1The biologically relevant rules of synaptic potentiation were investigated in hippocampal slices from adult rat by mimicking neuronal activity seen during learning behaviours. Synaptic efficacy was monitored in two separate afferent pathways among the Schaffer collaterals during intracellular recording of CA1 pyramidal neurones. The effects of pairing presynaptic single spikes or bursts with postsynaptic single spikes or bursts, repeated at 5 Hz (‘theta’ frequency), were compared.
2The pairing of ten single evoked excitatory synaptic events with ten postsynaptic single action potentials at 5 Hz, repeated twelve times, failed to induce synaptic enhancement (EPSP amplitude 95 % of baseline amplitude 20 min after pairing; n= 5). In contrast, pairing the same number of action potentials, but clustered in bursts, induced robust synaptic potentiation (EPSP amplitude 163 %; P < 0·01, Student's t test; n= 5). This potentiation was input specific, long lasting (> 1 h; n= 3) and its induction was blocked by an antagonist at NMDA receptors (20-50 μM D(-)-2-amino-5-phosphonopentanoic acid; EPSP amplitude 109 %; n= 6).
3Presynaptic bursting paired with postsynaptic single action potentials did not induce input specific synaptic change (113 % in the test input vs. 111 % in the control; n= 8). In contrast, postsynaptic bursting when paired with presynaptic single action potentials was sufficient to induce synaptic potentiation when the presynaptic activity preceded the postsynaptic activity by 10 ms (150 vs. 84 % in the control input; P < 0·01; n= 10).
4These results indicate that, under our conditions, postsynaptic bursting activity is necessary for associative synaptic potentiation at CA1 excitatory synapses in adult hippocampus. The existence of a distinct postsynaptic signal for induction of synaptic change calls for refinement of the common interpretation of Hebb's rule, and is likely to have important implications for our understanding of cortical network operation.
Associative long-term potentiation (LTP) is the dominant model of memory related synaptic modifications in the mammalian brain (Bliss & Lømo, 1973; Bliss & Collingridge, 1993). It has been studied mainly in the hippocampus, a structure of importance for memory (Morris et al. 1983; Squire & Zola-Morgan, 1991). Induction of associative LTP requires activation of the N-methyl-D-aspartate (NMDA) receptor (Collingridge et al. 1983; Bliss & Collingridge, 1993), which serves as a molecular coincidence detector, requiring both presynaptic release of glutamate and postsynaptic depolarization for its activation (Nowak et al. 1984; Mayer et al. 1984). Thus associative LTP obeys Hebb's learning rule (Hebb, 1949), which suggests that when the pre- and postsynaptic elements are active at the same time then the synapse between them will be strengthened. Indeed, pairing of presynaptic and postsynaptic activity can, under some experimental conditions, lead to synaptic potentiation (Wigström et al. 1986; Magee & Johnston, 1997; Markram et al. 1997). However, the physiological activity that occurs during learning behaviours and which produces the critical activation of NMDA receptors, leading to synaptic potentiation in adult hippocampus, has not been determined.
According to a common interpretation of Hebb's learning rule, synaptic potentiation would be expected to occur following temporal coincidence of presynaptic activity and postsynaptic single action potentials. However, when a rat learns about spatial relations during active exploration of an environment, neurons with appropriate place fields, i.e. coding for the current location of the rat in space, and therefore those neurons that are likely to be involved in associative memories, typically show bursting activity repeated at theta frequency (5-12 Hz) (e.g. O'Keefe & Recce, 1993). Perhaps postsynaptic bursts bear a special significance for associative synaptic modification.
We wanted to test directly the common interpretation of Hebb's rule, by investigating whether coincident single pre- and postsynaptic action potentials are sufficient to induce LTP in hippocampal slices from adult rat. In order to investigate whether bursts have a special role in associative synaptic modification, we compared the efficacy of pairing pre- and postsynaptic single action potentials and pre- and postsynaptic bursts in inducing synaptic change using neuronal activity seen during exploratory learning.
Transverse slices (400 μm) from the dorsal hippocampus were prepared from young adult Wistar rats (120-200 g) of both sexes after decapitation under isoflurane-induced anaesthesia. Slices were maintained at 32°C at the interface between humidified carbogen gas (95 % O2-5 % CO2) and artificial cerebrospinal fluid (ACSF) containing (mM): NaCl, 126; KCl, 3; NaH2PO4, 1·25; MgSO4, 2; CaCl2, 2; NaHCO3, 24; glucose, 10; pH 7·2-7·4; and bubbled with carbogen gas.
Synaptic efficacy was monitored in two separate excitatory input pathways onto individual CA1 pyramidal cells. Postsynaptic control was obtained by intracellular recordings made with glass microelectrodes (resistance 100-180 MΩ) containing 1·5 M KMeSO4. Presynaptic control was achieved by stimulating with two lacquer coated tungsten electrodes, placed in the stratum radiatum either side of the recording electrode, to evoke small excitatory postsynaptic potentials (EPSPs) (2-6 mV) at 0·05 Hz. To obtain better control of presynaptic activity, slices were limited to the CA1 field by removing the CA3 field and subiculum.
To monitor synaptic efficacy both EPSP amplitude and initial slope were measured; both measurements gave equivalent results (within ± 2 %). Following a period of stable responses of at least 15 min, a pairing protocol was implemented to the test pathway. The other pathway (control) was not activated during this time. The protocol involved pairing either presynaptic single or triple stimuli at 200 Hz with either postsynaptic single action potentials or bursts. Postsynaptic activity was elicited by intracellular current injections which produced either single postsynaptic action potentials (1 nA, 5 ms) or three postsynaptic action potentials (1 nA, 20 ms). In all cases the presynaptic activity preceded the postsynaptic activity by 10 to 20 ms in order to promote NMDA receptor activation (Debanne et al. 1998). Trains of ten pairings were made at a frequency of 5 Hz. After the pairing we resumed stimulation of each pathway alternately at 0·05 Hz. The stimulation strengths to both pathways remained unchanged throughout the experiment.
Drugs were purchased from Sigma (carbamylcholine chloride (carbachol)), and Tocris-Cookson (bicuculline methochloride and D(-)-2-amino-5-phosphonopentanoic acid (D-AP5)). When used, they were diluted from × 1000 stock solutions and added to the perfusate to the required concentration.
Data were recorded with an Axoprobe-1A amplifier, acquired on line and analysed using Igor Pro software. Some data were also stored on digital audio tapes for subsequent off-line acquisition. All data in the text are presented as percentage of baseline EPSP amplitudes 20 min after pairing. Student's t test was used for statistical analysis.
Experiments were made with both pre- and postsynaptic activity carefully controlled (see Methods). The pairing of single evoked excitatory synaptic events with postsynaptic single action potentials at theta frequency failed to induce synaptic enhancement (EPSP amplitude 95 % of baseline amplitude 20 min after pairing; n= 5; Fig. 1A and Ba). In contrast, pairing the same number of action potentials, but temporally clustered with three pre- and postsynaptic action potentials in each theta cycle, induced robust synaptic potentiation (163 % of baseline amplitude 20 min after pairing; P < 0·01; n= 5; Fig. 1A and Bb). This potentiation was input specific, long lasting (> 1 h; n= 3) and its induction was blocked by an antagonist at NMDA receptors (20-50 μM D-AP5; 109 % of baseline EPSP amplitude 20 min after pairing; n= 6; Fig. 1C). These features are shared with associative LTP induced by high frequency afferent stimulation (Bliss & Collingridge, 1993). Compared with LTP induced by high frequency stimulation (Gustafsson et al. 1989), however, the potentiation studied here developed more slowly. These results demonstrate that temporal coincidence of pre- and postsynaptic single action potentials at theta frequency is not sufficient to induce LTP, but rather that bursts of action potentials are essential.
We asked next whether both pre- and postsynaptic bursting are necessary, or whether bursting either presynaptically or postsynaptically is sufficient, when paired with single action potentials, to induce synaptic change (Fig. 2). We found that presynaptic bursting paired with postsynaptic single action potentials does not induce input-specific synaptic change (113 % in the test input vs. 111 % in the control; n= 8; Fig. 2B). In contrast, postsynaptic bursting when paired with presynaptic single action potentials is sufficient to induce synaptic potentiation when the presynaptic activity precedes the postsynaptic activity by 10 ms (150 vs. 84 % in the control input; P < 0·01; n= 10; Fig. 2C). This potentiation was as large as that induced by postsynaptic bursting paired with presynaptic bursting (151 % in the test input vs. 104 % in the control; P < 0·01; n= 15; Fig. 2D). However, with a longer delay of 20 ms between pre- and postsynaptic activity no change in synaptic efficacy was produced (100 vs. 99 % in the control; n= 5; data not shown). This might be due to the larger amount of inhibition recruited with the longer delay. Thus, the learning rule at CA1 excitatory synapses is that synaptic potentiation is caused by coincidence of presynaptic activity with postsynaptic bursts.
Inhibition is altered during exploration (Moser, 1996), and block of inhibition enhances synaptic plasticity (Wigström & Gustafsson, 1983). Perhaps postsynaptic single action potentials would be sufficient to induce synaptic potentiation under conditions of reduced inhibition. We tested this idea by repeating the experiments in the presence of a GABAA receptor blocker (10 μM bicuculline methochloride). Although under these conditions bursting occurred more readily, and in some cases hyperpolarizing current had to be applied through the recording electrode in order to prevent postsynaptic bursting, we never observed synaptic potentiation without bursting. No input specific synaptic potentiation was observed following pairing of postsynaptic single action potentials with presynaptic single stimuli (the paired input being 108 % of baseline values 20 min after pairing vs. 113 % in the control input; n= 6) or presynaptic bursts of stimuli (106 vs. 104 %; n= 3). By contrast, robust input specific potentiation was observed following pairing of postsynaptic bursts 20 ms after either presynaptic single stimuli (198 vs. 80 %; n= 4) or presynaptic bursts (201 vs. 95 %; n= 3). Therefore, the bursting requirement cannot be due to activation of either feedforward or feedback inhibitory loops during the pairing.
Cholinergic activity is critical for theta activity (Bland, 1986), and cholinergic activation has been reported to enhance synaptic plasticity (Huerta & Lisman, 1993). Pairing experiments performed in the presence of a cholinergic agonist, 20 μM carbachol, failed to reveal any potentiation without postsynaptic bursting. The mean EPSP amplitude from seven cells with only single action potentials postsynaptically during pairing was 101 % of baseline values 20 min after pairing, compared with 174 and 199 % of baseline EPSP amplitude for two cells which showed bursting activity postsynaptically during pairing. These results suggest that postsynaptic bursting is a genuine requirement for induction of associative LTP at these excitatory synapses.
We have demonstrated that temporal coincidence of pre- and postsynaptic single action potentials at theta frequency is not sufficient to induce synaptic potentiation, and rather that bursts of action potentials are essential under our experimental conditions. Furthermore, our results show that presynaptic bursting is neither necessary nor sufficient to induce synaptic plasticity. By contrast, it is postsynaptic bursting, when paired with presynaptic single or bursting stimuli, that is both necessary and sufficient to induce synaptic potentiation. This finding holds both under conditions of reduced GABAergic inhibition and activation of cholinoceptors, suggesting that postsynaptic bursting is a genuine requirement for induction of LTP at CA1 excitatory synapses in hippocampal slices from adult rat. Taken together with the requirement of postsynaptic bursting during induction of associative EPSP-spike potentiation (Jester et al. 1995), the results point towards a vital role for postsynaptic bursting in hippocampal plasticity.
This finding may seem surprising in view of the emphasis that has been placed on the role of presynaptic bursts in the induction of synaptic potentiation in vitro (Larson et al. 1986; Rose & Dunwiddie, 1986; Huerta & Lisman, 1993) as well as in vivo (e.g. Stäubli & Lynch, 1987). However, the postsynaptic activity was not monitored during these experiments, and the presynaptic network activity could not be carefully controlled, so that the essential requirements for synaptic potentiation remained uncovered. Our conclusion may also seem to contradict reports that prolonged episodes of afferent single stimuli delivered at theta frequency are sufficient to induce LTP in mouse hippocampus (Mayford et al. 1995). However, this method of inducing LTP may also require postsynaptic burst firing, since a correlation was found between the induction of LTP by 5 Hz stimulation and the occurrence of postsynaptic complex spike bursting (Thomas et al. 1998). One possible explanation for our failure to detect any potentiation with single spike pairing could be that stimulation of an afferent pathway at 5 Hz induced a weak depression which masked a weak potentiation produced by pairing with postsynaptic single spikes. However, it is unlikely that our afferent stimuli were sufficient to produce depression since long trains of low frequency activity are required to induce long-term depression (Dudek & Bear, 1992; Mulkey & Malenka, 1992). Moreover, consistent with our findings, in paired recordings from synaptically connected rat neocortical pyramidal cells, synaptic plasticity was induced only when the pairing of pre- and postsynaptic single action potentials was repeated at inter-spike intervals of 100 ms or less (Markram et al. 1997). It may well be, therefore, that this synaptic learning rule has validity beyond associative LTP in the CA1 region of the rat. In contrast, the rules seem to be different in developing tissue since associative LTP was induced between pairs of cultured hippocampal neurones using associations of single pre- and postsynaptic action potentials (Debanne et al. 1998; Bi & Poo, 1998). The reason for this difference remains unknown.
What is the mechanism underlying this requirement of burst firing? It is well established that LTP induced by a high frequency train of afferent stimulation depends on postsynaptic NMDA receptor activation (Collingridge et al. 1983; Bliss & Collingridge, 1993), which requires postsynaptic depolarization (Nowak et al. 1984; Mayer et al. 1984). Until recently, this postsynaptic depolarization was thought to arise primarily from the temporal and spatial summation of EPSPs in the dendrites (Bliss & Collingridge, 1993). However, the recent discovery that action potentials initiated at the axon initial segment can backpropagate in the dendrites (Stuart & Sakmann, 1994) provides an additional, or alternative, mechanism which could support the postsynaptic requirement. Our protocol which, unlike induction of LTP by long high-frequency trains of afferent stimulation, involves behaviourally relevant activity, also requires activation of NMDA receptors. It is thus likely that postsynaptic bursting satisfies the depolarization requirement necessary for a critical activation of NMDA receptors, whereas postsynaptic single action potentials do not. Although it might be argued that our afferent stimulation activated feedforward inhibition which might have prevented backpropagation of single action potentials, our findings hold in the presence of an antagonist at GABAA receptors: even under conditions of diminished inhibition single spike pairing was not sufficient to induce synaptic potentiation. Furthermore, under artificial conditions, fast sodium dependent action potentials are not even necessary for induction of LTP, since LTP could be induced by a pairing protocol in the presence of an intracellular blocker of sodium spikes (QX-222) provided that slow spikes were generated during pairing (Kelso et al. 1986), a result consistent with the requirement of burst firing rather than single action potentials during more physiological conditions. However, we cannot exclude the possibility that an additional local dendritic associative learning rule might exist, where a strong local depolarization in the dendrites could induce LTP in the absence of burst firing under conditions of strong perisomatic inhibition.
One might speculate that the rhythmic inhibition during theta activity serves to reset the membrane potential between each processing window (Paulsen & Moser, 1998), preventing depolarization during one theta cycle from influencing the next theta cycle, e.g. by influencing dendritic K+ channels controlling action potential backpropagation. Although we have found that the requirement for postsynaptic bursting holds both under conditions of reduced GABAergic inhibition and activation of one neuromodulatory input system using physiologically relevant activity patterns, we cannot exclude the possibility that under some conditions synaptic potentiation may occur without postsynaptic bursting, and it might be interesting to search for such conditions. Nevertheless, even if a small potentiation might be produced by single spike pairing under some conditions, we would argue that a small potentiation would not necessarily be of relevance for neuronal encoding during behavioural learning.
Whatever the underlying mechanism, if the necessity for postsynaptic bursting is a general requirement for synaptic change in adult animals, we can predict some important consequences for the study of cortical memory mechanisms. Our results challenge the validity of the conventional interpretation of the Hebbian learning rule, and this will have profound implications for our understanding of cortical network operation. Firstly, the bursting requirement might serve to prevent noise which would otherwise arise from spurious associative modifications due to random coincidences between pre- and postsynaptic activity. This requirement would restrict the synaptic changes to those neurons that are strongly activated. Secondly, a postsynaptic burst in one layer of the network presumably corresponds to a presynaptic burst at the next layer of the network, but the exclusive role of the postsynaptic mechanism may prevent interference between different layers of the network. And thirdly, if single action potentials carry information, the possibility exists that memories previously stored as changes in synaptic weights could be retrieved via single action potentials in parallel with the laying down of new memories by bursting activity. In conclusion, the discovery that a specific activity signals synaptic change offers many new opportunities for investigating the relationship between changes in synaptic efficacy and behavioural memory.
We are grateful to Professor A. David Smith and Professor Peter Somogyi for support and useful comments on the manuscript. We thank Dr John Bekkers for introducing us to programming in Igor Pro software and Dr Alan Larkman for kindly lending us micromanipulators. This research was supported by a Wellcome Trust project grant. Additional financial support from The Royal Society and the Medical Research Council is gratefully acknowledged. F.G. P. is an MRC Research Student in the Department of Pharmacology, Oxford, and O. P. holds the Christopher Welch Junior Research Fellowship in Biological Sciences at Wadham College, Oxford.