2, NO. 2, JUNE 2008

Fig. 10. Test results of the amperometric module. Fig. 12. Practical monitoring system with one amperometric module, one po- tentiometric module and one ohmic module.

Fig. 11. Test results of the ohmic module.

Fig. 13. Our perfused bioreactor with the (bio)sensors mounted. and a 6 1/2 digits multi-meter which has 1 V voltage reso- lution (Model 2000/E, Keithley Instruments Inc., Cleveland, processes have been optimized for enhanced biocompatibility OH). Varying the temperature of the tested water from 20 C and longevity. to 45 C, the corresponding voltage readings of the RTD are The amperometric sensors were constructed by mounting a recorded. Fig. 11 reports indicative test results for the ohmic platinum wire into a glass capillary. The electrical contact used module which confirm that the targeted resolution of 0.038 Wood’s metal. Subsequently the platinum wire was covered has been achieved. with polymer which is impermeable to liquids and gases. The tip of the wire was cut and a 200 micron disc electrode was III. CELL CULTURE MONITORING EXPERIMENTS formed. The potentiometric sensors were fabricated by dipping a pipette tip into an appropriate ion-selective membrane solu- A. Experiment Components tion. Then the tip was filled with the internal electrolyte and 1) Monitoring System: A complete practical monitoring the Ag/AgCl wire was assembled as an internal electrode. The system is shown in Fig. 12. Three data acquisition modules are dimensions of the sensing part are 0.2 mm, which is suitable mounted: one amperometric, one potentiometric and an ohmic for the small volume bioreactor used for stem cell culture. one. A USB cable connects the data acquisition modules to a Manufacturing and performance details of PEG-modified ion laptop which is used as the host PC. selective electrode based ammonia sensors used as part of our 2) Sensors and Bioreactor: A significant challenge in stem monitoring system can be found in [25]. cell culture is the interaction of the sensor with the host en- Our perfused bioreactor is shown in Fig. 13. The diameter of vironment (e.g., protein adsorption, cell adhesion), which is the bioreactor is 3.3 cm and it has a depth of 1.0 cm. Two inlet qualitatively described by the term sensor biocompatibility. and two outlet ports for perfusion are placed directly opposite The sensors designed for the monitoring stem cell culture to each other. The inlet ports are placed closer to the bottom of YUE et al.: REAL-TIME MULTI-CHANNEL MONITORING SYSTEM FOR STEM CELL CULTURE PROCESS 73

Fig. 14. Temperature experiment practical set-up. the chamber, at 0.1 cm from the bottom, while the outlet ports are at 0.1 cm from the top of the chamber. The positions of the inlet and outlet ports are such that the outlet will not allow cells Fig. 15. Recorded temperature time profiles at different locations within the to leave the bioreactor at ease while at the same time, the inlet perfused bioreactor. ports should allow cells to obtain fresh media readily. The cover of the bioreactor is made out of polydimethylsiloxane (PMDS), which allows the sensors to be conveniently inserted into the approximately 20 seconds later than sensor_1 and sensor_2. bioreactor at any location required. Sensor_3 has the lowest speed of temperature change. After 7 min, the temperature within the whole bioreactor tends to B. Experiments be uniform everywhere. Bearing in mind that both temperature 1) Temperature Monitoring Experiment: The temperature and mass are governed by the same form of dynamics [26], monitoring experiment described below aims to demonstrate these results show that with multi-channel measurements, tem- that additional useful culture information can be obtained by poral/spatial profiles within a bioreactor can be recorded reli- concurrent real-time multi-channel measurements. ably. These qualitative results show that our monitoring system The experiment set-up is illustrated in Fig. 14. Hot and cold can reliably provide additional, localized culture information water is simultaneously perfused at opposite ends of the biore- not easily obtainable by other means. actor at a flow rate of 0.5 mL/min. The hot and cold water tem- 2) Cell Culture Experiments: peratures are within 65 C–70 C and 18 C–19 C respectively. Experiment Set-up: Ammonia is the by-product of glu- It must be stressed that when the experiment starts the bioreactor tamine metabolism and is more toxic than lactate. Ammonia is already full of water of C. (In other words we do not levels are important in cell culture processes because high levels mix two equal volumes of water of two different temperatures of ammonia inhibit cell growth and basic cellular activities. In in an empty bioreactor. If that was the case then the resulting a stem cell culture, this reduces the expansion and prolifera- final temperature would be the average of the two temperatures). tion capability of stem cells, which results in a slower rate of Three sensors are placed within the bioreactor as shown in the cell growth. At extreme levels, ammonia can cause necrotic cell figure: one in the centre while the other two are located at the death. One study investigating the effects of ammonia and glu- edges of the bioreactor. Temperature readings are obtained by tamate on development indicated that the accumulation the system every 10 s for a total duration of 8 min. of ammonia affects embryo development by reducing blasto- The experimental results are shown in Fig. 15. It can be seen cyst cell number and cell mass, disrupting the metabolism and that the temperature rise is felt first by sensor_1 which is lo- intracellular pH regulation, and also altering expression cated closer to the hot water inlet than the two other sensors. The [27]. A close monitoring of ammonia levels is therefore crit- sensor_1 temperature recorded by the system reaches its max- ical in maintaining a healthy stem cell culture. In another study imum value of 23.2 C after approximately 200 seconds and that investigates the expansion of mouse embryonic stem cells then starts to taper off as the mixture in the bioreactor reaches in a stirred culture system, ammonia levels were monitored and a steady state temperature. Furthermore observe that though the kept below 3 mM at all times [28]. In a mammalian cell culture temperature change at the position of sensor_2 is felt almost si- study, ammonia level at 4 mM was found to reduce the specific multaneously with sensor_1, the speed of temperature change at growth rate of the cells by half [29]. In what follows we describe the position of sensor_2 is slightly lower than the corresponding preliminary experiments which focus upon the ammonia moni- speed of temperature change at the position of sensor_1. The toring of cell cultures by means of our biocompatible, long-life curve of sensor_2 becomes flat after 200 seconds of measure- (up to two weeks) potentiometric sensors [25]. ment. Also observe that sensor_2 reaches a lower than sensor_1 CD34+ cells were isolated from cryopreserved cord blood. maximum temperature value. This can be explained by the fact Culture media used was Iscove’s modified Dubelcco’s medium that sensor_1 is closer to the hot water inlet whereas sensor_2 is (IMDM) 10% fetal bovine serum (FBS) 4.5IU/ml erythro- closer to the colder water inlet. With respect of sensor_3 which poietin (EPO) and 75 ng/ml stem cell factor (SCF) 1% antibi- is located closer to the cold water inlet than the other two ob- otics. Perfusion culture were run for 7 days with a perfusion rate serve that its temperature initially drops and starts to increase of 7.2 mL/day while static cultures were also run for 7 days with 74 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 2, NO. 2, JUNE 2008

Fig. 16. Recorded raw ammonia concentration for the static cell culture case. replacement of the culture medium on day 4. Cells are seeded at a density of cells in each bioreactor, and a control (media only without cells) culture was also monitored for 7 days. At the end of the culture period, cells were examined under the micro- scope and enumerated to determine cell growth and viability. Throughout the culture period, cells were continuously moni- tored for ammonia levels. A similar experimental setup was per- formed for a leukaemic cell line, HL60, as a comparison in per- formance. Culture media used for HL60 cells was Dulbecco’s Modified Eagle’s medium (DMEM) 20% FBS 1% antibi- otics. HL60 cells were seeded at cells in each bioreactor, media volume and perfusion rate used were identical to the cord blood culture. Fig. 17. Real-time ammonia monitoring within the bioreactor at its center and Two sensors are used to determine variations of ammonia its edge. Top: Cord-blood CD34+ cell culture. Bottom: HL60 leukaemic cell concentrations in the bioreactor system. One sensor is placed in culture. the centre of the bioreactor whereas the other one is placed at the edge of the bioreactor. The data acquisition rate for these exper- iments was set at 10 minutes per reading, since this frequency of difference between the two curves reveals that cells grow better data collection is sufficient to capture any changes in the culture in the centre of the bioreactor than that at the edge of the biore- condition due to cell growth or other cellular activities. actor. A similar scenario is observed with the HL60 culture. The Results: Fig. 16 illustrates the raw ammonia concentration ammonia concentration at the centre of the bioreactor increases data of the last 24 hours of static culture in which no cells are faster than that at the edge [Fig. 17(bottom)]. The concentrations seeded. The upper curve represents the raw ammonia concen- eventually taper off to their respective steady-state values as the tration data at the edge whereas the bottom curve represents the production of ammonia is balanced by the constant removal of raw ammonia concentration data at the centre. There is practi- ammonia due to perfusion with media. In this case, we have also cally no concentration change during the whole process. After imaged the distribution of cells inside the bioreactor by a Leica the raw data are calibrated, both sensor traces coincide and an DMIL (Leica Microsystems Wetzlar GmbH, Wetzlar, Germany) ammonia concentration of 1.5 mM is detected. microscope as shown in Fig. 18. The pictures show a higher cell Fig. 17 illustrates the ammonia concentration data of the cell density at the center and middle areas of the bioreactor than at culture processes for cord blood [Fig. 17(top)] and HL60 cells the edge of the bioreactor consistent with the ammonia measure- [Fig. 17(bottom)] in the perfused system. In the case of the cord ments and demonstrate the ability of our monitoring system to blood culture, the upper curve represents the ammonia concen- detect and record with spatial resolution the metabolic activity tration at the centre of the bioreactor whereas the bottom curve of cells within a bioreactor. Bearing in mind the recorded am- represents the ammonia concentration at the edge [Fig. 17(top)]. monia concentration values of Figs. 16 and 17, it can be seen In the beginning, both traces start at 1.5 mM. Subsequently both that though the trend (increase, decrease, etc.) of the concentra- curves keep increasing which indicates that the cells are growing tion value recorded from each sensor is clear, the variations in the bioreactor and produce more ammonia. However, the am- of the recorded value might vary from sensor to sensor. monia concentration in the centre is higher than that in the edge This difference in noise (we adopt this term in the absence of and this tendency was preserved until the end of the culture a better one) could be attributed to sensor-to-sensor variability process. It can also be seen that the difference in ammonia con- (our sensors are hand-made), to the underlying chemistry or to centration between the two curves was increased with time and a combination of such factors. More importantly, however, it plateaus at a maximum value of approximately 1.2 mM. The should be stressed that the concentration is inferred from the YUE et al.: REAL-TIME MULTI-CHANNEL MONITORING SYSTEM FOR STEM CELL CULTURE PROCESS 75 recorded voltage value as where the terms are defined experimentally as follows: subsequently to the termination of a monitoring experiment each sensor is calibrated off-line against solutions of known concentrations. This calibra- tion aims at the experimental determination of the values and of the line .If is the electronics-induced recording error then the inferred concentration value is given by (when which is usually the case):

Consequently, for two sensors 1 and 2 whose operation is char- acterized by two different slope values and , respectively, will hold:

From this relation it is clear that depends on the value of the ratios

Fig. 18. Microscopic view of the cell distribution at different locations in the When and [case forFig. 17(top)] bioreactor. then . When and [case for Fig. 17(bottom)] then . Discussion: Local concentrations provide information on identification of conditions related optimally to the differentia- the spatial variation of essential culture parameters. Ultimately, tion of stem cells towards specific lineages. Currently we are in this could result to the identification of micro-concentration gra- the process of developing such novel sensors. dients, which in vivo represent the stem cell niches that regulate the proliferation and differentiation processes [30]. Although IV. CONCLUSIONS the practical ability to provide spatiotemporal information by A multi-parameter and multi-channel measurement system means of real-time multi-channel monitoring has been demon- capable of real-time physicochemical monitoring of hematopoi- strated by our experiments, it should be stressed that the exper- etic stem cell cultures and cell cultures in general has been iments presented in this paper are preliminary. For large-scale developed. The meaningful operation of the system as far multi-channel measurements, the basic issue of how many sen- as its ability to provide spatial and temporal resolution of sors should be used and where they should be placed within a sensed culture parameters, has been demonstrated by means bioreactor in order to generate reliable space-and time-depen- of preliminary experiments such as the real-time monitoring dent parameter profiles is still under investigation and is focused of temperature variations within a bioreactor and the real-time on the use of statistical design of experiments (DOE) [31], [32]. monitoring of ammonia concentration within a static and a However our preliminary experimental results presented here perfused bioreactor. The system is currently used to analyze demonstrate: in real-time stem cell cultures. Ultimately, the new modality a) ability of our system to interface with different types of could contribute to the optimal control of stem cell culture sensors providing real-time in-situ culture process infor- bioprocesses, by allowing the control and regulation of stem mation, and cell self-renewal, expansion, differentiation and death which b) ability of our system to monitor different cul- in turn could, perhaps, lead to the development of clinically tures. relevant culture systems capable of generating reproducible, Based on these facts it is believed that by introducing novel well-characterized, “designer” tissues and organs that meet the sensors which target parameters specific to stem cells (e.g., SCF, strict regulatory criteria for clinical applications. Furthermore, FL) the newly developed system will be able to provide the the real-time on-line culture process monitoring makes it pos- corresponding spatiotemporal profiles facilitating, perhaps, the sible to optimize and control cell culture feeding strategies. 76 IEEE TRANSACTIONS ON BIOMEDICAL CIRCUITS AND SYSTEMS, VOL. 2, NO. 2, JUNE 2008

ACKNOWLEDGMENT [24] G. L. Krauss, The Johns Hopkins Atlas of Digital EEG: An Interactive Training Guide. Baltimore, MD: Johns Hopkins Univ. Press, 2006. The authors wish to thank the U.K. BBSRC and EPSRC for [25] A. Radomska et al., “PET-modified ion selective electrode for mon- the financial support offered to the Intelligent Stem Cell Culture itoring metabolic activity during the growth and cultivation of stem cells,” Biosens. Biochem., to be published. Systems (ISCCS) project (B/BBS/17298). [26] A. F. Mills, Heat and Mass Transfer. New York: Irwin, 1995. [27] Y. He et al., “Glutamine synthetase is essential in early mouse embryo- genesis,” Development. Dynam., vol. 236, pp. 1865–1875, 2007. REFERENCES [28] A. M. Fernandes et al., “Mouse expension in a microcarrier-based stirred culture system,” J. Biotechnol., vol. 132, pp. [1] F. M. Watt and B. L. M. Hogan, “Out of Eden: Stem cell and their 227–236, 2007. niches,” , vol. 287, pp. 1427–1430, 2000. [29] S. S. Ozturk, M. R. Riley, and B. O. Palsson, “Effects of ammonia and [2] K. M. Sales, H. J. Salacinski, and N. Alobaid, “Advancing vas- lactate on hybridoma growth, metabolism, and antibody production,” cular tissue engineering: The role of stem cell technology,” Trends Biotechnol. Bioeng., vol. 39, pp. 418–431, 1992. Biotechnol., vol. 23, no. 9, pp. 461–467, 2005. [30] , M. A. Lichtman, Ed. et al., Williams Hematology, 7th ed. New York: [3] J. Audet et al., “Common and distinct features of cytokine effects on McGraw-Hill, 2005. hematopoietic stem and progenitor cells revealed by dose-response sur- [31] R. H. Myers and D. C. Montgomery, Response Surface Method- face analysis,” Biotechnol. Bioeng., vol. 80, no. 4, pp. 393–404, 2002. ology: Process and Product Optimization Using Designed Experi- [4] H. Yang, W. M. Miller, and E. T. Papoutsakis, “High pH promotes ments. New York: Wiley, 2002. megakaryocytic maturation and apoptosis,” Stem Cells, vol. 20, pp. [32] M. Lim et al., “Towards information-rich bioprocessing: Generation 320–328, 2002. of spatio-temporal profiles through the use of design of experiments [5] C. Proulx et al., “Increased megakaryopiesis in cultures in CD34-en- to determine optimal number and location of sensors—An example in riched cord blood cells maintained at 39 C,” Biotechnol. Bioeng., vol. thermal profiles,” Biochem. Eng. J., to be published. 88, pp. 675–680, 2004. [6] T. A. McAdams, W. M. Miller, and E. T. Papoutsakis, “pH is a potent modulator of erythroid differentiation,” Br. J. Haematol., vol. 103, pp. 317–325, 1998. [7] J. B. Christen and A. G. Andreou, “Design, fabrication and testing of a hybrid CMOS/PDMS microsystem for cell culture and incubation,” IEEE Trans. Biomed. Circuits Syst., vol. 1, no. 1, pp. 3–18, Mar. 2007. [8] M. Lim, H. Ya, and N. Panoskaltsis, “Intelligent bioprocessing for Xicai Yue received the B.Eng. degree in communica- haemotopoietic cell culture using monitoring and design of experi- tion engineering in 1985, and the M.Eng. and Ph.D. ments,” Biotechnol. Advances, vol. 25, pp. 353–368, 2007. degrees in biomedical engineering in 1995 and 1999, [9] F. Ulloa-Montoya, C. M. Verfaillie, and W. Hu, “Culture systems for respectively. pluripotent stem cells,” J. Biosci. Bioeng., vol. 100, no. 1, pp. 12–27, Since graduation, he has been a University 2005. Teaching Assistant and then a Lecturer in China. [10] P. A. Hammond, D. Ali, and D. R. S. Cumming, “Design of a single- From 1999 to 2004, he worked in Tsinghua Univer- chip pH sensor using a conventional 0.6 "M CMOS process,” IEEE sity, Beijing, China, and Oxford Brookes University. Sensors J., vol. 4, no. 6, pp. 706–712, Dec. 2004. He is currently working in the Department of [11] X. Xu, S. Smith, J. Urban, and Z. Cui, “An inline non-invasive optical Bioengineering, Imperial College London, London, system to monitor pH in cell and ,” Med. Eng. Phys., vol. U.K. His research interests include digital switching 28, pp. 468–474, 2006. and speech signal processing, joint time-frequency analysis, pattern recog- [12] E. Hwang et al., “Evaluation of the paratrend multi-analyte sensor nition with neural networks, auditory brainstem responses (ABR) and other for potential utilization in long-duration automated cell culture moni- biomedical signal processing, electrical impedance topography (EIT) for med- toring,” Biomed. Dev., vol. 3, no. 6, pp. 241–249, 2004. ical imaging, stem cell culture process monitoring, FPGA/embedded system [13] M. S. Kallos and L. A. Behie, “Inoculation and growth condition for design and low-power VSLI design for biomedical use. He has authored or high-cell-density expansion of mammalian neural stem cells in sus- co-authored more than 20 peer-reviewed journal papers. pension ,” Biotechnol. Bioeng., vol. 63, no. 4, pp. 473–483, Dr. Yue received an IEEE ISCAS Live Demo Special Session Award in 2007. 1999. [14] V. S. Whiffin, M. J. Cooney, and R. Cord-Ruwisch, “Online detection of feed demand in high cell density culture of Escherichia coli by mea- surement of changes in dissolved oxygen transients in complex media,” Emmanuel M. Drakakis (M’05) received the Biotechnol. Bioeng., vol. 85, no. 4, pp. 422–428, 2004. B.Sc. degree in physics and the M.Phil. degree [15] A. J. Bard and L. R. Faulkner, Electrochemical Methods: Funda- in electronic physics and radioelectrology from mentals and Applications, 2nd ed. New York: Wiley, 2001, ISBN Aristotle University of Thessaloniki, Macedonia, 0471043729. Greece, and the Ph.D. degree in analog IC design [16] J. Wang, Analytical Electrochemistry, 2nd ed. New York: Wiley, from the Department of Electrical and Electronic 2000, ISBN 0471678791. Engineering, Imperial College London, London, [17] C. Swanson, “Optimal temperature sensor selection: Achieving accu- U.K., in 2000 under the supervision of Dr. A. Payne. rate temperature measurement,” EuroAsia Semiconductor, vol. 29, no. He is a Senior Lecturer in the Department of 7, pp. 23–28, 2007. Bioengineering, Imperial College London, which [18] J. W. Quity et al., “Thermoluminescence apparatus using PT100 resis- he joined in October 2001. In the Department of tors as the heating and sensing elements,” Rev. Scientific Instrum., vol. Bioengineering, he founded the Bioinspired VLSI Circuits and Systems Group. 78, no. 8, p. 083905, 2007. The Group’s research focuses on circuits and systems “for and from .” [19] E. M. Spinelli, R. Pallas-Areny, and M. A. Mayosky, “AC-coupled He has authored or co-authored more than 70 peer-reviewed publications. front-end for biopotential measurement,” IEEE Trans. Biomed. Eng., Dr. Drakakis received a Prize from the Hellenic Army’s Research and vol. 50, no. 3, pp. 391–395, Mar. 2003. Technology Center in 1995. Between 1996-1998, he was sponsored by the [20] W. J. R. Dunseath and E. F. Kelly, “Multichannel PC-based data-ac- Micro-Electronics Research Center (MERC) of LM Ericsson, Kista, Stock- quisition system for high-resolution EEG,” IEEE Trans. Biomed. Eng., holm, Sweden. His Group received the IEEE MWSCAS Finalist Award in 2005 vol. 42, no. 12, pp. 1212–1217, 1995. and the IEEE ISCAS Live Demo Special Session Award in 2007. In 2006, he [21] G. W. Johnson and R. Jennings, LabVIEW Graphical Programming: received the Rector’s Award for Research Excellence, and in 2008 he received Practical Applications in Instrumentation and Control, 3rd ed. New a Human Frontier Science Program Award. He is a member of the BIOCAS York: McGraw-Hill, 2001. and CNNA IEEE Technical Committees, a past Associate Editor for both IEEE [22] J. Park and S. Mackay, Practical Data Acquisition for Instrumentation TRANSACTIONS ON CIRCUITS AND SYSTEMS—I: REGULAR PAPERS and IEEE and Control Systems. New York: Elsevier, 2003. TRANSACTIONS ON CIRCUITS AND SYSTEMS—II: EXPRESS BRIEFS, past Guest [23] D. Anderson, Universal Serial Bus System Architecture, 2nd Assistant Editor for IEE Electronics Letters and past Subject Editor for the ed. Reading, MA: Addison-Wesley, 2001. International Journal of Electronics (Taylor & Francis). YUE et al.: REAL-TIME MULTI-CHANNEL MONITORING SYSTEM FOR STEM CELL CULTURE PROCESS 77

Mayasari Lim received the B.Sc. degree in chem- Athanasios Mantalaris received the Ph.D. degree ical engineering from the University of at in from the University of Berkeley. Rochester, Rochester, NY, in bioprocess/tissue She is a Ph.D. research assistant in the Department engineering. of Chemical Engineering, Imperial College London, He is a Reader in the Department of Chemical En- London, U.K. Her current research interests include gineering, Imperial College London, London, U.K. haematopoietic stem cell bioprocessing and the ap- His interests are in multidisciplinary research with a plication of methods and strategies in experimental focus on applied bioprocessing that bridges science, designs for cell culture. engineering, and medicine. He has developed novel 3-D bioreactors for bone marrow tissue engineering and novel bioprocesses working with cord blood and embryonic stem cells. He has published over 50 papers in journals and books, and holds two U.S. . He sits on the MRC Stem Cell User Liaison Com- mittee.

Anna Radomska received the Ph.D. degree with dis- tinction in analytical chemistry from the Department Nicki Panoskaltsis received the M.D. degree from the , of Chemistry, Warsaw University, Warsaw, Poland, in Toronto, ON, , completed clinical training in internal medicine and 2004. haematology at the University of Rochester, Rochester, NY, and thereafter She is currently a Research Associate in the In- received the Ph.D. degree in immunology from Imperial College London, stitute of Biomedical Engineering, Imperial College London, U.K. London, London, U.K. During her Ph.D., she devel- She is an Assistant Professor in the Department of Haematology, Imperial oped several potentiometric and optical College London, and a Consultant Haematologist at Northwick Park and St. which were successfully used for monitoring, control Mark’s Hospital campus. She has been a member of the editorial board of the and assessment of haemodialysis therapy. The main journal Leukemia since 2003. fields of her research cover development and appli- cations of chemical sensors and biosensors in clinical and biomedical analysis. Her research has resulted in one book chapter and 11 journal publications as well as 12 conference publications. Anthony Cass is currently Professor of Chemical Bi- ology, Deputy Director and Research Director (Bio- nanotechnology) in the Institute of Biomedical En- gineering at Imperial College London, and a Fellow of the Royal Society of Chemistry. He trained origi- nally as a chemist with degrees from the Universities of York and Oxford. His research interests are in the Hua Ye received the degree in chemical engineering field of analytical and particularly in from Dalian University of Technology, China She re- the use of protein engineering and design to produce ceived the Ph.D. degree in biochemical engineering new reagents for biosensors and bioanalysis. He pi- from the University of Oxford, Oxford, U.K. oneered the use of synthetic electron transfer media- She joined the Imperial College London as a Post- tors for enzyme biosensors and his work in this area led to the development of doctoral Research Associate in the Chemical Engi- the first electronic blood glucose measuring system, commercialized by MediS- neering Department in March 2005. She is currently a ense Inc. (now part of Abbott Diagnostics), and the award of the Royal Society’s RCUK Academic Fellow in the Department of Engi- Mullard Medal (along with Prof. HAO Hill FRS and Dr. M. J. Green). Most of neering Science, University of Oxford. Her research his current research is focussed on using engineered and peptides in interests include tissue engineering and stem cell bio- a micro- and nano-structured materials and devices for both clinical and high processing, specifically bioreactor design, biomate- throughput analysis. In addition to his academic research, he is a member of sev- rials, stem cell ex vivo expansion and differentiation. eral Research Council Committees, a member of the Scientific Advisory Board of Oxford Biosensors and has acted as a consultant to several European and U.S. biotechnology companies. He is a member of the advisory board of International Pharmaceutical Training Ltd. He has published over 80 papers and edited three books, and is on the editorial boards of Biosensors and and IEE Proceedings . In addition, he is a Visiting Professor of the Chinese Academy of Sciences.