The dynamic clamp: artificial conductances in biological neurons

intention is to emphasize that scientists must be aware of their biases, because these biases can have a dramatic effect on the outcome of their research. If technical progress is to be made in the understanding of the possible genetic bases of mental illnesses, then it will be essential that the biases be explicitly acknowledged and that extreme efforts be made to overcome their effects on scientific research. Selected references 1 2 3 4 5 Sherrington,R. etal. (1988) Nature 336, 164-167 Baron, M. etaL (1987) Nature 326, 289-292 Egeland,J. A. etaL (1987) Nature 325,783-787 Pauls,D. L. (1993) Nature Genet. 3, 4-5 Barnes,D. M. (1989) Science 243,313-314 6 Baron, M. et al. (1993) Nature Genet. 3, 49-55 7 Baron,M., Endicott,J. and Ott, J. (1990) Br. J. Psychiatry 157, 645-655 8 Diagnostic and Statistical Manual of Mental Disorders, 3rd ed., Revised (1987), American Psychiatric Association, Washington DC, USA 9 Aldous, P. (1993) Science 259, 591-592 10 Billings, P. R., Beckwith, J. and Alper, J. S. (1992) Soc. Sci. Med. 35, 227-238 11 Kinney,D. K. (1990)in The Principles and Practice of Medical Genetics (2nd edn) (Emery, A. E. and Rimoin, D. L., eds), pp. 457-472, Churchill Livingstone 12 Alper, J. S. and Natowicz, M. R. (1992) Br. Med. J. 305, 666 13 Jensen,A. R. (1969) Harvard Educational Review 39, 1-123 14 Bailey, J. M. and Pillard, R. C. (1991) Arch. Gen. Psychiatry 48, 1089-1096 15 Goleman,D. (1992) New York Times, 15 September,C1, C7 Acknowledgements We wouldlike to thank membersof the GeneticScreening Study Group,Boston, MA for their valuable suggestions.This work wassupported in part by a grant from the Human GenomeInitiative through the US Departmentof Energy. techniques Thedynamicdamp: artificial condudancesin biological ReuroR$ A n d r e w A. Sharp, Michael B. O'Neil, L. F. A b b o t t and Eve Ma rd e r The dynamic clamp is a novel method that usescomputer simulation to introduce conductances into biological neurons. This method can be used to study the role of various conductances in shaping the activity of single neurons, or neurons within networks. The dynamic clamp can also be used to form circuits from previously unconnected neurons. This approach makes computer simulation an interactive experimental tool, and will be useful in many applications where the role of synaptic strengths and intrinsic properties in neuronal and network dynamics is of interest. A basic goal of neuroscience is to understand how membrane and synaptic conductances combine and interact to produce the behavior of neurons and neural circuits. The conventional method for altering the activity of single neurons or perturbing their activity in networks is to inject constant current using the current clamp. This allows the investigator to either depolarize or hyperpolarize a neuron, but does not correctly replicate the conductance changes produced by synaptic inputs or modified by neuromodulators. The primary tools used to determine the characteristics of individual neuronal conductances are the voltage clamp and the patch clamp. While essential for understanding the voltage- and time-dependence of a conductance, these methods are less useful for studying the interplay of conductances that determine how neurons act individually or in circuits. Most voltageand patch-clamp experiments halt the normal voltage excursions of the clamped neuron and they often employ pharmacological agents to isolate a single conductance. For these reasons, conventional voltage clamp methods do not allow the evaluation of the role of conductances during the normal dynamic evolution of the membrane potential. The classical solution to this problem is to simulate the electrical activity of a neuron using mathematical descriptions of its measured conductances 1. The limitation of this method is that it requires a detailed description of many, if not all, of the conductances in a neuron, and these data may be difficult or impossible to obtain. TINS, VoL 16, No. 10, 1993 The dynamic clamp is a new approach that allows an investigator to introduce artificial voltage- and time-dependent conductances into biological neurons (Refs 2 - 4 and Hutcheon, B. and Pull, E., unpublished observations). In a sense, the dynamic clamp uses biological neurons as simulators, allowing the investigator to evaluate the role of individual conductances in shaping the electrical activity of single neurons, as well as determining the consequences of synaptic strengths in networks. The dynamic clamp combines the control and flexibility of computer simulation with the accuracy and realism of electrophysiological recording, using computer modeling as an experimental tool. AndrewA. Sharp, MichaelB. O'Neil, L. F.Abbott and Eve Marderare at the Deptof Biologyand Centerfor Complex Systems,Brandeis University, Waltham, MA 02254, USA. The dynamic clamp produceschanges in conductance The basic set-up for the dynamic clamp is similar to a conventional voltage- or current-clamp rig. However, in the case of the dynamic clamp, the injected current is controlled by a computer program that duplicates the current that would flow through a real membrane or synaptic conductance (Box 1). Any conductance that can be modeled mathematically can be introduced into the neuron being studied. The capabilities and uses of the dynamic clamp are illustrated here, using examples from the stomatogastric ganglion (STG) of the crab Cancer borealis. The STG offers a small group of well-defined neurons, whose connections and circuit characteristics are well understood, including extrinsic inputs and neurotransmitters. Unlike current-clamp injection, the dynamic clamp duplicates both the voltage and the conductance changes caused, for example, by a neurotransmitter. In Fig. 1, the dynamic clamp mimics the response of an STG neuron to rapid bath application of the neurotransmitter ~,-aminobutyric acid (GABA). Hyperpolarizing constant current pulses were used to monitor the input impedance of the neuron. GABA increased the conductance of the neuron (seen as a decrease in the amplitude of the changes © 1993.ElsevieSci r encePublishersLtd, (UK) 389 .... : : ~ Box 1. Componentsof the dynamic clamp The dynamic clamp uses a standard electrophysiological set-up with computer interface to measure membrane potentials and control current injected through an intracellular electrode. The membrane potential V is recorded with an intracellular electrode and transmitted to the computer through an analog-to-digital converter. On the basis of V and the differential equations describing the desired conductance, the current I flowing through the simulated conductance is computed. This is converted back into an analog voltage V~, which controls the current injected into the neuron by the recording amplifier (see Figure). A mathematical model of the conductari~:e being simulated is programmed into the computer that is controlling the current injected by the dynamic clamp. The model current may be explicitly developed from voltage-clamp data describing a current from the preparation being studied. Typically, the membrane current ! is given by the classic form of the Hodgkin-Huxley modela: I = gmPhq(V-6) t l = gm~,hq(V -E~)II where s~ is given as a function of the presynaptic potential Vpr e by: s~(Vpre) = tanh [ (VpreAvth) ] where g is the conductance, p and q are integers, E, is the reversal potential of the current, Vis the membrane potential, and m and h are activation and inactivation variables described by the differential equations: • m(V) d._m= moo(V) - m dt and ~h(V) d h = h=(V) - h dt where ~m, ~h, m= and h= are measured functions of V. During the operation of the dynamic clamp these equations are integrated numerically in real time to generate the desired current. To construct an artificial chemical synapse, the dynamic clamp must be programmed with a mathematical model of the synapse being simulated (a previous method b used presynaptic potential to trigger current injection). For the examples described in this article, the synaptic current is given by: I = gs( %ost- Er) where Er is the synaptic reversal potential and Vpost is the. membrane potential of the postsynaptic neuron. The synaptic activation variable s varies between 0 and 1 and is determined by: [l-s= (Vpre) ] T's ds = s= (Vp,e) -s dt if Vwe > Vth; otherwise s= = 0. ~s, Vth and A are constants. An electrical synapse of conductance g between neurons with potentials V1 and V2 can be simulated by injecting a current, I = g(V2- V1), into neuron 1 and a current of equal magnitude but opposite sign into neuron 2 (see Refs c and d for another approach to building electrical synapses with an analog circuit). The configuration shown in the Figure allows the addition of artificial membrane conductances into either or both neurons, and construct on of artificial synapses between the two neurons. The dynamic clamp uses generally available hardware. In the system used for the examples cited in this article, analog-to-digital and digital-to-analog conversions were handled by a 'Scientific Solutions Lab Master DMA Board', and current calculations were performed by an Intel 80486DX-based PC running at 50MHz. An 'Axoclamp-2A' (Axon Instruments) run in discontinuous current clamp (DCC in Figure) mode (sampling frequency of 5 kHz) was used to record membrane potential and control current injection through an intracellular electrode. References a Hodgkin,A. L. and Huxley,A. F. (1952)J. Physiol. 117, 500-544 b Yarom, Y. (1991) Neuroscience 44, 263-275 c Joyner, R. W., Sugiwara, H. and Tau, R. C. (1991) Biophys. J. 60, 1038-1045 d Sharp,A. A., Abbott, L. F. and Marder, E. (1992)J. Neurophysiol. 67, 1691-1694 neuron. In addition, many neuromodulatory substances either activate or influence voltage- and time-dependent conductances. Because these conductances all have different voltage- and timedependent properties, it is often impossible to predict, without simulation, the effect on the activity of the neuron of changing one or more of these conductances. The dynamic clamp provides this tool directly. Figure 2 illustrates the use of the dynamic clamp to determine how a biophysically characterized, modulator-activated conductance influences Using the dynamic clamp to understand single the electrical activity of a single neuron. The peptide neuron behavior A large number of voltage- and time-dependent proctolin elicits a non-specific cation conductance 5 conductances may contribute to the activity of a in certain STG neurons that is maximal at membrane in membrane potential produced by the current pulses). When programmed to mimic the effect of GABA, the dynamic clamp also decreased the change in membrane potential produced by the current pulses, demonstrating that the dynamic clamp effectively changes the conductance of the neuron. In fact, the dynamic clamp behaves as if the channels described by the programmed equations were located at the tip of the microelectrode. 390 TINS, Vot. 16, No. 10, 1993 ,~ A ~ ~ ~i~ • • !~ , i~~¸ ~'~ B -69mY- 1' 30s /10mV 0.1mM GABA 4s t 30s GABA simulation Fig. 1. Artificial y-aminobutyric acid (GABA) conductance. (A) Intracellular recordings from a crab stomatogastric ganglion (5TG) neuron in primary cell culture. Hyperpolarizing current pulses (-0.05 nA) were applied every 3 s to monitor the input impedance of the neuron. At the arrow, the superfusion medium was changed from 30s to one containing O. 1 mM GABA. Recordings at different baseline potentials show that the reversal potential for this response was around - 8 0 mV. (B) Recordings from the same neuron shown in (A). At the arrow, the dynamic clamp was used to simulate the response of the neuron to GABA. The GABA response was modeled as having an 8 nS conductance with a reversal potential of - 7 5 m V and an exponential rise (t = 5s) and fall (t = 15s). (Taken from Ref. 2.) potentials close to the resting potential, but that clamp in conjunction with the biological AB neuron, decreases at hyperpolarized potentials because of a with all of its conductances intact, we were able to voltage-dependent block by extracellular Ca2+. The study the effects of the one conductance whose biophysical properties of this current are shown in properties we had obtained by conventional voltageclamp methods. Box 2. Bath application of real proctolin to a pacemaker neuron of the STG, the anterior burster (AB) neuron, Using the dynamic clamp to understand circuit increases both the frequency and the amplitude of dynamics its bursts6 (Fig. 2A). What features of the proctolinNeural circuit activity depends on the intrinsic activated current and the conductances of the AB properties of all the constituent neurons, as well neuron are important in these actions? Although as the synaptic connections among them. Many proctolin adds a depolarizing inward current to the neuromodulatory substances influence circuit neuron, its actions are not adequately mimicked by merely de- A Control 10-6 M proctolin polarizing the neuron (right-hand panel in Fig. 2B). Note that depolarization with conventional current clamp duplicates the in0.5s crease in burst frequency, but fails to replicate the increase in burst amplitude produced by proctolin and the artificial proctolin current. B Artificial proctolin Control Constant current This is because the voltage dependence of the proctolin current interacts with the other currents present in the AB neuron, so that the neuron undergoes oscillatory swings in membrane potential that alternately activate and inactivate the proctolin current. For this reason, the proctolin-activated current in5mV creases both the amplitude and ---J 0.5nA the frequency of the AB neuron ls burst. Because we do not have detailed descriptions of all the Fig. 2. Artificial proctolin conductance. (A) Intracellular recording from a lobster AB neuron (a voltage-dependent conductances pacemaker neuron of the stomatogastric ganglion) in control saline and in 10-6 M proctolin. (Taken in the AB neuron, these results from Ref. 6.) (B) Intracellular recordings from a crab AB neuron ( - 4 5 m V at dotted line). The could not be obtained using con- middle panel shows the effect of 2On5 of simulated proctolin current added to the cell. The final ventional modeling techniques. panel shows depolarization with constant current. The bottom traces are simultaneous recordings However, by using the dynamic of the injected current. TINS, Vol. 16, No. 10, 1993 391 Box 2. Testing the implementation of the dynamic clamp After the dynamic clamp has been programmed to mimic a given conductance it is important to test the accuracy of the program with a voltage-clamp experiment. A useful way to do this is to replace the neuron with a resistance-capacitance (RC) circuit. This approach retains the RC components of the cell membrane, but eliminates all the voltage- and timedependent conductances other than those created by the dynamic-clamp. The Figure illustrates this process for the proctolin current and for the hyperpolarization-activated inward current /H. The dynamic-clamp programs for these two conductances are based on mathematical descriptions a'b previously derived from voltage-clamp data obtained from the stomatogastric ganglion (STG)c,~. For the proctolin conductance, current traces (after leak subtraction) are shown after steps from a holding potential of - 1 0 0 m V to the eight test potentials listed at the right-hand side of the traces. Simultaneous recordings of the activation variable are shown below this. The steady-state I-V relationship measured from these recordings (plotted on the right) are in good agreement with those induced by real proctolin in the STG. The steady-state activation of the proctolin current from this experiment (data points in the proctolin current activation plot) fall properly on the theoretical curve. For /H, current traces produced by voltage steps from a - 4 0 m V holding potential are also shown. These current traces accurately represent the theoretical time- and voltagedependencies of activation plotted on the right. dynamics 7. To understand how this occurs ~t s necessary to study the effects of changes in synaptic strength, and of modifications of the properties of individual neurons. The use of the dynamic clamp to understand the role of a particular membrane current in network dynamics is shown in Fig. 3+ The current studied here is /H, a hyperpolarizationactivated inward current that is present in stomatogastric ganglion neurons 8'9. The dynamic-clamp description of this current is illustrated in Box 2. A Control LP ,, Ivn ',mr ............. , - r, D ,iLPY...E I I AB A I LP I I Proctolin current in voltage clamp 1-- Steady-state I-V curve of proctolin current I(nA) 35mV ,/ 5 - 10 B ' ° ~si . 20 __ I I 100 * ,1 ! 2* 20 20 60 *+ * . Ivn . . 7o . " llllH~ WlVr d . AB -- V(mV) .? 25 Artificial IH in LP I , IHllll__ . . . . . . . . Lu, 11111t!. . . . . . . . . . .d~ II~IIIF= - ' ' - ~w nnnr . . . . . . . . . r m ' I . 55 1.5nA 40 Instantaneous activation of proctolin current ] 1+0 LP Proctolin current activation 10 [ ~-~o~ ] ._ ] I / I I Artificial /H in AB C 0+5 Ivn 0.0 -- 120 80 40 M e m b r a n e potential (mV) 30ms AB ,, IILL.. ,, II "-.-m j__ IILIJ.,+~ till "''-+m I I. ., ~1'~+m a ..... Ill ..... . .~ I. . . . . Ir:--~ I I /H in voltage clamp 0mv / H Activation curve ( and relaxation rate ( . . . . 0"6hA 1L~s ~ -90 -1O0 ~11;0 } g °+b ___j 10mY \ ! o+t,.._L_ -140 -100 60 Membrane potential (mV) References a Golowasch, J., Buchholtz, F., Epstein, I. R. and Marder, E. (1992) J. Neurophysiol. 67, 341-349 b Buchholtz, F., Golowasch, J., Epstein, I. R. and Marder, E. (t992)J. Neurophysiol. 67, 332-340 c Go|owasch, J. and Marder, E. (1992) J. Neurosci. 12, 810-817 d Golowasch, J. and Marder, E. (1992) J. Neurophysiol. 67, 318--331 392 LP ~o~--~ -70 80 -- ) 0.5s Fig. 3. Artificial IH. (A) Control recordings from an intact stomatogastdc nervous system. The top trace is an extracellular recordin& from the lateral ventricular nerve (Ivn), which shows the periodic activity of the LP, PY and PD neurons. The next two traces are intraceltular recordin&s made from the AB and LP neurons. (B) Recordings from the same neuron as in (A) with 50n5 of artificial IH added to the LP neuron. IH was modeled as shown in Box 2. Notice that the LP neuron fires earlier and for a Ion&er period of time than in the control (C) Recordings from the same neuron as in (A) with 50n5 of IH added to the AB neuron. Notice that the frequency of the network has increased. The horizontal dashed lines indicate a membrane potential of -50mY. TIN& Vol. 16, No. 10, 1993 I'IIllIIII IIIII IIIll Illlll I techniques The pyloric rhythm of the A stomatogastric ganglion is shown in Fig. 3. The top extracellular nerve recording in Fig. 3A shows -60 mV -60 mV the rhythmic activity of the lateral pyloric (LP), pyloric (PY) and pyloric dilator (PD) motoneuron~. -49 mV -48 mV The other two traces in Fig. 3A are intracellular recordings from -61 mV -65 mV the AB neuron (which fires with and is electrically coupled to the _110 mV -85 m V - - ~ -80 mV PD neurons) and from the LP 0.2 s neuron. The LP and PY neurons -100 mV~ -100 mV are inhibited by the AB and PD neurons. What determines when the LP and PY neurons fire? B Previous work had suggested that the properties of both I A (the fast transient outward K÷current) and the time-courses and strengths of the inhibitory synapses ~°-~2 played a role. Figure 3 shows that l H is also important. Increasing IH in the LP neuron caused it to fire earlier and longer in the pyloric rhythm (Fig. 3B). Note that inc r e a s i n g I H in the AB neuron increases the frequency of the pyloric rhythm (Fig. 3C). IH has different effects on the AB and LP neurons because they have different intrinsic voltage- and time10 mV vth"~/ "k~,./ dependent conductances. Sys0.2 s tematic comparisons of the role of IH, IA and the strength and timecourses of the synapses in deter- Fig. 4. Artificial chemical synapses. (A) Construction of a single inhibitory synapse between two mining when the LP and other stomatogastric ganglion (STG) neurons. Depolarization of the presynaptic neuron generates a train neurons fire are now possible of action potentials, which in turn generate IPSPs in the postsynaptic neuron. The postsynaptic response is shown at several potentials. The synaptic reversal potential is programmed at - 6 5 mV using the dynamic clamp. Hutcheon and Puil (unpub- for the left panel and - 8 0 m Y on the right. (B) Recordings from the PD and AM neurons in the intact STG. There are no synaptic interactions between these neurons under control conditions lished observations) have used an (top). The bottom traces show the result of coupling these neurons with reciprocal inhibitory independently developed system synapses created using the dynamic damp (Vth for PD = - 4 5 mV, Vth for AM = - 3 0 m Y ; for both to study the role of I H in rat neurons g = lOOnS, Er = - B O m V , and A = -40mY). (Taken from Ref. 2.) neocortical neurons in slice preparations. Preliminary dynamic clamp experiments with the leech heartbeat prep- each PD neuron burst strongly inhibited the AM aration (another well-defined ganglionic circuit) neuron, each AM neuron action potential elicited an (Calabrese, R. H. and Sharp, A. A., unpublished IPSP in the PD neuron and, interestingly, the period observations) indicate that the strength and voltage- of the PD burst was dramatically altered. The modification of synaptic strengths is one of dependence of I H a r e crucial in generating the rhythm of the leech heartbeat oscillator, as well as the primary ways that neural circuits are modified, either by activity or by neuromodulators. By building contributing to its period. an artificial synapse in parallel with a real synapse, The dynamic clamp can be used to construct circuits the synaptic strength can be increased or decreased The dynamic clamp can also be used to create in a controlled manner and the effect on network artificial synapses between neurons. In order to activity can be observed. Artificial synapses can also be used to form novel construct an artificial chemical synapse, the dynamic clamp is programmed to modify the conductance of circuits from otherwise unconnected neurons. the postsynaptic neuron depending on the mem- Neurons grown in culture can develop natural brane potential of the presynaptic neuron (Box 1). synapses13,14, and these circuits have been Artificial synapses between STG neurons are shown studied ls-~7. However, in these experiments the in Fig. 4. The PD and anterior median (AM) neurons investigator is 'held hostage' by the serendipitous in the crab STG are not normally connected. When formation of synaptic contacts among the cultured an artificial synapse was constructed between them, neurons. By using the dynamic clamp to form Q ® TINS, VoL 16, No. 10, 1993 393 artificial synapses, circuits with precisely defined synaptic connectivity can be built and studied. In addition to constructing artificial synapses between neurons, a computer model of an entire neuron can be coupled through artificial synapses to a biological network (see Ref. 4). This allows interesting hybrid computer-biological networks to be built and studied. Accuracy and limitations of the dynamic clamp Acknowledsements We thank J. Rinze/ for early discussions of this idea, and R. Calabreseand M. Nusbaum for helping us troubleshoot the system. S. RenaudLeMasson and G. LeMasson collaborated with us on the parallel development of a similar system simulating full model neurons. We are grateful to B. Hutcheon and E. Pull for shanng unpublished work with us. Thisresearch was supported by NSF BNS9009251 and NIMH MH46742. 394 The primary limitation of the dynamic-clamp method is the same as that of the voltage and current clamp: the problem of measuring and clamping the potential in situations where the neuron is not electrotonically compact. Because the dynamic clamp simulates a point source of conductance, it does not accurately describe the normal distribution of channels in the membrane. In an extended neuron, this may limit the use of the technique, especially for fast currents. On the other hand, if the neuron is reasonably electrotonically compact and if the conductances are fairly slow, the dynamic clamp should provide a good simulation. We have successfully used the dynamic clamp for several slow currents in STG neurons. Use of the dynamic clamp technique requires that the current calculations are updated rapidly enough to simulate the real conductance while maintaining an accurate measurement of membrane potential. The maximum rate at which a conductance described by a single activation variable can be updated, with our present 50 MHz 486 computer, is about 5 kHz. This is sufficient to follow rapid voltage fluctuations such as action potentials, With our present system, the update rate for the simulation of eight conductances is - 1 kHz. Increases in hardware speed will increase these rates. The dynamic clamp will not replicate the effects of second messengers elicited by Ca 2÷ currents. To simulate Ca2+ entry using the dynamic clamp Ca 2÷containing electrodes can be used, but this may not duplicate the secondary effects of Ca 2÷ because the dynamic clamp injects current at the site of the electrode tip rather than where the Ca 2÷ channels are located. In principle, the dynamic clamp can be used to either add or subtract an existing conductance from a neuron. However, care must be exercised in the case of subtraction, especially if the investigator attempts to remove a conductance completely. If the conductance is not modeled accurately, a difference current with unpredictable properties may result from the mismatch between the real and the modeled conductance. Furthermore, adding a negative conductance that is too large will destabilize a neuron with disastrous consequences (for the neuron) if the maximum current output of the clamp is not restricted. Future prospects One of the most appealing features of this new method is the construction of novel circuits from isolated neurons in which the strength and timecourse of each synaptic connection can be varied at will. The experimenter can now study 'model' circuits without worrying about the limitations o[ over-simplified model neurons. The biological unknowns are built into such hybrid modeling studies because the many biochemical and metabolic processes that control neuronal activity are handled correctly by the neurons themselves. We expect the dynamic clamp will augment other conventional simulation, biophysical and pharmacological approaches. The dynamic clamp makes it possible to combine computer simulation with living neurons, and adds an interactive modeling technique to the tools available for studying neurons and nervous systems. Selected references 1 Hodgkin, A. L. and Huxley, A. F. (1952) J. Physiol. 117, 500-544 2 Sharp, A. A., O'Neil, M. B., Abbott, L. F. and Marder, E. (1993) J. Neurophysiol. 69, 992-995 3 Marder,E. etal. in Enabling Technologies ForCultured Neural Networks (Stenger, D. A. and McKenna, T. M., eds), Academic Press(in press) 4 Renaud-LeMasson,S., LeMasson,G., Marder, E. and Abbott, L. F. (1993) in Advances in Neural Information Processing Systems (Vol. 5) (Hanson,S. J., Cowan,J. D. and Giles,C. L., eds), pp. 813-819, Morgan Kaufmann Publishers 5 Golowasch, J. and Marder, E. (1992) J. Neurosci. 12, 810-817 6 Hooper, S. L. and Marder, E. (1987) J. Neurosci. 7, 2097-21 t 2 7 Marder, E. and Weimann, J. M. (1992) in Neurobiology of Motor Programme Selection (Kien, J., McCrohan, C. R. and Wintow, W., eds), pp. 3-19, PergamonPress 8 Golowasch, J. and Marder, E. (1992) J. Neurophysiol. 67, 318-331 9 Kiehn,O. and Harris-Warrick,R. M. (1992)J. Neurophysiol. 68, 496-508 10 Hartline, D. K. (1979) Biol. Cybernetics 33,223-236 11 Tierney, A. J. and Harris-Warrick, R. M. (1992) J. NeurophysioL 67, 599-609 12 Eisen, J. S. and Marder, E. (1984) J. Neurophysiot. 51, 1375-1393 13 Ready,D. F. and Nicholts,J. (1979) Nature 281, 67-69 14 Camardo, J., Proshansky, E. and Schacher, S. (1983) J, Neurosci. 3, 2614-2620 15 Syed, N. I., Bultoch, A. G. M. and Lukowiak, K. (1990) Science 250, 282-285 16 Kleinfeld, D., Raccuia-Behling, F. and Chiel, H. J. (1990) Biophys. J. 57, 697-715 17 Bulloch,A. G. M. and Syed,N. t. (1992) Trends Neurosci. 15, 422-427 Writing for TINS M o s t of the articles published in T r e n d s in N e u r o s c i e n c e s are specially invited by the editor. However, unsolicited articles will also be considered for publication, tf you wish to write for TINS, please contact the editor, or a member of the Editorial Advisory Board first, with an outline of y o u r intended article. TINS, Vol. 16, No. 10, 1993