Calcium is a second messenger, which can trigger the modification of synaptic efficacy. We investigated the question of whether a differential rise in postsynaptic Ca2+ ([Ca2+]i) alone is sufficient to account for the induction of long-term potentiation (LTP) and long-term depression (LTD) of EPSPs in the basal dendrites of layer 2/3 pyramidal neurons of the somatosensory cortex. Volume-averaged [Ca2+]i transients were measured in spines of the basal dendritic arbor for spike-timing-dependent plasticity induction protocols. The rise in [Ca2+]i was uncorrelated to the direction of the change in synaptic efficacy, because several pairing protocols evoked similar spine [Ca2+]i transients but resulted in either LTP or LTD. The sequence dependence of near-coincident presynaptic and postsynaptic activity on the direction of changes in synaptic strength suggested that LTP and LTD were induced by two processes, which were controlled separately by postsynaptic [Ca2+]i levels. Activation of voltage-dependent Ca2+ channels before metabotropic glutamate receptors (mGluRs) resulted in the phospholipase C-dependent (PLC-dependent) synthesis of endocannabinoids, which acted as a retrograde messenger to induce LTD. LTP required a large [Ca2+]i transient evoked by NMDA receptor activation. Blocking mGluRs abolished the induction of LTD and uncovered the Ca2+-dependent induction of LTP.
We conclude that the volume-averaged peak elevation of [Ca2+]i in spines of layer 2/3 pyramids determines the magnitude of long-term changes in synaptic efficacy. The direction of the change is controlled, however, via a mGluR-coupled signaling cascade. mGluRs act in conjunction with PLC as sequence-sensitive coincidence detectors when postsynaptic precede presynaptic action potentials to induce LTD. Thus presumably two different Ca2+ sensors in spines control the induction of spike-timing-dependent synaptic plasticity.
- synaptic plasticity
- two-photon microscopy
- spike-timing-dependent plasticity
Long-term changes in synaptic efficacy are thought to be the cellular basis of information storage and memory formation (Bliss and Collingridge, 1993; Whitlock et al., 2006). Modifications in the efficacy of transmission at synaptic contacts can be induced by coincident presynaptic and postsynaptic activity (Magee and Johnston, 1997; Markram et al., 1997; Debanne et al., 1998). The precise timing and the order of presynaptic and postsynaptic action potentials (APs) determine the magnitude and the direction of the change in synaptic strength (Markram et al., 1997; Bi and Poo, 1998; Feldman, 2000; Sjostrom et al., 2001; Froemke and Dan, 2002). A postsynaptic AP that follows a presynaptic AP within a time window of tens of milliseconds results in long-term potentiation (LTP), whereas the reverse order results in depression (LTD). Therefore, spike-timing-dependent plasticity (STDP) is one possible cellular model for the induction of local synaptic modifications, which could account for experience-driven changes of the connectivity in neuronal networks (Song and Abbott, 2001; Senn, 2002).
For STDP pairing protocols as well as for other plasticity induction protocols, like presynaptic tetanic stimulation (Lynch et al., 1983) or repetitive low-frequency synaptic stimulation (Mulkey and Malenka, 1992), the elevation of postsynaptic [Ca2+]i is essential. [Ca2+]i probably acts as one second messenger on downstream metabolic cascades that are responsible for the eventual modification of synaptic efficacy (Lisman, 1989). A long-standing hypothesis suggests that the peak amplitude of postsynaptic [Ca2+]i elevation determines the direction and the magnitude of such modifications (Bear et al., 1987; Artola and Singer, 1993; Hansel et al., 1997). Several studies have supported this hypothesis by measuring dendritic [Ca2+]i (Cormier et al., 2001; Ismailov et al., 2004; Gall et al., 2005). Nevertheless, the problem of how a single variable, the increase in global [Ca2+]i, can control the differential induction of changes in synaptic efficacy has not been explained conclusively. According to this “Ca2+ control hypothesis” the level of [Ca2+]i acts differentially on downstream protein cascades, activating either kinases or phosphatases, which, respectively, phosphorylate or dephosphorylate postsynaptic AMPA receptors (Lisman, 1989; Lee et al., 2000). Modified versions of this hypothesis include veto mechanisms for LTD at moderate Ca2+ levels (Rubin et al., 2005), differential microdomain Ca2+ signaling (Franks and Sejnowski, 2002), or the proposition of a second coincidence detector (Karmarkar and Buonomano, 2002) to account for the differential induction of LTD and LTP.
Here we characterize the Ca2+ signals in spines of basal dendrites of layer 2/3 (L2/3) pyramidal neurons in the somatosensory cortex during LTP- and LTD-inducing protocols. We found that postsynaptic elevation of [Ca2+]i in spines is necessary, but by itself it is not sufficient, to account for both potentiation and depression of synaptic strength. Induction of LTD requires, in addition, the activation of a metabotropic glutamate receptor-dependent (mGluR-dependent) signaling cascade resulting in the synthesis of endocannabinoids, which then act as a retrograde messenger. We conclude that the inductions of LTP and LTD both depend on a rise in [Ca2+]i but represent two separate processes triggered by two Ca2+ sensors. Thus [Ca2+]i elevation and the coactivation of mGluRs are necessary to account for the bidirectional changes in synaptic efficacy observed with STDP protocols.
Materials and Methods
Sagittal brain slices were prepared from postnatal day 13–15 Wistar rats. All experimental procedures were in accordance with the animal welfare guidelines of the Max Planck Society. Experiments were performed at physiological temperatures (32–36°C) in extracellular solution containing the following (in mm): 125 NaCl, 25 NaHCO3, 2.5 KCl, 1.25 NaH2PO4, 1 MgCl2, 25 glucose, 2 CaCl2, and 0.01 glycine, bubbled with 95% O2/5% CO2. Inhibitory inputs were blocked by the bath application of 10 μm bicuculline. Pyramidal neurons in L2/3 of the somatosensory cortex were visualized by using infrared (IR) gradient contrast video microscopy. Patch pipettes (5–7 MΩ) were filled with a low chloride intracellular solution containing the following (in mm): 130 K-gluconate, 10 K-HEPES, 10 Na-phosphocreatine, 4 Mg-ATP, 0.3 Na-GTP, 4 NaCl, and 10 Na-gluconate. Whole-cell current-clamp recordings were made with an AxoClamp-2B (Molecular Devices, Union City, CA) patch-clamp amplifier. Voltage signals were filtered at 3 kHz and digitized at 10 kHz with an ITC-16 (InstruTech, Port Washington, NY). Access resistance (10–20 MΩ) usually did not change during the course of the experiment. Input resistance was monitored constantly by a brief hyperpolarizing current pulse. Experiments were excluded if input resistance or membrane potential changed significantly over the time course of the experiment.
An extracellular stimulation pipette filled with extracellular solution was placed close to the basal dendrites of the L2/3 pyramidal neuron (50–150 μm from soma). Stimulation strength (3–10 μA; 100 μs duration) was adjusted to evoke baseline single component EPSPs with amplitudes between 1 and 3 mV. Baseline EPSPs were recorded for 10 min at 0.1 Hz stimulation. Then EPSPs were paired with one to three APs at different frequencies and variable onset times, Δt′. Pairings were repeated 60 times at 0.1 Hz stimulation. The time interval Δt′ was defined as the time between the onset of the AP burst (first AP in the burst) and the onset of the compound EPSP. The time interval Δt was defined as the time between the AP closest in time to the EPSP and the onset of the EPSP. The change in EPSP amplitude was evaluated 20–40 min after the end of the pairing period and normalized to the baseline EPSP amplitude. Data are presented as the mean ± SEM. Paired Student's t tests were applied as statistical tests if not indicated otherwise, and statistical significance was asserted for p < 0.05.
For two-photon excitation fluorescence microscopy an IR femtosecond-pulsed titanium sapphire laser (Mira 900, pumped by a 5 W Verdi; Coherent, Santa Clara, CA) was coupled directly to a confocal-scanning unit (LCS SP2RS, Leica Microsystems, Mannheim, Germany) attached to an upright microscope (DMLFS, Leica) equipped with a 63× objective (HCX APO W63× UVI; numerical aperture 0.9; Leica) (Rathenberg et al., 2003). Cells were filled with a combination of the Ca2+-insensitive dye Alexa 594 (50 μm; Invitrogen, Carlsbad, CA) and the Ca2+ indicator Oregon Green BAPTA-6F (500 μm; Invitrogen) added to the low chloride intracellular solution. Thus all Ca2+ imaging experiments represent a different set of experiments from the separately performed experiments addressing the change in synaptic strength. Dyes were excited at λ = 820 nm. Excitation IR laser light and fluorescence emission light were separated at 670 nm (excitation filter 670DCXXR, AHF Analysentechnik, Tübingen, Germany). The emission spectra were separated by a dichroic mirror at 560 nm (beam splitter 560DCXR, AHF) and corresponding bandpass (HQ525/50, HQ630/60, AHF) and IR-block filters (E700SP, BG39; AHF) and were detected by using non-descanned detection. In addition, the forward-scattered IR laser light was filtered spatially and imaged onto a photomultiplier tube to generate an IR-scanning gradient contrast image of the unstained brain slice (Wimmer et al., 2004). Line scans through a spine that responded with a [Ca2+]i transient to subthreshold synaptic stimulation were made to measure the [Ca2+]i transients evoked by the different pairing protocols. Relative changes in fluorescence are expressed as the following: ΔG/R = (G(t) − G0)/R, where G(t) is the fluorescence signal integrated from the region of interest covering the spine in green channels, G0 is the basal fluorescence averaged for 50 ms before electrical stimulation, and R is the averaged fluorescence of the red channels (Oertner et al., 2002). Single exponential fits to the decay phase of the [Ca2+]i transients yielded the peak amplitude (ΔG/R)max. The nonlinearity factor of Ca2+ signals was evaluated by normalizing the peak [Ca2+]i amplitude for a given pairing protocol to the expected linear sum of EPSP-evoked and AP-evoked [Ca2+]i transients (Nevian and Sakmann, 2004).
L-type voltage-dependent Ca2+ channels (L-VDCCs) were blocked by the bath application of nimodipine (10 μm) and T-type VDCCs (T-VDCCs) by NiCl2 (50 μm). NMDA receptors were blocked by the bath application of d-5-amino-phosphonopentanoic acid (d-APV) (50 μm) or intracellular application of (+)-5-methyl-10,11-dihydro-5H-dibenzo[a,d]cyclohepten-5,10-imine maleate (MK-801) (1 mm). mGluRs were blocked by the broadband mGluR antagonist (S)-α-methyl-4-carboxyphenylglycine (MCPG) (500 μm), phospholipase C (PLC) was blocked by 1-(6-[(17β-methoxyestra-1,3,5 -trien-17-yl) amino] hexyl)-1H-pyrrole-2,5-dione (U73122) (5 μm), and cannabinoid type 1 (CB1) receptors were blocked by N-(piperidin-1-yl)-5-(4-iodophenyl)-1-(2,4-dichlorophenyl)-4-methyl-1H-pyrazole-3-carboxamide (AM251) (2 μm). Inositol 1,4,5-triphosphate-mediated (IP3-mediated) Ca2+ release from internal stores was blocked by intracellular application of heparin (400 U/ml). Cytosolic elevations of Ca2+ were buffered by dialyzing the cells with EGTA and BAPTA at various concentrations (0.1–2 mm).
Burst-timing-dependent potentiation and depression in L2/3 pyramids
Whole-cell in vivo recordings from morphologically identified barrel-related L2/3 pyramidal neurons indicate large compound EPSPs in response to tactile stimuli but sparse AP responses (Zhu and Connors, 1999; Brecht et al., 2003), with rare burst activity (Svoboda et al., 1997, 1999). The induction of long-lasting changes in synaptic efficacy has been reported to depend strongly on the timing between presynaptic and postsynaptic APs (Markram et al., 1997; Bi and Poo, 1998). Therefore, we first investigated in L2/3 pyramidal neurons whether such a timing dependence also existed for pairing a single compound EPSP that was elicited by focal extracellular stimulation with a burst of three APs occurring at 50 Hz in the postsynaptic cell. We found that the degree of both potentiation and depression depended on the time interval Δt′ between onset of the AP burst and onset of the EPSP (Figs. 1, 2). When the AP burst preceded the EPSP by Δt′ = −90 ms or followed the EPSP by Δt′ = +50 ms, the pairing had no effect (Δt′ = −90 ms: 1.00 ± 0.09, p > 0.5, n = 6; Δt′ = +50 ms: 0.92 ± 0.11, p > 0.5, n = 4). Pairing at Δt′ = −50 ms resulted in a long-lasting depression of the EPSP amplitude, with a reduction of the EPSP amplitude by a factor of 0.68 ± 0.05 (p < 0.01; n = 10) as compared with control (Figs. 1C, 2B). Two APs before and one AP after the EPSP (Δt′ = −30 ms) did not change the EPSP amplitude (0.98 ± 0.12; p > 0.5; n = 8). In contrast, one AP before and two APs after the EPSP (Δt′ = −10 ms) resulted in moderate potentiation (1.42 ± 0.19; p < 0.05; n = 12). Strong potentiation by a factor of 2.01 ± 0.22 (p < 0.01; n = 11) was found if the three APs followed the EPSP by Δt′ = +10 ms (Figs. 1B, 2).
Second, we tested the requirement for postsynaptic bursting to induce a change in synaptic strength (Fig. 3A,B). The time interval Δt, defined as the time interval between the onset of the EPSP and the AP closest in time to the EPSP, was kept constant at Δt = −10 ms or Δt = +10 ms. Pairing an EPSP and one AP at Δt = +10 ms did not change the EPSP amplitude (1.04 ± 0.08; p > 0.5; n = 10), whereas the reverse order resulted in a decrease in EPSP amplitude (0.80 ± 0.07; p < 0.05; n = 5). A “minimal burst” of two APs (50 Hz) at Δt = +10 ms resulted in significant potentiation (1.95 ± 0.31; p < 0.05; n = 9). The reverse order (two APs and EPSP; Δt = −10 ms) resulted in depression (0.72 ± 0.12; p < 0.05; n = 5).
Synaptic stimulation without pairing with postsynaptic APs (EPSPs only) did not result in a change in EPSP amplitude (0.97 ± 0.08; p > 0.5; n = 5). Similarly, a burst of three APs at 50 Hz without pairing with an EPSP (three APs only) had no effect (1.05 ± 0.2; p > 0.9; n = 6).
Last, we varied the AP frequency in a burst of three APs between 20 and 100 Hz (Fig. 3C,D). A burst of three APs at 20 Hz that followed the EPSP by Δt = +10 ms had no effect on the EPSP amplitude after pairing (1.09 ± 0.27; p > 0.5; n = 5). A burst at 20 Hz preceding the EPSP by Δt = −10 ms resulted in depression (0.72 ± 0.14; p < 0.05; n = 7). A burst of three APs at 100 Hz that followed the EPSP by Δt = +10 ms induced strong potentiation (2.29 ± 0.48; p < 0.01; n = 7), whereas a burst of three APs at 100 Hz preceding the EPSP by Δt = −10 ms resulted in depression (0.52 ± 0.12; p < 0.05; n = 4).
From these results we conclude that LTP and LTD depend on the timing between the AP burst and onset of the compound EPSP. The induction of LTD is less sensitive to the properties of the burst, whereas LTP requires at least a burst of two APs at frequencies >20 Hz. These results indicate that small variations in postsynaptic AP firing with respect to a synaptically evoked EPSP can result in different directions of the change in synaptic strength.
LTP and LTD are equally sensitive to fast and slow Ca2+ buffers
Changes in synaptic strength depend on a transient [Ca2+]i elevation in the dendrites of the postsynaptic neuron. It is, however, unclear whether a putative “Ca2+ sensor” that may trigger the induction of changes in synaptic strength is sensitive to global, volume-averaged [Ca2+]i or whether it is sensitive to local [Ca2+]i transients in Ca2+ microdomains around sources of Ca2+ influx (Franks and Sejnowski, 2002). One way to differentiate between global and local [Ca2+]i signaling is to load the postsynaptic cell with different Ca2+ buffers, such as EGTA or BAPTA. These have a similar affinity (KD ≈ 200 nm) but faster (BAPTA) or slower (EGTA) equilibration kinetics.
Loading cells with EGTA or BAPTA at 2 mm was sufficient to block LTP induction (relative magnitude of LTP with EGTA: −0.05 ± 0.18, p > 0.7, n = 6; BAPTA: −0.16 ± 0.14, p > 0.7, n = 5) (Fig. 4A). Lower concentrations of EGTA and BAPTA blocked LTP induction less (0.5 mm EGTA: 0.64 ± 0.20, p < 0.05, n = 6; 0.5 mm BAPTA: 0.63 ± 0.3, p < 0.05, n = 6). A sigmoid fit to the concentration dependence indicated a half-effective concentration for blocking LTP of 0.6 mm for EGTA and 0.5 mm for BAPTA (Fig. 4B).
LTD induction was blocked completely by 1 mm EGTA (relative magnitude of LTD, −0.2 ± 0.25; p > 0.3; n = 3) (Fig. 4C), whereas 0.25 mm EGTA blocked LTD induction correspondingly less (−0.86 ± 0.35; p < 0.05; n = 6). BAPTA (1 mm) also blocked LTD induction completely (−0.09 ± 0.51; p > 0.5; n = 7), and 0.25 mm BAPTA blocked LTD induction less (0.92 ± 0.2; p < 0.05; n = 6). The half-effective concentration for the block of LTD was 0.39 mm by EGTA and 0.36 mm by BAPTA (Fig. 4D).
We conclude that postsynaptic elevation of [Ca2+]i is necessary for the induction of both LTP and LTD. The comparable concentration dependence of BAPTA and EGTA in blocking the induction or expression of potentiation or depression suggests that the putative Ca2+ sensor or sensors that trigger long-lasting changes in synaptic efficacy respond to the volume-averaged increase in [Ca2+]i. Either they are separated by a relatively large distance from the site of Ca2+ influx (Neher, 1998; Augustine et al., 2003) or they have slower Ca2+-binding kinetics than EGTA.
Ca2+ transients in spines evoked during STDP induction protocols
The results described above suggested that the induction of synaptic potentiation and depression depended on a rise in postsynaptic [Ca2+]i, presumably in dendritic spines. We used two-photon excitation fluorescence microscopy (Denk et al., 1990) to measure the [Ca2+]i transients in single spines evoked during the stimulation patterns described above (Fig. 5A–C).
First, varying the time interval Δt′ between an EPSP and a burst of three APs (50 Hz) allowed us to map the peak [Ca2+]i amplitude and the “nonlinearity” factor (peak [Ca2+]i amplitude, normalized to the expected linear sum of EPSP-evoked and AP-evoked [Ca2+]i transients) (Nevian and Sakmann, 2004) for the time intervals relevant for the induction of changes in synaptic strength (Fig. 5D). The plot of the peak [Ca2+]i amplitude for different time intervals Δt′ revealed that pairing an EPSP with three APs at Δt′ = +10 ms resulted in a significantly larger [Ca2+]i transient than pairing an EPSP with three APs at the other time intervals that were tested (p < 0.01; n = 4–39; ANOVA; Newman–Keuls). These other pairing protocols evoked [Ca2+]i transients with similar peak [Ca2+]i amplitudes (p > 0.2; n = 4–34; ANOVA; Newman–Keuls). The summation of the Ca2+ signals showed linear or supralinear summation, depending on the timing interval Δt′, similar to other types of cortical cells (Köster and Sakmann, 1998; Nevian and Sakmann, 2004). Pairing an EPSP with three APs at Δt′ = −10, +10, and +50 ms resulted in supralinear summation, with the largest degree of supralinearity for Δt′ = +10 ms (nonlinearity factor of 1.8 ± 0.1; p < 0.01; n = 39). Pairing at Δt′ = −30, −50, and −90 ms did not deviate significantly from linear summation of the Ca2+ signals (p > 0.1; n = 4–21).
Second, peak amplitude and nonlinearity of summation were analyzed as a function of the number of APs in a burst for a frequency of 50 Hz and a given time interval (Δt = −10 ms and Δt = +10 ms) between EPSP and AP burst. Peak amplitudes increased linearly with the number of APs for all of the stimulation protocols that were tested (linear regression lines, r2 > 0.98) (Fig. 5E). The summation of the Ca2+ signals was approximately twofold larger than the linear sum for stimulation protocols in which the APs followed the EPSP by Δt = +10 ms, independent of the number of APs in the burst (p < 0.01; n = 7–39). The summation of the Ca2+ signals for stimulation protocols in which the APs preceded the EPSP was independent of the number of APs in the burst and not different from linear summation (p > 0.5; n = 5–21).
Third, the peak [Ca2+]i amplitude increased linearly as a function of AP frequency in a burst of three APs for all of the stimulation protocols that were tested (Fig. 5F). The summation of the Ca2+ signals was larger than the expected linear sum for stimulation protocols in which the burst followed the EPSP by Δt = +10 ms with burst frequencies >20 Hz. The summation of the Ca2+ signals for stimulation protocols in which the burst preceded the EPSP was independent of the frequency of APs in the burst and not different from the expected linear sum (p > 0.5; n = 3–21).
Dependence of LTP and LTD induction on Ca2+ influx via NMDA receptors and VDCCs
The asymmetry in the peak [Ca2+]i amplitude timing curve (Fig. 5D) presumably is caused by the coincident activation of the NMDA receptor (NMDAR) channel after binding of glutamate and relief of the Mg2+ block caused by membrane depolarization attributable to backpropagating APs (Yuste and Denk, 1995; Köster and Sakmann, 1998; Nevian and Sakmann, 2004). We used the protocols consisting of an EPSP and three APs (50 Hz) at Δt = −10 ms and Δt = +10 ms to test the requirement of NMDAR activation for the timing-dependent difference in the Ca2+ signals and for the induction of potentiation and depression (Fig. 6). Bath application of the NMDAR antagonist d-APV (50 μm) significantly reduced the peak [Ca2+]i amplitude for both time intervals (reduction to 33 ± 9% for Δt = +10 ms, p < 0.05; n = 4 and reduction to 45 ± 19% for Δt = −10 ms, p < 0.05; n = 4). The [Ca2+]i transients were reduced to the level of AP-evoked Ca2+ influx, abolishing the timing dependence (Fig. 6B,C). Bath application of d-APV abolished potentiation and depression (Δt = +10 ms: 1.08 ± 0.11, p > 0.1, n = 3; Δt = −10 ms: 0.98 ± 0.11, p > 0.5, n = 6), indicating that NMDAR activation was necessary for the induction of LTP and LTD (Fig. 6D,E). The induction of spike-timing-dependent LTD, but not LTP, might be mediated by the activation of presynaptic NMDARs (Sjostrom et al., 2003; Bender et al., 2006). Postsynaptic NMDARs can be blocked specifically by loading the postsynaptic cell with MK-801 (1 mm), an open channel blocker for NMDARs, through the patch pipette. The induction of LTP was abolished (0.95 ± 0.19; p > 0.5; n = 6) with MK-801 present in the intracellular solution, similar to bath application of d-APV (Fig. 6F,H). In contrast, intracellular MK-801 had no effect on the induction of LTD (0.60 ± 0.11; p < 0.05; n = 7) (Fig. 6G,I). Therefore, we conclude that the induction of LTP depends on a large postsynaptic Ca2+ influx through NMDARs, whereas AP burst-pairing-induced LTD is independent of postsynaptic activation of NMDARs. In the latter case Ca2+ influx through VDCCs is presumably sufficient to induce LTD (Fig. 6B).
L-VDCCs might control the induction of changes in synaptic strength (Magee and Johnston, 1997, 2005). Therefore, we tested the contribution of L-VDCCs on the [Ca2+]i transients evoked by the protocols consisting of an EPSP and three APs (50 Hz) at Δt = −10 ms and Δt = +10 ms (Fig. 7A–C). Bath application of the L-VDCC blocker nimodipine (10 μm) had no effect on the [Ca2+]i transient evoked by pairing an EPSP and three APs at Δt = +10 ms. For the sequence of three APs and an EPSP at Δt = −10 ms, nimodipine significantly reduced the peak [Ca2+]i amplitude to 69 ± 1% of control (p < 0.05; n = 3). Next we tested the effect of nimodipine on plasticity. Pairing an EPSP with three APs (50 Hz) at Δt = +10 ms had no effect on LTP induction in the presence of nimodipine (Fig. 7D). The potentiation of the EPSP amplitude was 1.92 ± 0.31 (p < 0.01; n = 7), similar to control. Nimodipine also had no effect on the induction of LTD (0.70 ± 0.10; p < 0.05; n = 8), corresponding to the only minor reduction of the [Ca2+]i transient.
T-VDCCs have been suggested to be involved in spike-timing-dependent induction of LTD (Bender et al., 2006). Consistently, pairing one AP with an EPSP at Δt = −10 ms in the presence of Ni2+ abolished the induction of LTD (1.05 ± 0.16; p > 0.5; n = 3) (Fig. 7F). On the other hand, a burst of three APs (50 Hz) paired with an EPSP at Δt = −10 ms in the presence of Ni2+ resulted in LTD (0.70 ± 0.03; p < 0.05; n = 4) (Fig. 7G,I). Ni2+ reduced the [Ca2+]i transient evoked in this case to 65 ± 12% (n = 3) of control (Fig. 7H). Bath application of nimodipine together with Ni2+ resulted in a block of LTD by pairing three APs (50 Hz) with an EPSP at Δt = −10 ms (1.00 ± 0.05; p > 0.5; n = 3) (Fig. 7I).
These experiments show that VDCCs are necessary for the induction of LTD. Single-spike LTD requires Ca2+ influx through T-VDCCs; however, during pharmacological block of these channels, bursts of APs evoke a Ca2+ influx via other subtypes, which is sufficiently large to induce LTD. Therefore, the requirement for a specific VDCC-subtype for the induction of LTD is masked by AP burst pairing.
Spine Ca2+ transients cannot account for the direction of changes in synaptic efficacy
The analysis of the peak [Ca2+]i amplitude dependence on timing, number of APs, and frequency of APs suggested that pairing protocols, which induce either potentiation or depression, can raise [Ca2+]i to the same volume-averaged level. Therefore, we compared the average peak [Ca2+]i amplitude evoked by the different stimulation protocols to the average effect on long-term changes in synaptic efficacy. The resulting plot represented the average change in EPSP amplitude as a function of the average peak [Ca2+]i amplitude that a spine on a basal dendrite encounters during the pairing period (Fig. 8). The data showed no correlation between the peak [Ca2+]i amplitude and the direction of change in synaptic strength, because several stimulation protocols, which gave rise to similar peak [Ca2+]i amplitudes, induced either LTP or LTD (Fig. 8, shaded area). Thus although a postsynaptic increase in [Ca2+]i is necessary for the induction of changes in efficacy, the peak volume-averaged [Ca2+]i amplitude is not a unique determinant for the selective induction and expression of bidirectional synaptic plasticity. Additional factors, independent of the Ca2+ signal, must contribute to the induction pathway. The observation that protocols in which the APs preceded the EPSP resulted in LTD whereas the reverse order resulted in LTP (Figs. 2G, 3B,D) led to the clustering of the data into two groups according to the timing of the EPSP and APs. The clustering of stimulation protocols in which the APs either preceded or followed the EPSP indicated that the induction of LTD and LTP were two differential processes, both of which were controlled separately by the peak [Ca2+]i amplitude. Sigmoid fits to the two groups of data showed a high correlation between the induction of either LTD or LTP and the peak [Ca2+]i amplitude. The half-to-minimum value for the induction of LTD was (ΔG/R)max = 0.07, and the half-to-maximum value for the induction of LTP was approximately twofold larger [(ΔG/R)max = 0.13]. The stimulation patterns in which the EPSP was evoked within the burst of APs fell close to the fit of the Ca2+ dependence for the LTP group, indicating that the induction of LTD can be blanked by APs that follow the AP–EPSP sequence.
We conclude that the induction of LTP and LTD is controlled differentially by the elevation of [Ca2+]i in the postsynaptic spine. The sequence of EPSP and postsynaptic APs seems to activate additional mechanisms, which determine the direction of the change in synaptic strength.
LTD requires activation of mGluRs
In hippocampal as well as in cortical neurons the induction of some forms of LTD is sensitive to the activation of mGluRs (Otani and Connor, 1998; Anwyl, 1999, 2006; Egger et al., 1999; Cho et al., 2001). We tested whether LTD induced by pairing three APs (50 Hz) with an EPSP at Δt = −10 ms required the activation of mGluRs. Bath application of the broadband mGluR antagonist MCPG (500 μm) resulted in a complete block of LTD (1.06 ± 0.16; p > 0.5; n = 7) (Fig. 9E). In contrast, MCPG had no effect on LTP induction by pairing an EPSP and three APs (50 Hz) at Δt = +10 ms (1.79 ± 0.33; p < 0.05; n = 9) (Fig. 9D). The [Ca2+]i transients evoked by pairing an EPSP with three APs (50 Hz) at Δt = −10 ms and Δt = +10 ms were not different from control after the bath application of MCPG (p > 0.1; n = 4) (Fig. 9A–C), indicating that mGluRs had no direct effect on the [Ca2+]i transients (Brenowitz and Regehr, 2005) and that the block of LTD was attributable to inactivation of a G-protein-coupled signaling cascade (Piomelli, 2003).
The LTD-inducing protocol of three APs (100 Hz) paired with an EPSP at Δt = −10 ms evoked [Ca2+]i transients, which had a peak amplitude comparable to protocols that induced LTP. Bath application of MCPG also had no effect on the [Ca2+]i transient in this case (88 ± 6% of control; p > 0.1; n = 4). Surprisingly, pairing three APs (100 Hz) with an EPSP at Δt = −10 ms in the presence of MCPG resulted in the induction of LTP (1.53 ± 0.24; p < 0.05; n = 11) instead of inducing LTD, as under control conditions (Fig. 9F). We conclude that the activation of mGluRs is necessary for the induction of LTD. If the mGluR-dependent pathway is blocked, LTP can be induced, depending on the peak [Ca2+]i amplitude (Fig. 9G).
Signaling pathway for the induction of LTD
Next we investigated the downstream signaling cascade involved in the induction of LTD after mGluR activation. One product of this cascade is the synthesis of endocannabinoids, which can act as retrograde messengers, resulting in a decrease in synaptic efficacy (Sjostrom et al., 2003, 2004). The endocannabinoid 2-arachidonoylglycerol (2-AG) is synthesized via a PLC and diacylglycerol (DAG) lipase-dependent pathway. Bath application of the CB1 receptor antagonist AM251 (2 μm) resulted in the block of LTD by the protocol consisting of three APs (50 Hz) and an EPSP at Δt = −10 ms (1.09 ± 0.14; p > 0.5; n = 5), suggesting the involvement of endocannabinoid signaling (Fig. 10E). The PLC inhibitor U73122 (5 μm) also blocked LTD (1.11 ± 0.07; p > 0.1; n = 4) without any effect on the [Ca2+]i transient evoked by the sequence of three APs (50 Hz) and an EPSP at Δt = −10 ms (99 ± 11% of control; n = 4) (Fig. 10A,B).
A side product of the synthesis of DAG by PLC is IP3, which could result in the release of Ca2+ from internal stores. The observation that neither the block of mGluRs nor the block of PLC resulted in a significant change in the spine [Ca2+]i transients suggested that release from internal stores did not contribute to the [Ca2+]i transients that were required for the induction of LTD by a short burst of APs preceding synaptic activation.Another possibility might be that the concentration of IP3 within the spine gradually increases during the 60 pairings, resulting in Ca2+ release from internal stores only for later pairings. We recorded the [Ca2+]i transients in a spine evoked by three APs (50 Hz) and an EPSP at Δt = −10 ms during the pairing period. No correlation between the peak [Ca2+]i amplitude and the number of pairings was found (r2 = 0.2; n = 7), suggesting that the recorded spine [Ca2+]i transients did not change significantly during the pairing period (Fig. 10D). Finally, blocking IP3-dependent release of Ca2+ from internal stores by intracellular application of heparin (400 U/ml) through the patch pipette had no effect on LTD (0.68 ± 0.12; p < 0.05; n = 9) (Fig. 10C). The resulting pharmacological signature for LTD (Fig. 10F) suggests retrograde endocannabinoid signaling, mediated by G-coupled mGluR and PLC activation.
We tested several in vivo-like AP activity patterns in a local L2/3-to-L2/3 pyramid connection of somatosensory cortex for the induction of LTP or LTD and measured the corresponding [Ca2+]i transients in spines. The results indicate that synaptic potentiation and depression are two processes separately controlled by a rise in postsynaptic [Ca2+]i. The magnitude of the long-term change in synaptic transmission depends on the volume-averaged elevation of [Ca2+]i in a spine, but the direction of the change is controlled by the activation of a mGluR-dependent signaling cascade.
LTP induction required the pairing of an EPSP with a minimal burst of two APs at frequencies >20 Hz, whereas the induction of LTD was less sensitive to the properties of the AP burst. These results are comparable to findings in pairs of L5B pyramidal neurons in which low-frequency pairings of a unitary EPSP and a single AP failed to induce potentiation (Markram et al., 1997; Sjostrom et al., 2001). Pyramidal neurons in the CA1 region of the hippocampus show a developmental switch from single AP-induced LTP to burst LTP (Pike et al., 1999). In contrast, cultured hippocampal neurons (Bi and Poo, 1998), L2/3 pyramids in visual cortex (Froemke and Dan, 2002), and the L4-to-L2/3 spiny stellate-to-pyramid connection in somatosensory cortex (Feldman, 2000) are potentiated by a single AP that follows an EPSP. The [Ca2+]i transients evoked by pairing an EPSP with a single AP in our experimental conditions indicated that the rise in [Ca2+]i was not sufficient to induce LTP. This might be different for other connections, depending on the ion channel distribution in dendrites and spines, which influence the backpropagation of APs and Ca2+ signaling (Magee and Johnston, 1995; Schiller et al., 1995; Spruston et al., 1995; Sabatini and Svoboda, 2000; Sabatini et al., 2001; Waters and Helmchen, 2004; Gasparini and Magee, 2006). The requirement for AP bursts to induce synaptic modifications might provide stability of connectivity (Lisman and Spruston, 2005), because in vivo cortical activity is sparse and single APs occur more frequently than bursts (Lee et al., 2006; Waters and Helmchen, 2006).
Postsynaptic Ca2+ transients and synaptic plasticity
The induction of both LTP and LTD was equally sensitive to loading of the postsynaptic cell with the rapidly equilibrating Ca2+ buffer BAPTA or the slower buffer EGTA. Therefore, the volume-averaged elevation of spine [Ca2+]i is an important factor for the induction of changes in synaptic strength. The putative Ca2+ sensors that trigger the induction of LTP or LTD presumably are separated from the Ca2+ entry site by several tens of nanometers (Neher, 1998). They might be mobile Ca2+ buffers, which compete with Ca2+ extrusion and additional fixed Ca2+ buffers. Another possibility is that the Ca2+ sensors have slower binding kinetics for Ca2+ than EGTA. In this case the Ca2+ sensors might be localized in close proximity to Ca2+ channels, but still they would be sensitive mainly to the volume-averaged increases in [Ca2+]i. In the case of LTP, calmodulin is a possible candidate Ca2+ sensor. It can activate Ca2+/calmodulin-dependent protein kinase II (CaMKII), which translocates to the plasma membrane of spines (Bayer et al., 2001; Gleason et al., 2003; Otmakhov et al., 2004) and phosphorylates AMPA receptors (Malenka et al., 1989; Barria et al., 1997; Hayashi et al., 2000; Lisman et al., 2002). Its mobility and its presumably almost homogeneous distribution in the spine head could account for the similar effects of BAPTA and EGTA.
One hypothesis relating changes in synaptic efficacy and postsynaptic [Ca2+]i transients suggests that the level of postsynaptic [Ca2+]i elevation determines whether a synapse is potentiated or depressed (Lisman, 1989; Artola and Singer, 1993; Hansel et al., 1997). Below a threshold of [Ca2+]i the synaptic efficacy remains unaffected. In an “intermediate” range of [Ca2+]i elevations LTD is induced, whereas “large” increases in [Ca2+]i induce LTP (Zucker, 1999). A number of reports are in accordance with this view (Cummings et al., 1996; Cho et al., 2001). Uncaging experiments of Ca2+ showed a correlation between the peak level of volume-averaged [Ca2+]i and the direction of changes in synaptic strength (Yang et al., 1999). Ca2+ imaging in dendrites and somata also suggested a direct correlation (Hansel et al., 1996, 1997; Cormier et al., 2001; Ismailov et al., 2004; Gall et al., 2005). Simulations using the volume-averaged peak [Ca2+]i amplitude as a readout to determine the direction of change in synaptic efficacy reproduced in part the experimentally determined STDP curves (Karmarkar et al., 2002; Shouval et al., 2002; Shouval and Kalantzis, 2005). In contrast, other reports failed to find a correlation between peak [Ca2+]i and the direction of change in synaptic strength (Neveu and Zucker, 1996; Wang et al., 2005). Therefore, it was an unresolved issue whether and how a single variable, the global peak [Ca2+]i amplitude, could induce LTD or LTP differentially. It was suggested that the time course of the Ca2+ transient (Rubin et al., 2005) or that the stochastic opening of the NMDA receptor (Shouval and Kalantzis, 2005) can account for the differential induction of LTP and LTD.
Our results show clearly that induction protocols that increase volume-averaged [Ca2+]i to similar levels can induce either LTP or LTD. We conclude that the volume-averaged peak [Ca2+]i amplitude is not the only determinant for the direction of modification in synaptic efficacy. Thus the Ca2+ dependence of STDP induction protocols does not conform to a Ca2+ control hypothesis (Bear et al., 1987; Artola and Singer, 1993). However, we confirmed that both the inductions of LTD and LTP require the rise of [Ca2+]i above a threshold, which is approximately twofold higher for the induction of LTP than for LTD. The large Ca2+ influx during coincident activation of NMDARs by glutamate and depolarization by backpropagating APs resulted in LTP. In contrast, the induction of LTD required a transient increase in [Ca2+]i through VDCCs and the coactivation of mGluRs. In accordance with these results it was suggested that a second coincidence detector, triggering only LTD induction, could resolve the problem of bidirectional plasticity (Karmarkar and Buonomano, 2002; Bender et al., 2006). mGluRs are a component in several signaling cascades resulting in changes of synaptic strength (Anwyl, 1999, 2006; Bortolotto et al., 1999). Block of mGluRs did not modulate the postsynaptic [Ca2+]i transients (Brenowitz and Regehr, 2005), excluding a contribution of Ca2+ release from internal stores (Emptage et al., 2003). In agreement, blocking Ca2+ release from internal stores had no effect on burst-pairing-induced LTD in contrast to other forms of LTD (Otani et al., 2002; Bender et al., 2006). We suggest that mGluR activation can be regarded as a postsynaptic switch, which sets the sign for the direction of the change in synaptic strength. Activation of this switch results in the induction of LTD, independent of the [Ca2+]i levels that have been reached. If the switch is not activated and [Ca2+]i levels are sufficiently large, LTP is induced. The induction of LTD is triggered by AP-evoked Ca2+ influx through VDCCs and presumably the subsequent binding of Ca2+ to proteins of the mGluR-coupled signaling cascade before its activation. In support of this view, several proteins of the G-protein-coupled signaling cascade require the binding of Ca2+ and are thought to be involved in the induction of LTD (Artola and Singer, 1993; Daniel et al., 1998; Kemp and Bashir, 2001). The sequence for the induction of LTD is qualitatively different from the sequence of activation for the induction of LTP. In the case of LTP, synaptic activation and thus mGluR activation precede (or are simultaneous to) Ca2+ influx, rendering the LTD pathway silent. The requirement for Ca2+ influx preceding mGluR activation suggests the presence of a second coincidence detection mechanism in L2/3 pyramidal neurons. As a consequence, spines devoid of mGluRs might express only LTP or require a different mechanism for the induction ofLTD, like dephosphorylation or removal of AMPA receptors (Carroll et al., 1999; Lee et al., 2000; Froemke et al., 2005; O'Connor et al., 2005).
Signaling cascade for the induction of LTD
We identified the downstream signaling cascade for spike-timing-dependent LTD to involve PLC and endocannabinoids, similar to heterosynaptic depression of GABAergic transmission in hippocampus (Chevaleyre and Castillo, 2003). This suggested the signaling by group I mGluRs (Coutinho and Knopfel, 2002). Endocannabinoids act as a retrograde messenger for the induction of LTD (Sjostrom et al., 2003; Duguid and Sjostrom, 2006). They are synthesized via a PLC-dependent pathway, the efficiency of which is modulated greatly by a rise of [Ca2+]i (Hashimotodani et al., 2005; Maejima et al., 2005). This suggests that the occurrence of postsynaptic before presynaptic coincident activity is detected by PLC and finally results in the release of endocannabinoids to induce LTD. In this case the expression of LTD is presynaptic, and the changes in synaptic efficacy are attributed to a decrease in release probability (Sjostrom et al., 2003). Nevertheless, the trigger for these changes requires the postsynaptic elevation of [Ca2+]i coupling postsynaptic induction to presynaptic expression of LTD.
We conclude that postsynaptic Ca2+ influx is a necessary trigger for modifications of synaptic strength. However, the amplitude of postsynaptic [Ca2+]i elevation evoked by naturally occurring neuronal activity patterns is not sufficient to decode the direction of the change. Therefore, LTP and LTD are induced by two separate Ca2+ sensors. The selective activation of one of two coincidence detectors for the relative timing of presynaptic and postsynaptic activity can account for the capability of synapses for bidirectional modifications.
We thank Matthew Larkum and Hans-Rudolf Lüscher for their support and helpful comments on this manuscript and Marlies Kaiser and Karl Schmidt for excellent technical support.
- Correspondence should be addressed to Thomas Nevian at his present address: Institute for Physiology, Bern University, Bühlplatz 5, CH-3012 Bern, Switzerland.