379
NEURONAL NETWORKS AS BIOSENSORS JOURNAL OF APPLIED TOXICOLOGY J. Appl. Toxicol. 24, 379–385 (2004) Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/jat.1026
Cultured Neuronal Networks as Environmental Biosensors Thomas J. O’Shaughnessy,* Samuel A. Gray and Joseph J. Pancrazio Center for Bio/Molecular Science and Engineering, Code 6900, Naval Research Laboratory, Washington, DC 20375, USA
Key words: biosensor; microelectrode array; primary neuronal cultures; extracellular recording; action potentials.
Contamination of water by toxins, either intentionally or unintentionally, is a growing concern for both military and civilian agencies and thus there is a need for systems capable of monitoring a wide range of natural and industrial toxicants. The EILATox-Oregon Workshop held in September 2002 provided an opportunity to test the capabilities of a prototype neuronal network-based biosensor with unknown contaminants in water samples. The biosensor is a portable device capable of recording the action potential activity from a network of mammalian neurons grown on glass microelectrode arrays. Changes in the action potential firing rate across the network are monitored to determine exposure to toxicants. A series of three neuronal networks derived from mice was used to test seven unknown samples. Two of these unknowns later were revealed to be blanks, to which the neuronal networks did not respond. Of the five remaining unknowns, a significant change in network activity was detected for four of the compounds at concentrations below a lethal level for humans: mercuric chloride, sodium arsenite, phosdrin and chlordimeform. These compounds — two heavy metals, an organophosphate and an insecticide — demonstrate the breadth of detection possible with neuronal networks. The results generated at the workshop show the promise of the neuronal network biosensor as an environmental detector but there is still considerable effort needed to produce a device suitable for routine environmental threat monitoring. Copyright © 2004 John Wiley & Sons, Ltd.
INTRODUCTION Environmental contamination of water supplies, soil and air are a major concern of both military and civilian agencies. The sources for potential contaminations are many and varied, including industrial accidents and waste, agricultural use of fertilizers and pesticides, terrorist use of biological or chemical agents and even nature itself as exemplified by the algal blooms responsible for red tides. Traditional methods for detection of environmental contamination generally rely on techniques that look for specific, known contaminants. Although these methods are typically highly reliable and sensitive, the growing number of compounds of concern has quickly outpaced the ability of agencies to deploy and maintain specific tests. In addition, agencies involved with environmental monitoring must be concerned with as yet unknown toxic compounds and potentially hazardous synergistic effects between otherwise inert substances. These realities have led to the need for the development of generic environmental sensors capable of detecting toxic substances, both known and unknown, utilizing biological systems as functional references for toxicity.
* Correspondence to: T. J. O’Shaughnessy, Center for Bio/Molecular Science and Engineering, Code 6900, Naval Research Laboratory, Washington, DC 20375, USA. E-mail:
[email protected] Contract/grant sponsor: Defense Advanced Research Projects Agency. Contract/grant sponsor: Office of Naval Research. Copyright © 2004 John Wiley & Sons, Ltd. Copyright © 2004 John Wiley & Sons, Ltd.
Collaboration between the US Naval Research Laboratory and the University of North Texas has resulted in the development of a portable prototype of one such system: a neuronal network-based biosensor. This device utilizes networks of mammalian neurons as a function-based, generic threat detector. The basic principle behind this device is that neurons grown in culture form networks in which they communicate with one another via action potentials, also referred to as spikes. Action potentials are the basic units of communication among neurons within a living organism. The action potential itself is an all or nothing event, where the information content within the neural network is generally believed to be encoded in the rate of action potential firing and the strength and type of synapses between neurons, either excitatory or inhibitory. When cultured on glass microelectrode arrays (MEAs), it is possible to monitor the action potential activity from a select number of neurons in the network. Substances that affect the behavior of mammalian neurons in vivo are likely also to affect the firing patterns of the cultured neurons on MEAs, because neurons are the cellular unit within the brain responsible for cognition, motor control, sensory perception and behavior. In vivo neurons are protected by a series of mechanisms, including the blood–brain barrier and the liver, whereas the cultured neurons are directly exposed to the substance of interest, thus potentially providing greater sensitivity to threats than that resulting from oral ingestion. The use of neurons is advantageous because they can provide not only a measure of the potential toxicity of a substance but also a warning that a substance has the capacity to influence neurobehavior, a major aspect to operational readiness. J. Appl. Toxicol. 24, 379–385 (2004) Received 17 December 2003 Accepted 15 May 2004
380
T. J. O’ SHAUGHNESSY ET AL.
The EILATox-Oregon workshop held during September 2002 provided an opportunity to test the neuronal network-based biosensor against a panel of unknown substances. The workshop took place in a laboratory setting where a series of water samples, spiked either with model environmental toxins or blanks, was provided. The challenge of the workshop was twofold: to deploy successfully the portable sensor in a facility across the country from our laboratory; and to test a series of unknown samples and determine the ability of our system to detect their presence.
METHODS Neuronal network-based biosensor The biosensor is contained within an aluminum case measuring 45 cm long, 30 cm wide, 40 cm high and weighing ca. 15 kg (Fig. 1). Within the case are two separate electrical systems: one dedicated to recording the electrical activity from the microelectrode arrays and the other dedicated to controlling the temperature of the networks and providing the capacity to pump solutions over the neuronal networks. Both systems link to an interchangeable cartridge that contains the MEAs and neurons. Components of the
biosensor, including the temperature control and fluidics systems, have been described in detail elsewhere (Gray et al., 2001; Pancrazio et al., 2003). Likewise, the user interface, which provides real-time monitoring and limited quantitation of spike activity across the neuronal network, has been described previously (Pancrazio et al., 2003). The cartridge system for the microelectrode arrays used for this set of experiments consisted of a 5-mm thick silicone rubber gasket placed over the glass microelectrode array. A piece of glass coated with indium–tin oxide (ITO) formed the top of the sealed cell chamber. These pieces were held together using thumbscrews running through an aluminum top bracket and into an underlying aluminum base plate that supports the MEA. The bath volume for the assembled cartridge was ca. 5 ml and connection to the fluidics system of the biosensor was through 18-gauge hypodermic needles pushed through the rubber gasket. The ITO plating on the top glass plate served as the reference electrode for the bath. Heating resistors on the aluminum base plate were connected to the heating controller of the biosensor to maintain the chamber temperature. Cell culture and network shipping Both cultured mouse spinal cord networks and mouse frontal cortex networks were prepared at the University of
Figure 1. The neuronal network-based biosensor in operation at the EILATox-Oregon Workshop. The left panel shows the biosensor, with the computer monitor above displaying real-time data from a neuronal network. The right panel shows a closer view of the internal components of the biosensor. Copyright © 2004 John Wiley & Sons, Ltd.
J. Appl. Toxicol. 24, 379–385 (2004)
NEURONAL NETWORKS AS BIOSENSORS
North Texas (UNT) in Denton, TX. The techniques used to fabricate the MEAs and prepare the cultures on them have been described previously (Gross, 1979; Gross et al., 1985). Spinal cord networks were kept in Minimum Essential Medium (MEM) supplemented with 10% horse serum, whereas frontal cortex networks were maintained in Dulbecco’s Minimum Essential Medium (DMEM) supplemented with 5% horse serum. Both types of networks were maintained at 37 °C in a 90% air and 10% CO2 incubator. Just prior to shipment, the MEAs containing neuronal networks were assembled into the sealed cartridge format. Within the sealed cartridges, the networks are typically viable for 1–2 weeks without refreshing the media. After assembly, the cartridges were shipped by overnight commercial service from UNT to the Oregon State University in Corvalis, OR, where the workshop took place. Upon arrival, the received networks were stored in a 10% CO2 incubator at 37 °C until use. The experiments described herein were conducted 5–8 days after shipment. Preparation and testing of unknown samples For recording purposes, the shipping medium is typically replaced with a HEPES-buffered MEM consisting of MEM supplemented with 25 mM glucose, 40 mM N-(2-hydroxyethyl)piperazine-N-(2-ethanesulfonic acid) (HEPES) and 26 mM NaHCO3 with the pH adjusted to 7.4. The recording medium was serum free. The unknown samples at the workshop were provided in a synthetic soft water containing (in mg l−1): 48 NaHCO3, 30 CaSO4·H2O, 30 MgSO4 and 2 KCl, with a hardness of 40 mg l−1 (expressed as mg CaCO3 l−1) and a pH of 7.9. To introduce the unknown samples, we prepared our recording medium at double the normal concentration of each compound. This 2× medium was mixed with volumes of the unknown sample and control synthetic soft water to produce the desired dilution of the unknown sample for introduction to the neuronal network. The 2× medium mixed 1 : 1 with control synthetic water served as the control for all experiments. The unknown samples were diluted at a 1 : 100 or 1:200 ratio before addition to the neuronal networks. Depending on the results, a second application at 1 : 10 was performed. Two reasons underlie the choice to dilute the unknowns before use: the unknowns supplied were said to be at a concentration that would be of concern for human consumption, thus, to test for sensitivity below a level dangerous to humans, the samples were diluted; and dilution of the samples reduced the chance of a high concentration of an unknown toxicant killing or irreversibly harming the neurons and rendering the network useless for further experiments. The experimental paradigm utilized to test the unknown compounds began with the removal of a network cartridge from the incubator and its insertion into the biosensor. Once the temperature controller was attached to the cartridge and activated, a 15-min recording was run on all 64 channels at a sampling rate of 12 kHz per channel. These data were analyzed off-line to determine the active channels for the network. Up to 16 active channels were set in the real-time software for high-resolution recording at 40 kHz per channel. At this point the fluidics were attached to the network cartridge and control solution was perfused at 0.75 ml min−1. Efflux medium from the cartridge was collected in a waste container. Control medium perfusion continued for 30–60 min to establish a stable Copyright © 2004 John Wiley & Sons, Ltd.
381
baseline in the mean spike rate for the network. Once a stable baseline had been attained, an unknown sample was perfused at 0.75 ml min−1 across the network. Sample perfusion continued until the mean spike rate reached a new steady state typically of 20–30 min or, in the event of no apparent change in the mean spike rate, for a minimum of 30 min. Owing to limitations on time and available networks, multiple unknowns were tested on each network. Between each unknown, however, the network was washed for a minimum of 30 min with fresh control medium and a new stable baseline was established. Additional experiments were performed only if the new baseline was within 20% of the baseline before the unknown was applied. Data analysis During the experimental runs, the real-time display of mean spike rate was used to guide the experimental protocols. After the experiment, however, the data were run through high-fidelity spike detection algorithms, using custom programs for enhanced spike detection and analysis written in Visual C++ (Microsoft, Corp., Redmond WA) and as functions in the data analysis program Igor Pro (Wavemetrics, Inc., Lake Oswego, OR). The spike time stamps were binned at 1-min intervals to determine the mean spike rates for the networks. The mean and standard deviations for the mean spike rate were computed for the 6 min prior to the addition of the unknown and also for 6 min after addition of an unknown and establishment of a new plateau level, usually 30 min. Student’s t-test or one-way ANOVA, as appropriate, was used to determine if there was a significant (P < 0.05) change in the mean spike rate.
RESULTS A total of seven unknown samples were tested across three neuronal networks in a 2-day period. Two of the networks used for these experiments were frontal cortex and one was spinal cord. Two of the seven samples (nos 3 and 7) tested on a spinal cord network and a frontal cortex network, respectively, were later revealed to be blanks, i.e. the synthetic soft water. Neither network’s activity changed significantly over the 30-min period of perfusion with the blanks. The spinal cord network’s mean spike rate was 21.57 ± 0.59 Hz (mean ± SD) just prior to diffusion of the blank sample and 21.42 ± 0.49 Hz after 30 min; likewise, the frontal cortex network’s mean spike rates were 12.53 ± 0.85 Hz and 12.64 ± 1.15 Hz before and after, respectively. Four of the five remaining samples showed significant effects on the mean spike rate of at least one type of network. The most pronounced of these effects was produced by sample 5, later revealed to be mercuric chloride. Sample 5 was first introduced to a frontal cortex network at a dilution of 1 : 100 (1 mg l−1), producing a small inhibition of ca. 16% in mean spike rate. When the concentration was increased to 1 : 100 (10 mg l−1), the mean spike rate initially increased and then within 15 min was irreversibly lost (Fig. 2A). Frontal cortex networks typically exhibit a high degree of synchronized bursting of spikes across the network (Pancrazio et al., 2001). Interestingly, after the addition of the higher concentration of mercuric chloride, the initial increase in mean spike rate was due, in part, to J. Appl. Toxicol. 24, 379–385 (2004)
382
T. J. O’ SHAUGHNESSY ET AL.
Figure 2. (A) The mean spike rate for the frontal cortex network exposed to mercuric chloride (HgCl2) is shown for 5-s intervals. The frontal cortex was exhibiting a synchronous bursting activity and thus, at this level, the mean spike rate over a 5-s interval fluctuates between a low of ca. 2 Hz and a high at ca. 20 Hz. The mean spike rate prior to HgCl2 addition was 9.1 ± 1.2 Hz, whereas after addition of 1 mg l−1 HgCl2 the mean spike rate decreased to a stable level of 7.69 ± 0.9 Hz. After increasing the HgCl2 concentration to 10 mg l−1, the bursting activity gave way to tonic firing, with a distinct increase in mean spike rate prior to cessation of all spiking activity. (B) One minute of spike activity recorded from a single, representative channel before and after the addition of 10 mg l−1 HgCl2. (C) Expanded view of the region between the two vertical, dashed lines in (B), showing the spike distribution within one of the bursts. Copyright © 2004 John Wiley & Sons, Ltd.
J. Appl. Toxicol. 24, 379–385 (2004)
NEURONAL NETWORKS AS BIOSENSORS
Figure 3. Summary of the results for the unknown samples tested during the workshop (* P < 0.05).
a loss of the synchronized bursting activity in favor of tonic firing (Figs 2B and 2C). Sample 4 (phosdrin, an organophosphate) was tested on both a spinal cord and a frontal cortex network at a concentration of 40 mg l−1 (10 : 1 dilution). This resulted in significant reductions in the mean spike rate in both cases. The action of phosdrin was not reversible on the frontal cortex network, and resulted in discontinued use of that network. Sample 6 (sodium arsenite) had no effect on a frontal cortex network at 0.5 mg l−1 (200 : 1 dilution) but produced a 63.2% (P < 0.001) inhibition of mean spike rate at 10 mg l−1 (10 : 1 dilution). Chlordimeform, an insecticide, applied to a frontal cortex network at 1.3 mg l−1 (200 : 1 dilution) produced a small, reversible decrease in the mean spike of 18.0% (P = 0.01). The final compound tested, colchicine, did not produce a significant effect on the frontal cortex networks at either 2 or 40 mg l−1. However, at a concentration of 40 mg l−1 there was a clearly visible increase in the mean spike rate of ca. 15%, which was completely reversed upon washout of the colchicine. The complete reversibility of the noted effect suggests that colchicine does alter the firing rate of the neuronal networks, although it did not reach the level of significance as defined for this study at the concentrations tested. The results are summarized in Fig. 3.
DISCUSSION The EILATox-Oregon workshop presented a unique opportunity to test our prototype biosensor in a setting outside our laboratory with a set of unknown samples. The deployment of the biosensor from our laboratory in Washington, DC, to the workshop in Oregon went smoothly. The device was placed in a custom shipping crate and checked with the airline as standard luggage. More significantly, the shipment of sealed network cartridges has shown limited success. Shipments between UNT and the NRL have been routine with a success rate of ca. 50%, and the success rate was duplicated at the workshop with three of six networks arriving in a usable state. Copyright © 2004 John Wiley & Sons, Ltd.
383
The ability to ship networks is critical to the success of a cell-based sensor such as this. Preparation of the networks is a time- and skill-intensive task requiring the preparation of primary neuronal cultures. The ability to prepare cartridges at a centralized facility and ship them to the field is an important and significant step in moving this technology towards a successful transition. Of the results reported on the testing of the unknowns, failure to ‘detect’ either of the blanks is noteworthy. It is extremely important that a deployed environmental sensor has a very low rate of false positives; a system that alarms too often will not be considered reliable and will be taken out of service. The fact that the system did not respond to the blanks is an important first step. However, the conditions of this test were idealized and further validation must be performed to establish false positive rates. Future work must address the issue of sample preparation. The premise of the workshop was based on testing samples of a potable water supply, a stable matrix that is expected to be relatively free of impurities whereas testing of water from a river or lake could present additional sources of variation. The ion concentration and pH of these types of water sources are likely to vary by amounts that might affect the mean spike rate of a neuronal network as action potentials are generated based on ion gradients. For these types of water, preprocessing of the sample to correct for these types of variation, which are not toxic, will be necessary. Clearly, extension to more complex sample matrices to include soil or air requires attention. The experiments performed at the workshop were shortterm, acute tests. One potential application involves the use of the neuronal network-based biosensor for long-term monitoring. Establishment of stable baselines for 1–2 h is typical and is sufficient to perform experiments as presented here. However, over longer periods of time, slow drifts in the mean spike rate do occur under control conditions. Work is presently underway to determine how best to compensate for these slow drifts while not sacrificing the ability to detect slow-acting toxins. During the workshop the neuronal network-based biosensor showed sensitivity to a range of environmental toxicants, including two metal compounds often associated with industry (mercuric chloride and sodium arsenite), the insecticide chlordimeform and the organophosphate phosdrin. Mercury has well-known neurological effects ranging from behavioral abnormalities to vision and hearing loss. Multiple sites of action have been identified. A previous study using neuronal networks derived from mouse auditory cortex found that acute application of mercuric chloride produced an elevation in action potential bursting at concentrations below 20 mg l−1 and inhibition followed by irreversible loss of activity at 27–41 mg l−1 HgCl2 (Gopal, 2003). These results are of the same order of magnitude as seen in the present experiment, but, in the frontal cortex networks the elevation in firing rate at lower concentrations of mercury was not seen. Methylmercury (MeHg), another mercurial compound, has been implicated in the uptake of the excitatory neurotransmitter glutamate by astrocytes (Juárez et al., 2002). Both elemental mercury and MeHg have been found to inhibit voltage-gated calcium currents (Denny and Atchison, 1996; Sirois and Atchison 2000). While blocking voltage-gated calcium currents, mercurial compounds have been found also to alter calcium homeostasis, resulting in an elevation of the intracellular calcium concentration and increases of spontaneous J. Appl. Toxicol. 24, 379–385 (2004)
384
T. J. O’ SHAUGHNESSY ET AL.
transmitter release (Denny and Atchison, 1996). It is unclear, though, which of these effects might be responsible for the alterations seen in the network behavior. Unlike mercury, there are no reports of acute effects of arsenic on any of the synaptic or action potential generation machinery. Most reports focus on longer term chronic exposures. However, arsenite has been reported to induce apoptosis in rat cortical neurons, with activation of the initial components of key pathways within 30 mins of exposure to 1.3 mg l−1 arsenite (Namgung and Zhengui, 2000). The inhibition of spike rate observed in the neuronal network occurred at a concentration of 10 mg l−1 and was completely reversible; in fact, the network was used the following day for additional experiments. These results are difficult to reconcile with the reported apoptotic mechanism and suggest that sodium arsenite may have a novel action more closely related to neurotransmission. The insecticide chlordimeform, a member of the formamidine family of compounds, is a known α-adrenoceptor antagonist and has been shown to affect central nervous system excitability, sensory responsiveness and equilibrium when injected into rats (Moser et al., 1988). In addition, chlordimeform has been shown to inhibit calcium entry into muscle isolated from guinea-pig ileum in a manner similar to the L-type calcium channel blocker nifedipine (Candura et al., 1992). Other formamidines have been shown to inhibit directly the whole-cell voltage-dependent calcium currents in chick dorsal root ganglion (Mironov, 1994). The results reported here show phosdrin reducing the mean spike rate of the neuronal networks. Additional experiments performed in our laboratory after the EILATox-Oregon workshop have produced divergent results. Interestingly, organophosphates such as phosdrin are often associated with inhibition of anticholinesterase (AChE) activity, and a previous study (Keefer et al., 2001) that examined the effect of the AChE inhibitor eserine on mouse frontal cortex networks found that the mean spike rate was affected in a biphasic manner: increasing at low concentrations (10–25 µM) and being inhibited at higher concentrations (50–150 µM). The source of the phosdrin used in the later experiments (Sigma-Aldrich, St. Louis, MO) was different from that used for the workshop (Chem Service, Inc., West Chester, PA). It is possible that these two sources have different potencies, which might lead to the observed variation. Additional testing is needed to determine if this is the case or if other factors, such as hydrolysis of the compound, are responsible for the observations. What is clear, however, is that at the concentration tested at the workshop (40 mg l−1) phosdrin produces significant changes in the firing rate of the neurons. Interestingly, the final compound tested, colchicine, has been shown to inhibit GABAA receptors expressed in oocytes (Bueno and Leidenheimer, 1988). Approximately 60% inhibition of GABA-generated chloride currents was produced by 100 µM colchicine. This is consistent with the small increase in mean spike rate observed in the frontal cortex network exposed to 100 µM colchicine.
The breadth of detection, as demonstrated by the workshop experiments, is one of the key aspects of the neuronal network-based biosensor. By using mammalian neurons, the system should respond in a physiologically relevant manor to compounds with neurotoxic or neuropharmacological effects. For the four compounds found to have a significant effect on the mean spike rate, detection was at a level that should provide a warning prior to the compound becoming a serious acute threat to a human being. Although breadth of sensitivity is perhaps the greatest strength of this biosensor, one major concern that arises from this is that the neuronal networks may respond to benign compounds in addition to those of interest. Although this is a possibility, it is important to realize that from a detection standpoint most compounds that produce large changes in network activity are going to merit at least an increased awareness of their presence. Generic detectors, like the one described here, are useful as sentinel systems to indicate when heightened awareness is needed with more traditional, specific sensor technologies. As part of a suite of sensor technologies, the weakness of one system can be compensated for by the other sensors. Although potentially capable of detecting a wide variety of compounds, the neuronal network-based biosensor lacks the ability to determine the specific toxin present. By monitoring a single parameter — the mean spike rate — the system can only determine that a compound is present and indicate whether it is inhibitory or excitatory. Recently, a new parameter has been developed to describe the level of synchronization among the channels in the network (Selinger et al., 2004), which allows for additional information about a detected toxin regarding its properties as a convulsant/anticonvulsant. As additional parameters are added to the analysis regime, such as examination of interspike intervals to understand burst dynamics, and by taking into account neuron-specific effects it is hoped that additional information about the detected substance can be determined. The inability of the neuronal network-based biosensor to determine the exact toxin detected means that it will not serve as a replacement for the more specific sensors available, however its ability to respond to a wide range of both known and unknown threats in a physiologically significant manner should make it an excellent front end to an environmental monitoring system guiding the operation of the more specific detection strategies.
Acknowledgements This work was supported by the Defense Advanced Research Projects Agency and the Office of Naval Research. The authors gratefully acknowledge the support and encouragement from Professor Guenter W. Gross of the University of North Texas, the technical contributions to this effort from Mr Anthony Curran (University of North Texas) and Dr Nadezhda V. Kulagina (Naval Research Laboratory) for her follow-up efforts on phosdrin. T.J.O. held a National Research Council Research Associateship Award at the Naval Research Laboratory. The opinions and assertions contained herein are the private ones of the authors and are not to be construed as official or reflecting the views of the Department of the Navy or the military at large.
REFERENCES Bueno OF, Leidenheimer NJ. 1998. Colchicine inhibits GABAA receptors independently of microtubule depolymerization. Neurophamacology 37: 383–390. Candura SM, Marraccini P, Costa LG, Manzo L, Rossi A, Copyright © 2004 John Wiley & Sons, Ltd.
Coccini T, Tonini M. 1992. Calcium entry blockade as a mechanism for chlordimeform-induced inhibition of motor activity in the isolated guinea-pig ileum. Pharmacol. Toxicol. 71: 426–433. J. Appl. Toxicol. 24, 379–385 (2004)
NEURONAL NETWORKS AS BIOSENSORS Denny MF, Atchison WD. 1996. Mercurial-induced alterations in neuronal divalent cation homeostasis. Neurotoxicology 17: 47–61. Gopal KV. 2003. Neurotoxic effects of mercury on auditory cortex networks growing on microelectrode arrays: a preliminary analysis. Neurotoxicol. Teratol. 25: 69–76. Gray SA, Kusel JK, Shaffer KM, Shubin YS, Stenger DA, Pancrazio JJ. 2001. Design and demonstration of an automated cell-based biosensor. Biosens. Bioelectron. 16: 535– 542. Gross GW. 1979. Simultaneous single unit recording in vitro with a photoetched laser deinsulated gold multi-microelectrode surface. IEEE Trans. Biomed. Eng. 26: 273–279. Gross GW, Wen W, Lin J. 1985. Transparent indium–tin oxide patterns for extracellular, multisite recording in neuronal cultures. J. Neurosci. Methods 15: 243–252. Juárez BI, Martínez ML, Montante M, Dufour L, García E, Jiménez-Capdeville ME. 2002. Methylmercury increases glutamate extracellular levels in frontal cortex of awake rats. Neurotoxicol. Teratol. 24: 767–771. Keefer EW, Norton SJ, Boyle NAJ, Talesa V, Gross GW. 2001. Acute toxicity screening of novel AChE inhibitors using neuronal networks on microelectrode arrays. Neurotoxicology 22: 3–12.
Copyright © 2004 John Wiley & Sons, Ltd.
385
Mironov SL. 1994. Monovalent amidiniums block calcium channels in chick sensory neurons. J. Membr. Biol. 141: 231–237. Moser VC, McCormick JP, Creason JP, MacPhail RC. 1988. Comparison of chlordimeform and carbaryl using a functional observational battery. Fundam. Appl. Toxicol. 11: 189–206. Namgung U, Zhengui X. 2000. Arsenite-induced apoptosis in cortical neurons is mediated by c-Jun N-terminal protein kinase 3 and p 38 mitogen-activated protein kinase. J. Neurosci. 20: 6442–6451. Pancrazio JJ, Keefer EW, Ma W, Stenger DA, Gross GW. 2001. Neurophysiologic effects of chemical agent hydrolysis products on cortical neurons in vitro. Neurotoxicology 22: 393– 400. Pancrazio JJ, Gray SA, Subin YS, Kulagina N, Cuttino DS, Shaffer KM, Eisemann K, Curran A, Zim B, Gross GW, O’Shaughnessy TJ. 2003. A portable microelectrode array recording system incorporating cultured neuronal networks for neurotoxin detection. Biosens. Bioelectron. 18: 1339–1347. Selinger JV, Pancrazio JJ and Gross GW. 2004. Measuring synchronization in neuronal networks. Biosens. Bioelecton. 19: 675–683. Sirois JE, Atchison WD. 2000. Methylmercury affects multiple subtypes of calcium channels in rat cerebellar granule cells. Toxicol. Appl. Pharmacol. 167: 1–11.
J. Appl. Toxicol. 24, 379–385 (2004)
