Abstract
The Bienenstock, Cooper and Munro (BCM) computational model, which incorporates a metaplastic sliding threshold for LTP induction, accounts well for experience-dependent changes in synaptic plasticity in the visual cortex. BCM-like metaplasticity over a shorter timescale has also been observed in the hippocampus, thus providing a tractable experimental preparation for testing specific predictions of the model. Here, using extracellular and intracellular electrophysiological recordings from acute rat hippocampal slices, we tested the critical BCM predictions (1) that high levels of synaptic activation will induce a metaplastic state that spreads across dendritic compartments, and (2) that postsynaptic cell-firing is the critical trigger for inducing that state. In support of the first premise, high-frequency priming stimulation inhibited subsequent long-term potentiation and facilitated subsequent long-term depression at synapses quiescent during priming, including those located in a dendritic compartment different to that of the primed pathway. These effects were not dependent on changes in synaptic inhibition or NMDA/metabotropic glutamate receptor function. However, in contrast to the BCM prediction, somatic action potentials during priming were neither necessary nor sufficient to induce the metaplasticity effect. Instead, in broad agreement with derivatives of the BCM model, calcium as released from intracellular stores and triggered by M1 muscarinic acetylcholine receptor activation was critical for altering subsequent synaptic plasticity. These results indicate that synaptic plasticity in stratum radiatum of CA1 can be homeostatically regulated by the cell-wide history of synaptic activity through a calcium-dependent but action potential-independent mechanism.
Introduction
There is considerable evidence that activity-dependent synaptic plasticity such as LTP and LTD serve as memory mechanisms. As expected of memory mechanisms, there is not a fixed synaptic plasticity response to a given activation pattern, but instead plasticity is dynamically influenced by a number of state variables, including the trace left by prior neural activity, i.e., metaplasticity (Abraham and Bear, 1996; Abraham, 2008). The need for metaplasticity has been demonstrated in computational and neural network models which indicate that there must be control over both the extent and timing of plasticity induction (Grossman et al., 2002; Yeung et al., 2004). For example, the Bienenstock, Cooper and Munro (BCM) (Bienenstock et al., 1982) computational model of synaptic plasticity incorporates a parameter termed the modification threshold, θM, that dynamically varies with the history of prior postsynaptic cell firing, thereby homeostatically regulating the direction and degree of subsequent synaptic plasticity. Thus, following a period of increased cell activity, LTP becomes harder to induce and LTD more likely; conversely, a period of reduced activity has the opposite effect.
The activity-dependent regulation of θM in the BCM model is a form of metaplasticity (Abraham and Bear, 1996; Abraham, 2008) for which two key characteristics are posited: (1) it is calculated from a running time-average of postsynaptic action potential activity, and (2) it affects excitatory synapses throughout the cell (Bienenstock et al., 1982). However, recent variants of the BCM model have focused on the time-averaged intracellular free calcium concentration, or activity of a downstream enzyme such as calcium-dependent protein kinase II, as the biological signal that metaplastically controls the plasticity induction (Shouval et al., 2002a,b; Yeung et al., 2004).
BCM-like changes in plasticity induction account well for a variety of phenomena, including experience-dependent changes in visual and somatosensory neural activity (Benusková et al., 1994; Kirkwood et al., 1996; Philpot et al., 2003, 2007) and synaptic plasticity in the dentate gyrus (Benuskova and Abraham, 2007). BCM-like metaplasticity effects over a shorter timescale have also been observed in the hippocampus, providing a tractable experimental preparation for testing the two key postulates noted above. Because initial tests have been made using field potential recording techniques in the hippocampus (Wang and Wagner, 1999; Abraham et al., 2001), the precise role played by postsynaptic action potentials, for example, remains uncertain. Here we used the heterosynaptic metaplasticity experimental protocols established by Wang and Wagner (1999) as the starting point for making these tests. We found that repeated HFS caused a BCM-like shift in the plasticity induction function that could spread across the dendritic arbor of CA1 hippocampal pyramidal cells. Surprisingly, action potential generation was neither necessary nor sufficient for activity-dependent suppression of LTP, in contrast to the assumption of the original BCM model. Rather, in broad accord with calcium-focused derivatives of the model (Yeung et al., 2004), a calcium signal arising through release from intracellular stores, as triggered by M1 muscarinic acetylcholine receptor activation, mediated the metaplasticity effects.
Materials and Methods
Tissue preparation.
Young adult male Sprague-Dawley rats (6–8 weeks old) were anesthetized with ketamine (100 mg/kg, i.p.) and decapitated. Brains were then submersed in ice-cold sucrose dissection solution (containing, in mm: 210 sucrose, 26 NaHCO3, 2.5 KCl, 1.25 NaH2PO4, 0.5 CaCl2, 3 MgCl2, and 20 d-glucose) or in artificial CSF (ACSF, containing, in mm: 124 NaCl, 3.2 KCl, 1.25 NaH2PO4, 26 NaHCO3, 2.5 CaCl2, 1.3 MgCl2, and 10 d-glucose) equilibrated with carbogen (95% O2-5% CO2). Transverse hippocampal slices (400 μm) were prepared and transferred to a humidified submersion brain slice chamber. ACSF was continuously superfused (2 ml/min) by gravity feed and slices were allowed to equilibrate for at least 3 h before baseline recordings began, while the temperature was gradually brought to 32.5°C.
Extracellular recording techniques.
Synaptic potentials were evoked through stimulation of the Schaffer collateral-commissural pathway in stratum radiatum, and in some experiments stratum oriens, of area CA1 using 50 μm Teflon-insulated tungsten monopolar electrodes. Pulse half-wave duration was 0.1 ms. Recordings from the same stratum that was stimulated were made using glass micropipettes filled with 2 m NaCl (1–2 MΩ). In experiments using two stimulating electrodes in stratum radiatum, the electrodes were placed equidistant from the single recording electrode, on either side and at a similar central position in stratum radiatum. In the experiments where both stratum radiatum and stratum oriens were stimulated, two recording and two stimulating electrodes were used. These were arranged with the recording electrodes in line and perpendicular to the cell-body layer, to record from the same population of cells, while both stimulating electrodes were positioned on the same side of the recording electrodes. The criterion for use of a slice in an experiment was that the fEPSP responses for each pathway were 2 mV or greater at a stimulation intensity of 100 μA. Stimulation intensity was then adjusted to elicit a 1 mV response (25–70 μA). Baseline stimulation was delivered every 30 s, alternating between stimulating electrodes. Priming stimulation was given on one pathway (radiatum or oriens), using a protocol that consisted of two sets of three 100 Hz, 1 s trains (20 s between trains, 5–15 min between sets). Conditioning stimulation to induce plasticity was given on a second test pathway (always in stratum radiatum, unless otherwise noted) 30 min after priming. The conditioning protocol consisted of two trains of 600 pulses at stimulation frequencies of 10, 20, 50 or 100 Hz (10 min between trains) for initial experiments or a more standard LTP induction protocol of two 100 Hz 1 s trains (30 s between trains).
Intracellular recording techniques.
Because of the long duration of the experiments, and the consequent high probability of washout of biochemicals essential for LTP on the test pathway, single-cell recordings for LTP studies were made using intracellular sharp electrode techniques, rather than whole-cell patch-clamp methods. Synaptic potentials were evoked as for extracellular experiments using either two stimulating electrodes in stratum radiatum, or one each in strata radiatum and oriens. Intracellular recordings, using borosilicate micropipettes filled with 2 m KAc (1.0 mm outside diameter and 0.5 mm inside diameter; Sutter Instrument Co., 80–140 MΩ), were made in current-clamp mode from CA1 pyramidal cells. The criteria for use in an experiment were that a cell had a resting membrane potential of ≤ −60 mV, showed overshooting action potentials when depolarized and exhibited an input resistance (Rin) > 30 MΩ. A further criterion was that at −75 mV the EPSP amplitude was ≥ 8 mV with single pulse stimulation of 100 μA. For the experiments, the stimulation intensity was adjusted to produce a response consistently > 5 but < 8 mV (35–90 μA). The membrane potential of impaled cells was held constant at −75 mV during the baseline periods and depolarized to an initial −60 mV (radiatum-radiatum experiments) or −65 mV (oriens-radiatum experiments) during priming and conditioning stimulation by manually adjusting the holding current.
Pathway independence.
To confirm the independence of the two pathways, a paired-pulse stimulation test (50 ms interpulse interval) was used for all experiments. When the two pulses were given to the same pathway, paired-pulse facilitation (PPF) was observed. Experiments were excluded if PPF following the same-electrode stimulation was not >10%. When pulses were given to alternate pathways no slices showed cross-facilitation >10%, indicating pathway independence.
Data analysis.
The initial slopes of the fEPSPs and EPSPs were analyzed as a measure of synaptic efficacy. Baseline was calculated as the average response over the 10 min before conditioning. All response measurements were converted to a percentage of baseline. LTP was calculated as the average over the last 5 or 10 min of recording for extracellular and intracellular experiments, respectively. All group data are presented as mean ± SEM. Note that data are typically presented only for the heterosynaptic stratum radiatum test pathway (S2, Figs. 1A, 2A, insets). Priming stimulation (either in radiatum or oriens) routinely induced LTP homosynaptically. Unless otherwise noted statistical analysis for significant differences between control and primed groups was assessed by Student's t test or ANOVA (one-way or repeated measures with Greenhouse-Geisser corrected values, as appropriate). The critical value was set at p < 0.05.
Whole-cell recordings techniques.
To assess the effects of priming stimulation on NMDA receptor-mediated currents, hippocampal slices for whole-cell patch-clamp recordings were prepared as previously described (Ireland and Abraham, 2009). Standard techniques were used to record EPSCs from the soma of visualized CA1 pyramidal neurons. Patch electrodes were formed from soft borosilicate glass (WPI) and had resistances of 2–3 MΩ when filled with a K-based recording solution (containing, in mm: 140 KMeSO4, 10 HEPES, 4 Na2ATP, 0.4 NaGTP, 4 MgCl2, Na2 phosphocreatine, 0.2 EGTA-4Na, 1 QX-314, pH 7.25 when adjusted with KOH, ∼320 mOsm when adjusted with sucrose). Access resistance was continuously monitored throughout each experiment (10–20 MΩ at beginning of experiment) and experiments that differed from baseline values by >30% were discarded. EPSCs were evoked by electrical stimulation of Schaffer collateral fibers in stratum radiatum, and stratum oriens fibers, via broken off patch-electrodes filled with ACSF. EPSCs composed of currents mediated by both NMDA receptors and AMPA receptors (EPSCsNMDAR/AMPAR) were isolated by inhibiting GABAA and GABAB receptors with bath-applied picrotoxin (20 μm) and CGP55845 (1 μm), respectively. EPSCs mediated solely by NMDARs (EPSCsNMDAR) were isolated by the further addition of CNQX (10 μm). In all experiments, voltage-dependent calcium currents mediated by L-type channels were inhibited with bath-applied nimodipine (10 μm). All EPSCs were recorded in voltage-clamp mode at a holding potential of −30 mV using an Axopatch 1D amplifier and pCLAMP 9 software (Molecular Devices). They were sampled at 10 kHz and filtered at 2 kHz. Baseline stimulation (oriens and radiatum pathways) evoking single EPSCsNMDAR/AMPAR, priming stimulation delivered to stratum oriens, and stratum radiatum conditioning stimulation (100 Hz) evoking summated EPSCsNMDAR, were delivered with the same frequencies and intervals as in the sharp electrode intracellular recording experiments. Stratum oriens priming stimulation was delivered in current-clamp mode at a holding potential of −60 mV.
Patch-clamp data analysis.
Raw data were analyzed using pCLAMP 9 software. The amplitude of the NMDAR-mediated component of the EPSCNMDAR/AMPAR was measured at a fixed point on the EPSC after the AMPAR-mediated component had decayed. Peak amplitude and area (charge transfer) of summated EPSCsNMDAR recorded in response to stratum radiatum conditioning stimulation were normalized to the amplitude and area of single EPSCsNMDAR. Differences between the datasets were analyzed with unpaired Student's t tests at the p < 0.05 significance level. Data are reported as mean ± SEM.
Drugs and reagents.
All salts were obtained from BDH Laboratory Supplies. Drugs, including d-AP5, nimodipine, LY341495, CGP55845, CNQX, picrotoxin, QX-314, pirenzepine dihydrochloride, and cyclopiazonic acid (CPA), were obtained from Tocris Bioscience. Na2ATP, NaGTP, phosphocreatine, and EGTA-4Na were obtained from Sigma. KMeSO4 was obtained from Acros Organics. Xestospongin C was obtained from Cayman Chemical. AP5 and pirenzepine were dissolved in H2O. Nimodipine, xestospongin C, and cyclopiazonic acid were dissolved in DMSO and LY341495 was dissolved in NaOH (pH adjusted to 7 with HCl). Drugs were diluted by ≥500-fold and introduced into the ACSF through bath application. In experiments where the drug was dissolved in DMSO, control experiments were also exposed to DMSO during the corresponding time points.
Results
Priming stimulation heterosynaptically inhibits LTP and promotes LTD
We first aimed to confirm that the Wang and Wagner (1999) priming protocol reliably produced in our hands a BCM-like heterosynaptic regulation of plasticity at Schaffer collateral synapses in area CA1. Under control conditions, 100 Hz and 50 Hz conditioning stimulation induced stable LTP, 20 Hz conditioning did not change responses on average, and 10 Hz stimulation evoked LTD (Fig. 1A1,B1, control). As predicted, prior 100 Hz priming stimulation of a second stratum radiatum pathway (Fig. 1A2,B2) inhibited LTP and facilitated LTD induced 30 min later on the test pathway (Fig. 1A1,B1, primed). These results demonstrate that, in accordance with both the BCM model and previous results, a high level of prior activity can heterosynaptically inhibit LTP and facilitate LTD induced at synapses quiescent during the priming activity. Interestingly, there was no significant correlation between the magnitude of priming-induced LTP and the level of LTP/LTD subsequently induced on the test pathway (10 Hz group: r = −0.169, p = 0.72; 100 Hz group: r = 0.095, p = 0.86).
Surprisingly, the response near the LTD/LTP crossover point (θM of the BCM model) evoked by 20 Hz conditioning stimulation was only weakly shifted by the priming stimulation (Fig. 1C). To understand the lack of change at the LTD/LTP crossover point, we considered the possibility that our relatively weak stimulation strength used for conditioning stimulation may have resulted in a right-shifted crossover point under control conditions, making it difficult to shift it further with priming stimulation (compare our Fig. 1C with Wang and Wagner, 1999, their Fig. 3). To address this point, we increased the stimulation strength during the conditioning stimulation to one which elicited a 1.5 mV fEPSP, and found that it moved the crossover point from ∼20 Hz to < 10 Hz. Priming stimulation was then able to shift the crossover point to the right, in accord with the BCM model (Fig. 1D). All further experimentation was undertaken using the original conditioning stimulation strength, focusing on 100 Hz and 10 Hz stimulation for studies of LTP and LTD, respectively.
Heterosynaptic metaplasticity can occur cell-wide
In the above experiments we demonstrated BCM-like heterosynaptic interactions between independent sets of synapses terminating in a common dendritic field, i.e., the stratum radiatum. However, the BCM model predicts that prior activity will regulate the plasticity function of excitatory synapses throughout the cell. Since it was conceivable that the heterosynaptic metaplasticity in the present study only affected synapses located near the primed ones (Chevaleyre and Castillo, 2004), we tested the cell-wide nature of the present effect by giving priming stimulation either in the stratum oriens (basilar dendrites) or stratum radiatum (apical dendrites) and subsequently inducing LTP or LTD on the other side of the cell bodies. We observed LTP in stratum radiatum (30 ± 3%) was reliably inhibited by the prior delivery of priming stimulation to afferents in stratum oriens (16 ± 4%, p = 0.02; Fig. 2A). Correspondingly, LTD induced by low-frequency stimulation (LFS) was facilitated by priming stimulation (p = 0.01; Fig. 2B). In contrast, LTP induced in stratum oriens (42 ± 5%) was not inhibited by priming stimulation in stratum radiatum (61 ± 8%, p = 0.20; Fig. 2C). Increasing the intensity of the radiatum priming stimulation (to generate 1.5 mV responses) still failed to inhibit oriens LTP (42 ± 7%, p = 0.54; data not shown). These results demonstrate that metaplastic changes induced using our paradigm are compartmentally restricted when priming is delivered to afferents in stratum radiatum but cell-wide when delivered to afferents in stratum oriens. We do not, however, rule out the possibility that other priming paradigms may be effective in generating radiatum to oriens interactions.
Plasticity modification is not due to altered NMDAR currents or GABAergic transmission
In extensions of the BCM model, the expression of metaplastic change has been modeled as an alteration in the function of NMDARs (Gold and Bear, 1994; Shouval et al., 2002b). For this reason we predicted that there would be a change evident in the NMDAR-mediated EPSC following priming. To directly assess this possibility, single EPSCsNMDAR/AMPAR were recorded using whole-cell patch-clamp techniques. Note that AMPARs were not blocked at this stage to allow AMPAR-mediated depolarization to occur during the priming protocol, as would have occurred during the equivalent extracellular experiments. Following priming stimulation of stratum oriens (in current-clamp to allow cell depolarization), the amplitude of the NMDAR-mediated component of the EPSCNMDAR/AMPAR recorded on the same pathway (oriens) was indeed significantly reduced by −27 ± 8% (n = 5, p = 0.023 compared with unprimed controls; Fig. 3A). However, the NMDAR component recorded in stratum radiatum in the same slices was not significantly altered (4 ± 7%, p = 0.845; Fig. 3B). Note that since LTD of the EPSCNMDAR/AMPAR did occur homosynaptically in oriens synapses, it is unlikely that the lack of change in radiatum responses is due to washout of any biochemicals critical for inducing NMDAR change, as mediated by our whole-cell recording conditions.
Although the amplitudes of baseline radiatum NMDAR-mediated EPSCs were not changed by oriens priming stimulation, there may still have been a change in the summated EPSCNMDAR produced by radiatum conditioning stimulation for LTP induction. To address this issue, immediately at the conclusion of the previous experiment CNQX was applied to isolate the NMDAR-mediated EPSC and then the standard LTP induction protocol (2 trains of 100 Hz, 1 s, 30 s between trains) was applied to the stratum radiatum pathway. Summated EPSCsNMDAR for each burst were normalized to the single EPSCsNMDAR recorded during the 5 min before conditioning stimulation. There was no difference between control and primed cells for the summated EPSCNMDAR amplitudes or areas recorded during either the first or second HFS burst (Fig. 3C,D). Together these results indicate that a change in NMDAR function is not the mechanism by which the heterosynaptic metaplasticity is expressed.
An alternative mechanism by which priming could produce a very long-range effect is through a persistent facilitation of transmission at GABAergic synapses, or facilitated activation of GABAergic interneurons, consequently increasing synaptic inhibition on the heterosynaptic pathway during conditioning stimulation thus blocking LTP. To test this possibility, GABAA and GABAB receptor-mediated inhibition was blocked with 100 μm picrotoxin and 1 μm CGP55845, respectively, for the duration of the experiment. The priming stimulation and conditioning stimulation were given to separate stratum radiatum pathways, as per the methods of the previous extracellular experiments. The conditioning protocol for LTP induction, however, was reduced to 600 pulses at 50 Hz due to the increased slice excitability under these conditions. For control slices, conditioning induced 38 ± 5% LTP of S2 radiatum responses (n = 7; Fig. 3E). For primed slices, significantly less LTP was induced (15 ± 6%, n = 5, p = 0.014). These results indicate that neither GABAA nor GABAB receptors are involved in the induction or expression of the BCM-like metaplasticity.
Testing the role of postsynaptic cell firing
A critical element of the BCM model is that shifting of the modification threshold is determined by the time-averaged history of postsynaptic cell firing. Thus, we predicted that the induction of heterosynaptic metaplasticity in the present experiments would be mediated by somatic action potential firing in response to the priming stimulation. We tested this prediction using intracellular sharp electrode recording techniques. First, we demonstrated that the priming effect on LTP between two sets of radiatum synapses could be observed at the level of single cells, in the absence of any change in input resistance (Fig. 4A) or the amplitude of the peak AHP produced in response to LTP induction (control AHP: 7.64 ± 0.85 mV, n = 10; primed AHP: 6.73 ± 0.88 mV, n = 12, p = 0.448). Surprisingly, however, we observed that the postsynaptic neuron fired very few action potentials during the 600 pulses of priming stimulation (mean 5 total spikes, range 0–12 total spikes, n = 6 cells), and the number of action potentials fired did not correlate with the degree of LTP inhibition (Pearson r = −0.22; p = 0.663). The lack of firing is consistent with the generally weak depolarization generated under conditions of 100 Hz stimulation, whether as priming trains or conditioning trains (Fig. 1A, waveforms; mean 2.9 total spikes during the conditioning trains). These observations imply strongly that action potentials per se were not driving the metaplastic change. To test directly whether somatic action potentials during priming were necessary for the induction of priming in stratum radiatum, cells were hyperpolarized to a nominal −100 mV by manual application of negative holding current during priming stimulation to completely prevent cell firing. Despite this treatment, priming stimulation in stratum radiatum still significantly inhibited the degree of LTP (26 ± 9%) relative to control cells (54 ± 8%, p = 0.035; Fig. 4B). Hyperpolarization during priming, however, did prevent the induction of LTP on the priming (homosynaptic) pathway (data not shown).
Although somatic action potentials and LTP on the priming pathway were blocked in the preceding experiment, the possible contribution of local action currents cannot be excluded. Thus, we repeated the experiment but applied priming stimulation to the stratum oriens. Somatic hyperpolarization in this case would completely prevent the conduction of membrane depolarization or action currents between the basal and apical dendritic compartments. Accordingly, cells were held at −90 mV during oriens priming. In these experiments both regular priming (25 ± 5%, n = 8; data not shown) and priming during hyperpolarization (25 ± 10%, n = 8) significantly inhibited LTP in radiatum compared with control LTP (58 ± 5%, n = 8, p = 0.002; Fig. 4C). Routine measurement of the peak AHP showed no change following either priming or conditioning in any group (F(1.256,2.512) = 1.046, p = 0.383). Likewise, no lasting changes were observed in input resistance (data not shown).
From the above experiments, we concluded that somatic action potentials are not necessary for the metaplasticity effect. Nonetheless, it remained conceivable that action potentials were sufficient to trigger the effect, as posited in the BCM model and has, with a reduced number of action potentials, been previously demonstrated to be sufficient under some conditions to inhibit subsequent LTP induction (Yasuda et al., 2003). To test this prediction, depolarizing current pulses of sufficient magnitude (2 nA, 2 ms) to regularly fire a single action potential per pulse were injected postsynaptically in the same pattern as the afferent stimulation normally given during the priming protocol. Of the 600 priming pulses, action potentials occurred in response to ∼70% of them, on average. Contrary to the prediction of reduced LTP following action potential firing, cells primed by depolarizing pulses showed a significantly enhanced LTP (95 ± 15%; p = 0.019; Fig. 4B). These data demonstrate that somatic action potentials were neither necessary nor sufficient to produce heterosynaptic inhibition of LTP by priming stimulation, and thus represent a major departure from the assumption of the original BCM model.
Heterosynaptic metaplasticity is not mediated by NMDARs, mGluRs or L-type voltage-dependent calcium channels
What other mechanisms besides action potentials could integrate the degree of prior activity as required for BCM-like metaplasticity? Derivatives of the BCM model have proposed a critical role for postsynaptic Ca2+ in this regard (Artola and Singer, 1993; Shouval et al., 2002b; Yeung et al., 2004). Although NMDARs were considered the likely source of calcium, alternative sources were not excluded (Yeung et al., 2004). Given the ubiquitous nature of Ca2+ as a second messenger and the diverse number of ways in which rises in free intracellular Ca2+ can be produced, we tested various calcium sources as the mechanism mediating generation of the metaplastic state.
To investigate the contribution of NMDARs, d-APV (50 μm) was included in the ACSF from the beginning of the experiment, for both control and primed slices, and then washed out immediately following priming stimulation of stratum radiatum afferents (or at an equivalent time in control slices). In control slices, conditioning stimulation applied to the test pathway induced 35 ± 5% LTP (n = 7; Fig. 5A, black/gray dashed line). In the primed group, however, a mixed result was obtained: 6/9 slices showed inhibited LTP comparable to the level observed in Figure 1 (Primed (I): 14 ± 3%, p = 0.003 when compared with control); however, 3/9 slices showed dramatically facilitated LTP (Primed (F): 88 ± 5%; Fig. 5A,C inset). The group average of these slices was not significantly different to control LTP (38 ± 13%, p = 0.84; Fig. 5A). Interestingly, no intermediate (control) levels of LTP were observed in the primed group (Fig. 5C inset), which suggests that the inhibitory and facilitatory effects may be competitive. These results indicate that the inhibitory priming of LTP is probably not dependent on NMDAR activation during priming, although this conclusion is tempered by the observation that the presence of APV appears to allow some slices to show facilitated LTP. Note that in these experiments, LTP induction by the priming stimulation was completely blocked by APV (data not shown), indicating that the heterosynaptic priming effects are not dependent on the induction of homosynaptic LTP on the primed pathway.
Voltage-dependent calcium channels (VDCCs) are another major source of postsynaptic Ca2+ elevations in response to synaptic activity. One of the most prevalent postsynaptic VDCCs is the L-type channel which is present predominantly at the soma but is also expressed in dendrites (Christie et al., 1995; Magee and Johnston, 1995). Importantly, L-type VDCCs are critical for LTP induction in response to particularly high-frequency stimulation protocols (Grover, 1998; Freir and Herron, 2003). We considered it likely that VDCCs were activated by our high-frequency priming stimulation, and thus hypothesized that they could be an NMDAR-independent source of calcium driving the metaplasticity effect. To test this hypothesis, we repeated the above experiment, but included the dihydropyridine L-type VDCC blocker nimodipine (10 μm) with APV in the ACSF before and during priming. In control slices, conditioning induced 29 ± 5% LTP (n = 6; Fig. 5B, black/gray dashed line, C inset). For primed slices, there was once again a bimodal LTP response. Inhibited LTP was observed in 6/9 slices (Primed (I): 15 ± 2%; p = 0.045). However, 3/9 slices again showed dramatically facilitated LTP (Primed (F): 61 ± 3%) and thus the group mean was not significantly different to control LTP (31 ± 8%, p = 0.85; Fig. 5B). This pattern of results was similar to that seen in APV-only conditions, and demonstrates that L-type VDCC activation during priming does not mediate either the heterosynaptic inhibition of LTP or the facilitation of LTP that occurs in some slices exposed to APV.
Activation of Group I mGluRs can lead to elevations in free intracellular Ca2+ through triggering release from intracellular stores via a phospholipase C/Gq-coupled pathway. Moreover mGluRs are critically involved in some types of metaplasticity (Bortolotto et al., 1994; Cohen et al., 1998, 1999; Mellentin and Abraham, 2001; Gisabella et al., 2003). To test the contribution of mGluRs to the present metaplasticity effect, the global mGluR antagonist LY341495 (100 μm) was included with APV before and during the priming stimulation. In control slices, conditioning stimulation produced 37 ± 6% LTP (n = 5; Fig. 5C). Primed slices showed a consistently inhibited LTP (20 ± 3%, n = 5; p = 0.031). This demonstrates that heterosynaptic inhibition of LTP is not mediated by either mGluR or NMDAR activation during priming. It is notable that no primed facilitation was observed in the LY341495-treated slices, consistent with previous reports of Group I mGluR antagonists blocking primed facilitation of LTP (Cohen and Abraham, 1996; Cohen et al., 1998; Mellentin et al., 2007). Moreover, the priming effect was fully preserved in the face of a block of LTP on the primed pathway by APV/LY341495 (data not shown). Together, these results suggest that under normal (no drug) conditions, priming stimulation activates both NMDARs and mGluRs but that mGluR-mediated facilitation of LTP is prevented by the concurrent NMDAR activation. Most importantly, however, simultaneous block of both receptor types failed to prevent the inhibition of LTP by priming stimulation.
Role of release of Ca2+ from intracellular stores
An alternative mechanism for elevating intracellular Ca2+ could be the mobilization of Ca2+ from intracellular stores, which is the target of a number of signaling pathways in addition to Group I mGluRs. To test this possibility we used CPA (50 μm), a reversible sarcoplasmic/endoplasmic reticulum Ca2+-ATPase pump inhibitor which, by preventing refilling, effectively empties the intracellular stores over a period of 30 min baseline stimulation (Seidler et al., 1989; Mellentin et al., 2007). In control slices given CPA plus washout, conditioning stimulation induced 31 ± 8% LTP (n = 5; Fig. 6A). Priming stimulation in the presence of CPA had no significant effect on subsequent LTP (28 ± 3%, n = 5, p = 0.672; Fig. 6A). There was also no difference between control and primed LTP when priming stimulation was given in stratum oriens in the presence of CPA (Fig. 6B). Finally, CPA also prevented facilitation of LTD by stratum oriens priming stimulation (Fig. 6C).
To determine whether the release of calcium from intracellular stores was mediated by IP3 or ryanodine receptor activation, we used the membrane-permeable IP3R blocker xestospongin C. When priming stimulation was delivered to stratum oriens in the presence of 3 μm xestospongin, there was no significant difference between control (28 ± 8%, n = 4; Fig. 6D) and primed (25 ± 5%, n = 4; p = 0.80) levels of LTP. Together these results support the fundamental principle of calcium-focused derivatives of the BCM model by demonstrating that the metaplastic state is mediated by a calcium signal, in this case generated by IP3R-mediated mobilization from intracellular stores.
Muscarinic acetylcholine receptor activation is a trigger for heterosynaptic metaplasticity
Having eliminated the most likely triggers for the mobilization of Ca2+ from intracellular stores (mGluRs, NMDARs and L-type VDCCs), we considered the alternative hypothesis that it was triggered instead by activation of mAChRs during priming stimulation, as these receptors are known to couple via IP3 to calcium release from the endoplasmic reticulum. In control slices given the mAChR antagonist atropine (10 μm) followed by washout, conditioning stimulation induced 25 ± 3% LTP (n = 6; Fig. 7A). When stratum oriens priming was given in the presence of atropine, the priming effect on LTP was blocked (28 ± 9%, n = 6, p = 0.73). As atropine is a nonspecific AChR antagonist, we also used the specific m1-AChR antagonist pirenzepine. In control slices given pirenzipine (20 μm) followed by washout, conditioning stimulation induced 23 ± 4% LTP (n = 5; Fig. 7B). When oriens priming was given in the presence of pirenzipine, the priming effect on LTP was blocked (23 ± 9%, n = 5, p = 1.00).
Discussion
The BCM computational model of synaptic modification, originally designed to explain experience-dependent development of visual cortex synaptic organization (Bienenstock et al., 1982), has played a prominent role in conceptualizing the relation between different activity patterns and the resultant direction and degree of synaptic plasticity. Key features of the model are the biphasic plasticity induction function (the “φ” function) that relates the degree of correlated presynaptic and postsynaptic activity with the degree of LTD/LTP response, the cell-wide representation of the φ function, the homeostatic metaplastic control of the modification threshold, θM, and the ongoing role of postsynaptic cell firing in determining the position of θM. These specific features of the model have driven numerous experimental efforts to confirm them. For example, a φ-type function has been described for both visual cortical and hippocampal excitatory synapses, typically through manipulation of the frequency of afferent synaptic activity, which is however only an indirect way of manipulating the degree of correlated presynaptic and postsynaptic activity as modeled by the BCM theory (Dudek and Bear, 1993; Kirkwood et al., 1996; Philpot et al., 2003, 2007; but see also Ngezahayo et al., 2000). Furthermore, the modification threshold has been reported to shift in an activity-dependent fashion, through for example dark-rearing in the case of visual cortex synapses, or afferent stimulation in the case of hippocampal synapses (Wang and Wagner, 1999; Abraham et al., 2001). In the latter case, the use of afferent priming stimulation has allowed tests that confirm that increased afferent activity can drive θM to the right of the φ function not only for the stimulated synapses, but also heterosynaptically for neighboring quiescent synapses, in accord with the BCM postulate of a cell-wide sliding modification threshold. We have replicated and extended these findings by showing in the hippocampal CA1 model that synaptic activity in the basilar dendrites can affect the φ function even at synapses terminating on dendrites located on the opposite side of the cell body. Interestingly, the metaplastic state induced by apical dendrite stimulation did not extend to the basilar dendrites, revealing a site-specific compartmentalization of the effect.
Lack of a role for action potentials in heterosynaptic metaplasticity
A critical prediction of the BCM model that has resisted direct testing in previous studies is that postsynaptic action potentials mediate changes in θM. While recent derivatives of the BCM model have focused more on calcium signals as the activity “integrator,” such signals could still be dependent on back-propagating action potentials that open voltage-gated calcium channels (Spruston et al., 1995; Stuart et al., 1997; Yasuda et al., 2003). Some studies have reported that chronic infusions of the GABAA receptor agonist muscimol during monocular deprivation block the ocular dominance plasticity that normally occurs in visual cortex (Reiter and Stryker, 1988; Frank et al., 2006). Such treatments, however, will affect any depolarization-mediated events, and so will not be selective for action potentials per se. In the hippocampus, we reported that antidromic activation of dentate granule cells in vivo could inhibit LTP induction in lateral perforant path synapses for days (Abraham et al., 2001). However, the use of field potential recording techniques precluded a definitive conclusion regarding the role of postsynaptic action potentials.
Here, we took advantage of the CA1 slice as a tractable experimental preparation for investigating the necessity and sufficiency of postsynaptic action potential firing using single neuron recordings. We found that cell firing was neither necessary nor sufficient for integrating the neural activity driving heterosynaptic metaplasticity in this preparation. Indeed, firing postsynaptic action potentials actually facilitated subsequent LTP, supporting previous findings of Dudek and Fields (2002). Instead, the metaplasticity was dependent on a calcium signal arising from intracellular stores. This signal mediated not only radiatum-radiatum interactions, but also the long-range interactions from stratum oriens to stratum radiatum. Thus, our evidence provides the first direct support for the calcium-focused versions of the BCM model, and specifically rules out a critical role of either postsynaptic action potentials or NMDAR activation in governing θM, at least in CA1.
Comparison with other models of heterosynaptic modulation of plasticity
The heterosynaptic inhibition of LTP by priming stimulation, observed in the present study as well as others (Wang and Wagner, 1999; Abraham et al., 2001), contrasts with the heterosynaptic upregulation of LTP persistence revealed by synaptic tag and capture experiments (Frey and Morris, 1997, 1998a,b). In these latter experiments, a weak decremental LTP is converted into a long-lasting protein synthesis-dependent LTP if a strong heterosynaptic HFS that activates protein synthesis is given shortly before or after the induction of the weak LTP. In the present experiments, in contrast, the strong priming stimulation should have easily been sufficient to increase protein synthesis, yet it strongly inhibited LTP on the test pathway. One explanation for these conflicting findings could rest on the fact that the stimulation for generating weak LTP is set well below the threshold for activating protein synthesis, commonly involving only a single train of ∼20 pulses (Frey and Morris, 1997). In the present experiment, a stronger HFS paradigm was used for the test pathway (typically 200 pulses). Thus there appears to be a complexity of heterosynaptic interactions operating at CA1 synapses such that very weak synaptic potentiation can benefit from capture of newly synthesized proteins, while induction of a stronger LTP is impaired by prior intense activation of other synapses.
Our findings also contrast with previous studies of competitive interactions between pathways for LTP resources. For example, Fonseca et al. (2004) showed that LTP induction on one pathway could impair maintenance of prior LTP on a second pathway, presumably by competing for scarce raw materials required for LTP expression as caused by partial inhibition of protein synthesis. Subsequently it was shown that widespread saturation of LTP can inhibit subsequent LTP on a second nearby pathway (Roth-Alpermann et al., 2006). In our experiments, however, no LTP needed to be generated on the priming pathway in order for LTP to be inhibited on the test pathway (see APV and hyperpolarization experiments). Moreover, the degree of LTP in the primed pathway, when it did occur, did not correlate with the degree of LTP inhibition. Thus the heterosynaptic metaplasticity in the present model was completely independent of prior LTP induction at the priming synapses. These data are consistent with a more regulated homeostasis of synaptic plasticity as prescribed by the BCM model, which helps preserve plasticity capability in the face of large-scale changes in afferent activity.
Expression mechanisms of heterosynaptic metaplasticity
The expression mechanism ultimately responsible for the altered LTP/LTD remains elusive. The effect was not due to altered GABAA receptor-mediated synaptic inhibition, nor could we find evidence for a change in basal membrane properties, in contrast to other metaplasticity reports (Cohen et al., 1999; Le Ray et al., 2004). We also considered the possibility that priming stimulation might lead to a cell-wide change in NMDAR activation. However, voltage-clamp recordings revealed that while priming stimulation did cause an LTD of NMDAR-mediated currents in the primed synapses, heterosynaptically there was no change in the single or summated NMDAR current amplitudes or waveshapes (Xu et al., 2009; Lee et al., 2010). This leads us to conclude that the mechanism probably involves alteration in molecules downstream from membrane depolarization and NMDAR activation that are responsible for the decreased LTP and facilitated LTD.
Overall, our findings provide strong support for hippocampal CA1 exhibiting a novel form of heterosynaptic interactions between inputs. On the one hand, it is BCM-like in that high levels of activity at some inputs can reduce LTP and enhance LTD at other synapses spread widely across the cell (Wang and Wagner, 1999; Abraham et al., 2001; Fonseca et al., 2004). On the other hand, this form of heterosynaptic metaplasticity diverges from the computational models in (1) being compartmentally restricted when priming stimulation is delivered to afferents in stratum radiatum, (2) being independent of postsynaptic action potentials in the affected cell, and (3) being driven by muscarinic rather than glutamatergic receptor activation. Activation of muscarinic receptors elevates intracellular concentrations of calcium in both neurons and glia in the hippocampus (Egorov and Müller, 1999; Beier and Barish, 2000; Araque et al., 2002; Perea and Araque, 2005), and it will require additional approaches to determine which cell types and signaling pathways mediate the metaplasticity effects. Regardless, these experiments raise the question whether BCM-like metaplasticity in other brain regions such as visual cortex involve mechanisms similar to those shown here for the hippocampus. Indeed, muscarinic receptor activation can regulate concurrently induced synaptic plasticity in visual cortex, as in hippocampus (Huang et al., 2012; Kirkwood et al., 1999; Seol et al., 2007), and therefore it may serve an important metaplasticity function there as well. However, it remains important to directly test whether BCM-like metaplasticity responses to severe and prolonged changes in synaptic input, such as dark-rearing or eyelid closure, involves action potential-dependent mechanisms as proposed in the computational models.
Footnotes
This research was supported by a grant from the New Zealand Marsden fund to W.C.A., a postgraduate scholarship from the University of Otago to S.R.H., and a Neurological Foundation of New Zealand W.B. Miller postgraduate scholarship to O.D.J.
The authors declare no competing financial interests.
- Correspondence should be addressed to Sarah R. Hulme, Brain Health Research Centre and Department of Psychology, University of Otago, PO Box 56, Dunedin 9054, New Zealand. hulme{at}psy.otago.ac.nz