dual intracellular recordings and computational models of slow

DUAL INTRACELLULAR RECORDINGS AND COMPUTATIONALMODELS OF SLOW INHIBITORY POSTSYNAPTIC POTENTIALS IN

RAT NEOCORTICAL AND HIPPOCAMPAL SLICES

A. M. THOMSON*† and A. DESTEXHE‡*Department of Physiology, Royal Free and University College Medical School, Rowland Hill Street,

London NW3 2PF, U.K.

‡Department de Physiologie, Faculte de Medecine, Universite´ Laval, Quebec, Canada G1K 7P4

Abstract—Dual intracellular recordings in slices of adult rat neocortex and hippocampus investigatedslow, putative GABAB receptor-mediated inhibitory postsynaptic potentials. In most pairs tested in whichthe interneuron elicited a fast inhibitory postsynaptic potential in the pyramid, this GABAA receptormediated inhibitory postsynaptic potential was entirely blocked by bicuculline or picrotoxin (3:3 inneocortex, 6:8 in CA1, all CA1 basket cells), even when high-frequency presynaptic spike trains wereelicited. However, in three of 85 neocortical paired recordings involving an interneuron, although nodiscernible response was elicited by single presynaptic interneuronal spikes, a long latency ($20 ms)inhibitory postsynaptic potential was elicited by a train of$3 spikes at frequencies$50–100 Hz. Thisslow inhibitory postsynaptic potential was insensitive to bicuculline (one pair tested). In neocortex, slowinhibitory postsynaptic potential duration reached a maximum of 200 ms even with prolonged presynapticspike trains. In contrast, summing fast, GABAA inhibitory postsynaptic potentials, elicited by spike trains,lasted as long as the train. Between four and 10 presynaptic spikes, mean peak slow inhibitory postsynapticpotential amplitude increased sharply to 0.38, 2.6 and 2.9 mV, respectively, in the three neocortical pairs(membrane potential260 to265 mV). Thereafter increases in spike number had little additional effect onamplitude. In two of eight pairs in CA1, one involving a presynaptic basket cell and the other a putativebistratified interneuron, the fast inhibitory postsynaptic potential was blocked by bicuculline revealing aslow inhibitory postsynaptic potential that was greatly reduced by 100mM CGP 35348 (basket cell pair).The sensitivity of this slow inhibitory postsynaptic potential to spike number was similar to that ofneocortical ‘pure’ slow inhibitory postsynaptic potentials, but was of longer duration, its plateau phaseoutlasting 200 ms spike trains and its maximum duration exceeding 400 ms. Computational models ofGABA release, diffusion and uptake suggested that extracellular accumulation of GABA cannot aloneaccount for the non-linear relationship between spike number and inhibitory postsynaptic potential ampli-tude. However, cooperativity in the kinetics of GABAB transduction mechanisms provided non-linearrelations similar to experimental data. Different kinetic models were considered for how G-proteinsactivate K1 channels, including allosteric models. For all models, the best fit to experimental data wasobtained with four G-protein binding sites on the K1 channels, consistent with a tetrameric structure forthe K1 channels associated with GABAB receptors.

Thus some inhibitory connections in neocortex and hippocampus appear mediated solely by fastGABAA receptors, while others appear mediated solely by slow, non-ionotropic, possibly GABAB recep-tors. In addition, some inhibitory postsynaptic potentials arising in proximal portions of CA1 pyramidalcells are mediated by both GABAA and GABAB receptors. Our data indicate that the GABA released by asingle interneuron can saturate the GABAB receptor mechanism(s) accessible to it and that ‘spillover’ toextrasynaptic sites need not necessarily be proposed to explain these slow inhibitory postsynaptic potentialproperties.q 1999 IBRO. Published by Elsevier Science Ltd.

Key words: GABAB, interneuron, hippocampus, cerebral cortex, G-protein, spillover.

Since the characterization of the GABAB receptor5

and the development of the first selective antagonistsmany studies have described late, slow, bicuculline-insensitive inhibitory postsynaptic potentials(IPSPs) elicited by electrical stimulation, in a

number of brain regions, e.g.,19 that were sensitiveto Saclofen or 2-OH-Saclofen (see Ref. 23 forreview). These events are mediated by G-protein-activated K1 channels.2,12Typically, a higher stimu-lus threshold was required to elicit GABAB IPSPsthan GABAA IPSPs.12 Moreover, spontaneousGABAB inhibitory postsynaptic currents did notoccur unless 4-aminopyridine was added to thebathing medium,18,27presumably inducing repetitivefiring in the interneuron(s) activating the GABAB

receptors.Two hypotheses have been proposed to explain

1193

NeuroscienceVol. 92, No. 4, pp. 1193–1215, 1999Copyrightq 1999 IBRO. Published by Elsevier Science Ltd

Printed in Great Britain. All rights reserved0306-4522/99 $20.00+0.00PII: S0306-4522(99)00021-4

Pergamon

†To whom correspondence should be addressed.Abbreviations: ACSF, artificial cerebrospinal fluid; BF, burst

firing cells; CFS, classical fast spiking; EPSP, excitatory post-synaptic potential; FS, fast spiking cell; IPSP, inhibitory post-synaptic potential; MP, membrane potential; RS, regularspiking cell; SLM, stratum lacunosum moleculare; SO, stra-tum oriens; SP, stratum pyramidale; SR, stratum radiatum.

the apparent need for intense interneuronal activa-tion. First the ‘spillover’ hypothesis12 proposes thatmultiple GABA release would result in extracellularaccumulation of GABA. This GABA would reachlevels sufficient to activate extrasynaptically-locatedGABAB receptors only with sufficiently high inter-neuronal activity, or when GABA uptake wasblocked.17,34 Second the ‘G-protein’ hypothesis9

postulated that the nonlinear activation propertiesarise from intracellular mechanisms involving G-protein cascades that are responsible for K1 channelactivation. In this case, the GABAB response waspredicted to be sensitive to the number of presynap-tic spikes and a sufficiently high concentration of G-proteins was required for detectable IPSPs to beactivated. A need for a build up of activated G-proteins would also explain the minimal 10–20 msonset latency, sigmoidal rise and multi-exponentialdecay of GABAB IPSPs (as described by Ref. 28).

Whether mixed GABAA/GABAB IPSPs generatedby single interneurons occur in cortical regions, orwhether they are always separate has also been acontroversial issue for some time (for discussionsee Ref. 26). That the two types of IPSPs originatefrom quite separate interneurons has been proposedby several studies using indirect techniques, e.g.,4,32.Only two previous studies have, however, usedpaired recordings to study slow IPSPs. In hippocam-pus21 an apparently pure slow hyperpolarization wasinduced in a pyramidal cell by a train of spikes in asingle stratum lacunosum moleculare (SLM) inter-neuron. Conversely, in thalamus, a single GABA-ergic reticularis nucleus interneuron that elicited afast GABAA IPSP with single spikes could alsoelicit a bicuculline-insensitive, CGP 35348-sensitiveslow IPSP in a simultaneously recorded relay cell,provided prolonged, high-frequency firing waselicited in the interneuron.20 It is less clear whethermixed IPSPs occur in cortical regions. In somepaired recordings in neocortex, prolonged high-frequency firing in a single interneuron could elicita slowly developing and decaying event in additionto the summing fast GABAA receptor-mediatedIPSPs in a simultaneously recorded postsynapticpyramidal cell.36 This event was not apparent withspike trains briefer than 10–20 spikes. However,these events were not isolated pharmacologicallyand their dependence upon spike number andfrequency was not studied in any detail.

GABAB IPSP amplitude appears to be compro-mised by the concurrent activation of GABAA

IPSPs.8 This could be a postsynaptic phenomenon,blockade by raised intracellular Cl2 of the G-proteinactivation of K1 channels,22 or Cl2 channel shuntingof lower conductance events. The more effectiveactivation of GABAB IPSPs in the presence ofGABAA antagonists could equally result from themore intense activation of ‘disinhibited’ inter-neurons where electrical stimulation has been usedto elicit the IPSPs. It was therefore also of interest to

determine whether GABAB IPSPs elicited by asingle interneuron in which presynaptic firing canbe controlled, are masked by GABAA events, butrevealed by application of GABAA antagonists.

In this paper therefore, paired intracellular record-ings in slices of adult rat neocortex and hippocampuswere used to determine whether slow, putativeGABAB receptor-mediated IPSPs can arise fromactivity in a single cortical interneuron, whetherthey always, or conversely, never occur in theabsence of coincident GABAA receptor activation.To characterize the exact relationship between thepattern of presynaptic firing and the amplitude ofGABAB IPSPs and to test whether repetitive pre-synaptic firing is indeed required, paired recordingswere tested with a range of presynaptic firingpatterns. These results were then used in computa-tional models of GABAB responses to investigate theplausibility of extracellular and intracellular hypoth-eses to explain any observed presynaptic/postsyn-aptic relations.

EXPERIMENTAL PROCEDURES

Young adult male Sprague–Dawley rats (bred in-house),120–200 g in body weight were anaesthetized with Sagatal(pentobarbitone sodium, Rhoˆne Merieux, 60 mg/kg i.p.,substantially higher than the necessary anaesthetic dosefor surgery), or with Fluothane followed by Sagatal andperfused transcardially with an ice-cold artificial cerebro-spinal fluid (ACSF) in which the NaCl had been replacedwith 248 mM sucrose and to which Sagatal (60 mg/l) wasadded, and then decapitated. All procedures were performedon deeply-anaesthetised animals to prevent any suffering,experiments were performedin vitro and the minimumnumber of animals used. Coronal slices of neocortex orhippocampus, 450–500-mm-thick were cut (Vibroslice,Camden Instruments). During slice preparation, in ice-coldACSF, and for the first hour of incubation in the interfacerecording chamber (34–368C), the slices were maintained inthe sucrose containing medium (without Sagatal) equili-brated with 95%O2/5% CO2. This was then replaced withstandard ACSF containing 124 mM NaCl, 25.5 mMNaHCO3, 3.3 mM KCl, 1.2 mM KH2PO4, 1 mM MgSO4,2.5 mM CaCl2 and 15 mMd-glucose in which all recordingswere performed. Recordings commenced after a furtherhour in this medium.

Electrophysiological recordings

Intracellular recordings were made from pairs of neuronswith conventional sharp micro-electrodes filled with 2 MKMeSO4 (90–140 MV) and, in the majority of experiments,2% w/v Biocytin, using an Axoprobe. Typically, a stableintracellular recording, i.e. one in which the neuron exhib-ited a stable membrane potential, input resistance andresponses to injected current, was first obtained from oneneuron in a fresh region of the slice. A second electrode wasthen introduced and sequential recordings made from asmall number of neighbouring cells until a connection wasfound, or an interneuron located. If that interneuron was notsynaptically connected to the first cell, the first electrodewas withdrawn and a small number of other cells impaledand tested sequentially. The procedure was repeated in freshregions until a connection was apparent. Single spikes, pairsof spikes and spike trains were elicited in the presynapticneuron by injection of ramp and/or square wave currentpulses, delivered typically once every 3 s. The shape,

A. M. Thomson and A. Destexhe1194

amplitude and duration of the pulse and its repetition ratewere modified to elicit different patterns of presynapticfiring, interleaving single spikes and brief trains with longtrains of spikes and/or, low with high firing frequencies.Continuous analogue recordings from both neurons for theentire duration of the paired recording were made onmagnetic tape (Racal Store 4). Postsynaptic membranepotential (MP) was maintained within 2 mV of a presetvalue and after sufficient data had been collected at onepotential, could be changed. During continuous currentinjection, electrode balance was monitored by observingvoltage responses to small, brief current pulses injectedbefore, or after responses to presynaptic spikes.

In pairs in which a short latency IPSP was elicited bysingle presynaptic spikes, bicuculline or picrotoxin wasapplied after the fast IPSP had been characterized and trainsof presynaptic spikes were then elicited to determinewhether a slow, bicuculline-insensitive IPSP remained.Where no fast IPSP was apparent in control conditions,presynaptic spike trains were elicited to test for a ‘pure’slow IPSP. Drugs were applied dissolved in the bathingmedium; bicuculline methochloride (1–100mM, Tocris),picrotoxin (1–20mM Sigma), CGP 35348 (100mM,Novartis).

Data analysis

Dual recording data were collected from tape to disk andanalysed off-line (using in-house software, 5–10 kHz,voltage resolution 0.005–0.01 mV,36). Collected sweepswere 200, or 400 ms in duration, depending on the durationof the postsynaptic responses from that pair. Each datarecord (sweep) contained pre- and postsynaptic recordingsand trigger points that would be used to trigger subsequentdata analysis. Each sweep was then observed and eitheraccepted, edited if the trigger points were inaccurate, orrejected. Acceptable, or adequately edited sweeps werethose in which the rising phase of each presynaptic spikewas accurately recognised by the software as a trigger pointand where spike-elicited IPSPs were not contaminated byartifacts or by large spontaneous postsynaptic potentials.Editing consisted of removing false trigger points, or adjust-ing inaccurate trigger points. Data were then divided intosubsets according to the postsynaptic MP, the presence orabsence of drugs and the number and frequency of pre-synaptic spikes.

Fast IPSPs elicited by the first action potential in a trainwere analysed using the rising phase of the first spike totrigger data collection and analysis and compared withresponses to single spikes. Typically, a point close to thepeak of the spike was chosen and within the limits of thesample frequency, an identical point used for each spike inany data set. Analysis of 2nd, 3rd...5th spike responsesrequired triggering of data analysis from the 2nd, 3rd...5thspike. Second spike responses in brief trains were comparedwith 2nd spike responses to spike pairs with the same inter-spike interval, 3rd spike responses with responses to spiketriplets of the same duration and so on. Where these weresimilar, a composite average of the response to a brief trainof spikes could be generated and the relative amplitudes anddurations of successive IPSPs in trains assessed.

To analyse slow IPSPs, the number of spikes in the pre-synaptic trace and the average spike frequency for the entirespike train and for the first four spikes in each train, thelatency to onset of the IPSP from the 1st spike, the IPSPrise time, width at half amplitude, duration and amplitudewere measured for each single sweep (see Fig. 1). Thesemeasurements were plotted one against another to determinewhether any correlations existed between these parameters.Average IPSP amplitudes for data subsets involving thesame MP, the same number of presynaptic spikes with asimilar firing frequency were also computed and plottedagainst spike number and frequency. In the figures presented

of averaged postsynaptic responses, presynaptic spike arti-facts were often removed or reduced graphically from aver-aged IPSPs. Averages of some slow IPSPs were smoothedby three or five point averaging.

Computational models

A previous model of the diffusion of GABA in the extra-cellular space and of GABAB receptor activation was usedhere with some modification.9 Release, diffusion and uptakeof GABA were simulated using a two-dimensional grid of0.5× 0.5mm to approximate the thin extracellular spacesurrounding neurons. Release was simulated by a suddenincrease in the concentration of transmitter in one of thecompartments, reaching a peak of 3 mM. Extracellulardiffusion of GABA was simulated with a diffusion coeffi-cient of D�8× 1026 cm2 s21, based on values ofcompounds of similar molecular weight.3 Tortuosity of theextracellular space may reduce the effectiveness of extra-cellular diffusion significantly.29 A lower diffusion coeffi-cient of D� 3× 1026 cm2 s21 was therefore also used(corresponding to a tortuosity factor of about 1.6). GABAuptake was simulated according to the following scheme:

GABA 1 U O B

B! U 1 GABAi

The first step describes the reversible binding of GABA tothe unbound transporter (U), leading to bound transporter(B). The second step represents the irreversible translocationof GABA into the intracellular space (GABAi). Thus, theconcentration of GABA in the extracellular space isdescribed by the following equations:

2T� �x; t�2t

� 2k1T� �x; t��Bm 2 B� �x; t��1 k21B� �x; t�

1 D72T� �x; t�

2B� �x; t�2t

� 2k1T� �x; t��Bm 2 B� �x; t��1 �k21 1 k2�B� �x; t�whereT� �x; t� andB� �x; t� are the concentration of GABA andof bound transporter, respectively at point�x and timet. k1andk21 are the forward and backward binding rate of GABAto transporter, respectively,k2 is the rate constant for thetranslocation of GABA, and Bm is the total extracellularconcentration of available transporter. A value of(k21 1 k2)/k1� 4mM was estimated experimentally,6 show-ing a high affinity of the transporter for GABA. For gluta-mate, binding to the transporter is very fast, while thetranslocation step is slower.11 Assuming the same transloca-tion rate as estimated for glutamate31 leads to the followingset of parameters: k1� 30 mM21 ms21, k21� 0.1 ms21 andk2�0.02 ms21. The total concentration of transporter mole-cules is difficult to estimate. It was assumed constantthrough the extracellular space and was tested in the rangeof Bm�0 to 1 mM.

Equations of GABA release, diffusion and uptake weresimulated using a first-order implicit integration methodimplemented in NEURON14 (see details in Ref. 9).

Activation of GABAB responses by GABA was simulatedusing the following kinetic equations:

R0 1 GABA N Rp �1�

R p 1G0 N RG! R p 1Gp �2�

Gp! G0 �3�

K0 1 nGp N Kp �4�where GABA binds to the inactive form of the GABAB

GABAB IPSPs in neocortex and hippocampus 1195

receptor (R0), leading to the activated form of thereceptor (R*; reaction 1). R* catalyses the formation ofactive G-proteins (G*) from their inactive form (G0) accord-ing to a Michaelis–Menten scheme (reaction 2). Activated

G-protein (G*) is hydrolysed to its inactive form (G0; reac-tion 3). Finally, n activated G-proteins bind to the closedform of the K1 channel (K0), leading to opening of K1

channels (open form K*; reaction 4).Several approximations in this scheme, such as consider-

ing that intermediate enzymatic forms are at steady-state,that G-proteins are in excess and that open K1 channels areat steady state due to fast binding of G* to K0, lead to thefollowing simplified equations:

dr=dt � K1:T�1 2 r�2 K2:r

dg=dT � K3:r 2 K4:g

IGABAB � �gGABAB × �gn=�gn

=Kd�� × �Vpost 2 EK �where r is the fraction of receptors in the active form, T isthe GABA concentration in the synaptic cleft, g is thenormalized concentration of G-proteins in the active state.K1� 0.18 ms21 mM21, K2� 0.0096 ms21, K3� 0.19 ms21,K4� 0.060 ms21 are kinetic rate constants for reactions 1–4above and Kd�17.83 is the dissociation constant for acti-vated G-proteins bound to K1 channels. IGABAB is the currentactivated by GABA andGABAB is the maximal conductanceof the K1 channel, Vpost is the postsynaptic membranepotential and EK the potassium equilibrium potential.These equations were discussed in detail in previouspapers.9,10

Fitting of the model to experimental data was performedusing a simplex procedure.30 The model was run at eachinteraction of the simplex procedure and the parametersoptimized so as to minimize the mean square error betweenmodel and data. Several initial conditions were tested toensure that the optimal fit was unique. The behaviour wasalways robust within some domain of parameter valuesaround the optimal fit. The corresponding range of para-meter values are given in the text.

In the model shown above,n G-proteins bind simulta-neously to act on K1 channels. In addition to this simplemodel, several other models were considered to simulate thebinding of activated G-proteins to open K1 channels.

Hodgkin–Huxley-like model

A possible model for K1 channel activation is analogousto the model introduced by Hodgkin and Huxley:15

IGABAB � �gGABAB :mn�V 2 EK�

dm=dt � ag�1 2 m�2 bm

where K1 channels open according ton independent gates,m is the fraction of gates in the open state,a andb are rateconstants and m is given by:

m� g=�g 1 Kd�where Kd�b/a is the dissociation constant of G-proteinbinding on each gate.

Allosteric model

Another possible model is derived from the allostericscheme introduced by Monod, Wyman and Changeux:24

C O O

O 1 G O O1

O1 1 G O O2…

O�n21� 1 G O On

where the closed (C) and open (O) form of the channel are atequilibrium, and activated G-proteins bind only to the open


Fig. 1. Original single sweep records of a neocortical slow IPSPunder control conditions. Five dual traces illustrating responsesto trains of 14, 10, 6, 4 and 1 presynaptic action potentials areillustrated. The dotted lines indicate the positions at whichcursers are placed for measurements of the latency from thefirst presynaptic action potential, of the duration of the IPSP,of the IPSP amplitude and of its rise time. Capacitance couplingspike artifacts coincident with presynaptic spikes are apparent.These artifacts are removed or reduced graphically in subse-quent figures showing averaged postsynaptic responses.

(960509B2.)

state. In this case, the total concentration of all open forms atsteady-state is given by:24

Y � 1=�1 1 L=�1 1 g=Kd�n�where L is the allosteric constant (the equilibrium constantof COO transition) and Kd is the dissociation constant ofG-protein binding. The current is then given by:

IGABAB � �gGABAB :Y�V 2 EK �

Histological processing

Following the paired recording, cells were filled withbiocytin by passing 0.5 nA depolarizing current pulses ina 50% duty cycle at 1 Hz for 5–15 min. In earlier neo-cortical experiments, slices were incubated for at least afurther hour after leaving the cell before fixation. Even-tually, however, it became apparent that while this couldresult in good axonal filling, it typically resulted in a lossof biocytin from the soma, dendrites and the most proximalportion of the axon of interneurons (but not of pyramids),making unambiguous identification and reconstructionimpossible. Probably for this reason, only one of the pre-synaptic neocortical interneurons discussed here wasadequately recovered for reconstruction. In more recenthippocampal experiments therefore, slices were incubatedfor no more than a further 5–15 min after leaving the cells.Provided recordings had exceeded 45–60 min, good axonalfilling resulted. Slices were then fixed overnight in 0.1 Mphosphate buffer containing 1.25% glutaraldehyde and 3%paraformaldehyde. After embedding in gelatin, the sliceswere sectioned at 60mm on a Vibratome. Injected biocytinwas localized using the Vector ABC kit (Elite), incubatedwith 0.1% Triton overnight. The injected biocytin was thenvisualized using 30,30-diaminobenzidine (Sigma) and thesections dehydrated and embedded in Durcupan resin(Fluka) on slides. Well-filled and characterized interneuronswere fully reconstructed using a drawing tube and a×100objective.

RESULTS

During recordings, cells were identified as inter-neurons by two criteria; firstly by the fast spikingbehaviour that typifies several classes of neocorticaland hippocampal interneurons. Secondly, some ofthe interneurons elicited in a simultaneouslyrecorded cell an IPSP with a brief and constantlatency. Some regular spiking interneurons thatwere not recorded simultaneously with a postsyn-aptic partner are likely to have been missed. In addi-tion a number of the recorded interneurons weresubsequently recovered histologically (see below).

In 24 experiments in neocortex (96 slices) and 17experiments in the CA1 region of the hippocampus(68 slices) a cell that was identifiable at the time ofrecording as an interneuron was recorded simulta-neously with a pyramidal cell in 85 neocorticalpaired recordings and 179 hippocampal pairedrecordings. This constitutes approximately 10% ofall paired recordings made in these regions, themajority of which involved two pyramidal cells.Fourteen of the neocortical (14:85, or 1:6) and 39of the hippocampal (39:179, or 1:4.5) paired record-ings yielded a monosynaptic fast IPSP. The majorityof neocortical interneurons in this study were inlayer V (including all those eliciting slow IPSPs)

and the somata of all reported CA1 interneuronswere located in, or close to stratum pyramidale(SP). Fast IPSPs recorded in neocortex have beenreported in some detail previously,36 the fast IPSPsin hippocampus form part of another study, and onlythose tested with bicuculline will be discussedfurther here.

Inhibitory postsynaptic potentials apparentlymediated only by GABAA receptors

Three paired recordings in neocortex (two in layerV, one in layer III) and seven in hippocampus inwhich a single spike in the presynaptic interneuronelicited a short latency IPSP, were challenged withbicuculline. An additional hippocampal pair waschallenged with picrotoxin. In control conditions,these IPSPs had constant latencies around 1 ms,and durations between 20 and 100 ms. In responseto high-frequency spike trains, GABAA IPSPssummed (see Fig. 2, Fig. 5, Fig. 10). The secondand subsequent IPSPs were usually smaller thanthe first (measured as the difference between thepeak of that IPSP and the falling phase of the preced-ing IPSP, see Fig. 2, Fig. 5, Fig. 10A) and plateaupotentials reached during prolonged activity werebetween 50% and 150% of the average amplitudeof the first IPSP. These summing plateau eventswere maintained throughout long spike trains andthen decayed at least as rapidly as the single spikeIPSP. In all pairs tested, single spike responses wereblocked with a low dose (1–5mM) of bicucullinewithin 3–15 min. However, with these low doses,when brief, high-frequency presynaptic spike trainswere elicited, a small, longer latency response waselicited in some pairs. This response, which wastypically ,15% of the amplitude of the controlsingle spike IPSP appeared however to be mediatedby GABAA receptors. Its occurrence was accuratelytime-locked to the second or third presynaptic spikeand its latency, from that spike, was very constantand brief (analysis using the rising phase of thesecond or third spike as the trigger was essentialfor this procedure). When its shape was comparedwith the shape of the control IPSP, scaled to matchin amplitude, no difference was apparent. Moreover,increasing the bicuculline concentration blocked thisresponse entirely and even long trains of presynapticspikes now elicited no detectable response in thethree neocortical pairs or in five of the seven hippo-campal pairs tested (the bicuculline resistant, slowIPSPs in the remaining two hippocampal pairs aredescribed below). Picrotoxin blocked the IPSPtotally in the one CA1 pair tested. These then werepure fast GABAA receptor-mediated responses.

One of the neocortical interneurons elicitinga ‘pure’ fast IPSP was a fast spiking cell whichwas fully recovered histologically and was asmall, smooth layer III interneuron with a denselocal axonal arbour. The other two neocortical


interneurons that elicited demonstrably pure fastIPSPs, were not fully recovered. One was a regularspiking and one a fast spiking cell (see Ref. 36 forfiring pattern descriptions). The six CA1 inter-neurons that elicited pure fast IPSPs were fullyrecovered histologically and all had extensive axonramifying throughout the depth of SP and extendinglittle beyond its borders, i.e. a basket cell-like axonalarbour (see Fig. 3. for an example). Three of thesewere fast spiking cells with a rounded spike after-hyperpolarization, one a regular spiking cell (RS),and two were burst firing cells (BF, see Ref. 1 fordescription of these firing patterns in CA1 basketand bistratified interneurons).

Interneurons eliciting pure slow inhibitorypostsynaptic potentials in neocortex

In three of the 85 paired recordings in neocortex

that involved an interneuron, a single interneuronalaction potential elicited no discernible response inthe simultaneously recorded pyramidal cell, evenwhen the postsynaptic pyramidal cell was depolar-ized up to and beyond spike threshold and whenlarge spontaneous fast IPSPs were apparent.However, trains of spikes elicited a long-latencyIPSP. In one of these three examples the classicalfast spiking (CFS) presynaptic interneuron was fullyrecovered histologically (960509B2, reconstructedin Fig. 7). In the other two examples, what appearedto be interneuronal axon was apparent in therecorded region and beaded aspiny interneuronaldendrites were also recovered histologically fromone fast spiking presynaptic cell (950209A1).Since one of these interneurons was a CFS cell(960509B2), one a regular spiking cell (RS,960716A1) and one a fast spiking cell (FS, not clas-sical, 950209A1), the presynaptic firing pattern did


Fig. 2. Two inhibitory connections in CA1 that appear to be mediated solely by GABAA receptors. In controlconditions (A and C), single spikes and double spikes (C) elicit brief, short latency IPSPs. Following addition ofbicuculline (B and D), all postsynaptic response, even to high-frequency spike trains, is blocked. In D the controlresponse to a similar high-frequency spike train (normal trace) is compared with the response in bicuculline (boldtrace). The dotted lines in C and D indicate the time-course of the single spike control IPSP. In each case, a typicalsingle sweep record of the presynaptic interneuron’s firing is shown above an average (30–100 sweeps) of thepostsynaptic response. A reconstruction of the interneuron that elicited the IPSPs in C and D is shown in Fig. 3

(970618B1, 970619C1.)

Fig. 3. The IPSPs illustrated in Fig. 2. were both elicited by CA1 interneurons with basket-like axonal arbours.One is reconstructed here in its entirety at× 100 magnification. The axon (blue) ramified throughout the depth ofstratum pyramidale (SP), but extended little beyond its borders and its dendrites (red) extended through stratumoriens (SO) and through stratum radiatum (SR) and into stratum lacunosum moleculare (SLM). The dotted linesindicate the borders of these hippocampal layers. The insert indicates the position of this cell in the slice from

which it was recorded. (970618C1.)


Fig

.3.

not appear to predict the ability of an interneuron toelicit a slow IPSP.

Since two of the presynaptic neurons were fastspiking cells and sufficiently recovered histologi-cally for identification, these two slow IPSPs weredemonstrably elicited by interneurons. Fast spikingbehaviour has never been reported for a pyramidalcell, but is regularly described in identifiable inter-neurons. The third presynaptic cell was regular spik-ing and therefore not so easily distinguishable from apyramidal cell by its electrophysiological character-istics and only its axon was recovered histologically.This pair (960716A1) was however tested with bicu-culline to exclude the possibility of a disynaptic fastIPSP resulting from a pyramid–interneuron–pyra-mid circuit. The slow IPSP was not reduced by bicu-culline (10mM, Fig. 4), which blocked spontaneousfast IPSPs in the postsynaptic pyramidal cell within3 min of application. In addition, in all other respectsthis slow IPSP resembled those elicited by the moreeasily identifiable interneurons.

Inputs to ‘pure slow’ inhibitory postsynapticpotential generating interneurons

Before the connection involving the slow IPSPwas studied, two of these interneurons were foundto receive inputs from other cells. One was postsy-naptic to a second interneuron which elicited in it ashort latency IPSP (Fig. 5) with a width at halfamplitude of 9 ms, i.e. briefer than any of theIPSPs reported in pyramidal cells to date (range 10to .100 ms,36). This IPSP summed on repetitiveactivation of the presynaptic interneuron. Anotherof the interneurons presynaptic to a slow IPSP waspostsynaptic to a layer III pyramidal cell, which

elicited in it an excitatory postsynaptic potential(EPSP) about 1 mV in amplitude. This EPSP exhib-ited pronounced paired pulse depression (Fig. 5). Togenerate the$3 spike bursts necessary for activationof the slow IPSP, such an excitatory input wouldprobably therefore be inadequate alone.

Amplitude of pure slow inhibitory postsynapticpotentials and effect of presynaptic spike number

Data subsets in which the postsynaptic MP andaverage presynaptic firing rate were the same wereselected and where at least 10 and up to 30 sweepsinvolved the same number of presynaptic spikes,average IPSP amplitudes were calculated. Withpresynaptic spike trains of$100 Hz, and at least10 spikes, mean IPSP amplitudes were 0.38^0.3(S.D., 950209A1), 2.91 0.8 (960509B2) and2.6^0.81 mV (960716A1), respectively, for thethree pairs at postsynaptic MPs between260 and265 mV. In all three pairs, IPSP amplitude wasstrongly influenced by spike number between threeto four and 10 spikes. No discernible IPSP waselicited with single or double presynaptic spikesand only a barely detectable response to three spikes,while between four and 10 spikes, IPSP amplitudeincreased sharply with spike number. However,beyond 10 spikes little additional increase in ampli-tude was apparent (Fig. 6, see Fig. 15B for the modelfit to these data). The presynaptic interneuron for thepair illustrated in Fig. 6 is reconstructed in Fig. 7.

Effect on slow inhibitory postsynaptic potentials ofpresynaptic spike train duration

Provided spike frequency was.50–100 Hz, the


Fig. 4. Slow IPSPs elicited in a layer V neocortical pyramidal cell by spike trains in a regular spiking layer Vinterneuron are insensitive to bicuculline (bold traces). Averaged postsynaptic responses to two types of spiketrains, in the absence (normal trace) and presence of bicuculline (bold trace) are compared. (A) spike trainsconsisting of nine or 10 spikes and (B) spike trains of 13 to 15 spikes (10–30 sweeps each average). Preceding

each averaged IPSP is the response to a small brief current pulse injected postsynaptically. (960716A1.)

duration of the presynaptic spike train had little or noinfluence on the amplitude of the IPSP, when spikenumber was held constant. For example, trains of 10spikes at 100 Hz lasting 90 ms and 10 spike trains at250 Hz lasting 36 ms elicited similar amplitudeIPSPs. Spike train duration also had only a smallinfluence on IPSP duration (see Fig. 8D) and thecorrelation between these parameters was weak. Inone pair (960509B2), spike trains briefer than 60 msgenerated IPSPs with a mean duration of117^ 38 ms (n�54), while trains 100 to 160 msin duration resulted in IPSP durations of171^ 36 ms (n�38). However, responses to moreprolonged trains peaked and had almost decayedbefore the spike train ended (Fig. 9B). Thus, incontrast to the summing GABAA receptor-mediatedIPSPs elicited in other pairs by long spike trains,these slow IPSPs lasted no longer than 200 msafter the first spike, even with long spike trains.

Effect of presynaptic spike frequency on slowinhibitory postsynaptic potential time-course

Presynaptic spike frequencies below 50 Hzappeared to generate little postsynaptic responseand were not studied systematically. Between 75and 300 Hz presynaptic firing frequency had onlya small, non-significant influence on IPSP ampli-tude, when trains with similar numbers of spikeswere compared. Frequency had a small influenceon latency however (see Figs 8A and 9A). In one

pair (960716A1), frequencies of 170–200 Hzresulted in latencies of 41 5.5 ms (n�13),frequencies of 200–220 Hz, latencies of36^ 3.6 ms (n�19), frequencies of 220–250 Hzlatencies of 33 3.2 ms (n�13) and frequenciesof 250–300 Hz latencies of 26 4 ms (n� 4). Inanother pair (950509B2) frequencies of 140–180 Hz yielded latencies of 39 10 ms (n�83)and frequencies of 200–230 Hz, latencies of31^ 9 ms (n� 49). In one pair firing frequencyhad no significant effect on IPSP rise time, inanother frequencies between 75 and 150 Hz resultedin rise times of 42 12 ms (n� 58) and rise timedecreased at frequencies above 180 Hz to30^ 13 ms (n�56). Single sweep events in thethird pair were too small to analyse in this way. Todetermine whether repetition rate influenced IPSPamplitude, slow IPSPs were elicited at two repetitionrates (0.33 and 0.1 Hz) and events elicited by spiketrains of similar duration and frequency compared.There was no significant difference in the amplitudesof the IPSPs at these two rates.

The largest data subset in which all parameterswere constant was selected and an averaged IPSPcomputed. Neither the rising phase nor the decayof this averaged IPSP was well fit by a single expo-nential, the decay approximating to the sum of twoto three exponentials and the rising phase to asigmoidal curve (as reported previously,28). Amodel fit to the time-course of this IPSP is shownin Figs 14 and 15.


Fig. 5. Synaptic inputs to two layer V interneurons that subsequently elicited slow IPSPs in pyramidal cells. (A)One of these interneurons received a descending excitatory input from a layer III pyramidal cell. Compositeaverages of EPSPs elicited by brief trains of three presynaptic spikes and recorded at two postsynaptic membranepotentials (MPs) are shown. This EPSP exhibited strong brief train depression. (B) The other interneuron receiveda fast inhibitory input from another layer V interneuron. A composite average of the summing fast IPSPs elicitedby a spike train in the presynaptic interneuron is illustrated, components of the composite average were triggeredfrom the 1st, 2nd, 3rd, 4th and 5th spikes in the train. Later spikes in the trains were not used as trigger points foraveraging and the decay of the average does not therefore accurately represent the time-course of the decay of thesumming IPSP (20–60 sweeps per average in the composite). The dotted lines indicate the time-course of the

postsynaptic responses to single (and in A, double) presynaptic spikes. (960716A1; 960509B2.)

Voltage relations of slow inhibitory postsynapticpotentials

In one pair (960509B2), slow IPSPs generated by

trains of $10 spikes were measured at four MPs.The amplitudes of these IPSPs are plotted againstMP in Fig. 8C. A linear regression through thesepoints yielded an extrapolated reversal potential of


Fig. 6. Slow IPSPs elicited by a single fast-spiking interneuron increase in amplitude with increasing numbers ofpresynaptic spikes. Spike trains and postsynaptic responses to them were collected into data subsets according tothe number of spikes and the spike frequency and averaged. The number of spikes in the train that elicited eachresponse is indicated to the left of each average and the mean frequency of the spike trains included to the right ofeach averaged IPSP. Single sweep examples of typical presynaptic spike trains are shown on the far left. The

interneuron that was probably responsible for this IPSP is reconstructed in Fig. 7. (960509B2.)

Fig. 7. A light level (× 100 magnification) drawing tube reconstruction of the layer V interneuron that elicited theslow IPSPs shown in Fig. 6. (960509B2). A level of ambiguity as to the identity of this interneuron exists becausetwo interneurons were recorded in this region of the slice (see Fig. 5. for the IPSPs elicited in the slow IPSP-generating interneuron by another interneuron). However, the other interneuron was recorded only relativelybriefly, not actively filled with biocytin and then incubated for. 1.5 h in the bath before fixation. Such long post-filling incubation times typically lead to loss of biocytin from soma, dendrites and most proximal axon even ofactively and well-filled interneurons (but not of pyramids). The slow IPSP generating interneuron on the otherhand, was recorded for. 1 h, actively filled by current injection and fixed within 20–25 min of filling. The somaand dendrites of the reconstructed interneuron (red) were completely, but less intensely stained than pyramidalcells filled during the same experiment and the axon (blue) was intensely stained. The insert shows the position of

this interneuron in the slice from which it was recorded. (960509B2.)

Fig

.7.

288 mV. A similar voltage relation was apparentin the other two pairs, but was studied lesssystematically.

Inhibitory postsynaptic potentials mediated by bothGABAA and GABAB receptors

The sensitivity of the neocortical slow IPSPs tocurrent injected at the soma might suggest that thesewere relatively proximal events. However, thepresynaptic neocortical interneurons were eithernot adequately recovered, or their treatment withTriton precluded confirmation of synaptic siteswith correlated ultrastructural morphology. Todetermine therefore whether proximal inhibitoryinputs could be mediated by GABAB receptors,paired recordings were made in SP of the CA1region of the hippocampus where the subcellularpostsynaptic targets of filled interneurons can bemore readily identified than in neocortex. None ofthe paired recordings to date involving SP inter-neurons has yielded a demonstrably pure slowIPSP (.100 connected pairs), but two yielded amixed GABAA/GABAB IPSP. Both of these pre-synaptic CA1 interneurons were BF cells. One was

only partially recovered. Its axon appeared to ramifyin both stratum oriens (SO) and stratum radiatum(SR), but less in SP, i.e. it resembled a bistratifiedcell.1 The slow, bicuculline-resistant component ofthe IPSP generated by this putative bistratified inter-neuron was small (0.2–0.3 mV) and it was notstudied in detail, but its shape and time-coursewere similar to those of the other mixed IPSP.

The other CA1 presynaptic interneuron was fullyrecovered and had a basket cell-like axonal arbour(Fig. 12). Under control conditions, single presynap-tic basket cell spikes elicited an IPSP 1.07 mV inaverage amplitude, latency 1 ms, rise time 9 msand width at half amplitude 33 ms in this pair(970319A). Trains of presynaptic spikes elicitedsumming IPSPs. The average amplitude of secondand subsequent IPSPs in trains was, however, smallerthan that of first IPSPs. The amplitude of the secondIPSP (measured from its peak to the falling phase ofthe 1st IPSP) was 33% of the amplitude of the first atinterspike intervals between 7 and 9 ms. SubsequentIPSPs were further depressed (Fig. 10A), but duringprolonged trains a plateau was attained and main-tained. At the time of recording, a mixed fast/slowIPSP was suspected, but the slow component is not


Fig. 8. (A) Slow IPSP latency is weakly inversely correlated with the presynaptic firing frequency preceding theIPSP (average frequency of the first four spikes in each sweep). (B) IPSP duration is weakly correlated with spiketrain duration. Each point was obtained from a single sweep record. (C) The extrapolated reversal potential forone of the slow IPSPs in neocortex is close to290 mV. Maximum IPSP amplitudes for sweeps with presynapticfiring rates.100 Hz and 10 or more spikes in the train are plotted against postsynaptic MP, for one pair

(960509B2).

readily apparent in averaged recordings undercontrol conditions. Indeed, as illustrated in Fig.10A, the slow component coincided with therebound depolarization that followed short trainsof fast IPSPs.

The single spike fast IPSP was completelyblocked by 10mM bicuculline within 5 min.Complete blockade of the summing GABAA IPSPelicited by longer spike trains took 20 min. GABAA

receptor blockade revealed a long-latency, slowIPSP (Fig. 10A) which was greatly reduced by addi-tion of 100mM CGP 35348 within 20 min (Fig.10B), at which time the recording was lost. Likethe ‘pure’ slow IPSPs in neocortex, the amplitude

of this IPSP was strongly influenced by the numberof presynaptic spikes that elicited it (see Fig. 11). Nodiscernible postsynaptic response was elicited byone or two presynaptic spikes. Between three tofour and 10 spikes IPSP amplitude sharply increasedand little additional increase in amplitude wasachieved with larger numbers of spikes. The stereo-typical firing pattern of this BF interneuron, whosethreshold response was a high-frequency burst of atleast two and more usually four or five spikes,precluded assessment of the influence of a widerange of tonic firing frequencies. This slow IPSPwas, however, longer in duration than the ‘pure’slow IPSPs in neocortex and the response to200 ms spike trains did not decline significantlyuntil the train had ended and it lasted$400 ms.Fig. 12 shows the reconstruction of the presynapticinterneuron of this pair. Its axonal arbour spans theentire depth of and is almost entirely restricted to SPindicating that this was a basket cell. The dendriticarbour is, however, unusual for a basket cell(compare Fig. 12 with Fig. 3), having a single‘apical’ dendrite that does not branch until it hastraversed approximately half the depth of SR andthen branches significantly in SLM.

Computational models

‘Spillover’ hypothesis. Models first investigatedwhether the ‘spillover’ hypothesis explained thenonlinear spike number-dependencies apparent forthe slow IPSPs. A model of extracellular release,diffusion and uptake of GABA (see Fig. 13A andExperimental Procedures) was used to determinewhether extracellular mechanisms can lead to accu-mulation of GABA only when multiple high-frequency presynaptic spikes occur. Simulation ofa terminal releasing at 200 Hz in the presence ofuptake led to a very brief presence of GABA inthe extracellular space immediately adjacent to therelease site and a limited duration presence 1mmand 2mm away from the release site (Fig. 13B,Control). Without uptake, the time-course was stillbrief in the cleft but the spillover of GABA outsidethe cleft was enhanced (Fig. 13B, No uptake). With asmaller diffusion coefficient to account for extracel-lular space tortuosity, and with uptake, the peakconcentration of GABA was diminished extrasynap-tically, but its accumulation was enhanced (Fig.13B, Low diffusion).

If an extracellular mechanism is to account for theobserved spike number-dependency of GABAB

responses, a similar dependency on spike numbershould be apparent in the model at extrajunctionalsites. To test this, the total integrated amount ofGABA at a distance of 2mm from the release sitewas computed. The relative amount is representedfor different numbers of presynaptic spikes at200 Hz in Fig. 13C. In control conditions and with-out uptake, the relation can be seen to be perfectly


Fig. 9. Effect of presynaptic firing frequency on slow IPSPlatency (A) and of spike train duration on IPSP duration (B).In each case, data subsets involving the same number andfrequency of presynaptic spikes were selected and averaged(30–40 sweeps each average). In A averaged IPSPs elicitedby two types of spike trains are compared. A high firingfrequency preceding the IPSP (bold traces) results in a brieferlatency than the slower train (normal trace). In B spike trainswith similar firing frequencies, but different durations arecompared. The longer spike train (normal traces) elicits a longerlasting IPSP, but the event peaks and begins to decline beforethe spike train ends. These data also indicate that spike numberhas a greater influence on IPSP amplitude than does spike

frequency. (960509B2.)

linear. However, a significant deviation from linear-ity occurred for high concentrations of GABA trans-porter (Bm�1 mM). Variations in other parameters,such as the size of the grid, the diffusion coefficient,the uptake parameters and the amount of GABAreleased led to similar behaviour: the relation waslinear for most parameter values. Non-linear rela-tions occurred only with high concentrations oftransporter and only at a substantial distance fromthe release site (2mm here). Saturation of theresponse with larger spike numbers (as observedexperimentally) was, however, not obtained withthis model.

The G-protein hypothesis.To investigate apossible intracellular origin of the observed non-linearity, we also investigated a kinetic model ofthe G-protein transduction mechanisms underlyingGABAB responses (see details in ExperimentalProcedures and references therein). The principleof G-protein transduction is schematized in Fig.14A: GABA binds to the receptor, leading to itsactivated form. The activated receptor catalysesthe activation of G-proteins (GDP-GTP exchange)on the intracellular side of the membrane. Active G-proteins (G*) can either affect the gating of thechannel or be degraded back to the inactive form.Simulation of this mechanism is shown in Fig. 14B,with the typical slow time course of G-protein illu-strated for a single presynaptic spike as well as for atrain of 10 spikes at 200 Hz. The peak G-proteinconcentration was proportional to the numberof presynaptic spikes, according to a standardMichaelis–Menten saturation curve (Fig. 14C).This model can result in the saturation of the


Fig. 10. A mixed GABAA/GABAB IPSP in the hippocampus.(A) A burst firing presynaptic CA1 interneuron with a basketcell-like axon (see Fig. 12) elicits in a pyramidal cell, a shortlatency, summing fast IPSP (composite average of responses tosingle spikes, double, triple and quadruple spikes, and brief fivespike trains triggered from the 1st...5th presynaptic spikes,respectively) in control conditions (bold trace). After additionof bicuculline, the fast IPSP is blocked and a slow IPSP, gener-ated by a similar spike train is revealed. (B) The slow IPSPrevealed by blockade of GABAA receptors (normal trace,labelled bicuculline) was greatly reduced by addition of100mM CGP 35348, a selective GABAB antagonist. Two repre-sentative presynaptic spike trains of the types elicited in the

presence of CGP 35348 are shown above. (970319A1.)

Fig. 11. Spike number dependence of the slow component ofthe mixed GABAA/B IPSP. Three typical presynaptic spike trainsare illustrated above and averaged responses to two (normaltrace), five (bold), eight or nine (normal trace), and 10 to 13(bold) presynaptic spikes are compared below. Two presynapticspikes elicited almost no postsynaptic response. An almostmaximal response was elicited by eight or nine spikes. Theresponse to the longer spike trains was, however, prolongedand the IPSP elicited by the longest spike trains (10–13 spikes)

hardly decayed before the end of the train. (970319A1.)

Fig. 12. The CA1 basket cell that elicited a mixed GABAA/B IPSP is drawn in its entirety (× 100 magnification).The axon (blue) ramifies throughout the depth of stratum pyramidale (SP) and the dendrites (red) extend throughstratum oriens (SO), and through stratum radiatum (SR) branching in stratum lacunosum moleculare (SLM). Thedotted lines indicate the borders of these hippocampal layers. The insert shows the position of the interneuron in

the slice. (970319A1.)


Fig

.12

.

response with large numbers of spikes. However, ifthe K1 currents were simply proportional to the peakG-protein concentration, this mechanism alonewould be insufficient to account for the non-linearityobserved experimentally with low spike numbers.

A model with a nonlinear transduction mechan-ism has been proposed based on multiple bindingsites for G-proteins on the K1 channels.9 Using asimilar model here (see Experimental Procedures),the experimental data were used to constrain themodel directly and to estimate (i) to what extend

this model can account for the data and (ii) whatare the optimal values for parameters such as thenumber of binding sites. Fitting was provided byusing directly the information from the paired intra-cellular recordings. The train of action potentialsrecorded in the presynaptic interneuron was usedin the presynaptic compartment of the model, andthe time-course of the simulated postsynapticresponse was compared with the averaged IPSPrecorded in the postsynaptic pyramidal cell (largestdata set available with consistent MP, spike number


Fig. 13. Simulation of the ‘Spillover hypothesis’ for GABAB responses. (A) Extracellular space was simulatedusing a 12× 12 two-dimensional grid consisting of 0.5× 0.5mm extracellular compartments. A release site forGABA is indicated in light grey, as well as two sites in which the transmitter concentration was analysed. (B)Time-course of [GABA] following successive releases occurring at the same site at 200 Hz, shown in the threesites indicated in A. The time-course of GABA is compared in control conditions (Control), following block ofuptake (No uptake) and using a smaller diffusion coefficient (Low diffusion). (C) Transmitter presence 2mm fromthe release site shown as a function of the number of presynaptic spikes. The extracellular GABA concentrationwas integrated and normalized to the value obtained with 20 spikes. The relation was approximately linear forcontrol conditions (Bm�0.1 mM), low diffusion and without uptake (Bm�0). However, a significant deviation

from linearity occurred for high concentrations of GABA transporter (Bm�1 mM).

and frequency was used). Best fits of the model fordifferent numbers of binding sites (n) are shown inFig. 15 (left panel). The minimal error was attained

with n�2, while n� 4 andn�8 showed relativelysmall errors. In addition, a complete spike-number/response amplitude curve, for this pair was used toconstrain the model further (Fig. 15B, mean IPSPpeak amplitudes for a series of data sets with similarMP and firing frequencies, but different numbers ofspikes are plotted). In this case, the optimal fits wereobtained with n�4 and n�8. Using bothconstraints therefore, the number of binding sitesthat was most consistent with the data wasn�4.

The observation, that in neocortex (though not inCA1) slow IPSP duration reached a maximum whichcould not be increased by increasing spike trainduration, was also tested in the model. Althoughthe increase in simulated IPSP duration withincreases in train duration was not linear, it did notsaturate. Thus while the present model reproduceswell the CA1 mixed IPSP data, another componentis required to describe fully the neocortical slowIPSP durations.

For the pooled data plot illustrated in Fig. 16,single sweep records were first separated intosubsets in which spike number and firing frequencywere the same and peak amplitudes of each singlesweep IPSP were measured. Mean peak IPSP ampli-tudes were then plotted against spike number (sepa-rately for each frequency range and for each pair)and peak IPSP amplitudes normalized using thepeak amplitude achieved with a sigmoidal curve fitto the data. The optimal sigmoid for the pooled,normalized data is shown in Fig. 16A. The modelfits to these pooled data show thatn� 4 binding sitesreproduces the data from all three pairs relativelywell (continuous line in Fig. 16B) while a linearmodel (n�1; dotted line in Fig. 16B) had a Michae-lis–Menten type of saturation curve (as in Fig. 14C)and did not capture well the GABAB response curve.The optimal fit shown in Fig. 16B was refined fromFig. 15 (n� 4) by optimizing the value of Kd withsame kinetic parameters (K1...K4, see ExperimentalProcedures and Fig. 15 legend). In this case, themodel captured well both the dependency on thenumber of spikes and the 10-spike IPSP time-course(inset Fig. 16B).

Other models of activation of K1 channels withG-proteins were also tested to fit the nonlinear rela-tionship between number of presynaptic spikes andIPSP amplitude. A Hodgkin–Huxley15 (1952) likemodel, in which each K1 channel hadn independentgates and each gate was opened by G-protein bind-ing (see Experimental Procedures), fit the data opti-mally whenn�4 (not shown). Another model testedwas the allosteric activation of the K1 channel bybinding of G-proteins on the open state of thechannel (see Experimental Procedures). Hereagain,n�4 binding sites gave the best fits to thedata (Fig. 16C). For the three models tested, comput-ing the dose–response relation between simulatedIPSP amplitude and GABA concentration during a1 s extracellular GABA application, led to Hill


Fig. 14. Simulation of the G-protein transduction mechanismunderlying GABAB responses. (A) Scheme of the release ofGABA, activation of the receptor by binding of GABA, andcatalysis of G-proteins into activated form (G*) by the boundreceptor. Subsequently, activated G-proteins may affect thegating of ion channels as well as participating in other biochem-ical mechanisms, or be degraded into inactive form (indicatedby the dashed arrow). (B) Typical slow time-course of G-proteinconcentration following one or 10 presynaptic spikes at 200 Hz.(C) Peak G-protein concentration as a function of the number ofpresynaptic spikes (same parameters as in Experimental

Procedures).

coefficients of 1.8 0.2 for n�4 binding sites (notshown), consistent with the values of 1.4–2.1measured experimentally.33 These models all indi-cate thatn�4 G-protein binding sites can optimallyaccount for the GABAB non-linearity, and that thisconclusion is not dependent on the particularkinetics assumed for the models, within the contextof the present approximations.

DISCUSSION

This paper reports for the first time ‘pure’ slowIPSPs elicited in neocortical pyramidal cells andthe first demonstrably mixed, GABAA/GABAB

receptor-mediated IPSP in a cortical region usingdual intracellular recordings. It also answers anumber of relatively controversial issues. Firstly,the firing of a single hippocampal interneuron canrelease sufficient GABA to activate GABAB recep-tors. Simultaneous firing in several interneuronswith very close terminal zones might activateextra-synaptic receptors, but simultaneous activityis not always a requirement. Secondly, confirmingsuggestions made in studies using less directmethods, some IPSPs in both neocortex and in theCA1 region of the hippocampus were found to bemediated solely by GABAA receptors. Evenprolonged trains of high-frequency firing inducedno discernible postsynaptic response when GABAA

receptors were blocked. Whether the GABAreleased by these terminals canaccess extra-synapticGABAB receptors, but activate them only whenadditional GABA is released by other terminals,was not determined. Thirdly, ‘pure’ slow IPSPscan be elicited in neocortex. In 85 dual recordingsin which one of the neurons was a neocortical inter-neuron, three yielded a slow IPSP, i.e. with a randomselection of interneurons and of putative pyramidaltargets in layer V, a slow IPSP was elicited in one of30 pairs. Since 14 of the 85 pairs in this seriesyielded a fast IPSP, ‘pure’ slow IPSPs occurred inthree of 17 (or approximately one in five) mono-synaptic inhibitory connections studied in layer Vof the neocortex in this series. No mixed IPSPswere encountered in neocortex in this series(although only three fast IPSPs were challengedwith a GABAA antagonist) and no ‘pure’ slowIPSPs have yet been seen in CA1 when the outputsof SP interneurons are studied. Finally, these experi-ments demonstrated that single axon CA1 IPSPs canbe mixed. In two of the eight SP connections chal-lenged with bicuculline or picrotoxin a slow IPSP,which was later blocked by CGP 35348 (in one pair),was revealed.

Disynaptic inhibitory postsynaptic potentials?

One obvious concern is that long latency events,particularly those elicited by repetitive presynapticfiring, might be disynaptic. Disynaptic IPSPs have

been studied extensively in both regions (unpub-lished data), but have distinctly different propertiesfrom the slow IPSPs reported here. Although disyn-aptic IPSPs latencies are a few milliseconds longerand somewhat more variable (2–5 ms after the 1st,2nd or 3rd spike) than those of monosynaptic fastIPSPs (1 ms), fast rising events can clearly be seenon single sweeps. Moreover, the disynaptic IPSPsstudied to date have been readily blocked by 1–5mM bicuculline. Unlike the present slow IPSPs,large, disynaptic IPSPs in neocortex are readilyelicited by two to three spikes and in CA1 by oneto two presynaptic pyramidal spikes in appropriatepaired recordings. Moreover, the amplitude of thesedisynaptic events is not increased by increasingspike number. In addition, with disynaptic IPSPrecordings the presynaptic neuron displays typicalpyramidal behaviour and no filled interneuronalaxons or dendrites have been apparent on subse-quent histological processing.

In this study, both CA1 interneurons mediatingmixed GABAA/GABAB IPSPs and one of theneocortical interneurons eliciting a slow IPSPwere fully recovered histologically and filled inter-neuronal axons and in one case dendrites, wereapparent in the recorded region in the two remainingexperiments. Two of the presynaptic interneuronsin neocortex were fast-spiking layer V cells andunlikely therefore to be pyramidal cells and theIPSP elicited by the third (RS) neocortical inter-neuron was insensitive to bicuculline. The neuronseliciting slow IPSPs in the present study are there-fore likely to be GABAergic interneurons. The ques-tion therefore arises as to whether an interneuron–interneuron–pyramid disynaptic connection couldhave resulted in the present data.

Although a strong IPSP can result in ‘rebound’activation of a postsynaptic cell,7,36 it is unlikelythat this would occur before the presynaptic spiketrain and therefore the inhibition had ended. If,therefore, these slow IPSPs had resulted from adisynaptic connection involving two seriallyconnected interneurons, latencies longer than thepresynaptic spike train would have resulted (for anexample of a summing IPSP recorded in an inter-neuron see Fig. 5). This was clearly not the case.Moreover, single spikes and very brief trainswould have been just as effective as longer trains.

The final possibility explored was electrical gapjunction coupling between the recorded interneuronand others nearby. Two observations suggest thatthis possibility can be discounted in relation to thepresent data; firstly the extremely tight correlationbetween recorded presynaptic spike number andslow IPSP amplitude even when trains with differentnumbers of spikes were elicited by identical currentpulses and secondly the absence of any ‘dye-coupling’ between well filled interneurons and anyother cells in the slice. Gap junctions could be avery effective way of recruiting interneuronal


populations, but do not appear to contribute to thepresent results. It seems likely therefore that the‘pure’ slow IPSPs in neocortex and the mixedIPSPs in CA1 originated from monosynapticconnections.

A proximal location for some GABAB receptor-mediated inhibitory postsynaptic potentials?

The relatively steep voltage sensitivity of one ofthe neocortical slow IPSPs and the basket-like


Fig. 15. Model predictions and fit to experimental data. (A) An averaged IPSP elicited by relatively high-frequency trains each of 10 spikes was best fit by the model whenn was 2 (n is the number of G-protein bindingsites on the K1 channel). A poor fit was obtained whenn was 1. (B) For this pair (960509B2), spike trains ofsimilar frequencies were selected and averaged IPSP amplitudes calculated for subsets that included trains withthe same number of spikes. IPSP mean amplitude was plotted against the number of presynaptic spikes. Themodel best predicted the spike-number dependence of the IPSP amplitude whenn was 4 or 8. The optimalparameters were: Kd�8.52 in all cases. K1�0.024 ms21 mM21, K2�0.033 ms21, K3�0.33 ms21,K4�0.031 ms21 for n�1; K1�0.066 ms21 mM21, K2�0.017 ms21, K3�0.27 ms21, K4�0.044 ms21 forn�2; K1�0.18 ms21 mM21, K2�0.0096 ms21, K3�0.19 ms21, K4�0.060 ms21 for n�4;

K1�0.24 ms21 mM21, K2�0.0066 ms21, K3�0.15 ms21, K4�0.070 ms21 for n�8.

axonal arbour of one of the interneurons mediating amixed CA1 IPSP, demonstrate that slow IPSPs canbe elicited by proximal inputs. Whether GABAB

IPSPs are more likely to occur in distal dendriticregions must await a systematic study using postsy-naptic dendritic recordings. The small amplitude ofthe bicuculline-insensitive component of the mixedIPSP elicited by a bistratified interneuron in CA1may have resulted from its less proximal location,although its time-course cannot be entirelyexplained by dendritic filtering since the GABAA

IPSP from this connection was no slower (width athalf amplitude 42 ms) than many of the ‘fast’ IPSPselicited by basket cells. The present estimate, thatone in five of the relatively proximal inhibitory

connections in layer V of the neocortex results in apure slow IPSP and may be mediated solely byGABAB receptors, indicates that these may not beuncommon occurrences.

GABAA inhibitory postsynaptic potentials indicate arelatively high release probability

Slow IPSPs elicited in thalamic relay cellsrequired longer presynaptic spike trains20 thanthose required in the present study in cortex andhippocampus. Several explanations could be pro-posed to account for this difference. Perhaps themost likely derives from the differences apparentin the frequency-dependent characteristics of fastIPSPs elicited by spike trains in the two studies. Incortical regions, both in this and in our previousstudies,36 fast IPSPs typically displayed paired-pulse depression, indicating, a relatively high prob-ability of GABA release in response to the firstspike. Moreover, failures of transmission are veryrare when large fast IPSPs are studied. In contrast,in thalamus, strong facilitation of fast IPSPs wasapparent, indicating that each contributory synapticrelease site has a low probability of release. Clearly,if repetitive release from the same terminal isrequired for the extracellular GABA transient to beextended and for receptor activation to be prolongedthereby, higher frequencies of firing, or longerpresynaptic spike trains would be required whererelease probabilities were low. That postsynapticcortical inhibitory sites are exposed to repetitiveGABA release at GABAA receptor-mediatedconnections was indicated by the results obtainedwith threshold antagonist doses of bicuculline.These doses could block all response to singlespikes, but a second or third spike in quick succes-sion appeared to release sufficient additional GABA


Fig. 16. Sigmoidal relationship between presynaptic spikenumber and postsynaptic slow IPSP amplitude. In this compo-site plot mean IPSP amplitudes for data subsets involving thesame postsynaptic MP, the same firing frequency and the samenumber of spikes were normalized to the maximum IPSP ampli-tude predicted for that data set (same MP, same firingfrequency), by a sigmoidal curve fit to the spike number/IPSPamplitude plot. (A) Best sigmoid fit to the pooled data. Thetemplate function was y�1/(11 exp[-(x-x0)/K]) and yieldedan optimal fit for x0�7.1 and K�1.4 (units of number ofspikes). (B) Best fit of the standard model withn�4 bindingsites. The optimal fit shown in B was refined from Fig. 15.(n�4) by optimizing the value of Kd with same kinetic para-meters (K1...K4). In this case, the model captured well both thedependency on the number of spikes and the 10-spike IPSPtime-course. The best fit was obtained with the same K1...K4

kinetics as used in Fig. 15. and with Kd optimized(Kd�17.83). (C) Best fit of the allosteric model withn�4binding sites (the optimal parameters were:K1�0.17 ms21 mM21, K2�0.013 ms21, K3�0.17 ms21,K4�0.047 ms21, L�17621, Kd�0.15). Although the kineticmodels in B and C were different, they both gave optimal fits forn�4 binding sites (in both cases, the best fit to the 10-spike

IPSP is shown in the inset).

to compete with the antagonist and partially relievethe blockade.

Strength and time-course of slow, putative GABAB

receptor-mediated inhibitory postsynaptic potentials

The largest single sweep slow IPSPs reported herewere between 4 and 5 mV in amplitude at post-synaptic MPs of257, or252 mV. They can there-fore be large enough to hyperpolarize thepostsynaptic cell significantly and, indeed, at theseMPs could delay spontaneous postsynaptic firing. Inthis they were, however, less effective than manyproximal GABAA IPSPs. What is perhaps rathersurprising is the relatively brief and very restrictedduration of the neocortical ‘pure’ slow IPSPs. Notonly was the entire event only twice the duration ofsome of the broader GABAA IPSPs recorded in theseareas,36 but in neocortex, the slow IPSP could not beprolonged by longer spike trains.

The model that best reproduces the other charac-teristics of the slow IPSP does not predict thisrestricted time-course. An additional mechanism isrequired. Desensitization/inactivation of one of theevents in the cascade might occur, or the pattern ofneurotransmitter release at these slow neocorticalconnections might differ from that at other corticalGABAergic synapses. In a mixed GABAA/B connec-tion, the pattern of GABA release could be assessedfrom the fast IPSPs elicited. Although releasedeclined during high-frequency activity, it rapidlyreached a plateau that was then maintained forlong spike trains. Interestingly, at these mixedCA1 connections, the slow IPSP was not asrestricted in duration as ‘pure slow’ neocorticalIPSPs. Release at these ‘pure slow’ connectionscannot be assessed in the same way and mighthave declined significantly towards the end of longerspike trains.

For significant, long-lasting inhibition, repetitiveactivation and summation of GABAA IPSPs may bemore effective than activation of slow IPSPs. Thefunctional relevance of a brief GABAB componentof a mixed IPSP is unclear. This component wasalmost completely masked under control conditionsby the higher conductance summing fast compo-nents in CA1 pyramids. It remains possible thatthe activation of GABAB receptors results in anadditional postsynaptic effect that is not readilyapparent in electrophysiological recordings.

Interneurons mediating slow inhibition

It is unclear at present whether specific inter-neurons elicit postsynaptic effects only via GABAB

receptors. The three presynaptic neocortical cellsreported here each had different firing characteris-tics. If these correlate with other parameters, morethan one ‘class’ of neocortical interneuron couldactivate pure slow IPSPs. Moreover, the slow

IPSPs recorded in neocortex were ‘pure’ only inthe sense that no GABAA receptor-mediated compo-nent was apparent in the postsynaptic target that wasrecorded. Other targets, that were not recorded, mayhave responded via GABAA receptors or via bothGABAA and GABAB receptors. It remains a possi-bility that some targets are rapidly and profoundlyinhibited with low-frequency presynaptic firing viaGABAA receptors, while others are unaffected in theabsence of repetitive firing and even with long trainsof interneuronal spikes inhibited only for a relativelybrief time. Answers to these questions must awaitmultiple target recordings and/or immunolocaliza-tion of the postsynaptic GABAB receptors. However,the relatively brief events reported here are not theonly inhibitory events elicited by repetitive firing ofa single interneuron in neocortex. Although thepharmacology of the more slowly activating anddecaying events elicited by some fast spiking basketcells that also elicited fast IPSPs, was not deter-mined,36 it remains possible that these events repre-sent a more long-lasting GABAB receptor-mediatedinhibition that can outlast the summed fast IPSP.Conversely, either the present relatively brief, orthe previously reported more prolonged slowcomponents, might have been mediated by peptider-gic receptors.

Computational models of GABAB receptor-mediatedinhibitory postsynaptic potentials

Slow IPSPs were never observed to follow singleor double presynaptic spikes. This observation alonemay explain why more intense activation of inter-neurons has been required in previous studies12

using less direct methods. It is also consistent withthe absence of GABAB components in GABAergicminiature events,27,35suggesting that a single releaseevent is insufficient to activate detectable GABAB

currents. The present observation is also similar toprevious reports of the activation of slow versus fastEPSPs in the myenteric plexus, the slow eventsrequiring at least two and the fast events only onepresynaptic action potential.25 In addition, betweenthree and 10 spikes, all reported slow IPSPsincreased steeply in amplitude, but beyond 10 spikeslittle additional increase in amplitude was apparentin any. Experiments therefore reveal an unusualsigmoidal relation between spike number andresponse amplitude for putative GABAB-mediatedIPSPs.

Two possible hypotheses were investigated toaccount for this spike-number dependence. Firstly,according to the ‘spillover’ hypothesis,12,23 GABAB

receptors would be located extrasynaptically andtheir activation would require significant extracellu-lar accumulation of GABA and therefore the intenseactivation of interneurons. Simulated diffusion ofGABA in a 2-dimensional extracellular space withuptake indicated that accumulation of GABA could


indeed occur at extrasynaptic sites (Fig. 13B).However, the integrated amount of GABA accumu-lated extrasynaptically increased linearly withnumber of spikes for most of the parameters tested.One exception was that with high concentrations oftransporter (, 1 mM) the accumulation of GABA ata distance of 2mm from the cleft showed nonlinearsummation in the micromolar range (Fig. 13C,Bm�1).

Thus, if GABAB receptors are located extrasynap-tically and are sensitive to the micromolar GABAconcentration range and if GABA transporters arepresent at concentrations as high as 1 mM, then the‘spillover’ mechanism may account for some non-linearity in the relationship between GABAB IPSPamplitude and number of presynaptic spikes. Predic-tions of the ‘spillover’ hypothesis are therefore thatGABAB receptors should be located at a significantdistance outside the cleft and that the GABA trans-porters should be present all through the extracellu-lar space at concentrations in the millimolar range. Itmust be noted that this mechanism does not accountfor the saturation of GABAB IPSP amplitudes with10–12 spikes evidenced in the present experiments.A ‘spillover’ model would need additional mechan-isms to explain this observation.

Secondly, the ‘G-protein’ hypothesis9 postulatedthat the nonlinear stimulus-dependence of GABAB

responses was due to the properties of the transduc-tion mechanisms linking GABAB receptors to K1

channels via G-proteins. This mechanism was testedhere using different possible transduction mechan-isms. All models tested, including Hodgkin–Huxleyand allosteric types, predicted that, if the binding ofseveral G-proteins is needed to activate each K1

channel, an optimal number of four binding sitesaccounts for the present experimental data. In addi-tion, this model also reproduced the sigmoidal riseand multi-exponential decay of GABAB IPSPs.

Thus, the ‘G-protein’ hypothesis accounts forboth the shape of the GABAB IPSPs (reported bothhere and previously) and the present spike number/IPSP amplitude relationship, including its satura-tion, with no need for an extrasynaptic localizationof the receptors. If there are multiple G-protein bind-ing sites on K1 channels, a sufficient G-proteinconcentration must be accumulated to evoke detect-able K1 currents. We therefore interpret the absenceof GABAB components in miniature GABAergicevents27,35 and the absence of any discernibleGABAB response following one or two presynapticspikes (present data) as attributable to the fact that asingle release event does not produce sufficientlevels of G-proteins to activate the K1 currents.On the other hand, with multiple high-frequencyspikes, G-proteins reach membrane concentrationsthat can activate K1 currents.

Predictions of the ‘G-protein’ hypothesis wouldbe that the K1 channels underlying GABABresponses should express several binding sites for

G-proteins and that activation of G-protein-depen-dent K1 currents in membrane patches should revealnonlinear dose–response relations. A Hill coefficientgreater than unity has indeed been observed forGABAB responses in dissociated hippocampalneurons33 and is similar to the Hill coefficient of1.8 obtained in the present model with four G-protein binding sites.

This model is also consistent with other modellingstudies of the activation of K1 channels. Since thework of Hodgkin and Huxley,15 a tetrameric struc-ture has been demonstrated for most K1 channelsand shown to account for voltage-dependent orcalcium-dependent properties of various K1 channelsubtypes.13 In the present model, the K1 channelassociated with GABAB receptors could bedescribed as a multimer composed of four identicalsubunits, with one G-protein binding site on eachsubunit. A similar conclusion has indeed beenreached by a combined patch-clamp and modellingstudy of the K1 channels associated with muscarinicreceptors.16 In this case, the channel was found toactivate according to an allosteric model with fourbinding sites. Our results are in perfect agreementwith this type of model, indicating perhaps a prin-ciple valid for all G-protein-activated K1 channels.

CONCLUSIONS

We conclude that some IPSPs in neocortex andhippocampus are mediated solely by GABAA recep-tors, others in neocortex appear mediated solely bynon-ionotropic, possibly GABAB receptors, whilesome in CA1 are mediated by both GABAA andGABAB receptors. The GABA released by a singleinterneuron, provided it fires at least three spikes athigh frequency, is sufficient to activate postsynapticGABAB receptors. The nonlinear relationshipbetween spike number and slow IPSP amplitude isbest predicted by a model in which the non-linearityis due to the transduction mechanisms underlyingthe putative GABAB-mediated responses. The multi-plicity of G-protein binding sites on K1 channels isalone sufficient to account for experimental data. Inthis case, the precise location of the receptors; sub-or peri-synaptic, does not matter if they have highsensitivity to GABA, which does indeed seem to bethe case.33 GABA uptake may contribute an addi-tional non-linearity if transporters are present at highdensity, but an extrasynaptic location for the recep-tors need not be proposed to account for the presentdata.

Acknowledgements—This work was supported by theMedical Research Council, Novartis Pharma (Basel), theWellcome Trust and by The Medical Research Council ofCanada (MT-13724). CGP 35348 was a kind gift fromNovartis Pharma. We are extremely grateful to Mr JoelHahn for histological processing of early neocortical tissueand to Mr Pete Bannister for hippocampal processing andlight level reconstructions.


REFERENCES

1. Ali A. B., Deuchars J., Pawelzik H. and Thomson A. M. (1998) CA1 pyramidal to basket and bistratified EPSPs: dualintracellular recordings in rat hippocampal slices.J. Physiol.507,201–217.

2. Andrade R., Malenka R. C. and Nicoll R. A. (1986) A G-protein couples serotonin and GABA receptors to the same channelsin hippocampus.Science234,1261–1265.

3. Atkins P. W. (1986)Physical Chemistry. 3rd edn. Freeman, New York, NY.4. Benardo L. S. (1994) Separate activation of fast and slow inhibitory postsynaptic potentials in rat neocortexin vitro. J. Physiol.

476,203–215.5. Bowery N. G., Hill D. R. and Hudson A. L. (1983) Characteristics of GABAB receptor binding sites on rat whole brain synaptic

membranes.Br. J. Pharmac.78, 191–206.6. Clark J. A. and Amara S. G. (1994) Stable expression of a neuronal gamma-aminobutyric acid transporter, GAT-3, in

mammalian cells demonstrates unique pharmacological properties and ion dependence.Molec. Pharmac.46, 550–557.7. Cobb S. R., Buhl E. H., Halasy K., Paulsen O. and Somogyi P. (1995) Synchronization of neuronal activity in hippocampus by

individual GABAergic interneurons.Nature378,75–78.8. Crunelli V., Haby M., Jassik-Gerschenfeld D., Leresche N. and Pirchio M. (1987) Cl2 and K1 dependent inhibitory post-

synaptic potentials evoked by interneurones of the rat lateral geniculate nucleus.J. Physiol.399,153–176.9. Destexhe A. and Sejnowski T. J. (1995) G-protein activation kinetics and spill-over of GABA may account for differences

between inhibitory responses in the hippocampus and thalamus.Proc. natn. Acad. Sci. U.S.A.92, 9515–9519.10. Destexhe A., Mainen M. and Sejnowski T. J. (1998) Kinetic models of synaptic transmission. InMethods in Neuronal

Modelling (eds Koch C. and Segev I.), 2nd edn, pp. 1–26. MIT, Cambridge, MA.11. Diamond J. S. and Jahr C. E. (1997) Transporters buffer synaptically released glutamate on a submillisecond time scale.

J. Neurosci.17, 4672–4687.12. Dutar P. and Nicoll R. A. (1988) A physiological role for GABAB receptors in the central nervous system.Nature332,156–158.13. Hille B. (1992)Ionic Channels of Excitable Membranes. Sinauer Associates INC, Sunderland, MA.14. Hines M. L. and Carnevale N. T. (1997) The NEURON simulation environment.Neural Comput.9, 1179–1209.15. Hodgkin A. L. and Huxley A. F. (1952) A quantitative description of membrane current and its application to conduction and

excitation in nerve.J. Physiol.117,500–544.16. Hosoya Y., Yamada M., Ito H. and Kurachi Y. (1996) A functional model for G-protein activation of the muscarinic K1

channel in guinea pig atrial myocytes. Spectral analysis of the effect of GTP on single-channel kinetics.J. gen. Physiol.108,485–495.

17. Isaacson J. S., Solis J. M. and Nicoll R. A. (1993) Local and diffuse synaptic actions of GABA in the hippocampus.Neuron10,165–175.

18. Jarolimek W. and Misgeld U. (1993) 4-Aminopyridine-induced synaptic GABAB currents in granule cells of the guinea-pighippocampus.Pflugers Arch. Eur. J. Physiol.425,491–498.

19. Kehl S. J. and McLennan H. (1983) Evidence for a bicuculline insensitive long lasting inhibition in the CA3 region of thehippocampal slice.Brain Res.279,278–281.

20. Kim U., Sanchez-Vives M. V. and McCormick D. A. (1997) Functional dynamics of GABAergic inhibition in the thalamus.Science278,130–134.

21. Lacaille J. C. and Schwartzkroin P. A. (1988) Stratum lacunosum-moleculare interneurons of hippocampal CA1 region. II.Intrasomatic and intradendritic recordings of local circuit synaptic interactions.J. Neurosci.8, 1411–1424.

22. Lenz R. A., Pitler T. A. and Alger B. E. (1997) High intracellular Cl- concentrations depress G-protein-modulated ionicconductances.J. Neurosci.17, 6133–6141.

23. Mody I., DeKoninck Y., Otis T. S. and Soltesz I. (1994) Bridging the cleft at GABA synapses in the brain.Trends Neurosci.17,517–525.

24. Monod J., Wyman J. and Changeux J. P. (1965) On the nature of allosteric transitions: a plausible model.J. molec. Biol.12,88–118.

25. Morita K. and North R. A. (1985) Significance of slow synaptic potentials for transmission of excitation in guinea pigmyenteric plexus.Neuroscience14, 661–672.

26. Nurse S. and Lacaille J. C. (1997) Do GABAA and GABAB inhibitory postsynaptic responses originate from distinct inter-neurons in the hippocampus?Can. J. Physiol. Pharmac.75, 520–525.

27. Otis T. S. and Mody I. (1992) Differential activation of GABAA and GABAB receptors by spontaneously released transmitter.J. Neurophysiol.67, 227–234.

28. Otis T. S., DeKoninck Y. and Mody I. (1993) Characterization of synaptically elicited GABAB responses using patch-clamprecordings in rat hippocampal slices.J. Physiol.463,391–407.

29. Perez-Pinzon M. A., Tao L. and Nicholson C. (1995) Extracellular potassium, volume fraction, and tortuosity in rat hippo-campal CA1, CA3, and cortical slices during ischemia.J. Neurophysiol.74, 565–573.

30. Press W. H., Flannery B. P., Teukolsky S. A. and Vetterling W. T. (1986)Numerical Recipes. The Art of Scientific Computing.Cambridge University Press, Cambridge, MA.

31. Rusakov D. and Kullmann D. (1998) Extrasynaptic glutamate diffusion in the hippocampus: ultrastructural constraints, uptakeand receptor activation.J. Neurosci.18, 3158–3170.

32. Segal M. (1990) A subset of local interneurons generate slow inhibitory postsynaptic potentials in hippocampal neurons.BrainRes.511,163–164.

33. Sodickson D. L. and Bean B. P. (1996) GABAB receptor-activated inwardly rectifying potassium current in dissociatedhippocampal CA3 neurons.J. Neurosci.16, 6374–6385.

34. Solis J. M. and Nicoll R. A. (1992) Pharmacological characterization of GABAB mediated responses in the CA1 region of therat hippocampus.J. Neurosci.12, 3466–3472.

35. Thompson S. M. and Ga¨hwiler B. H. (1992) Effects of the GABA uptake inhibitor tiagabine on inhibitory synaptic potentials inrat hippocampal slice cultures.J. Neurophysiol.67, 1698–1701.

36. Thomson A. M., West D. C., Hahn J. and Deuchars J. (1996) Single axon IPSPs elicited in pyramidal cells by three classes ofinterneurones in slices of rat neocortex.J. Physiol.496,81–102.

(Accepted8 January1999)


Documents

dual intracellular recordings and computational models of slow