Department of Biological Sciences and Center for Network Neuroscience, P.O. Box 5218, University of North. Texas, Denton, TX 76203, USA. Hassan M.E. ...
Biosensors & Bioelectronics 10 (1995) 553-567
The use of neuronal n e t w o r k s on m u l t i e l e c t r o d e arrays as biosensors Guenter W. Gross* & Barry K. Rhoades Department of Biological Sciences and Center for Network Neuroscience, P.O. Box 5218, University of North Texas, Denton, TX 76203, USA
Hassan M.E. Azzazy & Ming-Chi Wu Department of Biochemistry and Molecular Biology, University of North Texas Health Science Center, Fort Worth, TX 76107, USA (Received 1 September 1994; accepted 11 November 1994)
Abstract: Mammalian spinal neuronal networks growing on arrays of photo-
etched electrodes in culture provide a highly stable system for the long-term monitoring of multichannel, spontaneous or evoked electrophysiological activity. In the absence of the homeostatic control mechanisms of the central nervous system, these networks show remarkable sensitivities to minute chemical changes and mimic some of the properties of sensory tissue. These sensitivities could be enhanced by receptor up-regulation and altered by the expression of unique receptors. The fault-tolerant spontaneous network activity is used as a dynamic platform on which large changes in activity signify detection of chemical substances. We present strategies for the expression of novel supersensitivities to foreign molecules via genetic engineering that involves the grafting of ligand binding cDNA onto truncated native receptor DNA and the subsequent expression of such chimeric receptors. Keywords: sensomimes, nerve cell biosensors, chimeric receptors, transfection, extracellular recording, liposomes.
INTRODUCTION The rapid and reliable measurement of harmful substances in the environment, of medically pertinent metabolites and pharmacological substances, and of compounds important to the food
* Author to whom correspondence should be addressed. Present address: Clinical Chemistry Laboratory, University of Maryland Medical Center, Baltimore, MD 21201, USA. 0956-5663/95/$09.50 (~) 1995 Elsevier Science Ltd
industry, necessitates the development of a great variety of sophisticated sensor systems (c.f. D'Amico, 1992). However, the analytical demands for a chemical sensor--selectivity, sensitivity, accuracy, stability, freedom from maintenance, reliability, and rapidity--are not easily achieved for every compound of interest (Cammann, 1992). Biosensors have evolved to expand the sensing domain and accelerate crucial reactions. Within this group, whole cell-based sensors are receiving increasing attention because they have the potential to perform unique sensory 553
G.W. Gross et al.
functions (Mazzei et al., 1992; Varfolomeyev et al., 1994) by relying on the response amplification and reliability clearly provided by populations of living sensor units. In this area, an additional interesting step could be taken by using more complex cell systems such as neurons in cell culture. Neuronal networks are intrinsically nonlinear dynamic systems which create complex electrophysiological spatio-temporal patterns through rapid communication between neurons in the network. In animals, such tissues are capable of remarkable discrimination and sensitivity. However, animals cannot be used for a large range of applications such as continual online monitoring, quantitative analysis, or performing in hostile environments. Therefore, the reconstruction of sensory systems in vitro is an attractive concept that has, however, received only marginal research attention. Two approaches can be taken: (1) the exact reconstruction of sensory systems; (2) the artificial induction of sensory functions into non-sensory neural tissue, which is robust and reliable under culture conditions. This paper discusses the latter approach and presents strategies for the transformation of non-sensory neural tissue in vitro into biosensors. The approach uses the spontaneous network activity, generated by neuronal cultures, as the basic dynamic platform on which large changes in activity represent detection events. Knowledge of complex (and still unknown) sensory coding is, therefore, unnecessary. Neurons are also homeostatic systems that constantly regenerate molecules and may, therefore, be considered renewable "receptor platforms". A network of such neurons should provide reliability and greater reproducibility. Although not yet understood on a quantitative level, neuronal networks are clearly fault tolerant and maintain activity patterns despite the loss of network components and alteration of the circuitry. In this manner, a spontaneously active neuronal ensemble might be engineered to sense a variety of chemical compounds, with the potential to provide sensitivity, accuracy, and long-term on line performance.
G E N E R A L CHARACTERISTICS OF SPONTANEOUS N E T W O R K ACTIVITY Nerve cells can be grown in cell culture on photoetched electrode arrays that allow the 554
Biosensors & Bioelectronics
simultaneous monitoring of spike activity from many neurons (Gross et al., 1977; Gross, 1979; Pine, 1980; Gross & Kowalski, 1991; Gross, 1994). The introduction of transparent indiumtin oxide as a viable material for recording (Gross et al., 1985) as well as for stimulating (Gross et al., 1993b) neural activity provides a miniaturized electrode system which does not interfere with optical observation of the network, even at high electrode densities. This technical step allows the correlation of network development with continual multichannel monitoring of network activity. It also allows studies of fault tolerance with laser cell surgery (Gross et al., 1983), whereby the targeted removal of specific cells or the random transection of neurites can be correlated with changes in the spontaneous or evoked activity of the network. Figure 1 shows a phase contrast micrograph of a low density neuronal network growing on the 1 mm 2 recording matrix of a multimicroelectrode plate (MMEP), produced in our laboratory. Methods for MMEP fabrication, cell culture, and recording are summarized in Gross & Kowalski (1991). Most networks show initial spiking activity at approximately one week after cell seeding. With few exceptions they generate complex, often coordinated burst patterns at two to three weeks. The spontaneous activity obtained from cultured spinal networks ranges from apparently stochastic spiking to organized bursting and even stable, long-term (days) synchronized oscillatory activity. Since the initial days of neuronal cell culture, survival times in vitro have increased enormously (Conn, 1990). Spinal monolayers, for example, survive for more than nine months in culture and routinely show vigorous electrical activity at all ages above two weeks (Gross, 1994). The growth of neurons and glia on a microelectrode surface creates a remarkably stable recording environment, primarily because electrodes do not have to invade the tissue and do not vibrate or slip in relation to the neural components. Although definitive, long-term studies of signal-to-noise ratios (SNRs) at individual electrodes have not yet been completed, the preliminary data obtained is encouraging. Spike profiles formed by superimposed action potentials recorded in a closed flow chamber from the same electrode over a period of 7 days have revealed minimal changes (Gross, 1994). Although the number of active electrodes and the SNRs achieved still vary substantially from culture to
Biosensors & Bioelectronics
Use o f neuronal networks on multielectrode arrays as biosensors
Fig. I. Neuronal network on the recording array o f a 64 microelectrode culture plate. Transparent indium-tin oxide (ITO) conductors in the centre area are 8 ~ in diameter and terminate in four rows o f 16 electrodes each. Spacing between columns is 40 txm," between rows 200 ~m. Craters in the 2 txm thick polysiloxane insulation layer are made with single laser shots followed by gold plating o f the exposed ITO. The 1 m m thick glass carrier plate measures 5 x 5 cm and has 32 contact strips on either side for connection to amplifiers.
culture, and are influenced by the neuronal cell density, glial density, and size of the adhesion island over the recording matrix, there is no obvious trend of a decrease in these variables with days in vitro (Fig. 2). This suggests at least a statistical stability of cell-electrode coupling with time. Figure 3 depicts typical multichannel data after digitizing and printout on a chart recorder. The left panel (a) shows a 1.3 s burst episode with burst initiation at the beginning of the record and clear phase delays between channels. Individual action potentials (spikes) are visible during low firing frequency episodes. Clearly, on the spike level, detailed analyses are difficult. However, the complexities of multichannel spike corre-
lations can be avoided by feature extraction methods, such as simple integration. The right panel (b) shows such integrated data over a 2.5 min period, and emphasizes that temporally coordinated bursting across channels is often seen in the native, pharmacologically unmodified state. The degree of coordination usually increases with increasing spike frequency within a burst, a phenomenon termed "intensity recruitment" (Gross & Kowalski, 1991). For applications such as biosensors, spontaneously bursting networks are recommended because pattern changes can be more readily identified.
555
G.W. Gross et al.
Biosensors & Bioelectronics
a tonic excitatory bias that potentiates synaptic excitation in the networks, increasing burst amplitude and burst rate. At high concentrations (>20/xM), NMDA may lead to burst fusion and tonic high frequency spiking, followed by excitotoxicity. (3) Norepinephrine and, to some degree, acetylcholine are n e u r o m o d u l a t o r s with primarily excitatory effects that enhance the existing patterns.
100-
_m __
•
•
i;;.i;.-
70
5o
7",-..'..:.
.
'°o t ........................................... ....... 7 7 7 7 7 7 - 7 7 7 i
........................... . . . . i
Glycine and G A B A (y-aminobutyric acid) act as inhibitory neurotransmitters. If applied to the 30 culture medium, they provide an exogenous tonic ~...........................................................................I........^,SNR .............I-" .......................... inhibition to most network neurons and globally 261 . ............ i" M = ~ " I~ 22 . . . . reduce spike activity and burst rates. Sodium ~20 :-barbital, carbamazipine, and valproate each 18 potentiates intrinsic inhibition at G A B A recep~ 14 tors. The observation that each of these com12 pounds can shut down network activity implies ~ 1o that intrinsic inhibition is strong in these cultures. ~ 6 " -~ I :-*--~ Magnesium, 2APV, MK801, ketamine, and etha4 "." ". nol all decrease synaptic excitation by blocking 0 0 20 40 60 80 100120140160180200220240260280300 or antagonizing NMDA receptors or the associDAYS IN VITRO ated ion channel. Magnesium, cobalt, and high Fig. 2. (A) Percent electrodes with measurable electroextracellular calcium levels decrease excitability physiological activity as a function o f culture age (days and reduce spike activity and burst amplitude, in vitro). (B) Mean and maximum signal-to-noise ratios duration and frequency. Tetrodotoxin blocks (SNRs) as a function o f culture age. SNRs were derived sodium action potentials and reversibly silences from slow sweep data which contain neural noise. Low all networks at 0.5 p,M. cutoff was 1.5:1. Total system noise varied between 30 The neurotransmitter G A B A stops sponand 45 tzV. taneous bursting between 15 and 30/xM in all cultures (Fig. 4). This effect is highly reproducible NETWORK SENSITIVITY TO for repeated trials in the same culture as well P H A R M A C O L O G I C A L AGENTS as between different cultures (Jordan, 1992). Despite the decreasing burst rate, burst synchronExcitation, inhibition, and disinhibition ization among channels, integrated burst amplitude and burst duration do not change; this Table 1 summarizes pharmacological manipuindicates that this inhibitory transmitter affects lations performed on spinal monolayer networks burst initiation but not burst synchronization, in culture via bath applications of neuroactive maintenance and termination. Glycine responses substances. Network excitatory influences fall are more complex and bursting stops at approxiinto three groups. mately 60 to 100 ~M with some random spiking (1) The excitatory amino acids glutamate and remaining under these conditions. G A B A is aspartate, which provide an exogenous the first transmitter we have used to explore tonic excitation to most network neurons interculture variability. This variability is remarkand increase spike rates. At high concenably low. The precise shut-off point is more trations (10 tzM), they overwhelm the dependent on the pre-application, native activity effects of synaptically released transmitter, than on culture cell density and age (Jordan, disrupt the temporal excitatory network 1992): the greater the initial activity, the more interactions, and generate irregular high GABA is usually required to terminate activity. frequency spiking. It should be noted that the dose response (2) N-methyl-D-aspartate (NMDA) provides curves of Fig. 4 may be considered calibration 0
20
40
60
80 100120140160180200220240260280300
DAYS IN VITRO
•
oo
~
_
•
o-
~
.
,
556
-
l _
o
"
:
~
.
*
~ . ~ _
-
,
.
,
-
.
-
,
-
,
.
,
.
,
.
,
.
,
-
.
.
,
-
,
-
Biosensors & Bioelectronics a
Use of neuronal networks on multielectrode arrays as biosensors
..i, liillllt L.'h-- ~;:,ail,llli _|tJ • __~..,~i.~.,....li~*
....
I
.t..
l...
Ira
....
tt,.~l
i
' r u ' r.J' " 5~ ' r - ' 7 " iildmnlit
~
/
l
l
i
l
d~i~...t.kl
-
" ? " " M~"1""r~"r~r":~ U
l
l
U d~lkl/I}
"hi ~
~
~
Lm
I.Imu
! m~mvmnliinuluw,tllt~ I •I, lllll.~iR],~dlml~lln ~uj liillihl lu.ll,.ltl,] .:;_:i_ mI..tJ_!. I
|
t l , . ,L mti~., ~ i . ~ . L L
!
, ,ilili,
: i
il'lnWl!lillr'l;t-l'rillqllT'l
•
'...... L............
I
. i . , .. L . L • .... i , l ; .
i
i
~ 1 ~ , ,IlL
li
" ; ' l "~" l'- r ! - i ' ' T " ' l ~ ; 7
1 .....
l,luiimitim
.
I"
ii,
,,,.,,.~,_.__
iI
--
....
~
=
' i
6
i
l
i~
.!
Fig. 3. (A ) Digitized multichannel action potential data showing partial burst coordination with phase delays among channels (bar: 100 ms). (B) Slower speed printout of integrated data showing coarse grain burst synchronization and typical temporal patterns produced spontaneously by spinal networks (bar: 20 s).
curves of a sensor system. In a stable culture, one could measure an unknown quantity of GABA in the low micromolar range if all other parameters are kept constant. Clearly, network response repeatability and variability must be explored extensively before this approach is considered seriously. However, as our knowledge of network dynamics and the culture environment increases, there is little doubt that methods will be adjusted to control variability and improve repeatability. Regardless of the initial native activity, organized coordinated activity can be obtained in almost all cases via disinhibition with strychnine or bicuculline which serve to block the inhibitory glycine and GABAA receptors, respectively. This response generally results in stable oscillatory behaviour. From observations of over 700 cultures, it is quite clear that the tendency for a temporal coordination of spike clusters (or bursts) reported by different electrodes is ubiquitous and represents the most general activity feature of these monolayer networks. It is also evident that such spike clusters are usually not synchronized but show variable phase delays on different channels. On a coarse-grained level, such as a
stripchart record, the term burst "synchronization" is used with the understanding that the phase differences are ignored. A network transition from the native state to an oscillatory state, induced by 10/xM strychnine, is shown in Fig. 5. Such massive transitions do not require complex spike correlation methods for analysis. It is clear that a major event has occurred. Initially, bioengineering efforts should be directed at utilizing large transitions as the sensory response. More subtle pattern changes, which most likely carry additional information, can be left to future efforts. It is interesting that excitation and disinhibition usually do not lead to the same network activity modes, although both increase total spike production. There is a clear asymmetry in the way networks respond: whereas excitation increases spiking within highly variable patterns, disinhibition almost always regularizes burst durations and periods and generates oscillatory states. Other compounds, such as 4-aminopyridine and cesium, increase burst rate and regularity, in a manner qualitatively matched by elevating extracellular potassium (Rhoades & Gross, 1994). 557
G.W. Gross et al.
Biosensors & Bioelectronics
TABLE 1 Network responses to pharmacological agents* Compound
Minimum Detectable Concentration
Normal Experimental Range Used
Toxicity Level**
General Network Responses
EXCITORS
10 nM
10--40/xM 10-50 IzM 10-100/zM 30 txM 1-10/zM
5/zM
10-40/zM
picrotoxin 4-amino pyridine 8-Br-cAMP
100/zM
10-100/xM 20/zM 1-5 mM
penicillin NMDA
1/zM
100-1000 U/ml 1-50/zM
acetylcholine glutamate aspartate noradrenaline strychnine
bicuculline
1/xM 5/xM
50/xM
30 IzM
50 tzM
50/~M
excitatory, disrupts existing pattern excitatory, destroys existing pattern excitatory, destroys existing pattern excitatory, accelerates existing pattern periodic, coherent burst patterns, burst stretching and increased integrated burst shape regularity periodic, coherent burst patterns, high frequency bursting increased non-periodic activity acceleration of existing burst pattern increases burst frequency, decreases spikes/burst increased non-period activity concentration-dependent excitatory effects ranging from pattern acceleration, to pattern disruption, burst fusion, and intense spiking
INHIBITORS glycine GABA
1 ~M
20-100/zM 5-50/zM
diltiazem
40/zM
100/~M
ketamine 2-APV MK-801 phenytoin
5/zM
5-300~M 30-75~M 5-15~M 40-100 ~m
carbamazepine valproic acid magnesium ethanol
5 mM ?
20--60 tzM 300-700/xM 10-12 mM 0.5%
barbitol
20/xM
50-100 ~M
tetrodotoxin
200 nM
300-500 nM
ouabain veratridine barium chloride
10/zM 2/zM 1 mM
20/~M
500/xM (5 mM) 1 mM
1%
stops all bursting at 50 tzM stops bursting at 15-30 tzM (in all cultures) shortens burst duration, slows burst frequency inhibitory inhibitory, no cell death at 5 mM inhibitory inhibitory, slows activity and stops bursts at 100/~M inhibitory, stops bursts at 60/zM inhibitory, slows existing pattern total inhibition at 10-12 mM slows activity in extreme to random spiking slows burst rate and shortens bursts
NEUROTOXINS stops all spike activity rapidly at 500 nM stops all activity gradually (10 min) stops activity at 6 tzM stops activity at 5 mM
*Murine spinal tissue dissociated at day 14 gestation and investigated at culture ages ranging from 3 weeks to 9 months. **Concentration at which cell death begins under present culture conditions.
558
Biosensors & Bioelectronics 60
I
I
Use of neuronal networks on multielectrode arrays as biosensors I
50]_
,c
I
i
i
5
10
i
,
15
20
I
30 20
40 ¢~.@
.......
~'.
"~-"~,
='~
30
•
20
I
".,..
0
~
",,',~
~
