Spike timing, calcium signals and synaptic plasticity - Semantic Scholar

305

Spike timing, calcium signals and synaptic plasticity Per Jesper Sjöström and Sacha B Nelson* Plasticity at central synapses depends critically on the timing of presynaptic and postsynaptic action potentials. Key initial steps in synaptic plasticity involve the back-propagation of action potentials into the dendritic tree and calcium influx that depends nonlinearly on the action potential and synaptic input. These initial steps are now better understood. In addition, recent studies of processes as diverse as gene expression and channel inactivation suggest that responses to calcium transients depend not only their amplitude, but on their time course and on the location of their origin. Addresses Department of Biology and Volen Center for Complex Systems, Brandeis University, Mailstop 008, 415 South Street, Waltham, Massachusetts 02454-9110, USA *e-mail: [email protected] Current Opinion in Neurobiology 2002, 12:305–314 0959-4388/02/$ — see front matter © 2002 Elsevier Science Ltd. All rights reserved. DOI 10.1016/S0959-4388(02)00325-2 Abbreviations AMPA α-amino-3-hydroxy-5-methyl-4-isoxazole proprionate APs action potentials CaM calmodulin CaMKII CaM-dependent protein kinase II CICR Ca2+-induced Ca2+ release CREB cAMP response element binding protein EGTA ethylene glycol-bis(β-aminoethyl ether)-N,N,N’,N’-tetraacetic acid EPSP excitatory postsynaptic potential FRET fluorescence resonance transfer IA transient potassium current IH hyperpolarization-activated cation current IPSPs inhibitory postsynaptic potentials L2/3 cortical layers 2 and 3 L5 cortical layer 5 LTD long-term depression LTP long-term potentiation MAPK mitogen-activated protein kinase mGluR metabotropic glutamate receptor NMDA N-methyl-D-aspartate STDP spike timing-dependent plasticity VDCCs voltage-dependent Ca2+ channels

Introduction It seems obvious that the mechanisms underlying memory should permit us to remember the sequential order in which events occur. Given the importance of remembering sequences for everything from recalling phone numbers to playing the piano, it is perhaps surprising that it is only within the past few years that temporal order has been appreciated as a key determinant of synaptic plasticity. Although the role of timing was appreciated in some older studies [1–3], the current reawakening of interest in this topic stems from a pair of papers that appeared in 1997 [4,5]. Together, these papers and some important additional studies by other groups [6–9] demonstrate that the same presynaptic and postsynaptic action potentials (APs) can

lead either to long-term potentiation (LTP) or long-term depression (LTD) depending on their order. At most synapses studied, LTP is produced when presynaptic firing repeatedly precedes postsynaptic firing by 10–15 ms or less (pre-before-post), whereas LTD is produced when postsynaptic firing repeatedly occurs before presynaptic firing (post-before-pre). This temporal relationship ‘rewards’ presynaptic inputs that contribute to causing subsequent postsynaptic firing, while ‘punishing’ inputs that, because they occur later, do not contribute to causing postsynaptic firing. These and other important functional consequences and computational properties of ‘spike timing-dependent plasticity’ (STDP) have recently been explored [10,11,12••,13••,14,15••]. In this review, we focus on the signaling mechanisms that might permit the same two events — an excitatory postsynaptic potential (EPSP) and a postsynaptic AP — to have dramatically different consequences depending on their temporal order.

Regulation of back-propagating action potentials Because postsynaptic APs are usually initiated at the axon initial segment, they must first propagate through the dendritic tree — in a manner termed ‘back-propagation’ — before they can influence events at the synapse (reviewed in [16,17]). Recent work has clarified some of the mechanisms that regulate this key initial step in the induction of STDP. Modulation of dendritic currents

Dendritic membranes contain a rich array of voltage-gated channels that allow them to actively propagate APs. Initial studies of how APs and EPSPs interact focused on the role of the transient potassium current IA [18,19]. IA attenuates back-propagating APs in CA1 neurons of the hippocampus and is inactivated by depolarizing synaptic input. This provides a mechanism for appropriately timed EPSPs to ‘boost’ the back-propagating spike [4]. This view has been elaborated upon in the past year. First, it was demonstrated that, in CA1 dendrites, IA channels inactivate faster than do Na+ channels, so that the Na+-to-IA current ratio increases with the duration of a subthreshold depolarization [20]. Brief excitatory synaptic activity thus promotes AP back-propagation, enabling dendritic AP amplitude to reflect the level of excitatory synaptic activity. Second, the reliability of AP back-propagation into distal CA1 dendrites depends critically on the expression levels of Na+ and IA channels: AP back-propagation is unreliable below a threshold Na+-to-IA channel ratio [21]. These studies highlight the involvement of IA channels in regulating AP back-propagation, but also emphasize the interplay between multiple dendritic conductances in controlling propagation. Other conductances likely to contribute include the hyperpolarization-activated cation current IH [22] and Ca2+ currents [23••].

306

Signalling mechanisms

Figure 1

Figure 1 legend AP amplification, LTP of L5-to-L5 synapses, and excitatory all-or-none responses exhibit supralinear summation. (a) The back-propagating AP attenuates and fails as it invades the L5 apical dendrite. Propagation can be rescued by depolarization sufficient to recruit INa [26••]. In the dendrite, the threshold for AP boosting is approximately 13 mV, which corresponds to 3.6 mV in the soma [26••]. (b) Coincidence of EPSPs and APs produces LTP at L5-to-L5 synapses, but only if depolarization measured at the soma exceeds 2.3 mV [27••]. Strong synapses, which provide sufficient depolarization, potentiate at low frequency [6–8,34], whereas weak synapses need high-frequency spiking [5]. The net result is a novel form of cooperativity [27••]. (c) Glutamate uncaging produces all-or-none responses in terminal dendrites mediated by NMDA and VDCCs [41,42••]. The responses become supralinear above ∼3 mV depolarization measured at the soma. Reproduced with permission from [26••],  2001 Nature Publishing Group, [27••], and [42••],  2001 American Association for the Advancement of Science.

(a)

Amplitude (mV)

60

30

0 0

10

20

Dendritic EPSP amplitude (mV)

propagation [23••] but the mechanism relies primarily on Na+ channel activation, rather than on IA channel inactivation [26••]. The timing requirements for AP amplification by EPSP–AP coincidence closely match those previously reported for LTP in these neurons [5], suggesting an involvement of AP amplification in long-term plasticity. Further support for this notion comes from the observation that both AP amplification (Figure 1a) and LTP [27••] (Figure 1b) and require depolarization that exceeds a threshold value. Because of dendritic attenuation, this threshold is quite modest when measured at the soma.

(b)

LTP [%]

150

100

0.25

1

4

16

Total depolarization [mV]

(c)

Not only can synaptic activity enhance back-propagation, it can also block back-propagation. This can occur through dendritic hyperpolarization [28] or by reducing input resistance [28,29]. The emerging view is that a complex balance between multiple opposing conductances determines the integration of synaptic inputs and back-propagation. The significance of dendritic morphology

∆V (mV)

10

5

0 1

10 Laser duration (ms)

In cortical layer 5 (L5) neurons, IA channels are expressed at lower levels than in the hippocampus and do not show the spatial gradient previously observed in CA1 [24,25]. In these neurons, EPSPs are also capable of boosting

The dendritic arborization of different cell types differs widely. Vetter et al. [30•] reported that when computational models of eight different neuronal reconstructions were given the same conductance distributions, the reliability of AP back-propagation varied extensively. For example, in substantia nigra dopamine neurons, back-propagating APs hardly attenuated at all, whereas in Purkinje cells, APs failed within 50 µm of the soma. Furthermore, the mode of propagation in these same two cell types was relatively insensitive to changes in the ratio of Na+ and K+ conductances, suggesting the AP back-propagation mode was largely determined by dendritic morphology. Conduction can also vary significantly in different portions of the dendritic arbor of a single cell. The key determinant appears to be the degree of dendritic branching. Highly branched dendrites are more prone to failure, whereas relatively unbranched dendrites more readily conduct APs (Figure 2a). Interestingly, pyramidal neuron morphologies were among the most sensitive to changes in the distribution of dendritic ion channels, suggesting an enhanced capacity for modulation of propagation in these neurons [17,30•].

Spike timing, calcium signals and synaptic plasticity Sjöström and Nelson

Consistent with this, apical dendrites of pyramidal neurons both in hippocampus [21] and neocortex [23••] vary widely in the efficiency with which they propagate APs, and this variance is not accounted for by individual differences in morphology.

Coincidence detection At excitatory synapses onto hippocampal and cortical pyramidal neurons, pre-before-post spiking within a 10 ms timing window gives rise to LTP, whereas post-before-pre firing produces LTD [5–8,27••]. Mechanistically, the timing requirement for LTP is thought to depend on supralinear summation of the Ca2+ responses due to coincidence of EPSPs and back-propagating APs [4,31–33]. In contrast, when EPSPs follow APs, sublinear Ca2+ summation has been observed, although the timing of these nonlinear effects does not always perfectly match the timing windows over which plasticity is induced [31]. What mediates the supralinear response? As noted above, the voltage produced when back-propagating APs and EPSPs interact is itself highly nonlinear. At individual spines of pyramidal neurons, the source of the Ca2+ influx is primarily through N-methyl-D-aspartate (NMDA) receptors and voltage-dependent Ca2+ channels (VDCCs) [4,31–33]. Due to the multiple gating properties of these channels, the flux of Ca2+ is a nonlinear function of time and voltage. It is not yet clear, however, whether or not these nonlinearities are sufficient to account for the features of the spike timing curve (Figure 3b). For example, given the fact that glutamate can remain bound to the NMDA receptor for ≥ 50–200 ms (depending on subunit composition), why is the LTP window so short? One as yet untested possibility is that the window is set in part by the time course of the AMPA (α-amino-3-hydroxy-5-methyl-4isoxazole proprionate) receptor-mediated EPSP. Another potential site of coincidence detection entails Ca2+-induced Ca2+ release (CICR) from internal stores. Although there is some evidence that internal Ca2+ stores are involved in hippocampal LTD [34], a role for internal stores in hippocampal LTP remains controversial (for a recent review, see [35]).

307

fiber stimulation levels were small and physiologically relevant; with larger stimulation levels, VDCCs produced Ca2+ signals in whole branchlets, obviating the need for Ca2+ release from internal stores. Multiple mechanisms have also been observed for long-term plasticity in other systems. In the hippocampus, large EPSPs favor an NMDA receptor-dependent form of LTD, whereas smaller EPSPs favor an mGluR-dependent form of LTD [39]. Similarly, in young Xenopus laevis tadpoles, LTP by theta burst is not input-specific, whereas LTP by AP–EPSP coincidence is [40•]. These studies highlight the importance of distinguishing plasticity mechanisms that may be engaged at a unitary synaptic connection from those that may require coordinated activation of many synapses. Strong excitatory synaptic activation of cortical L5 basal dendrites was previously shown to produce localized NMDA spikes that are subthreshold to somatic APs [41]. In the past year, Wei et al. [42••] demonstrated that glutamate uncaging produced similar all-or-none Ca2+ responses that remained localized to distal CA1 apical dendritic branchlets, although these responses depended on VDCCs in addition to NMDA receptors (Figure 1c). Although most have focused on the increased computational power obtained by such binary responses (e.g. [43,44]), it is possible — considering the involvement of the NMDA receptor — that strong excitatory afferent coincidence and binary subthreshold responses can produce long-term plasticity in the absence of postsynaptic spiking. Although a majority of studies on STDP report that pre-before-post spiking results in LTP, and post-before-pre spiking causes LTD [5–9,27••], there are a few notable exceptions to this rule [45,46]. Interestingly, in a recent report concerning cortical layers 2 and 3 (L2/3) pyramidal neurons, the timing requirements of inhibitory connections were demonstrated to be opposite to those of most excitatory connections [47•]. It remains elusive how this ‘inverse’ learning rule is produced mechanistically, but perhaps it makes sense, from a functional point of view, that inhibitory connections employ a learning rule of opposite temporal requirements as compared to excitatory connections.

Reading out a calcium coincidence In cerebellar Purkinje cells, on the other hand, CICR from internal stores appears to be crucial for supralinear Ca2+ summation and for the expression of cerebellar LTD (reviewed in [35]). Ca2+ responses to trains of parallel fiber stimulation are mediated by an early influx through VDCCs and Ca2+-permeable non-NMDA glutamate receptors, and by a later metabotropic glutamate receptor (mGluR)-dependent Ca2+ release from inositol triphosphate (IP3)-sensitive stores [36,37]. Climbing fiber stimulation at the end of trains of such parallel fiber stimulation produces supralinear spine Ca2+ responses that depend on mGluRs and internal Ca2+ stores [38]. Strong spine Ca2+ signals lead to LTD induction. Importantly, Ca2+ responses were only dependent on release from internal stores when parallel

Synaptic plasticity relies on the timing and rate of calcium influx

Regardless of the precise molecular mechanism of coincidence detection, it is virtually certain that the primary output of this detection is a change in [Ca2+]i. A widely held view is that the sign of plasticity is determined by the amplitude, and perhaps time course of the change: smaller, slower increases in [Ca2+]i give rise to LTD, whereas larger, more rapid increases cause LTP (Figure 3a) [48–51]. Two groups have recently tried to test this idea directly. The first [52•] stimulated neurons in perirhinal cortex at low frequency while varying depolarization and the concentration of internally perfused Ca2+ buffers. Although uncertainty remained about buffer equilibration given the short perfusion

308


Figure 2

(a)

(b) Hyperpolarizing

Depolarizing

GABA

NMDA

ICa

IK

INa IA

IH

Current Opinion in Neurobiology

Dendritic branching patterns and ion channel distributions regulate AP back-propagation and resulting Ca2+ influx. (a) In L5 neurons, the back-propagating AP fails ∼450 µm from the soma (black arrow), where dendritic branching is less prominent (green circle). This AP attenuation is a prerequisite for AP amplification by depolarization (Figure????••]. The dependence of L5-to-L5 LTP on depolarization (Figure 1b) [27••] could be explained by a mechanism such as AP amplification. However, L5-to-L5 connections are mostly found on basal dendrites [80], so this hypothesis predicts that APs fails more proximally in basal dendrites. The more finely branched basal arborization (compare red and green circles) may dissipate current, causing APs to attenuate more rapidly [30•]. (b) Dendritic currents dynamically regulate AP back-propagation [16,17]. Inactivation of the

transient K+ current IA, which otherwise attenuates APs, may be involved in AP boosting by concurrent EPSPs [18–21]. Fast Na+ channels are necessary and sufficient for AP amplification in cortical L5 neurons [26••]. More generally, interactions between APs and EPSPs depend upon a balance of inward and outward currents: relative levels of INa and IA control the reliability of back-propagation [21] and this ratio is dynamically regulated by depolarization [20]. Temporal summation of synaptic activity may be regulated by the hyperpolarization-activated cation current IH [22]. Inhibitory γ-amino butyric acid (GABA) synapses can block back-propagation [28], whereas NMDA [41] and Ca2+ [42••] channels can produce all-ornone subthreshold responses (Figure 1c). Finally, Ca2+ currents may be involved in linking the soma and the distal apical dendrite [23••].

times, Cho et al. [52•] found a U-shaped relationship between the strength of LTD and the presumed [Ca2+]i. LTD was absent with very strong buffering, or with strong stimulation in the presence of minimal buffering. Their results also suggest the existence of a ‘no-man’s land’ region, in which Ca2+ levels are too high to cause LTD but too low to cause LTP. The second group [53•] varied the strength of glutamate iontophoretically applied to CA1 neurons and directly measured [Ca2+]i. They also concluded that smaller increases in [Ca2+]i (180–500 nM) produced LTD, larger increases (>500 nM) caused LTP, whereas intermediate increases (450–500 nM) more often resulted in no plasticity.

If strong, rapid Ca2+ elevations produce LTP and slow, prolonged increases in Ca2+ concentration cause LTD [48–51], how can neurons distinguish the two messages produced by the same messenger? Recent evidence suggests the intriguing possibility that different temporal profiles of [Ca2+]i may be read out differentially by calmodulin (CaM). DeMaria et al. [54••] demonstrated that the two lobes of CaM respond to [Ca2+]i elevations with different kinetics and initiate opposing effects on P/Q Ca2+ channels. The carboxyl (C) terminal lobe is sensitive to fast, strong [Ca2+]i elevations and promotes facilitation of Ca2+ channels, whereas the amino (N) terminal responds best to slower, smaller increases in [Ca2+]i and promotes Ca2+ channel inactivation.


309

Figure 3 Ca2+ levels and long-term plasticity. (a) Low-frequency stimulation produces small Ca2+ elevations, whereas high-frequency stimulation evokes rapid, strong Ca2+ increases [48–51]. Although LTD typically requires low-frequency stimulation, highfrequency stimulation results in LTP [81,82], consistent with the Ca2+ levels evoked by these stimuli. (b) Pre-before-post spiking produces supralinear summation of EPSPs and APs, resulting in rapid, strong Ca2+ elevation [4,31–33] and LTP (e.g. [5,9]), whereas sublinear Ca2+ responses may arise from post-before-pre firing [31]. It should be noted that if Ca2+ concentration is what

(a)

(b)

[Ca2+] LTP LTP LTD

LTD

Frequency

Postbeforepre

LTP LTD

Prebeforepost ?

Rest Current Opinion in Neurobiology

matters regardless of source or influx rate, then pre-before-post timings outside the LTP window should produce LTD (region illustrated

Key targets of CaM or plasticity are, of course, the CaM-dependent kinases, especially CaM-dependent protein kinase II (CaMKII) and the CaM-dependent phosphatase PP2B (calcineurin). It has long been hypothesized that the different frequency ranges over which LTP and LTD are readily induced reflect a frequency-dependent balance between the activity of CaMKII and associated phosphatases [49]. According to this view, kinase activity exceeds phosphatase activity at high frequencies, causing autophosphorylation of the enzyme and persistent Ca2+independent activity [55,56]. Although this idea has been tested in vitro [57], it has only this year been directly addressed in neurons. The experiments reveal that, at least in cultured sensory neurons, Ca2+-independent activity was nearly maximal at all but the lowest firing frequencies tested (0.1–1 Hz) [58]. It remains to be seen, however, whether or not a similar frequency range holds for central neurons and for all pools of CaMKII within the cell. STDP protocols emphasize timing, rather than rate, but it is now clear that both these aspects of activity jointly determine plasticity. At low frequency, LTP induction requires multiple inputs or strong somatic depolarization [27••]. This novel form of cooperativity may reflect a need for sufficient dendritic depolarization to ‘boost’ the backpropagating spike (see above). The absence of this cooperativity in culture may reflect the much larger unitary synaptic inputs typically observed [6,7]. At higher frequencies, a similar ‘boosting’ is supplied by the residual depolarization provided by preceding spikes within a burst. Removing this depolarization by hyperpolarizing the neuron between APs blocked plasticity even when presynaptic and postsynaptic firing occurred with optimal timing and frequency. Frequency-dependent boosting could also underlie recent reports that postsynaptic bursting is required for LTP induction in hippocampus (reviewed in [59]). Intriguingly, spike-timing dependent LTD shows a very different pattern of voltage and frequency dependence. Low-frequency post-before-pre pairing results in robust

by the questionmark). This second LTD window has been found in CA1 [34], but has not been reported in other regions [6,7–9,27••].

LTD regardless of the size of the input, whereas high-frequency (≥40 Hz) pairing fails to produce LTD, instead it produces LTP [27••]. The absence of LTD at high frequencies reflects the fact that although each postsynaptic spike occurred 10 ms prior to a presynaptic spike, it also occurred 15 ms after the preceding presynaptic spike, hence the timing requirements for LTP were also met. These results are consistent with the idea of distinct Ca2+ thresholds for LTD and LTP. Biophysical models of synaptic plasticity

One means of exploring the implications of multiple Ca2+ plasticity thresholds is through the construction of detailed biophysical models [56,60]. A particularly detailed set of simulations [61] has addressed the question of why STDP is frequency-dependent. In these simulations, Ca2+ influx, though timing-dependent, was found to be roughly independent of frequency. On the other hand, the kinetics of endogenous Ca2+ buffering caused CaM binding and therefore subsequent CaMKII activation to be strongly frequency-dependent. There are, however, some reasons to suspect that this simple ‘multi-level’ hypothesis will not adequately account for all aspects of STDP. One problem is that it predicts that strong postsynaptic depolarization without accompanying presynaptic activity should be sufficient for inducing plasticity, yet this is not observed [62,63]. A second, subtler prediction of the ‘multi-level’ hypothesis is that there should be not just a single LTD timing window, but two LTD timing windows. One in which the postsynaptic cell fires first and a second in which the presynaptic cell fires first but at longer delays than give rise to LTP (Figure 3b). This is expected because, at increasingly longer delays between EPSP and the subsequent spike, the current through the NMDA channel will diminish (due to unbinding, and perhaps to inactivation). One recent report provided evidence for a second LTD window at CA1 synapses [34], but other STDP studies failed to observe this. It should be pointed out, however, that some of these studies may not have adequately

310


sampled the longer pre-before-post intervals, and hence may have missed the putative second window. The importance of effector molecules

An attractive resolution to some of the problems raised by the ‘multi-level’ hypothesis is the idea that plasticity induction is regulated not merely by the level and timing of Ca2+ entry, but also by its geometrical relationship to effector molecules. A wealth of recent evidence suggests that signaling molecules activated by Ca2+ are organized into macromolecular complexes that permit Ca2+ entry through different routes to have potentially different effects. In one technically impressive study, a new triple-labeling fluorescence resonance transfer (FRET) technique was used to show that CaM, previously shown to regulate both activation and inactivation of Ca2+ channels [64,65], is preassociated with VDCCs [66••]. CaM also participates in regulation of NMDA receptor desensitization [67], a process that appears to place Ca2+ entry through NMDA receptors under direct, rapid feedback control [68•]. The NMDA receptor binds directly to CaMKII, and this has now been shown to enhance translocation and Ca2+-independent activity of the enzyme [69••]. Local Ca2+ gradients near NMDA receptors regulate activation of extracellular signal-regulated kinase, an important signal for long-term activity-dependent changes in nuclear gene expression [70•]. A similar CaM-dependent local mechanism has recently been shown to control mitogen-activated protein kinase (MAPK)-activated gene expression following Ca2+ entry through L-type Ca2+ channels [71••]. Taken together, the observation that CaM produces opposing effects depending on the kinetics of the Ca2+ signal [54••], and that it participates in multiple local and therefore source-specific signaling complexes, may help resolve the conundrum of how millisecond differences in the timing of EPSPs and APs can produce opposing forms of plasticity. Differences in associated signaling molecules may account for the observation that LTP can be induced by activation of synaptic NMDA receptors in culture, but that following activation of nonsynaptic NMDA receptors only, LTD results [72•]. Moreover, such differences may explain the observation that uncaging glutamate in slice preparations, which is expected to activate nonsynaptic receptors, readily produces LTD, but does not cause LTP, even when paired with strong postsynaptic depolarization [73,74]. Despite growing support for a more local view of the effects of Ca2+ on plasticity, some pieces of evidence remain hard to account for. Most notable is the fact that EGTA (ethylene glycol-bis[β-aminoethyl ether]-N,N,N’,N’-tetraacetic acid), a Ca2+ chelator with relatively slow binding kinetics, readily blocks induction of LTP at many synapses [52•,75]. This may imply that Ca2+ should have ample time to diffuse away from the site of entry prior to activating downstream effectors responsible for LTP induction. Alternatively, LTP induction may require both local and

more global Ca2+ signaling. For example, EGTA, by changing [Ca2+] more globally, may disrupt Ca2+-dependent binding or phosphorylation within synaptic signaling complexes required for, but not sufficient for, expression of LTP. Resolution of these questions will require a combination of physiology, imaging and the targeted disruption of binding between channels and complexed signal transduction molecules. Detailed simulations, particularly those in which geometric relationships within the synapse are preserved, may also contribute [61].

Conclusions and future directions Spike timing based protocols are attractive because they hold the potential to link physiological patterns of presynaptic and postsynaptic activity to functional plasticity such as sequence learning [76] and receptive field plasticity [12••,13••,15••]. Progress in making such a linkage compelling depends in part on further understanding of how the diversity of protocols and preparations used in plasticity studies relate to one another. For example, how are STDP and ‘classical’ forms of LTP and LTD related? What molecular mechanisms account for the diversity of learning rules apparent at different classes of synapses even within the same neural structure? Although elevation of intracellular Ca2+ plays a nearly ubiquitous role, it remains unclear which forms of plasticity are mediated by global changes in dendritic Ca2+ and which require local signaling through specific macromolecular complexes. Significant progress has been made in identifying the signaling mechanisms, most dependent on CaM, that permit highly selective responses to Ca2+ signals that differ in amplitude or time course, or that arise from different sources. However, the precise roles of these signaling mechanisms in STDP remain unknown. Finally, it is worth noting that, with the exception of timing-dependent LTD in the cerebellum [77,78•] and at one class of neocortical synapse [46], virtually nothing is known about the expression mechanisms of STDP. Future studies will be required to identify these expression mechanisms and the relationship between the various plasticity protocols that engage them. It is likely that progress, like the plasticity itself, is only a matter of time.

Update Two recently published imaging studies address the dynamics of spine Ca2+ transients. Sabatini et al. [83••] focus on the important issue of the Ca2+ buffering effects of Ca2+ dyes. Because these dyes must bind Ca2+ to fluoresce, they can also distort the time course of changes in intracellular Ca2+, complicating the interpretation of experimental data (reviewed in [35]). Endogenous Ca2+ buffers are relatively immobile, so the addition of the Ca2+ indicator promotes diffusion. At the same time, by reducing the concentration of free Ca2+, the indicator slows down Ca2+ extrusion. Using Ca2+ indicators of different concentrations and affinities, Sabatini et al. [83••] extrapolate


their data to biological Ca2+ buffering conditions and find that physiological Ca2+ transients are very large and rapid. In the intact neuron, termination of the Ca2+ event is completely dominated by Ca2+ pumps, because diffusion works two orders of magnitude more slowly. This means that the spine Ca2+ response is isolated from that of the parent dendrite. Importantly, it also means that spine Ca2+ kinetics are largely shaped by the Ca2+ source, so that AP-evoked Ca2+ events are rapid, whereas synaptically evoked Ca2+ responses are largely determined by the time course of the NMDA receptor-mediated Ca2+ influx. Sabatini et al. [83••] argue that this last observation might explain why postsynaptic spiking alone does not induce long-term plasticity, whereas sufficient NMDA receptor activation often does. The finding that, under physiological buffer conditions, Ca2+ dynamics are dominated by extrusion mechanisms is at odds with the conclusions reached by Holthoff et al. [85]. Using indicator concentrations that produce significant buffering, this group previously demonstrated that spine Ca2+ kinetics are heterogenous in hippocampal CA1 neurons. Specifically, spines could be classified as ‘diffusers’ or ‘pumpers’, according to whether the Ca2+ response appeared to be terminated largely by diffusion through the spine neck or by Ca2+ extrusion mechanisms, respectively [84]. Holthoff et al. [85] use similar methods and report that these same two spine classes exist in L5 neurons. Interestingly, a proximal-to-distal gradient of the two spine types occurs: ‘diffusers’ are found closer to the soma, whereas ‘pumpers’ are largely located in the distal apical dendrite. However, for the reasons discussed above, this distinction may not persist under more physiological conditions. The idea of spatial gradients in Ca2+ dynamics holds interesting implications for plasticity, but needs to be reexamined under conditions that more closely approximate physiological levels of Ca2+ buffering. In another recent paper, Froemke and Dan [86•] study the effects of natural firing on the induction of STDP in visual cortical L2/3. This type of study is important, because in vivo firing patterns are more random than the stereotyped ones typically used in plasticity studies [1–7,9,45,46] (but see also [8,27••]). Froemke and Dan [86•] find that with brief spike triplets or quadruplets lasting a few tens of milliseconds, the first spike pairing appears to carry the most weight, and, in fact, suppresses subsequent spike inter-actions. They develop a computational model that accurately predicts the amount of LTP or LTD during low-frequency random firing. Unfortunately, the model predicts that higher frequencies should produce little LTP, which is contrary to what has been observed in L5 [5,27••] and L2/3 (Figure 2 in [46]).

Acknowledgements We thank John Lisman, Gina Turrigiano, Leslie Griffith, Alanna Watt and Wade Regehr for helpful discussions. This work was funded by the National Institutes of Health.

311

References and recommended reading Papers of particular interest, published within the annual period of review, have been highlighted as:

• of special interest •• of outstanding interest 1.

Debanne D, Gähwiler BH, Thompson SM: Asynchronous pre- and postsynaptic activity induces associative long-term depression in area CA1 of the rat hippocampus in vitro. Proc Natl Acad Sci USA 1994, 91:1148-1152.

2.

Gustafsson B, Wigström H, Abraham WC, Huang YY: Long-term potentiation in the hippocampus using depolarizing current pulses as the conditioning stimulus to single volley synaptic potentials. J Neurosci 1987, 7:774-780.

3.

Levy WB, Steward O: Temporal contiguity requirements for longterm associative potentiation/depression in the hippocampus. Neuroscience 1983, 8:791-797.

4.

Magee JC, Johnston D: A synaptically controlled, associative signal for Hebbian plasticity in hippocampal neurons. Science 1997, 275:209-213.

5.

Markram H, Lübke J, Frotscher M, Sakmann B: Regulation of synaptic efficacy by coincidence of postsynaptic APs and EPSPs. Science 1997, 275:213-215.

6.

Bi GQ, Poo MM: Synaptic modifications in cultured hippocampal neurons: dependence on spike timing, synaptic strength, and postsynaptic cell type. J Neurosci 1998, 18:10464-10472.

7.

Debanne D, Gähwiler BH, Thompson SM: Long-term synaptic plasticity between pairs of individual CA3 pyramidal cells in rat hippocampal slice cultures. J Physiol 1998, 507:237-247.

8.

Feldman DE: Timing-based LTP and LTD at vertical inputs to layer II/III pyramidal cells in rat barrel cortex. Neuron 2000, 27:45-56.

9.

Zhang LI, Tao HW, Holt CE, Harris WA, Poo M: A critical window for cooperation and competition among developing retinotectal synapses. Nature 1998, 395:37-44.

10. Abbott LF, Nelson SB: Synaptic plasticity: taming the beast. Nat Neurosci 2000, 3:1178-1183. 11. Bi GQ, Poo MM: Synaptic modification by correlated activity: Hebb’s postulate revisited. Annu Rev Neurosci 2001, 24:139-166. 12. Schuett S, Bonhoeffer T, Hubener M: Pairing-induced changes of •• orientation maps in cat visual cortex. Neuron 2001, 32:325-337. Together with [13••,15••], this study demonstrates a role for STDP in the plasticity of sensory maps in the mammalian neocortex. Intrinsic signal optical imaging was used to monitor plasticity in the orientation preference of populations of neurons in kitten visual cortex. Pairing a briefly flashed oriented grating with electrical stimulation of the cortex caused an expanded representation of the orientation of the grating when it preceded the electrical stimulus, and caused a contraction of that representation when the grating followed the electrical stimulus. 13. Song S, Abbott LF: Cortical development and remapping through •• spike timing-dependent plasticity. Neuron 2001, 32:339-350. Correlation-based Hebbian learning has previously been used in theoretical studies of activity-dependent aspects of developmental plasticity. However, additional constraints have been needed to provide stability and ensure competition among synapses. Here, the authors show that, in a computational model, STDP can account for activity-dependent plasticity of columns and maps during development and adulthood, even in the absence of such additional constraints. Because of its temporal asymmetry, STDP itself provides the competition. Importantly, STDP leads to the formation of stable cortical structures even if plasticity is never inactivated. See also [12••,15••]. 14. van Rossum MCW, Bi GQ, Turrigiano GG: Stable Hebbian learning from spike timing-dependent plasticity. J Neurosci 2000, 20:8812-8821. 15. Yao H, Dan Y: Stimulus timing-dependent plasticity in cortical •• processing of orientation. Neuron 2001, 32:315-323. This study demonstrates shifts in the orientation tuning of single neurons in cat visual cortex following repeated presentation of strong (optimally oriented) and weak (nonoptimally oriented) visual stimuli. The direction and magnitude of the shift depended on the relative timing of the two stimuli in a manner consistent with STDP. Additional psychophysical experiments demonstrated an analogous shift in the orientation perception of human subjects following a similar pairing procedure. See also [12••,13••].

312


16. Sourdet V, Debanne D: The role of dendritic filtering in associative long-term synaptic plasticity. Learn Mem 1999, 6:422-447. 17.

Stuart G, Spruston N, Sakmann B, Häusser M: Action potential initiation and backpropagation in neurons of the mammalian CNS. Trends Neurosci 1997, 20:125-131.

18. Hoffman DA, Magee JC, Colbert CM, Johnston D: K+ channel regulation of signal propagation in dendrites of hippocampal pyramidal neurons. Nature 1997, 387:869-875. 19. Migliore M, Hoffman DA, Magee JC, Johnston D: Role of an A-type K+ conductance in the back-propagation of action potentials in the dendrites of hippocampal pyramidal neurons. J Comput Neurosci 1999, 7:5-15. 20. Pan E, Colbert CM: Subthreshold inactivation of Na+ and K+ channels supports activity-dependent enhancement of backpropagating action potentials in hippocampal CA1. J Neurophysiol 2001, 85:1013-1016. 21. Golding NL, Kath WL, Spruston N: Dichotomy of action-potential backpropagation in CA1 pyramidal neuron dendrites. J Neurophysiol 2001, 86:2998-3010. 22. Berger T, Larkum ME, Luscher HR: High Ih channel density in the distal apical dendrite of layer V pyramidal cells increases bidirectional attenuation of EPSPs. J Neurophysiol 2001, 85:855-868. 23. Larkum ME, Zhu JJ, Sakmann B: Dendritic mechanisms underlying •• the coupling of the dendritic tree with the axonal action potential initiation zone of adult rat layer 5 pyramidal neurons. J Physiol 2001, 533:447-466. Using up to quadruple recordings from the same cortical L5 neuron, the authors study forward-propagation and back-propagation in the apical dendrite. They find that propagation in different neurons is either reliable or unreliable and that it depends on depolarization sigmoidally (c.f. [21,26••]). Finally, there are two distinct spike initation zones: one distal dendritic and one axonal. Interaction between the two zones depends on the degree of proximal depolarization, producing a wide variety of firing patterns under different conditions. 24. Bekkers JM: Properties of voltage-gated potassium currents in nucleated patches from large layer 5 cortical pyramidal neurons of the rat. J Physiol 2000, 525:593-609. 25. Korngreen A, Sakmann B: Voltage-gated K+ channels in layer 5 neocortical pyramidal neurones from young rats: subtypes and gradients. J Physiol 2000, 525:621-639. 26. Stuart GJ, Häusser M: Dendritic coincidence detection of EPSPs •• and action potentials. Nat Neurosci 2001, 4:63-71. The authors demonstrate that failing back-propagating APs in distal apical dendrites of cortical L5 neurons can be amplified by depolarization that sufficiently recruits Na+ channels. Although dendritic IA channels take part in AP attenuation, their inactivation is not involved in the amplification process. Amplification depends critically on the precise timing between EPSP and AP, with timing requirements similar to those previously reported for STDP in these cells [5], suggesting an involvement in long-term plasticity. See also [23••]. 27. ••

Sjöström PJ, Turrigiano GG, Nelson SB: Rate, timing, and cooperativity jointly determine cortical synaptic plasticity. Neuron 2001, 32:1149-1164. Here, the authors examine the effects of timing, rate, and cooperativity on cortical L5-to-L5 long-term plasticity. LTP depends on a threshold level of depolarization, which can be provided by synchronous excitatory inputs or by preceding APs in spike trains of sufficiently high frequency. LTD, on the other hand, does not depend on depolarization or on frequency. Random firing produces LTD at low frequencies and LTP at high frequencies, showing that plasticity appears only rate-based when timing is imprecise. The authors develop a numerical model, which accurately predicts the effect of random firing on long-term plasticity. 28. Tsubokawa H, Ross WN: IPSPs modulate spike backpropagation and associated [Ca2+]i changes in the dendrites of hippocampal CA1 pyramidal neurons. J Neurophysiol 1996, 76:2896-2906. 29. Paré D, Shink E, Gaudreau H, Destexhe A, Lang EJ: Impact of spontaneous synaptic activity on the resting properties of cat neocortical pyramidal neurons in vivo. J Neurophysiol 1998, 79:1450-1460. 30. Vetter P, Roth A, Häusser M: Propagation of action potentials in • dendrites depends on dendritic morphology. J Neurophysiol 2001, 85:926-937. Using simulations based on detailed reconstructions of eight different neuronal morphologies, the authors examine dendritic AP propagation. With the same current distributions, the mode of AP back-propagation in the

different models is largely determined by the branching pattern. In heavily branched Purkinje cells, APs fail almost regardless of ion channel densities, whereas spikes in dopaminergic neurons attenuate only moderately over a wide range of Na+-to-K+ conductance ratios. 31. Koester HJ, Sakmann B: Calcium dynamics in single spines during coincident pre- and postsynaptic activity depend on relative timing of back-propagating action potentials and subthreshold excitatory postsynaptic potentials. Proc Natl Acad Sci USA 1998, 95:9596-9601. 32. Schiller J, Schiller Y, Clapham DE: NMDA receptors amplify calcium influx into dendritic spines during associative pre- and postsynaptic activation. Nat Neurosci 1998, 1:114-118. 33. Yuste R, Denk W: Dendritic spines as basic functional units of neuronal integration. Nature 1995, 375:682-684. 34. Nishiyama M, Hong K, Mikoshiba K, Poo MM, Kato K: Calcium stores regulate the polarity and input specificity of synaptic modification. Nature 2000, 408:584-588. 35. Sabatini BL, Maravall M, Svoboda K: Ca2+ signaling in dendritic spines. Curr Opin Neurobiol 2001, 11:349-356. 36. Finch EA, Augustine GJ: Local calcium signalling by inositol-1,4,5trisphosphate in Purkinje cell dendrites. Nature 1998, 396:753-756. 37.

Takechi H, Eilers J, Konnerth A: A new class of synaptic response involving calcium release in dendritic spines. Nature 1998, 396:757-760.

38. Wang SS, Denk W, Häusser M: Coincidence detection in single dendritic spines mediated by calcium release. 2000, 3:1266-1273. 39. Oliet SH, Malenka RC, Nicoll RA: Two distinct forms of long-term depression coexist in CA1 hippocampal pyramidal cells. Neuron 1997, 18:969-982. 40. Tao HW, Zhang LI, Engert F, Poo M: Emergence of input specificity • of LTP during development of retinotectal connections in vivo. Neuron 2001, 31:569-580. LTP induced by theta burst at developing Xenopus retinotectal connections is nonspecific in young tadpoles; converging unstimulated inputs also cause LTP. However, timing-dependent LTP at this age is input-specific, highlighting the importance of induction protocol. Dendritic Ca2+ responses due to theta burst are more localized in older tadpoles, when theta burst LTP becomes specific, than in younger tadpoles. 41. Schiller J, Major G, Koester HJ, Schiller Y: NMDA spikes in basal dendrites of cortical pyramidal neurons. Nature 2000, 404:285-289. 42. Wei DS, Mei YA, Bagal A, Kao JP, Thompson SM, Tang CM: •• Compartmentalized and binary behavior of terminal dendrites in hippocampal pyramidal neurons. Science 2001, 293:2272-2275. In distal CA1 pyramidal branchlets, glutamate uncaging produces all-or-none responses that are subthreshold to the AP, in a manner similar to the NMDA spikes previously reported for cortical L5 neurons [41]. Here, activation of VDCCs is also involved. This localized dendritic coincidence detection can vastly expand the computational power of an individual neuron. 43. Euler T, Denk W: Dendritic processing. Curr Opin Neurobiol 2001, 11:415-422. 44. Poirazi P, Mel BW: Impact of active dendrites and structural plasticity on the memory capacity of neural tissue. Neuron 2001, 29:779-796. 45. Bell CC, Han VZ, Sugawara Y, Grant K: Synaptic plasticity in a cerebellum-like structure depends on temporal order. Nature 1997, 387:278-281. 46. Egger V, Feldmeyer D, Sakmann B: Coincidence detection and changes of synaptic efficacy in spiny stellate neurons in rat barrel cortex. Nat Neurosci 1999, 2:1098-1105. 47. •

Holmgren CD, Zilberter Y: Coincident spiking activity induces long term changes in inhibition of neocortical pyramidal cells. J Neurosci 2001, 21:8270-8277. Long-term plasticity of unitary inhibitory postsynaptic potentials (IPSPs) onto cortical L2/3 pyramidals is timing dependent: coincidence of the IPSP with a postsynaptic burst produces LTD, whereas IPSPs arriving approximately 300 ms after the end of the burst are potentiated. This timing relationship is the opposite to that reported for most excitatory connections [5,6,9]. 48. Hansel C, Artola A, Singer W: Relation between dendritic Ca2+ levels and the polarity of synaptic long-term modifications in rat visual cortex neurons. Eur J Neurosci 1997, 9:2309-2322.


49. Lisman J: A mechanism for the Hebb and the anti-Hebb processes underlying learning and memory. Proc Natl Acad Sci USA 1989, 86:9574-9578. 50. Neveu D, Zucker RS: Postsynaptic levels of [Ca2+]i needed to trigger LTD and LTP. Neuron 1996, 16:619-629. 51. Yang SN, Tang YG, Zucker RS: Selective induction of LTP and LTD by postsynaptic [Ca2+]i elevation. J Neurophysiol 1999, 81:781-787. 52. Cho K, Aggleton JP, Brown MW, Bashir ZI: An experimental test of • the role of postsynaptic calcium levels in determining synaptic strength using perirhinal cortex of rat. J Physiol 2001, 532:459-466. Together with [53•], this study provides evidence for the induction of LTD and LTP by independent levels of [Ca2+]i. Different concentrations of the Ca2+ buffer EGTA were included in the pipette during whole cell recordings in perirhinal cortex. The results support the hypothesis that modest elevations of [Ca2+]i are required for LTD, higher elevations are required for LTP, and at intermediate levels no plasticity is produced. 53. Cormier RJ, Greenwood AC, Connor JA: Bidirectional synaptic • plasticity correlated with the magnitude of dendritic calcium transients above a threshold. J Neurophysiol 2001, 85:399-406. In this study, the authors manipulated [Ca2+]i by varying the amount of glutamate iontophoresis applied to CA1 neurons. They provide support for two ranges over which plasticity of concurrently activated synapses can be evoked: a lower range in which LTD is observed, and a higher range in which LTP is observed. Between these two ranges, little or no plasticity occurs. A strength of the study is that [Ca2+]i was measured directly using the calcium indicator fura-2. See also [52•]. 54. DeMaria CD, Soong TW, Alseikhan BA, Alvania RS, Yue DT: •• Calmodulin bifurcates the local Ca2+ signal that modulates P/Q-type Ca2+ channels. Nature 2001, 411:484-489. By coexpressing mutant CaMs and P/Q type Ca2+ channels, the binding of Ca2+ to the two lobes of CaM is here shown to have opposing effects on Ca2+ channel gating. Binding to the C-lobe promotes facilitation, whereas binding to the N-lobe promotes inactivation. Importantly, it appears that C-lobe binding reflects local, rapid changes in [Ca2+] near the channel, whereas N-lobe binding reflects slower, more spatially averaged changes in [Ca2+]. This provides a mechanism by which different kinetic profiles of a single messenger can lead to opposing downstream changes. Because CaM is a widespread signaling molecule this principle of ‘bifurcation’ of the Ca2+ signal may have more general application to mechanisms underlying synaptic plasticity. See also [71••]. 55. Lisman JE, Zhabotinsky AM: A model of synaptic memory: a CaMKII/PP1 switch that potentiates transmission by organizing an AMPA receptor anchoring assembly. Neuron 2001, 31:191-201. 56. Holmes WR: Models of calmodulin trapping and CaM kinase II activation in a dendritic spine. J Comput Neurosci 2000, 8:65-85. 57.

De Koninck P, Schulman H: Sensitivity of CaM kinase II to the frequency of Ca2+ oscillations. Science 1998, 279:227-230.

58. Eshete F, Fields RD: Spike frequency decoding and autonomous activation of Ca2+-calmodulin-dependent protein kinase II in dorsal root ganglion neurons. J Neurosci 2001, 21:6694-6705. 59. Paulsen O, Sejnowski TJ: Natural patterns of activity and long-term synaptic plasticity. Curr Opin Neurobiol 2000, 10:172-179. 60. Castellani GC, Quinlan EM, Cooper LN, Shouval HZ: A biophysical model of bidirectional synaptic plasticity: dependence on AMPA and NMDA receptors. Proc Natl Acad Sci USA 2001, 98:12772-12777. 61. Franks KM, Bartol TM, Sejnowski TJ: An MCell model of calcium dynamics and frequency-dependence of calmodulin activation in dendritic spines. Neurocomput 2001, 38-40:9-16. 62. Kullmann DM, Perkel DJ, Manabe T, Nicoll RA: Ca2+ entry via postsynaptic voltage-sensitive Ca2+ channels can transiently potentiate excitatory synaptic transmission in the hippocampus. Neuron 1992, 9:1175-1183. 63. Wang Y, Rowan MJ, Anwyl R: LTP induction dependent on activation of Ni2+-sensitive voltage-gated calcium channels, but not NMDA receptors, in the rat dentate gyrus in vitro. J Neurophysiol 1997, 78:2574-2581. 64. Peterson BZ, DeMaria CD, Adelman JP, Yue DT: Calmodulin is the Ca2+ sensor for Ca2+-dependent inactivation of L-type calcium channels. Neuron 1999, 22:549-558. 65. Lee A, Wong ST, Gallagher D, Li B, Storm DR, Scheuer T, Catterall WA: Ca2+/calmodulin binds to and modulates P/Q-type calcium channels. Nature 1999, 399:155-159.

313

66. Erickson MG, Alseikhan BA, Peterson BZ, Yue DT: Preassociation of •• calmodulin with voltage-gated Ca2+ channels revealed by FRET in single living cells. Neuron 2001, 31:973-985. Here, a novel three-cube FRET technique was developed and used to demonstrate that apoCam (labeled with cyan fluorescent protein) is preassociated with each of the major classes of high-threshold VDCCs (labeled with yellow fluorescent protein). Elegant control experiments ruled out nonspecific causes for the observed FRET. The results demonstrate that, as for several other channel types, the binding of the Ca2+ sensor, CaM, to the channel itself enables rapid and local Ca2+ channel modulation. 67.

Zhang S, Ehlers MD, Bernhardt JP, Su CT, Huganir RL: Calmodulin mediates calcium-dependent inactivation of N-methyl-D-aspartate receptors. Neuron 1998, 21:443-453.

68. Umemiya M, Chen N, Raymond LA, Murphy TH: A calcium • dependent feedback mechanism participates in shaping single NMDA miniature EPSCs. J Neurosci 2001, 21:1-9. Prior work demonstrated several forms of Ca2+-dependent inhibition of NMDA currents. Here, the authors combine physiology and high-resolution imaging in cultured cortical neurons to show that Ca2+ influx through NMDA receptors shortens the decay of miniature NMDA currents. This negative feedback reduces the trial-to-trial variability of synaptic Ca2+ transients. In addition, rapid, Ca2+-dependent inactivation of NMDA currents could have implications for the mechanism underlying STDP [79]. 69. Bayer KU, De Koninck P, Leonard AS, Hell JW, Schulman H: •• Interaction with the NMDA receptor locks CaMKII in an active conformation. Nature 2001, 411:801-805. The authors show that binding of CamKII to NMDA receptors, previously thought to require autophosphorylation of the enzyme, can occur at a second site on the NR2B subunit of the NMDA receptor in response to Ca2+–CaM binding. This binding site is unique in that it appears to prevent binding of the autoinhibitory domain of the kinase, thus allowing for at least partial autonomous (i.e. CaM-independent) kinase activity even in the absence of autophosphorylation. This binding of CamKII to NR2B also enhances translocation of the enzyme to the synapse, and increases CamKII’s ability to ‘trap’ CaM, thereby reducing the rate of other CaMdependent processes. 70. Hardingham GE, Arnold FJ, Bading H: Nuclear calcium signaling • controls CREB-mediated gene expression triggered by synaptic activity. Nat Neurosci 2001, 4:261-267. The authors demonstrate that cAMP response element binding protein (CREB) activation does not depend on nuclear translocation of activated CaM, as previously hypothesized. Instead, burst-triggered Ca2+ influx produces Ca2+ release from internal stores, creating a Ca2+ wave that propagates to the nucleus, where CREB is activated by nuclear CaM kinases. The bursting frequency is converted into a nuclear Ca2+ level, providing a means for activity to shape genomic responses. 71. Dolmetsch RE, Pajvani U, Fife K, Spotts JM, Greenberg ME: •• Signaling to the nucleus by an L-type calcium channel-calmodulin complex through the MAP kinase pathway. Science 2001, 294:333-339. L-type Ca2+ channels play a critical role in activating CREB and other transcription factors. By blocking endogenous channels with dihydropyridines and transfecting cultured cortical neurons with various mutant channels not sensitive to the blocker, the authors show that signaling depends upon CaM binding to a specific binding site on the channel and subsequent activation of MAPK. This study, together with [54••], demonstrates the ability of CaM binding at sites of Ca2+ entry to engage highly localized and source-specific signaling responses. 72. Lu W, Man H, Ju W, Trimble WS, MacDonald JF, Wang YT: Activation • of synaptic NMDA receptors induces membrane insertion of new AMPA receptors and LTP in cultured hippocampal neurons. Neuron 2001, 29:243-254. In an interesting variation on prior studies of chemically induced LTP and LTD in cultured neurons, the authors show that selectively activating synaptic NMDA receptors by applying glycine results in LTP, mediated by addition of new AMPA receptors. The activation of nonsynaptic NMDA receptors (or of both synaptic and nonsynaptic receptors), however, leads exclusively to LTD. 73. Dodt H, Eder M, Frick A, Zieglgänsberger W: Precisely localized LTD in the neocortex revealed by infrared-guided laser stimulation. Science 1999, 286:110-113. 74. Kandler K, Katz LC, Kauer JA: Focal photolysis of caged glutamate produces long-term depression of hippocampal glutamate receptors. Nat Neurosci 1998, 1:119-123. 75. Lynch G, Larson J, Kelso S, Barrionuevo G, Schottler F: Intracellular injections of EGTA block induction of hippocampal long-term potentiation. Nature 1983, 305:719-721. 76. Rao RP, Sejnowski TJ: Spike-timing-dependent Hebbian plasticity as temporal difference learning. Neural Comput 2001, 13:2221-2237.

314

77.


Wang YT, Linden DJ: Expression of cerebellar long-term depression requires postsynaptic clathrin-mediated endocytosis. Neuron 2000, 25:635-647.

78. Linden DJ: The expression of cerebellar LTD in culture is not • associated with changes in AMPA-receptor kinetics, agonist affinity, or unitary conductance. Proc Natl Acad Sci USA 2001, 98:14066-14071. Together with [77], this paper makes a strong case for a postsynaptic reduction in the number of AMPA receptors as the mechanism underlying expression of LTD at synapses between cultured cerebellar neurons. 79. Senn W, Markram H, Tsodyks M: An algorithm for modifying neurotransmitter release probability based on pre- and postsynaptic spike timing. Neural Comput 2001, 13:35-67. 80. Markram H, Lübke J, Frotscher M, Roth A, Sakmann B: Physiology and anatomy of synaptic connections between thick tufted pyramidal neurones in the developing rat neocortex. J Physiol 1997, 500:409-440. 81. Dudek SM, Bear MF: Homosynaptic long-term depression in area CA1 of hippocampus and effects of N-methyl-D-aspartate receptor blockade. Proc Natl Acad Sci USA 1992, 89:4363-4367. 82. Mulkey RM, Malenka RC: Mechanisms underlying induction of homosynaptic long-term depression in area CA1 of the hippocampus. Neuron 1992, 9:967-975. 83. Sabatini BL, Oertner TG, Svoboda K: The life cycle of Ca2+ ions in •• dendritic spines. Neuron 2002, 33:439-452. Here, two-photon microscopy is employed in the study of AP-evoked and NMDA receptor-evoked Ca2+ transients in CA1 spines. The authors use

Ca2+ dyes of different concentrations and Ca2+ affinities to extrapolate to physiological buffering conditions. They find that the physiological Ca2+ buffer capacity is very small, which allows for large and rapid Ca2+ events. With endogenous Ca2+ buffers alone, diffusion across the spine neck is insignificant, because Ca2+ pumps work two orders of magnitude faster, thus functionally isolating the spine from the parent dendrite. For slow events, Ca2+ kinetics are largely determined by the time course of the Ca2+ source. This makes the slow NMDA receptor-evoked Ca2+ elevation much more sustained than that of the back-propagating AP. The authors argue that this may explain why Ca2+ from NMDA receptors, but not from APs, is involved in LTP and LTD. 84. Majewska A, Brown E, Ross J, Yuste R: Mechanisms of calcium decay kinetics in hippocampal spines: role of spine calcium pumps and calcium diffusion through the spine neck in biochemical compartmentalization. J Neurosci 2000, 20:1722-1734. 85. Holthoff K, Tsay D, Yuste R: Calcium dynamics of spines depend on their dendritic location. Neuron 2002, 33:425-437. 86. Froemke RC, Dan Y: Spike-timing-dependent synaptic • modification induced by natural spike trains. Nature 2002, 416:433-438. Using extracellular stimulation, the authors study the effects of natural, random firing patterns on the induction of STDP in visual cortical L2/3. They find that the first spike in spike triplets and quadruplets, lasting a few tens of milliseconds, best predicts the amount of LTP/LTD, and thus overrides the impact of the subsequent spike pairings. A numerical model is developed that accurately predicts the build-up of plasticity due to random spike trains. The authors conclude that plasticity depends not only on spike timing but also on interspike interval.