ARTICLES

Cell-specific, spike timing–dependent plasticities in the dorsal cochlear nucleus

Thanos Tzounopoulos1,Yuil Kim1,Donata Oertel2 & Laurence O Trussell1

In the dorsal cochlear nucleus, long-term synaptic plasticity can be induced at the parallel fiber inputs that onto both fusiform principal and cartwheel feedforward inhibitory . Here we report that in mouse fusiform cells, spikes evoked 5 ms after parallel-fiber excitatory postsynaptic potentials (EPSPs) led to long-term potentiation (LTP), whereas spikes evoked 5 ms before EPSPs led to long-term depression (LTD) of the synapse. The EPSP-spike protocol led to LTD in cartwheel cells, but no synaptic changes resulted from the reverse sequence (spike-EPSP). Plasticity in fusiform and cartwheel cells therefore followed Hebbian and anti-Hebbian learning rules, respectively. Similarly, spikes generated by summing EPSPs from different groups of parallel fibers produced LTP in fusiform cells, and LTD in cartwheel cells. LTD could also be induced in glutamatergic inputs of cartwheel cells by pairing parallel-fiber EPSPs with depolarizing glycinergic PSPs from neighboring cartwheel cells. Thus, synaptic learning rules vary with the postsynaptic cell, and may require the interaction of different http://www.nature.com/natureneuroscience transmitter systems.

The mammalian dorsal cochlear nucleus (DCN) is the site of conver- The mammalian DCN resembles the electrosensory lobe of gence of multiple streams of neural information1,2.Disruption of this mormyrid fish, in which STDP has been documented14,15.Both are region causes deficits in the ability to orient the head to sounds3,4. -like structures with similar cell types and possibly similar Spectral information is brought by auditory nerve fibers to the deep computational potential. In the present study, we examined the layer of the DCN; information from diverse sources, notably proprio- dependence of synaptic plasticity on spike timing in the DCN at both ceptive signals that convey head and ear position, are brought principal cells and interneurons. Whereas fusiform cells showed a through granule cells and their parallel fibers to the molecular layer. conventional STDP, plasticity in the cartwheel cell resembled that Fusiform cells, the principal neurons of the DCN, combine input observed in homologous cells in the electric fish. In exploring how from auditory nerve fibers and associated inhibitory interneurons on activation of different groups of synaptic inputs may induce STDP,we smooth basal , as well as from the of granule cells (the also found that depolarizing glycinergic contacts between cartwheel parallel fibers) and associated inhibitory interneurons on spiny apical cells may provide the spike trigger necessary for induction of LTD at © 2004 Nature Publishing Group dendrites (Fig. 1a). Cartwheel cells are glycinergic interneurons parallel fiber . whose sole glutamatergic contacts are from the parallel fibers onto their spiny dendrites5.They are also excited by glycinergic input from RESULTS other cartwheel cells5,6.These cells provide powerful feedforward STDP differs in fusiform and cartwheel cells inhibition to fusiform cells7.Using conventional stimulus protocols Recordings were made from single cartwheel and fusiform cells in coro- that pair parallel fiber activation with various levels of postsynaptic nal slices of mouse brainstem. These cell types were identified by their depolarization, a previous study demonstrated both LTP and LTD in location in the densely cellular fusiform cell layer and by their charac- cartwheel and fusiform cells8.How more natural patterns of activity teristic responses to current injection (Fig. 1a,b). The larger cell bodies trigger synaptic plasticity is not known. in the fusiform cell layer are cartwheel and fusiform cells. Unlike In some neurons, long-term plasticity may be induced by pairing fusiform cells that fire regularly when they are depolarized, cartwheel EPSPs and postsynaptic action potentials in a critical timing relation- cells fire mixtures of simple and complex action potentials5,16,17. STDP ship, a process termed spike-timing dependent plasticity (STDP)9–12. was induced by pairing EPSPs with postsynaptic action potentials Although exact timing relationships vary, a common theme is that evoked by current injection through the recording electrode. Test EPSPs followed within a few milliseconds by spikes lead to LTP, EPSPs were delivered at 0.1 Hz. A pairing protocol consisted of a shock whereas reversing the sequence leads to LTD. In interneurons of the to parallel fibers followed 5 ms later by a suprathreshold current pulse electrosensory lobe of mormyrid electric fish, a different learning rule to the postsynaptic cell, causing the peak of an action potential to be holds, as yet undescribed in any mammalian circuit. In these neurons, produced ∼5 ms after the onset of the EPSP. For the 10-Hz protocol, spikes immediately following EPSPs lead to depression, but reversal of these pairs were delivered five times in 100-ms intervals, and followed the sequence does not produce an associative plasticity13. by a 5 s rest; this set was then repeated nine times (Fig. 2a,b).

1Oregon Hearing Research Center and Vollum Institute, Oregon Health & Science University, 3181 SW Sam Jackson Park Road, Portland, Oregon 97239, USA. 2Department of Physiology, University of Wisconsin Medical School, 1300 University Ave., Madison, Wisconsin 53706, USA. Correspondence should be addressed to T.T. ([email protected]).

Published online 20 June 2004; doi:10.1038/nn1272

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 7 | JULY 2004 719 ARTICLES

a Figure 1 Parallel fibers and their targets in the DCN. (a) Parallel fibers, the axons of granule cells, contact spines on the dendrites of cartwheel cells and apical dendrites of fusiform cells. Cartwheel cells terminate locally with extensive terminal arbors, contacting fusiform cells and other cartwheel cells. (b) Characteristic responses to current injection in cells of the fusiform cell layer: cartwheel cells fire combinations of simple and complex spikes whereas fusiform cells fire regular simple spikes. Cartwheel cells are the only neurons of the DCN that show complex spikes16,17. b

plasticity to parallel fiber stimuli (Fig. 3e). Thus, the learning rules for STDP at parallel fiber synapses depend upon the identity of the post- synaptic . STDP in cortical pyramidal cells and in the mormyrid elec- trosensory lobe requires activation of NMDA receptors10,12,13,18,19. We repeated the EPSP-spike protocol with a 5-ms interval that produced LTP in fusiform and LTD in cartwheel cells under con- At synapses between parallel fibers and fusiform cells, pairing of trol conditions, while NMDA receptors were blocked by 100 µM EPSPs and postsynaptic spikes resulted in bidirectional plasticity. DL-APV. In the presence of DL-APV, no STDP was obtained (cart- When the onset of EPSPs preceded spikes by 5 ms, LTP was induced wheel, 104 ± 4%, n =6;fusiform, 97 ± 4%, n =4,data not shown). that lasted for the duration of the recordings ((post-pairing/con- This result suggests that increases in intracellular Ca2+ are trol)×100 = 134 ± 5%, n = 6; Fig. 2c,d). When the spike preceded the required to induce opposing plasticity in the two different cell EPSP by 5 ms, LTD was observed (73 ± 10%, n = 4). This synaptic types. In STDP of the electrosensory lobe, LTD required pairing plasticity is thus associative, requiring for its induction a specific synaptic stimuli with a broad Ca2+-dependent spike13.Since our http://www.nature.com/natureneuroscience timing and order between pre- and postsynaptic activity. Tests of the induction stimuli triggered broad, Ca2+-dependent, complex timing limits between EPSP and spike required for LTP and LTD spikes in cartwheel cells5,16,20,we asked if simple spikes could also revealed that no synaptic plasticity resulted when the interval trigger plasticity. To induce simple spikes without occasional sec- between EPSP and spike was longer than 20 ms (Fig. 2e). Mean ondary peaks, we followed the depolarizing current step with a delays between the onset of EPSPs and the peaks of the spikes are strong hyperpolarizing one. Figure 3g illustrates the waveform of a shown in Figure 2e. representative simple spike in the same temporal relationship as A different form of STDP was found at synapses between parallel the complex spike shown in Figure 3b.The simple spike supported fibers and cartwheel cells. LTD was observed when the conditioning LTD equally well as the complex spike (Fig. 3d,f). EPSP preceded the spike by ∼5 ms (71 ± 9%, n =9;Fig. 3a–d). When We further explored the properties of cartwheel SDTP by shorten- the spike preceded the EPSP by ∼5 ms, no change in synaptic strength ing the interval between delivery of EPSP-spike pairs from 100 ms to was observed (97 ± 4%, n = 6), despite the similar time course of the 25 ms (40 Hz). At this higher frequency, LTP was observed when the two events. The timing limits between EPSP and spike required for spike followed the EPSP by 5 ms (Fig. 4a,b). Thus, cartwheel cells can LTD were sharp. For 10-Hz pairs, even small deviations from the 5-ms show LTP with EPSP-spike pairs, providing a potential mechanism © 2004 Nature Publishing Group delay reduced or eliminated LTD (Fig. 3e). Comparison of the timing for reversing LTD in these cells8.However, when pairs were delivered limits for plasticity in cartwheel and fusiform cells emphasizes that with the reverse sequence, LTP was still obtained (Fig. 4c,d). Thus, for the 5-ms spike delay, the two cells exhibited opposite long-term STDP is dependent on the frequency of stimulation12.

a b

c Figure 2 STDP in fusiform cells. (a) Example of a cell’s responses to pairing of a subthreshold EPSP with a current-evoked spike delivered 5 ms later. (b) Plasticity was induced by a protocol comprising 5 such pairs delivered at 100 ms intervals followed by a 5-s pause, and repeated a total of 10 times. (c) Examples of averaged EPSPs before and 15–20 min after pairing. (d) The d e average of every 6 responses delivered at 0.1 Hz amplitude was normalized to the average over the pre-pairing time period. Data from six experiments were then averaged and their time course plotted. (e) Timing rule for STDP in the fusiform cell when pairs were delivered at 100-ms intervals. Time points are shown as the average interval between EPSP onset and spike peak measured in each experiment. Horizontal error bars are smaller than the symbols. Data are from 3–6 cells per point.

720 VOLUME 7 | NUMBER 7 | JULY 2004 NATURE NEUROSCIENCE ARTICLES

Figure 3 STDP in cartwheel cells. (a) Example of a cell’s responses to a b pairing of a subthreshold EPSP with a current-evoked spike delivered 5 ms later. (b) Plasticity was induced by a protocol comprising five such pairs delivered at 100 ms intervals followed by a 5-s pause, and repeated 10 times. (c) Examples of averaged EPSPs before and 15–20 min after pairing. c (d) Time course of 9 experiments determined as in Figure 2. (e) Timing rule for STDP in the cartwheel cell (solid line) when pairs were delivered at 100 ms intervals. Time points are shown as the average interval between EPSP onset and spike peak measured in each experiment. Horizontal error bars are smaller than the symbols. Data are from 4–9 cells per point. Timing rule for d e fusiform cells is in gray. (f) Time course for seven experiments using the timing rule in b, but with a simple spike as the trigger. (g) Example of the simple spike generated by following a 1 ms, 700 pA step with a 5 ms, –1 nA step to prevent generation of a complex spike.

STDP induction by synaptically evoked spikes If STDP can be induced by pairing EPSPs with current-evoked spikes, f g can it also be induced by pairing EPSPs with synaptically evoked spikes? Two non-overlapping groups of parallel fibers (‘pathways’) were stimulated with two stimulus pipettes (Fig. 5a). The amplitude of the stimulus to each group was adjusted so that the synaptic responses were subthreshold in current clamp. To assess facilitation,a presynaptic consequence of repetitive stimulation, we then measured synaptic currents under voltage clamp. In all of the pathways tested, the second of a pair of stimuli at a 50-ms interval evoked a larger http://www.nature.com/natureneuroscience synaptic current than the first (Fig. 5b). The ratio of amplitudes of synaptic currents (responses to pulse 1/pulse 2) in cartwheel cells was that pathways were independent, one pathway (defined as Path 1) was 2.0 ± 0.1 (n = 5), and in fusiform cells it was 1.7 ± 0.1 (n = 5). Our cri- stimulated at 0.1 Hz to measure the amplitude of the response in volt- terion for the independence of the two pathways was the absence of age clamp. Pairing was done in current clamp; a stimulus to this path- facilitation between pathways (Fig. 5b) and was confirmed in each way was followed by a stimulus to the other pathway so that summation experiment. We then tested whether LTP or LTD induced by pairing of the two responses always resulted in a spike (Fig. 5c,d). When the EPSPs and spikes in one pathway led to long-term plasticity in a sec- interval between the onset of the Path 1 EPSP and the peak of the spike ond, unpaired pathway, and found that it did not (fusiform cells, 103 was 4–5 ms (4.3 ± 0.2 ms), this pairing protocol induced LTD in Path 1 ± 12%, n =3;cartwheel cells, 96 ± 10%, n =3;data not shown). in cartwheel cells (Fig. 5c,e; 64 ± 9%, n = 5). When the interval between The possibility of manipulating plasticity separately in two pathways the Path 1 EPSP and spike was raised to 8.1 ± 0.1 ms, no plasticity was raised the question of whether interactions between different gluta- observed within 20 min after pairing (91 ± 8%, n =5;data not shown). matergic contacts could induce long-term plasticity. After establishing In fusiform cells, LTP of 127 ± 9% (n = 4) resulted when the delay between Path 1 EPSP and the spike was 8–9 ms (8.6 ± 0.1 ms; upper © 2004 Nature Publishing Group panels in Fig. 5d,f). Unlike the immediate reduction in amplitude observed when EPSPs were paired with current-evoked spikes, the LTD induced in the cartwheel cells by summed EPSPs from two pathways developed more gradually (Fig. 3d). LTD in Path 1 EPSPs was, however, similarly sensitive to 100 µM DL-APV (100 ± 10%, n =3,data not shown). These data indicate that the learning rules for STDP apply also when spikes are generated by the summation of subthreshold synaptic inputs from independent pathways. This conclusion was further tested by examining the plasticity of the EPSPs generated by stimulation of Path 2 in each cell type. The EPSPs generated by Path 2 were succeeded by a spike after a shorter time than those of Path 1 (Fig. 5c,d). The latency of the peak of the spikes after the onset of the Path 2 EPSP was 1.5 ± 0.4 ms for cartwheel cells and 1.3 ± 0.4 ms for fusiform cells. As shown in the lower panels of Figure 5d and f, no persistent, significant plasticity in the EPSP was seen (cartwheel cell, 93 ± 6%; fusiform cell, 107 ± 7%; P > 0.05), as predicted from the EPSP-spike timing relationship defined in Figures 2 and 3 for these cell

Figure 4 Frequency-dependence of STDP. (a) Example of LTP induced with an EPSP-spike latency of 5 ms, and an interval between pairs of 25 ms (see inset). (b) Average data from 5 neurons for the paradigm shown in a. (c) In five neurons, LTP was induced when the spike preceded the EPSP by 5 ms. (d) Average data for final 5–10 min of recordings with EPSP-spike sequence of ±5 ms, and pairs delivered at either 100-ms or 25-ms intervals.

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 7 | JULY 2004 721 ARTICLES

a b

c

d e http://www.nature.com/natureneuroscience

Figure 6 Excitatory glycinergic connections among cartwheel cells. (a) Schematic illustration of connections between cartwheel cells and Figure 5 STDP induced by pairing glutamatergic inputs from different groups their cartwheel and fusiform cell targets. (b) Voltage-clamp, whole- of parallel fibers. (a) Schematic representation of the placement of stimulus cell, patch-recordings from pairs of cartwheel cells reveal synchronous © 2004 Nature Publishing Group pipettes to activate independent groups of parallel fibers to cartwheel (left) or synaptic activity. The bursting pattern suggests that other cartwheel fusiform (right) cells. Path 1 was stimulated with an electrode placed cells are the source of the input5. (c) In extracellular, cell-attached superficially in the molecular layer; Path 2 was stimulated with an electrode recordings, neurons showed either complex (i) or simple (ii) spiking placed deep in the molecular layer. (b) The independence of pathways was activity. Local application of 50 µM glycine with a pressure pulse tested in every experiment by paired-pulse facilitation. The second of a pair of caused increased firing in the complex spiking cell (a presumptive shocks induced facilitation in the responses to stimuli of the same path but cartwheel cell) but shut off spontaneous firing in the simple spiking not when pathways were separate and independent. (c,d) Examples of pairs of cell (a presumptive fusiform cell). (d) In a synaptically coupled pair of responses to independent pathways 1 and 2, in cartwheel and fusiform cells cartwheel cells, a complex spike in the presynaptic cell led to inward respectively. (e) In cartwheel cells such a protocol led to LTD of the Path 1 synaptic currents in the postsynaptic neuron. The presynaptic neuron (upper panel), but no persistent effect in Path 2 (lower panel). (f) In fusiform was recorded in current-clamp; the postsynaptic neurons was held in cells such pairing of parallel fiber inputs evoked LTP of Path 1, but no voltage clamp at –70 mV. The pipette contained 16.7 mM Cl– so that plasticity in Path 2. ECl = –53 mV. Overshooting spikes, but not the later smaller spikes were successfully conducted to the terminals. (e) Current clamp recording from another pair of cartwheel cells shows that the types. Thus, the experiments of Figure 5 show that STDP may arise postsynaptic cell responded with multiple complex spikes to a simple from a critical sequence and timing of activation of different popula- spike in the presynaptic cell. tions of synaptic inputs. Moreover, these experiments show that the learning rules for STDP apply independent of whether spikes are trig- gered by current injection or by the summation of subthreshold EPSPs. common complex-spiking cell (Fig. 6a,b). Microelectrode recordings6 show that the reversal potential for glycine is suprathreshold in cart- LTD induced by pairing glutamatergic and glycinergic PSPs wheel cells (–53 mV) but hyperpolarizing in fusiform cells (–68 mV), Previous studies suggest that cartwheel cells make glycinergic contact resulting from differing intracellular Cl– concentration ([Cl–]) in the on one another5,21.Indeed, we found that in recordings from pairs of two cell types. This difference is maintained in mice as old as 22 d after cartwheel cells while glutamatergic transmission was blocked, it was birth and so is unlikely to reflect the more global elevation of intracel- common to detect synchronous, spontaneous bursts of synaptic cur- lular [Cl–] of immature brain, which declines in auditory neurons by 1 rents in both cells as would be expected if both receive input from a week after birth22.We observed that application of 50 µM glycine

722 VOLUME 7 | NUMBER 7 | JULY 2004 NATURE NEUROSCIENCE ARTICLES

Figure 7 Pairing glutamatergic inputs from a b parallel fibers and glycinergic inputs from cartwheel cells triggered LTD. (a) Schematic illustration of the position of stimulating and recording pipettes. S1 activated glutamatergic parallel fibers in the molecular layer; S2 activated fibers in the deep layer that included the axons of glycinergic cartwheel cells. (b) Example of responses in a cartwheel cell. The two pathways were stimulated in current clamp 4–5 ms apart at c 10 Hz, such that the second response, the glyPSP, always became superthreshold. (c) This protocol d led to LTD. (d) The responses to S1 stimuli were blocked by 20 µM of DNQX, an AMPA receptor antagonist, but not by 0.5 µM of strychnine, a glycine receptor antagonist, indicating that they were glutamatergic and therefore arose from parallel fibers. Responses to S2 were blocked by strychnine, indicating that they were glycinergic and therefore probably arose from cartwheel cells. The slice was bathed in a solution of 1 µM DNQX and 50 nM strychnine to reduce the amplitudes of all synaptic inputs to minimize polysynaptic activity. stimulus (S2) was sensitive to strychnine and that the glutamatergic EPSC evoked by S1 was unaffected (Fig. 7d). We also always confirmed that the EPSC was blocked by DNQX, an antagonist of AMPA receptors increased the firing rate of most cartwheel cells in extracellular record- (Fig. 7d). These experiments demonstrate the generality of the learning ings that did not affect the intracellular [Cl–] (5 of 7 cells; Fig. 6c, rules for STDP; they show that the learning rules apply, regardless of the part i). In contrast to cartwheel cells, application of glycine on pre- spike triggering method. http://www.nature.com/natureneuroscience sumptive fusiform cells inhibited the generation of spontaneous action potentials (6 of 8 cells; Fig. 6c, part ii). To determine whether DISCUSSION cartwheel cells can evoke action potentials in other cartwheel cells, Cell-specific STDP dual recordings were made from pairs of synaptically coupled cart- Long-term plasticity in synaptic responses to parallel fiber activation wheel cells loaded with a pipette whose composition was designed to in the DCN differs as a function of the postsynaptic cells. Although the give a Cl– reversal potential of –53 mV. Figure 6d illustrates an exam- parallel fiber inputs to both fusiform and cartwheel cells are regulated ple of a synaptically coupled cell pair in the presence of glutamate by the temporal relationship of the activation of synapses and postsy- receptor antagonists. The presynaptic cell responded to a brief current naptic firing, they are regulated differently. Parallel fiber inputs to pulse with a complex spike, whereas the postsynaptic cell, whose volt- fusiform cells are strengthened when paired with subsequent postsy- age was clamped at –70 mV, produced short-latency inward synaptic naptic firing and obey Hebbian learning rules, whereas in cartwheel currents after each of the overshooting presynaptic spikes. Figure 6e cells they are weakened when paired with subsequent firing over a sim- shows a recording from a cartwheel-cartwheel pair in which the presy- ilar time period and obey anti-Hebbian learning rules. We have shown naptic cell was hyperpolarized so that the current stimulus produced that the learning rules apply generally and irrespective of how spiking © 2004 Nature Publishing Group only a simple spike. In the current-clamped postsynaptic neuron, the in postsynaptic cells is triggered. Although they are opposite in resulting depolarizing glycinergic PSPs evoked complex spikes, show- fusiform and cartwheel cells, the STDP learning rules nevertheless ing that glycinergic transmission is an effective means of driving net- drive the firing in the principal cells of the DCN in similar directions work interactions among many cartwheel cells. because fusiform cells are a major target of cartwheel cells and are Building on the observation that depolarizing, glycinergic PSPs (here inhibited by them5.The net effect of opposing STDPs in principal cells termed glyPSPs) from cartwheel cells can produce spikes in neighboring and interneurons is that parallel fiber activity is excitatory to fusiform cartwheel cells, we tested whether glyPSP-driven action potentials, when cells either by direct activity on fusiform cell apical dendrites or indi- paired with EPSPs from parallel fibers, are capable of inducing LTD at rectly through reduction of inhibition by cartwheel cells. The fact that the glutamatergic inputs. We stimulated the glutamatergic and glyciner- differential SDTPs may work together highlights the importance of gic inputs independently. One stimulating pipette was positioned in the examining plasticities throughout the circuit as a whole, in order to molecular layer to stimulate the glutamatergic parallel fibers, and the understand the significance of modification of any one synapse. other stimulating pipette was positioned in the deep layer of the DCN to Our results support the general conclusion that short- and long- stimulate axons of other cartwheel cells (Fig. 7a). After independence of term synaptic plasticity of transmission from a common afferent the two pathways was tested, the glutamatergic pathways were stimu- pathway is determined by the postsynaptic target. For example, in the lated at low frequency to obtain control EPSC responses. The two path- hippocampus, different mechanisms of presynaptic plasticity are ways were then stimulated in current clamp 4–5 ms apart at 10 Hz, such found in mossy fiber inputs depending on their postsynaptic target23. that the second response, the glyPSP, always became suprathreshold Moreover, repetitive action potentials in cortical pyramidal cells (Fig. 7b). The mean latency between the onset of the first EPSP and the resulted in facilitation in bitufted GABAergic interneurons and in peak of the spike was 4.9 ± 0.2 ms. This pairing protocol induced LTD depression in multipolar interneurons24. (74 ± 3%, n = 4) in the parallel-fiber input to cartwheel cells (Fig. 7c) Millisecond differences in the relative timing of pre- and postsy- that was similar in magnitude to that observed with LTD induced by naptic activity are critical for the induction, polarity and magnitude paired glutamatergic inputs (Fig. 5e) or by EPSPs and current injection of activity-dependent, long-lasting changes in synaptic efficacy at (Fig. 3) At the end of each experiment, the glycine receptor antagonist other synapses25.In general, synaptic modification is maximal follow- strychnine was applied to verify that the glyPSC evoked by the second ing conditioning protocols that have small differences (0–10 ms)

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 7 | JULY 2004 723 ARTICLES

between pre- and postsynaptic spikes. The temporal window for plas- cues resulting from repositioning of the head or ears1,2.A negative ticity varies in different parts of the brain. In most cases, the relative image may be formed in the DCN, but this would have to involve the order of pre- and postsynaptic spikes determines whether the proto- fusiform cell and its Hebbian SDTP as well. Thus, it may be possible col induces LTP or LTD: LTP is induced when the presynaptic spike that the DCN instead creates a memory trace that sensitizes the sys- precedes the postsynaptic spike, and LTD is induced when the spikes tem to particular profiles of subsequent sensory stimuli. occur in the reverse order (Hebbian plasticity)10,11,18,19,25–27.In some cases LTD depends on the relative timing of pre- and postsynaptic Inhibitory interneuronal networks and anti-Hebbian STDP spikes, but not on their precise order28.By contrast, in medium gan- The involvement of the glyPSP suggests a role for anti-Hebbian STDP. glion cells of the cerebellum-like electrosensory lobe of mormyrid The reversal potential for glycinergic activity is depolarizing in many fish, cells that occupy a similar position in the circuit and bear consid- cartwheel cells but hyperpolarizing in fusiform cells, as a result of dif- erable resemblance to cartwheel cells, LTD is induced when the post- fering intracellular [Cl–] in the two cell types6 (Fig. 6). It has been synaptic spike immediately follows the EPSP13.In those cells, LTP is suggested that depolarizing glyPSPs might allow for context-depend- elicited in the absence of spike-pairing, that is, non-associatively. ent synaptic action6, initiating activity when the cells are quiet, as Elevation of intracellular Ca2+ is required for STDP. In the DCN, shown here, but tempering activity when cells are active (see also induction of LTP and LTD by the pairing of EPSPs and spikes was ref. 37). The unusually depolarized chloride reversal potential (ECl) prevented by blocking NMDA receptors, echoing similar results in represents a unique means to generate a network of inhibitory other mammalian systems as well as in fish electrosensory lobe and interneurons. It differs from that used in subtypes of interneurons in frog tectum10,13,26,29.In the electrosensory lobe, STDP requires a the cerebral cortex and in amacrine cells of retina in which gap junc- broad, Ca2+-induced action potential. LTD in cartwheel cells could tions mediate coupling of otherwise inhibitory cells38,39.The compu- be elicited by simple as well as complex spikes, but electrically remote tational power of feedforward inhibition of principal cells in a neural dendritic compartments might support Ca2+ action potentials even circuit is enhanced by spread of activity among interneurons. On the when a simple spike is recorded at the cell body30.Why do fusiform other hand, excitatory recurrent connections are inherently destabi- and cartwheel cells differ in the polarity of STDP despite a common lizing. In the DCN, instability may be lessened both by the intermedi- stimulus? It has been suggested that the extent of the rise in postsy- ate nature of ECl and the rules that govern bidirectional synaptic http://www.nature.com/natureneuroscience naptic Ca2+ determines whether LTP or LTD is induced31,32. plasticity. Intense activity of parallel fibers should lead to LTP regard- Although it is possible that Ca2+ elevation differs in the two cell types less of spike-EPSP timing relationship; under such conditions cart- with our stimulus protocols, differences are probably more compli- wheel cells will act to temper each other’s activity. When subsets of cated. Fusiform and cartwheel cells differ not only in the shape of parallel fibers are moderately active, their synapses on cartwheel cells their spikes, but also in the expression of calmodulin-binding pro- will be depressed given a particular EPSP/spike timing, selectively teins, mechanisms for sequestration and release of Ca2+ from intra- promoting activation of fusiform cells. Recent work suggests that cellular stores, and other related receptor proteins8,21,33–35.Thus, it antiHebbian mechanisms also serve to equalize synaptic strength seems likely that they may have distinctly different induction mecha- along extended dendrites40. nisms, downstream from the activation of NMDA receptors. Future It is intriguing that quiescent cartwheel cells can be effectively efforts will be needed to define the site of expression of STDP in cart- driven, even by a single presynaptic cartwheel cell, to trigger plasticity wheel and fusiform cells, as well as their dependence on different at the parallel fiber inputs. By having a strong glycinergic input to pro- transduction systems. vide a trigger, the requirement for cooperativity among parallel fibers is reduced. Moreover, the trigger provided by activity in an interneu- © 2004 Nature Publishing Group Functional roles of STDP in the DCN ronal network should also permit cartwheel cells to modify the The role of STDP is likely to depend both on its learning rule and on strength of their parallel fiber inputs in a manner that is dependent on the type of information that is integrated by a given circuit. In visual the timing of activation of other cartwheel cells some distance away. cortical systems, Hebbian STDP seems to strengthen synapses that Thus, STDP may enable these neurons to be sensitive to the particular reflect a particular direction of movement in the visual field36.By sequence of activation of diverse groups of parallel fiber inputs. contrast, in the electrosensory lobe, anti-Hebbian SDTP has been shown to support the storage of temporally precise negative images of predictable sensory input produced by an animal’s own move- METHODS ments14.This negative image is useful in sensitizing the organism to Coronal brain slices were made from ICR mice (postnatal day 18–22). The care unexpected sensory patterns. and use of animals for the study was approved by the Institutional Animal The neuronal organization of the electrosensory lobe of mormyrid Care and Use Committee of OHSU. Single cells were visualized with differen- fish resembles that of the DCN; medium ganglion cells occupy a sim- tial interference contrast optics and recorded using patch pipettes in either ilar position in that circuit as cartwheel cells do in the DCN15.It is voltage- or current-clamp modes. Cells in the fusiform cell layer of the DCN therefore of interest that LTD of parallel fiber-cartwheel cell synapses were identified on the basis of morphological and electrophysiological criteria can be induced by patterns of activity like those used in the elec- (Fig. 1). This layer comprises cartwheel cells that are round and tend to lie trosensory lobe, the first observation of such precise anti-Hebbian superficially, as well as fusiform cells that are larger, often spindle-shaped and SDTP in mammals. The fusiform cell integrates acoustic with tend to lie on the deep edge of the fusiform cell layer. The external solution somatosensory input and is one of the primary output neurons of the contained 130 mM NaCl, 3 mM KCl, 1.2 mM KH2PO4, 2.4 mM CaCl2, 1.3 mM MgSO ,20 mM NaHCO ,3 mM HEPES and 10 mM glucose, and was satu- DCN. In the mammalian auditory system, the function of the integra- 4 3 rated with 95% O2/5% CO2. Pipettes were filled with a potassium-based solu- tion of somatosensory and acoustic inputs is less well understood. It 8 tion containing 113 mM potassium gluconate, 4.5 mM MgCl2, 14 mM has been proposed that it could modify the response to spectral cues trisphosphocreatine, 9 mM HEPES, 0.1 mM EGTA, 4 mM Na-ATP and to take account of the position of the shoulders, head and pinnae, or 0.3 mM tris-GTP at pH 7.3. For experiments in Figures 6 and 7, in which to separate expected, self-generated noise (e.g.,by movement of the glycinergic postsynaptic potentials were recorded, the pipette solution had an ears) from unexpected sounds, or to temper alterations in spectral elevated [Cl–] of 16.7 mM, with parallel reduction in gluconate. Glycinergic

724 VOLUME 7 | NUMBER 7 | JULY 2004 NATURE NEUROSCIENCE ARTICLES

µ µ 12. Sjostrom, P.J., Turrigiano, G.G. & Nelson, S.B. Rate, timing, and cooperativity jointly and GABAAergic inputs were blocked by 0.5 M strychnine and 20 M SR- 95531 during all other experiments. Fiber tracts were stimulated with voltage determine cortical synaptic plasticity. Neuron 32, 1149–1164 (2001). µ 13. Han, V.Z., Grant, K. & Bell, C.C. Reversible associative depression and nonassocia- pulses (100 s, 7–30 V). All experiments were performed at room temperature tive potentiation at a parallel fiber synapse. Neuron 27, 611–622 (2000). (22–25 °C). For extracellular recordings, patch pipettes contained bath solu- 14. Roberts, P.D. & Bell, C.C. Computational consequences of temporally asymmetric tion, and glycine was applied by a nearby pressure pipettes. For current-clamp learning rules: II. Sensory image cancellation. J. Comput. Neurosci. 9, 67–83 experiments, negative current (–10 to –30 pA) was injected to cartwheel cells (2000). 15. Bell, C.C. Evolution of cerebellum-like structures. Brain Behav. Evol. 59, 312–326 to prevent them from firing spontaneously. Input resistance was monitored (2002). from a response to a hyperpolarizing step during each sweep. Experiments 16. Manis, P.B., Spirou, G.A., Wright, D.D., Paydar, S. & Ryugo, D.K. Physiology and were not included if the input resistance changed more than 20% over 50– morphology of complex spiking neurons in the guinea pig dorsal cochlear nucleus. 60 min. Stimulus intensity was set to evoke subthreshold, single-component J. Comp. Neurol. 348, 261–276 (1994). 17. Zhang, S. & Oertel, D. Cartwheel and superficial stellate cells of the dorsal cochlear EPSPs. Action potentials were evoked by somatic current injection. For volt- nucleus of mice: intracellular recordings in slices. J. Neurophysiol. 69, 1384–1397 age-clamp experiments, series resistance was monitored throughout the (1993). experiment from the whole-cell, fast capacitive transient in response to a 18. Debanne, D., Gahwiler, B.H. & Thompson, S.M. Long-term synaptic plasticity 5-mV hyperpolarization after compensation of the fast capacitive transient. between pairs of individual CA3 pyramidal cells in rat hippocampal slice cultures. J. Physiol. 507, 237–247 (1998). Input resistance was calculated from the sustained response to the same step. 19. Feldman, D.E. Timing-based LTP and LTD at vertical inputs to layer II/III pyramidal EPSC amplitude was defined as the mean amplitude during a 1–2 ms window cells in rat barrel cortex. Neuron 27, 45–56 (2000). at the peak of the EPSC minus the amplitude during a similar window imme- 20. Molitor, S.C. & Manis, P.B. Dendritic Ca2+ transients evoked by action potentials in diately before the stimulus artifact. rat dorsal cochlear nucleus pyramidal and cartwheel neurons. J. Neurophysiol. 89, 2225–2237 (2003). Mean, baseline, EPSP slope or amplitude was calculated from 50–100 sweeps 21. Berrebi, A.S. & Mugnaini, E. Distribution and targets of the cartwheel cell axon in immediately before the start of pairing. Post-pairing slope or amplitude was the dorsal cochlear nucleus of the guinea pig. Anat. Embryol. 183, 427–454 calculated from 20–30 min after the end of pairing. Data are reported as mean (1991). ± s.e.m. Statistical comparisons were made by unpaired two-tailed Student’s 22. Ehrlich, I., Lohrke, S. & Friauf, E. Shift from depolarizing to hyperpolarizing glycine action in rat auditory neurones is due to age-dependent Cl– regulation. J. Physiol. t-test in comparing average EPSPs before pairs with a mean of a 7–10 min win- 520, 121–137 (1999). dow taken 20 min after pairing. Identification of LTP or LTD was always based 23. Maccaferri, G., Toth, K. & McBain, C.J. Target-specific expression of presynaptic on changes in mean EPSPs with a significance of P < 0.05. mossy fiber plasticity. Science 279, 1368–1370 (1998). 24. Reyes, A. et al. Target-cell-specific facilitation and depression in neocortical cir- cuits. Nat. Neurosci. 1, 279–285 (1998). http://www.nature.com/natureneuroscience ACKNOWLEDGMENTS This work was supported by National Institutes of Health grant NS28901. We 25. Sjostrom, P.J. & Nelson, S.B. Spike timing, calcium signals and synaptic plasticity. Curr. Opin. Neurobiol. 12, 305–314 (2002). thank G. Awatramani, C. Bell, A. Klug, T. Lu and N. Golding for discussions about 26. Magee, J.C. & Johnston, D. A synaptically controlled, associative signal for Hebbian this study. plasticity in hippocampal neurons. Science 275, 209–213 (1997). 27. Zhang, L.I., Tao, H.W., Holt, C.E., Harris, W.A. & Poo, M. A critical window for coop- COMPETING INTERESTS STATEMENT eration and competition among developing retinotectal synapses. Nature 395, The authors declare that they have no competing financial interests. 37–44 (1998). 28. Egger, V., Feldmeyer, D. & Sakmann, B. Coincidence detection and changes of Received 30 December 2003; accepted 11 May 2004 synaptic efficacy in spiny stellate neurons in rat barrel cortex. Nat. Neurosci. 2, Published online at http://www.nature.com/natureneuroscience/ 1098–1105 (1999). 29. Tao, H.W., Zhang, L.I., Engert, F. & Poo, M. Emergence of input specificity of LTP dur- ing development of retinotectal connections in vivo. Neuron 31, 569–580 (2001). 1. Young, E.D. & Davis, K.A. Circuitry and functions of the dorsal cochlear nucleus. in 30. Pouille, F. et al. Dendro-somatic distribution of calcium-mediated electrogenesis in Integrative Functions of the Mammalian Auditory Pathway (eds. Oertel, D., Popper, Purkinje cells from rat cerebellar slice cultures. J. Physiol. 527, 265–282 (2000). 2+ A.N. & Fay, R.R.) 160–206 (Springer-Verlag, New York, 2002). 31. Neveu, D. & Zucker, R.S. Postsynaptic levels of [Ca ]i needed to trigger LTD and 2. Oertel, D. & Young, E.D. What’s a cerebellar circuit doing in the auditory system? LTP. Neuron 16, 619–629 (1996). Trends Neurosci. 27, 104–110 (2004). 32. Lisman, J. A mechanism for the Hebb and the anti-Hebb processes underlying learn- 3. Sutherland, D.P., Masterton, R.B. & Glendenning, K.K. Role of acoustic striae in hear- ing and memory. Proc. Natl. Acad. Sci. USA 86, 9574–9578 (1989). © 2004 Nature Publishing Group ing: reflexive responses to elevated sound-sources. Behav. Brain Res. 97, 1–12 (1998). 33. Ryugo, D.K., Pongstaporn, T., Wright, D.D. & Sharp, A.H. Inositol 1,4,5-trisphos- 4. May, B.J. Role of the dorsal cochlear nucleus in the sound localization behavior of phate receptors: immunocytochemical localization in the dorsal cochlear nucleus. cats. Hear. Res. 148, 74–87 (2000). J. Comp. Neurol. 358, 102–118 (1995). 5. Golding, N.L. & Oertel, D. Physiological identification of the targets of cartwheel 34. Bilak, S.R. & Morest, D.K. Differential expression of the metabotropic glutamate cells in the dorsal cochlear nucleus. J. Neurophysiol. 78, 248–260 (1997). receptor mGluR1alpha by neurons and axons in the cochlear nucleus: in situ 6. Golding, N.L. & Oertel, D. Context-dependent synaptic action of glycinergic and hybridization and immunohistochemistry. Synapse 28, 251–270 (1998). GABAergic inputs in the dorsal cochlear nucleus. J. Neurosci. 16, 2208–2219 (1996). 35. Molitor, S.C. & Manis, P.B. Voltage-gated Ca2+ conductances in acutely isolated 7. Davis, K.A., Miller, R.L. & Young, E.D. Effects of somatosensory and parallel-fiber guinea pig dorsal cochlear nucleus neurons. J. Neurophysiol. 81, 985–998 (1999). stimulation on neurons in dorsal cochlear nucleus. J. Neurophysiol. 76, 3012–3024 36. Yao, H. & Dan, Y. Stimulus timing-dependent plasticity in cortical processing of ori- (1996). entation. Neuron 32, 315–323 (2001). 8. Fujino, K. & Oertel, D. Bidirectional synaptic plasticity in the cerebellum-like mam- 37. Chavas, J. & Marty, A. Coexistence of excitatory and inhibitory GABA synap1ses in malian dorsal cochlear nucleus. Proc. Natl. Acad. Sci. USA 100, 265–270 (2003). the cerebellar network. J. Neurosci. 23, 2019–2031 (2003). 9. Bell, C.C., Han, V.Z., Sugawara, Y. & Grant, K. Synaptic plasticity in a cerebellum- 38. Veruki, M.L. & Hartveit, E. Electrical synapses mediate signal transmission in the like structure depends on temporal order. Nature 387, 278–281 (1997). rod pathway of the mammalian retina. J. Neurosci. 22, 10558–10566 (2002). 10. Bi, G.Q. & Poo, M.M. Synaptic modifications in cultured hippocampal neurons: 39. Beierlein, M., Gibson, J.R. & Connors, B.W. A network of electrically coupled dependence on spike timing, synaptic strength, and postsynaptic cell type. interneurons drives synchronized inhibition in neocortex. Nat. Neurosci. 3, 904–910 J. Neurosci. 18, 10464–10472 (1998). (2000). 11. Markram, H., Lubke, J., Frotscher, M. & Sakmann, B. Regulation of synaptic efficacy 40. Rumsey, C.C. & Abbott, L.F. Equalization of synaptic efficacy by activity- and tim- by coincidence of postsynaptic APs and EPSPs. Science 275, 213–215 (1997). ing-dependent synaptic plasticity. J. Neurophysiol. 91, 2273–2280 (2004).

NATURE NEUROSCIENCE VOLUME 7 | NUMBER 7 | JULY 2004 725