Abstract
Synapses onto distal dendritic tufts are believed to function by modulating time-locked proximal inputs; however, the role of these synapses when proximal inputs are asynchronous or silent is unknown. Surprisingly, we found that activation of apical tuft synapses alone resulted in heterosynaptic potentiation of proximal synapses. In mouse adult hippocampal CA1 pyramidal neurons, we show that activation of distal inputs from the entorhinal cortex (EC) specifically strengthens proximal synapses projecting from CA3. This slow AMPA receptor-mediated potentiation is accompanied by increased synaptic GluN2B-containing NMDA receptors, which are normally restricted to juvenile animals. These two synaptic modifications interact to generate striking bidirectional metaplastic changes. Heterosynaptically potentiated synapses become resistant to subsequent long-term potentiation (LTP) as the two forms of AMPA receptor-mediated potentiation occlude. However, this is only true when the LTP induction protocol is relatively weak. When it is strong and repeated, the magnitude of LTP after heterosynaptic plasticity is greatly increased, specifically through the activation of GluN2B-containing NMDA receptors. Thus, CA1 neurons expressing heterosynaptic potentiation induced by external sensory input from the EC become more strongly driven by internally generated environmental representations from CA3. Furthermore, subsequent SC LTP in this ensemble is shifted to potentiate only strongly activated CA3 inputs, while endowing these synapses with enhanced potentiation. These results show that one set of inputs can exert long-lasting heterosynaptic control over another, allowing the coupling of two functionally and spatially distinct pathways, thereby greatly expanding the repertoire of cellular and network plasticity.
Introduction
Synaptic currents originating in the apical dendritic tuft are greatly attenuated by electrotonic decay, and even distal dendritic spikes evoked by the strong simultaneous activation of tuft synapses are unable to reliably drive somatic action potentials (APs) (Schiller et al., 1997; Golding and Spruston, 1998; Jarsky et al., 2005). Distal dendritic signals can interact with coincident proximal inputs to either boost or inhibit activity, but this modulation requires coactivation of both pathways within a narrow time window (Larkum et al., 1999; Remondes and Schuman, 2002; Dudman et al., 2007; Takahashi and Magee, 2009). Thus, it remains unclear how, or even if, distal synapses can impact neuronal function when proximal input is not coincident or absent.
We studied this question in the output layer of the hippocampus, CA1, which receives two well-segregated synaptic inputs carrying two distinct types of information. Entorhinal cortical inputs synapse on the distal apical tuft and carry path integration signals such as grid cell firing from the medial entorhinal cortex (Sargolini et al., 2006) as well as nonspatial information from the lateral entorhinal cortex (Hargreaves et al., 2005). Schaffer collateral (SC) synapses from hippocampal CA3 innervate the proximal apical and basal dendrites and carry internally generated spatial maps.
Previous studies suggest that the direct entorhinal cortical inputs onto the distal apical tuft have the unusual ability to heterosynaptically control synaptic plasticity at distant SC synapses. These studies activated tuft synapses at low frequency (1 Hz for 900 s) and found intriguing, though somewhat contradictory, results. One study found that low-frequency stimulation produced heterosynaptic potentiation at SC synapses (Wöhrl et al., 2007), whereas another found that the same stimulation protocol did not affect basal synaptic strength at SC synapses, but rather depotentiated previously potentiated SC synapses (Izumi et al., 2008). Rather than using low-frequency stimulation, we used high-frequency activation, similar to the firing of these afferents in vivo (Fyhn et al., 2008), to investigate how entorhinal cortical inputs may control SC synapses.
Using a combination of slice electrophysiology and calcium imaging, we found that activation of dendritic tuft synapses resulted in the long-lasting modification of proximal apical synapses, by both potentiating AMPA receptor-dependent synaptic transmission and increasing the contribution of synaptic GluN2B, the predominant NMDA receptor subunit in juvenile animals. As a result, SC synapses in the apical dendrite are strengthened but also shift their input selectivity to specifically potentiate strong afferents from CA3 via the activation of GluN2B.
Materials and Methods
Slice preparation.
Horizontal hippocampal slices (400 μm) were made from young male adult C57/BL6 mice (aged 6–15 weeks for field recordings and 6–8 weeks for whole-cell recordings), in compliance with the animal use and care guidelines of the Salk Institute. Slicing was done on a Leica VT1200S in ice-cold sucrose-substituted artificial cerebrospinal fluid (ACSF) containing the following (in mm): 25 NaHCO3, 3 KCl, 85 NaCl, 1.25 NaH2PO4, 75 sucrose, 25 glucose, 0.5 sodium ascorbate, 0.5 CaCl2, and 4 MgCl2. Slices recovered at 32–33°C for 1.5 h in recording ACSF containing (in mm): 25 NaHCO3, 3 KCl, 125 NaCl, 1.25 NaH2PO4, 25 glucose, 1.3 CaCl2, and 1 MgCl2, and then held at room temperature. Slices were microdissected before recording by removing the dentate gyrus and CA3 and making small cuts in the stratum lacunosum moleculare (SLM) and SC to isolate stimulation of the SC and SLM pathways (Fig. 1A).
Electrophysiology.
Recordings were performed at 32–33°C at least 3 h after slicing since baseline synaptic strength is not completely stable at earlier times. Concentric bipolar electrodes delivered electrical stimulation (100 μs duration) at 0.0167 or 0.0333 Hz. For extracellular field potential recordings, recording pipettes ranged from 1 to 3 MΩ and were filled with ACSF. SLM stimulation intensity was set at 40–50% of maximal field EPSP (fEPSP) peak, and the same intensity was used to deliver SLM theta-burst stimulation (TBS). Slices with a half-maximal SLM fEPSP smaller than 0.1 mV were not used. SC stimulation was set so fEPSP peak was 0.3–0.4 mV. For whole-cell experiments, borosilicate glass pipettes ranged from 2 to 4 MΩ.
For repatch experiments, intracellular recording solution contained the following (in mm): 140 K+-gluconate, 10 NaCl, 10 HEPES, 0.1 EGTA, 3 MgATP, 0.3 LiGTP, 10 phosphocreatine-Tris, and 0.05 Alexa 555, pH 7.24, osmolarity 310–320. Baseline recordings (5 min duration) began 10 min after break in and were performed in voltage clamp with the cell held at the calculated Cl− reversal potential (−67 mV) after correcting for the calculated liquid junction potential (15.4 mV). After pulling off the cell, we waited at least 20 min before delivering an SLM TBS to allow the cell to recover from patching. Alexa-filled neurons were targeted for repatch one to three times to obtain a satisfactory series resistance, Rs (±20% of baseline Rs, all recordings, <20 MΩ). BAPTA experiments were done with the same methodology except internal solution substituted 5 mm BAPTA-tetrapotassium salt for EGTA, and K+-gluconate was lowered to 115 mm.
For loose-patch experiments, SLM and SC field stimulation intensity was set as described above for extracellular field recordings. We whole-cell patched a CA1 pyramidal neuron, recorded EPSCs, and pulled off the cell, as described for repatching. We then loose-patched the same CA1 pyramidal neuron using a borosilicate glass pipette (2–4 MΩ) filled with external solution. We continuously monitored membrane resistance, capacitance, and series resistance to ensure that we did not accidentally break into the cell. We never spontaneously broke into a cell while loose patched. To verify the quality of recording, we looked for high signal-to-noise capacitative transients representing APs by first giving a short train of three SC stimuli at 100 Hz. If we did not see an evoked AP, we simply waited to observe a spontaneous AP. Then we delivered a typical SLM TBS (10 stimuli at 100 Hz, repeated 10 times at 5 Hz) to look for somatic spikes. In three neurons, after recording the SLM TBS in loose-patch configuration, we pulled off the loose patch and waited >60 min to allow expression of heterosynaptic plasticity. Then we repatched the same neuron and measured the SC EPSC.
For AMPA/NMDA EPSC ratio recordings, heterosynaptic potentiation was induced while recording fEPSPs. Approximately 60 min after SLM TBS (or controls), we washed in 6-Imino-3-(4-methoxyphenyl)-1(6H)-pyridazinebutanoic acid hydrobromide (SR95531; 10 μm; Ascent Scientific) and patched a neuron [internal solution contained (in mm) 140 Cs+-gluconate, 10 CsCl, 10 HEPES, 0.5 EGTA, 3 MgATP, 0.3 LiGTP, and 10 phosphocreatine-Tris, pH 7.24, with osmolarity 310–320]. After 10 min, we recorded AMPA receptor-mediated currents in voltage clamp at −70 mV for 10 min at one stimulation per minute. Stimulation intensity was set so EPSCs were 100–200 pA. Then we washed in NBQX (10 μm; Ascent Scientific), held the cell at +30 mV, and recorded NMDA receptor-mediated currents at one stimulation per minute. AMPA/NMDA ratios were calculated using EPSC peaks. For measuring ifenprodil sensitivity, after 10 min of baseline recording of NMDA receptor-mediated currents, we washed in ifenprodil (3 μm; Tocris Bioscience) for 35 min, then measured the NMDA receptor-mediated EPSC for 15 min, then washed in d,l-APV (100 μm; Ascent Scientific). Only experiments where Rs was low and stable were analyzed (±20% of baseline Rs; all recordings, <20 MΩ). For all experiments using ifenprodil (see Figs. 8, 9), we used paired-pulse stimulation with a 50 ms interval to accelerate GluN2B block by the use-dependent ifenprodil (Barria and Malinow, 2005). Only the first pulse was measured for the experiments shown. For experiments looking at metaplasticity, SC stimulus intensity was turned down after heterosynaptic potentiation so fEPSPs were approximately the same size as before potentiation. For field experiments examining SC LTP after heterosynaptic plasticity (see Fig. 7), we stimulated two independent SC pathways in each slice. For quantification of heterosynaptic plasticity, we averaged both SC pathways, since the expression of heterosynaptic plasticity is not input specific. Subsequent SC LTP was induced in one pathway, and the resulting synaptic strength was normalized to the naive pathway.
Calcium imaging.
SLM and SC field stimulation intensity was set as described above for extracellular field recordings. We then whole-cell patched a neuron with a pipette containing K+-gluconate internal and 20 μm Alexa 555, with either 100–200 μm Oregon Green BAPTA-1 (OGB-1) or 200 μm Fluo-5F. We allowed dye to fill distal dendrites of the neuron for 50–60 min before imaging. Images near the stratum radiatum (SR)/SLM were collected 40× magnification at 30 Hz with a cooled CCD camera (Andor iXon DU885) triggered by pClamp to synchronize imaging with electrophysiology. Excitation light was supplied by an Exfo Xcite system. The approximate border between the SR and SLM was identified based on the dark appearance of the SLM due to the myelinated fibers projecting from the entorhinal cortex.
Data acquisition and analysis.
Recordings were obtained with a Multiclamp 700B (Molecular Devices) under the control of pClamp (Molecular Devices), filtered at 2 kHz and digitized at 10 kHz by a Digidata (Molecular Devices). All data analysis was performed in Matlab, Excel, and Prism. For calcium imaging, data were analyzed using custom-written Matlab routines. Regions of interest along the apical dendrite were selected, and three nearby background regions were averaged and used to subtract baseline fluorescence before calculating the ΔF/F. Average values are means ± SEM. Statistical comparisons were made using nonparametric tests: Mann–Whitney for pairwise comparisons, and Kruskal–Wallis with a post hoc Dunn's multiple comparison as indicated in the text.
Results
Initial demonstration of heterosynaptic potentiation
We recorded from acute hippocampal slices from young adult mice (6–15 weeks) that were microdissected to allow independent stimulation of the distal dendritic pathway in SLM and proximal pathway in stratum radiatum (Fig. 1A) (Remondes and Schuman, 2002). Stimulation in the SLM activates perforant path fibers originating in layer III of the entorhinal cortex and a smaller projection from the thalamic reuniens (Wouterlood et al., 1990; Otmakhova et al., 2002; Remondes and Schuman, 2002), whereas stimulation in the stratum radiatum activates SC axons from CA3 of the hippocampus. Inhibition was intact, and recordings performed at 32–33°. We monitored synaptic transmission by measuring the initial slope of the extracellular fEPSP, which reports AMPA-receptor mediated synaptic strength.
Theta-burst stimulation of the SLM (five stimuli at 100 Hz, repeated 10 times at 5 Hz, set at 40–50% of maximal intensity) had no long-lasting effect on SLM synaptic strength compared to control slices that did not receive a TBS (measured at 50–60 min after TBS and normalized to −20–0 min baseline; controls 1.02 ± 0.04, n = 7; TBS, 0.99 ± 0.03, n = 11; p = 0.47; Fig. 1B,C, left). In contrast, the SC pathway exhibited a slow increase in synaptic transmission (controls, 1.04 ± 0.04, n = 7; TBS, 1.23 ± 0.05, n = 11; p < 0.05; Fig. 1B–D). Synaptic strength in the SC pathway was unchanged immediately after SLM TBS, indicating no posttetanic potentiation at SC synapses, suggesting that SC axons were not cross-stimulated by the SLM stimulating electrode (Fig. 1B,C). See Figure 5 for more details on the pathway-specificity of SLM stimulation.
Further characterization of heterosynaptic potentiation
This exclusively heterosynaptic potentiation of SC synapses was restricted to the apical dendrite as two-pathway experiments in the SR and stratum oriens (SO) showed significant potentiation in only the SR pathway (SR with SLM TBS, 1.28 ± 0.03, n = 5, p < 0.01; SO with SLM TBS, 1.01 ± 0.03, p = 0.66; Fig. 2A). Heterosynaptic potentiation was not caused by test pulses in either the SLM or SC (no SLM test pulses, 1.34 ± 0.09, n = 3 vs with SLM test pulses, 1.23 ± 0.05, n = 11; p = 0.29; potentiation of SC pathway stimulated with test pulses throughout experiment, 1.31 ± 0.03, n = 4 vs SC pathway with test pulses shut off until 50–60 min after TBS, 1.47 ± 0.08, n = 4; p = 0.26, Fig. 2C). Finally we investigated the pathway specificity of the induction of heterosynaptic potentiation. Is SLM stimulation required for heterosynaptic plasticity or can it be induced by SC synapses? One unusual aspect of the SLM TBS we use to induce heterosynaptic potentiation is that it is too weak to produce homosynaptic LTP at SLM synapses (Fig. 1B,C, left). Perhaps stimulation that is subthreshold for LTP is sufficient to induce heterosynaptic potentiation. To test this hypothesis, we delivered a weak TBS to one SC pathway and looked for heterosynaptic potentiation in an independent SC pathway. Administering a TBS protocol subthreshold for inducing homosynaptic LTP to one SC pathway (three stimuli at 100 Hz, repeated eight times at 5 Hz) did not produce heterosynaptic potentiation in an independent SC pathway (at 30–40 min after TBS, stimulated SC pathway, 0.99 ± 0.06, n = 5; no TBS SC pathway, 0.93 ± 0.05, n = 5; p = 0.54; Fig. 2D). We also note that suprathreshold (i.e., producing LTP) stimulation of one SC pathway does not produce heterosynaptic potentiation in an independent SC pathway (Fig. 7B). To summarize, heterosynaptic potentiation at SC synapses in response to SLM stimulation is restricted to the apical dendrite and does not spread to basal dendrites, does not depend on continuous test stimulation in either the SLM or SC pathway, and cannot be produced by stimulation of SC synapses that is subthreshold for LTP induction.
Heterosynaptic potentiation measured in single neurons
These experiments were done using extracellular field recordings that average signals from many local synapses; however, what is the response of individual neurons? There are two possibilities: (1) no potentiation in some cells averaged with large potentiation in others, or (2) uniform potentiation across the population. To distinguish between these possibilities, we turned to whole-cell patch clamp. To prevent washout of LTP, we developed a whole-cell repatching technique that allows for low series resistance (Rs) recording (in contrast to perforated-patch recordings), while limiting dialysis of intracellular components (Fig. 3A). Cells were patched with a K+-gluconate-based internal solution containing Alexa 555 to label the cell and held at Cl− reversal potential in voltage clamp to record responses to SC stimulation. After recording baseline EPSCs, the pipette was gently pulled off the cell, allowing the membrane to reseal. After a 20 min interval to allow the cell to recover from patching, slices either received SLM TBS or no stimulation (controls). Approximately 60 min later, to allow expression of plasticity, the same cell was targeted for repatch and EPSCs measured again. Only recordings where the Rs was stable within ±20% between the initial patch and repatch were included. There was no difference in the initial Rs of control and SLM TBS neurons (controls, 10.99 ± 0.81 MΩ, n = 6; SLM TBS, 12.21 ± 1.01 MΩ, n = 8; p = 0.57; Fig. 3D, inset). In addition, we also monitored the extracellular field potential in the same slice, allowing us to compare potentiation in a single cell versus the rest of the population.
Control neurons showed no change in the SC EPSC after repatching (sample experiment in Fig. 3B), verifying the stability of the repatch technique (control EPSC, 1.03 ± 0.05, n = 6; control fEPSP, 1.10 ± 0.10, n = 6; Fig. 3D, open triangles). In contrast, most neurons that received SLM TBS showed significant SC potentiation (sample experiment in Fig. 3C; TBS EPSC, 1.41 ± 0.11, n = 8, p < 0.05 compared to control EPSC; TBS fEPSP, 1.42 ± 0.09, n = 8, p < 0.05 compared to control fEPSP; Fig. 3D, open circles). Plotting response amplitude in the patched neuron versus the potentiation in the population revealed a good correspondence between the averaged single-cell and population data (Fig. 3D). Thus, we conclude that heterosynaptic potentiation is not restricted to a subset of cells, but rather occurs in a majority of neurons in the population. We note that in this experiment we failed to induce heterosynaptic plasticity in a relatively high percentage of slices (2 of 8, or 25%), in which case both the fEPSP and EPSC remained near baseline levels. This failure rate is somewhat higher than that observed in experiments with only extracellular recordings (∼10%). This is likely because in extracellular recording-only experiments, we continuously monitor SLM field strength before taking the baseline and are able to turn up stimulation strength in case the SLM fEPSP falls below 40% of maximal. For experiments using whole-cell patch, we set the SLM stimulation strength using field recording, but then switch to using that manipulator for patching a cell, so we are unable to monitor SLM field strength just before taking baseline synaptic strength measurements. Thus, the failure to induce heterosynaptic plasticity in some slices is likely due to subthreshold stimulation of distal dendritic synapses.
Role of intracellular Ca2+
Within an individual neuron, elevated Ca2+ is a common trigger for synaptic plasticity (Malenka et al., 1988). To probe the role of cell-autonomous Ca2+ signaling, we used the repatch technique to specifically load only patched neurons with the Ca2+ chelator BAPTA (5 mm) during the baseline recording period. Indeed we found that BAPTA blocked heterosynaptic plasticity in loaded cells (control BAPTA EPSC, 1.07 ± 0.08, n = 5 vs TBS BAPTA EPSC, 1.05 ± 0.12, n = 4; p = 1; Fig. 3E), while the remaining population of unloaded neurons in the same slice displayed normal potentiation (control fEPSP, 1.15 ± 0.05, n = 5 vs TBS fEPSP, 1.61 ± 0.17, n = 4; p < 0.05; Fig. 3E,F). There was no difference in the initial Rs of control and SLM TBS neurons (controls, 14.71 ± 1.31 MΩ; SLM TBS, 10.87 ± 1.05 MΩ; p = 0.29; Fig. 3F, inset). These results indicate that cell-autonomous Ca2+ signaling is required for heterosynaptic potentiation.
Role of somatic APs
One way SLM activation could elevate neuronal calcium is by evoking somatic action potentials. To assess the role of somatic APs in triggering heterosynaptic plasticity, we used a variant of the repatch technique. First we whole-cell patched a neuron, recorded EPSCs, and pulled the pipette off the cell, identical to a repatch experiment. Then we brought in a pipette filled with ACSF and loose patched the same cell. In this configuration, the pipette tip is nestled up against the cell, but not broken through the membrane, allowing the recording of APs as capacitative transients without disturbing the intracellular contents of the cell. To verify the quality of the recording, we tried to evoke an AP with three stimulations of the SC pathway at 100 Hz. If that did not evoke an AP, we waited for a spontaneous AP to gauge recording quality. All recordings showed APs with a high signal-to-noise ratio (Fig. 4A). None of the recorded neurons exhibited spiking in response to SLM TBS (0/18; Fig. 4B,C). We repatched three of the 18 neurons, verifying that they were still healthy, and showed normal heterosynaptic potentiation (1.45 ± 0.16, n = 3; Fig. 4D). These data indicate that somatic APs are not required for the induction heterosynaptic plasticity.
Spatial extent of Ca2+ influx during SLM TBS stimulation
Together, our data suggest that stimulation of distal tuft synapses generates Ca2+ transients required for heterosynaptic potentiation. These Ca2+ transients could produce potentiation in SC synapses in one of two ways: (1) transients are spatially restricted to the SLM region and active a long-range signaling cascade targeted at distant SC synapses or (2) transients generated in the SLM extend into the SR, producing direct calcium-mediated potentiation at SC synapses. To distinguish between these two possibilities, we directly measured the spatial extent of Ca2+ influx during SLM TBS using Ca2+ imaging. We patched a CA1 neuron, filling with either OGB-1 (100–200 μm, n = 6; Fig. 5A) or Fluo-5F (200 μm, n = 3), and allowed dye to load distal dendrites of the neuron for 50–60 min before imaging. We marked the approximate SLM/SR border based on the dark appearance of the myelinated axons from the entorhinal cortex versus the phase bright axons from CA3. We imaged a section of the apical dendrite straddling the SLM/SR border and recorded both Ca2+-induced fluorescence signals and membrane potential (Vm) in current clamp. We found that SLM TBS generated large Ca2+ transients that were largely restricted to the distal apical tuft (Fig. 5B, top left). The simultaneously recorded somatic Vm showed a very modest depolarization (Fig. 5B, bottom left), consistent with our earlier finding that SLM stimulation does not generate somatic APs (Fig. 4). As a positive control for detecting fluorescence changes in the proximal apical dendrite, we also gave a short burst of SC stimulation resulting in a backpropagating AP (bAP; Fig. 5B, bottom right). Ca2+ transients associated with the bAP were easily detectible throughout the apical dendrite (Fig. 5B, top right). Pooled data from all slices fit with a fourth-order polynomial (R2 = 0.65) show that the maximal intracellular Ca2+ signal in response to SLM stimulation decreased dramatically at the SLM/SR border (Fig. 5C). To verify that the dF/F reported by the high-affinity Ca2+ sensor OGB-1 (Kd = 170 nm) was not decreased by Ca2+ saturation, we compared OGB-1 measurements to the low-affinity sensor Fluo-5F (Kd = 2300 nm). We binned the dF/F for each analysis window measured from 20 μm within the SLM to the putative SLM/SR border and compared the two sensors. There was no difference in the dF/F reported by the two sensors (OGB-1, 12.02 ± 2.49%, n = 8 analysis windows from six neurons vs Fluo-5F, 6.95 ± 2.03%, n = 5 from three neurons), nor was there a difference in the mean position of the analysis windows between the two sensors (OGB-1, −7.09 μm ± 2.46 from the SLM/SR border, n = 8 vs Fluo-5F, −8.31 μm ± 2.89, n = 5), indicating that OGB-1 was not appreciably saturated by Ca2+ binding. These data rule out direct calcium influx at SC synapses and suggest a long-distance signaling pathway induced in the SLM and directed to SC synapses. We also note that these data showing no Ca2+elevation in the SR, together with the lack of post-tetanic potentiation at SC synapses in response to SLM TBS (Fig. 1B), illustrate that the SLM stimulation used for generating heterosynaptic potentiation is restricted to SLM synapses and does not produce significant direct activation of SC synapses.
Signaling pathways required for the induction of heterosynaptic potentiation
Next we investigated the signaling pathways required for heterosynaptic potentiation. NMDA receptors play a critical role in the induction of many forms of synaptic plasticity (Morris et al., 1986). We found that blocking NMDA receptors with d,l-APV (100 μm) or 3-((R)-2-carboxypiperazin-4-yl)-propyl-1-phosphonic acid (R-CPP; 100 μm) did not significantly inhibit heterosynaptic potentiation (1.17 ± 0.03, n = 5, p > 0.05 compared to TBS; Fig. 6A,E). Instead, previous studies have implicated R-/T-type Ca2+ channels in the induction of synaptic plasticity evoked by distal dendritic stimulation (Golding et al., 2002; Takahashi and Magee, 2009). Indeed we found that blocking R-/T-type Ca2+ channels with NiCl (50 μm) completely prevented heterosynaptic potentiation (0.99 ± 0.04, n = 5; p < 0.05 compared to TBS; Fig. 6B,E), without affecting basal synaptic transmission in the SLM pathway.
Next we examined the role of GABAA receptor-mediated inhibition on heterosynaptic potentiation. In contrast to homosynaptic LTP which does not require inhibition, we found that blocking GABAA receptors throughout the experiment with SR95531 (3 μm) prevented heterosynaptic potentiation (1.00 ± 0.04, n = 3, p > 0.05; Fig. 6C,E). This result raises two possibilities for the role of GABAA-mediated inhibition: (1) it is required for the induction of heterosynaptic potentiation, or (2) the heterosynaptic potentiation we observe is actually due to the slow disinhibition of SC synaptic strength and we do not observe “potentiation” in these experiments because there is no inhibition to be lost. The second possibility is unlikely for two reasons. First, to measure our extracellular field potentials (Fig. 1), we measured the initial slope of the field PSP which is not contaminated by feedforward inhibition. Second, blocking intracellular Ca2+ elevation with BAPTA prevented heterosynaptic potentiation (Fig. 3). Still, to directly rule out a role for disinhibition in the expression of heterosynaptic potentiation, we gave SLM TBS in normal ACSF and quickly washed in SR95531. Because the kinetics of expression of heterosynaptic potentiation are slow, we were able to fully block GABAA receptors before any significant expression of heterosynaptic potentiation. We renormalized SC synaptic strength during this period (5–10 min after SLM TBS) and found that wash-in of SR95531 after SLM TBS did not block heterosynaptic potentiation (1.27 ± 0.07, n = 9, p < 0.05 compared to no SLM TBS, 0.99 ± 0.09, n = 7; Fig. 6B,C,E). Thus, we conclude that GABAA receptor-mediated inhibition is required for the induction of heterosynaptic potentiation.
Next we tested the contribution of several signaling pathways implicated in synaptic plasticity. We found that heterosynaptic potentiation did not depend on L-type Ca2+ channels (nimodipine, 1.23 ± 0.05, n = 6), Ca2+ release from internal stores (thapsigargin, 1.20 ± 0.08, n = 4), dopamine signaling through D1/D5 receptors (SCH23390, 1.19 ± 0.07, n = 5), muscarinic acetylcholine receptors (atropine, 1.31 ± 0.14, n = 4), or mGluR5 signaling (MPEP, 1.23 ± 0.10, n = 5; all conditions compared including SLM TBS with no inhibitors; p = 0.99 by Kruskal–Wallis test; Fig. 6E). To summarize, in marked contrast to homosynaptic LTP, the induction of heterosynaptic potentiation does not depend of NMDA receptors, but instead requires both R-/T-type Ca2+ channels and GABAA receptor-mediated inhibition.
Expression mechanisms of heterosynaptic potentiation
Paired-pulse ratio
While the induction mechanisms of heterosynaptic potentiation are distinct from LTP, how similar are their expression mechanisms? Homosynaptic LTP is believed to be postsynaptically expressed by the insertion of AMPA receptors, whereas NMDA receptor-mediated transmission remains unchanged. First we looked for changes in presynaptic function after heterosynaptic potentiation by using extracellular field recording to measure the paired-pulse ratio (PPR), an indicator of the presynaptic probability of transmitter release. We found that the PPR was not altered after heterosynaptic potentiation (SC PPR baseline, 1.60 ± 0.09 vs PPR at 50–60 min after TBS, 1.66 ± 0.06; n = 4, p = 0.25; Fig. 7A, left), while normal heterosynaptic potentiation was seen in these experiments (1.17 ± 0.05, n = 4; Fig. 7A, right), consistent with little or no change in the presynaptic probability of release. However, PPR is only an approximation for presynaptic function and does not always correlate with probability of release (Deng et al., 2011), so although our results are consistent with no presynaptic alterations, we cannot rule out the possibility that changes in presynaptic function contribute to heterosynaptic potentiation.
Occlusion by prior LTP
Next we examined how similar our AMPA receptor-mediated heterosynaptic potentiation was to homosynaptic AMPA receptor LTP by testing whether LTP could occlude the expression of heterosynaptic potentiation. To do this, we stimulated two independent SC pathways and saturated homosynaptic LTP in one pathway using a TBS protocol repeated four times at 5 min intervals, and then induced heterosynaptic potentiation by SLM TBS and looked for differences in heterosynaptic potentiation in the saturated and the naive pathways (Fig. 7B, top). For display purposes only, we renormalized synaptic strength in the saturated pathway to facilitate comparison to the naive pathway. SC LTP remained saturated throughout the experiment as another 4× TBS delivered to the saturated pathway at the conclusion of the experiment did not induce further LTP (data not shown). Control slices did not receive SLM TBS, and both SC pathways were stable until the end of the experiment (after renormalization, SCsaturated at the end of the experiment, 1.02 ± 0.01 vs SCnaive, 1.04 ± 0.04; n = 7, p = 0.61; Fig. 7B, bottom left). SLM TBS induced normal heterosynaptic potentiation in the naive SC pathway, but heterosynaptic potentiation was significantly occluded by SC LTP in the saturated pathway (SCnaive, 1.23 ± 0.06 vs SCsaturated, 1.08 ± 0.03; n = 6, p < 0.05; Fig. 7B, bottom right). This occlusion suggests that heterosynaptic and homosynaptic AMPA receptor potentiation share some common expression mechanism. It remains unclear whether heterosynaptic potentiation depends on AMPA receptor insertion, or if potentiation is occluded at an earlier step in the signal transduction pathway.
AMPA/NMDA EPSC ratio after heterosynaptic potentiation
To further probe the relationship between heterosynaptic potentiation and homosynaptic LTP, we looked for changes in the AMPA/NMDA ratio. Homosynaptic SC LTP is characterized by an increase in the AMPA/NMDA ratio, reflecting the potentiation of AMPA receptor-mediated signaling, whereas NMDA receptor-mediated signaling is unchanged (Kauer et al., 1988). Since we found that the AMPA-receptor mediated heterosynaptic potentiation was occluded by homosynaptic LTP, we expected the AMPA/NMDA ratio to increase after heterosynaptic potentiation. We measured the AMPA/NMDA ratio by inducing heterosynaptic plasticity, and then after the expression of potentiation, patching an individual neuron and measuring the AMPA (at −70 mV) and NMDA (at +30 mV) components (Fig. 7C, top left). To our surprise, we found that the AMPA/NMDA ratio of peak EPSCs was unchanged after heterosynaptic potentiation (AMPA/NMDA ratio in control neurons, 5.49 ± 0.77, n = 6 vs after SLM TBS, 5.64 ± 0.62, n = 5; p = 0.93; Fig. 7C, right, top, bottom), suggesting that NMDA signaling had increased concomitantly with AMPA signaling.
NMDA receptor contribution to heterosynaptic potentiation
NMDA receptors consist of a mixture of GluN2A and GluN2B subunit-containing receptors, with GluN2A dominating in adult neurons (Monyer et al., 1992; Williams et al., 1993). Increased NMDA receptor signaling could be achieved by preferentially increasing current mediated either by GluN2A subunits, by GluN2B, or by scaling up the contribution of both subunits equally. We distinguished between these possibilities by recording synaptic NMDA currents and then washing in the GluN2B-specific antagonist ifenprodil (3 μm) (Williams et al., 1993). Ifenprodil had little effect on NMDA EPSC peak current and area in control slices, inhibiting 10.84 ± 3.87% and 10.36 ± 5.28%, respectively (n = 6; Fig. 8A,C), consistent with the downregulation of this receptor subunit in mature animals. In contrast, neurons that received SLM TBS showed increased sensitivity to ifenprodil, with NMDA peak current inhibition of 33.78 ± 7.30% and NMDA area inhibition of 37.67 ± 6.26% (n = 5), significantly more inhibited by ifenprodil than control slices (p < 0.05; Fig. 8B,C). These data show that NMDA receptor signaling at SC synapses is shifted after stimulation of the direct cortical inputs, from a mature, GluN2A-dominated NMDA current to a more juvenile composition with significant contribution from GluN2B-containing receptors.
To summarize, the expression of heterosynaptic potentiation consists of both AMPA and NMDA receptor-mediated components. The AMPA receptor-mediated component is postsynaptic and occluded by previous homosynaptic LTP, suggesting shared expression mechanisms. However, unlike homosynaptic LTP, heterosynaptic potentiation has an additional NMDA receptor-mediated component, specifically, the increased contribution of juvenile GluN2B subunits in these mature mice.
These findings predict opposite metaplastic effects of heterosynaptic plasticity on subsequent SC signaling. On the one hand, the increase in synaptic GluN2B predicts a greater magnitude of SC LTP (Kirkwood et al., 1996) via greater Ca2+ influx and coupling to downstream effectors (Barria and Malinow, 2005; Sobczyk et al., 2005). On the other hand, homosynaptic SC LTP occludes heterosynaptic potentiation (Fig. 7B), suggesting less SC LTP after heterosynaptic potentiation. We hypothesized that these two effects might be differentially engaged as a function of SC stimulus intensity and tested this prediction using either weak (one TBS of 30 stimuli) or strong stimulation (four TBS at 5 min intervals, 50 stimuli per burst, 200 stimuli total) to trigger homosynaptic SC LTP after prior heterosynaptic potentiation. In control slices that did not receive SLM TBS, we found that the weak TBS protocol produced stable SC LTP (1.20 ± 0.04, n = 6; Fig. 9A). In slices that received SLM TBS, heterosynaptic potentiation developed normally (1.15 ± 0.04, n = 7 vs controls, 1.01 ± 0.03, n = 6; p < 0.05); however, subsequent weak TBS produced LTP that was significantly occluded (1.03 ± 0.02, n = 6, p = 0.01; Fig. 9A), consistent with the occlusion seen in Figure 7B, where SC LTP was expressed before heterosynaptic plasticity.
In marked contrast, when SC LTP was induced using the strong TBS protocol, SC LTP was significantly larger in slices that had undergone prior heterosynaptic plasticity compared to LTP from control slices (control LTP, 1.25 ± 0.03, n = 10 vs LTP after SLM TBS, 1.45 ± 0.05, n = 10; p < 0.01 by Kruskal–Wallis; p < 0.05 by Dunn's multiple comparison; Fig. 9B). Enhanced SC LTP in slices after heterosynaptic plasticity was dependent on the activity of GluN2B-containing NMDA receptors, as blocking these receptors with ifenprodil before SC TBS decreased LTP to 1.12 ± 0.06 (n = 6), significantly smaller than LTP in untreated SLM TBS slices (p < 0.01 by Dunn's multiple comparison; Fig. 9B). Ifenprodil itself had no effect on SC LTP in control slices (control LTP, 1.25 ± 0.01, n = 6 vs LTP with ifenprodil, 1.26 ± 0.08, n = 4; p = 0.91; Fig. 9C).
To illustrate the differential effects of heterosynaptic plasticity on moderate and strong SC stimulation, we plotted the amount of LTP versus the number of stimulations in the SC theta burst (Fig. 9D, left). The resulting graph clearly shows that heterosynaptic plasticity inhibits SC LTP induced by moderate SC stimulation, while a longer and repeated TBS protocol results in larger SC LTP. Plotting the effects of heterosynaptic plasticity as the percentage of inhibition or enhancement of SC LTP further demonstrates that prior heterosynaptic plasticity inhibited SC LTP induced by moderate SC TBS by 84.18%, but increased SC LTP by 77.38% when induced by strong SC TBS (Fig. 9D, right). These data highlight the powerful metaplastic control that heterosynaptic plasticity exerts over subsequent SC LTP.
Discussion
Our current understanding of distal apical tuft synapses is that their main function is to modulate time-locked proximal input (Remondes and Schuman, 2002; Jarsky et al., 2005; Takahashi and Magee, 2009). We show that strong distal synaptic activation alone has long-lasting effects on neuronal activity through heterosynaptic potentiation of proximal synapses. This heterosynaptic potentiation is expressed by both increased AMPA and NMDA receptor-mediated synaptic strength, the latter specifically by enhanced GluN2B signaling. Furthermore we demonstrate that these two mechanisms interact to both shift the input selectivity of SC synapses and expand the dynamic range of LTP. This finding reveals a new organizational level for the regulation of synapses and circuit function: pathway-specific control of synaptic strength where signals at distal inputs can produce long-lasting modulation of proximal synapses.
Heterosynaptic plasticity has been observed in diverse preparations and can be expressed in various forms. Many types of heterosynaptic plasticity act to homeostatically regulate synaptic strength; for example, homosynaptic LTP can result in heterosynaptic LTD at nearby nonstimulated synapses (Nishiyama et al., 2000; Royer and Paré, 2003; Chevaleyre and Castillo, 2004). However, the opposite is also seen where homosynaptic potentiation results in heterosynaptic potentiation (Engert and Bonhoeffer, 1997; Reissner et al., 2010). One common characteristic of previously described forms of heterosynaptic plasticity is the short spatial spread of heterosynaptic effects, on the order of tens of microns. This is due to the nature of the induction signal for these forms of plasticity: the short-range diffusion of signaling molecules like endocannabinoids or the intracellular release of Ca2+ from internal stores. Functionally, these forms of heterosynaptic plasticity indiscriminately modify nearby synapses, favoring the coregulation of clustered groups of synapses. In contrast, the heterosynaptic plasticity we describe works across a much longer spatial scale and is both directional and input specific; that is, induction at one set of inputs can influence another set of anatomically and functionally distinct inputs. In this way, activity in the entorhinal cortex can act as a switch to control information flow within the hippocampus.
The cellular/molecular mechanisms underlying this plasticity remain unknown, although there are two scenarios by which activation of distal synapses could potentiate proximal synapses. First is the possibility that dendritic spikes, generated in the apical tuft, propagate some distance into the proximal dendrite to trigger plasticity directly at SC synapses. However, our Ca2+ imaging experiments showing that SLM-evoked Ca2+ signals drop precipitously at the SLM/SR border suggest that local activation of SC synapses is not likely. In addition, the long latency for the expression of heterosynaptic potentiation further argues against direct Ca2+-dependent activation of SC synapses. Second is the possible activation of a signaling molecule in the distal dendrite that diffuses or is transported proximally to generate plasticity. This scenario is compatible with the slow time course for heterosynaptic potentiation, but if this is the case, there must exist a yet unknown mechanism for preventing potentiation at SLM synapses or targeting the putative activation factor specifically to SC synapses. Examining heterosynaptic potentiation with single-synapse resolution to precisely define the spatiotemporal dynamics of potentiation will greatly aid in answering these questions.
While we have shown that intracellular Ca2+ is required for heterosynaptic potentiation, the role of dendritic spikes in the apical tuft remains unclear. Ca2+ imaging shows that the SLM stimulation protocol used for inducing heterosynaptic plasticity produces large Ca2+ transients in the apical tufts, but we have not shown that these transients result from supralinear summation of synaptic inputs. One intriguing possibility is that SLM stimulation strong enough to repeatedly drive dendritic spikes is required for the induction of heterosynaptic plasticity. Future experiments varying SLM activation while measuring dendritic spikes and subsequent heterosynaptic plasticity should resolve this possibility.
Another unusual aspect of this heterosynaptic plasticity is its dependence on GABAA receptor-mediated inhibition. There are at least two possibilities for the role of inhibition in the induction of heterosynaptic potentiation. One is that GABAA receptor-mediated hyperpolarization may be required to deinactivate low-threshold-activated T-type calcium channels located in the apical tuft, and the resulting rebound firing after the release of inhibition is required for generating large Ca2+ transients. This possibility is consistent with our result showing that R- or T-type Ca2+ channels are required for heterosynaptic potentiation. Continued pharmacological dissection of the contribution of these two types of calcium channels will resolve this possibility. Second, it has been shown that presynaptic GABAA receptors can unexpectedly enhance LTP induction by depolarizing the synaptic terminal in hippocampal mossy fiber boutons (Nakamura et al., 2007; Ruiz et al., 2010). Further work will be required to reveal exactly how GABAA receptor-mediated inhibition contributes to the induction of heterosynaptic plasticity. Although we ruled out the contribution of many molecular pathways to heterosynaptic potentiation (Fig. 6E), we did not exhaustively test all possibilities. In particular, we did not investigate the role of GABAB inhibition, which has been shown to mediate cross-pathway inhibition at these synapses on a millisecond-to-second time scale (Dvorak-Carbone and Schuman, 1999). It remains to be seen what role, if any, it plays in long-lasting heterosynaptic changes.
The heterosynaptic plasticity we describe is distinct from, and strikingly complementary to, Hebbian LTP. While LTP expresses quickly with no change in NMDA receptor-mediated current, heterosynaptic plasticity develops slowly and features a metaplastic shift of NMDA receptor subunits. Computational modeling of neuronal networks has found that engaging multiple plasticity mechanisms featuring this specific combination of offset time courses coupled with metaplastic modification of synapses dramatically improves memory retention by preventing interference between plasticity mechanisms (Fusi et al., 2005). Thus, while the fast time course of LTP is well suited for the rapid encoding of memories, slow heterosynaptic plasticity may come online to capture a later phase of memory formation. If heterosynaptic plasticity does contribute to memory encoding, its distinctly slow kinetics should be observable in the development of CA1 firing in vivo. Indeed, several studies have found that although CA3 firing rates stabilize within minutes after an animal has been introduced to a new environment, CA1 firing continues to evolve over hours to days (Lever et al., 2002; Leutgeb, 2006; Karlsson and Frank, 2008; Dupret et al., 2010).
How might memory be encoded in heterosynaptically activated neurons? One specific form of network activity that is uniquely suited to match these dual requirements for delayed activation and strong, repeated afferent input is sharp wave/ripple complexes. These events, believed to drive memory consolidation, occur constantly during pauses in active exploration and slow-wave sleep and are associated with the repeated replay of previous experience (Lee and Wilson, 2002; Girardeau et al., 2009; Nakashiba et al., 2009; Dupret et al., 2010; Jadhav et al., 2012). Furthermore, sharp waves/ripples, generated by the intense, synchronized discharge of CA3 neurons, are the strongest known form of hippocampal network depolarization, driving firing in CA1 neurons and activity throughout the cortex (Buzsáki, 1986; Siapas and Wilson, 1998; Ji and Wilson, 2007). Thus, strong entorhinal cortical input activating heterosynaptic plasticity may prime an ensemble of CA1 neurons for memory consolidation via sharp wave/ripples occurring during exploration or sleep. This proposed role for heterosynaptic plasticity in memory consolidation is consistent with a study showing that lesion of the direct entorhinal cortical pathway to the distal dendrites of CA1 prevented memory consolidation (Remondes and Schuman, 2004).
Synaptic inputs onto electrotonically remote distal dendritic tufts occur throughout the brain, raising the possibility that heterosynaptic plasticity may operate in other neuron types. Unfortunately, the lack of well-segregated afferent input in other brain regions and the washout of plasticity during whole-cell recording make studying this plasticity difficult. However, the use of the repatch technique described here combined with optogenetic stimulation of defined afferent inputs should reveal whether heterosynaptic plasticity is a common signaling mechanism for distal dendritic synapses.
Footnotes
This work was supported by an NIA training grant (E.B.H.) and an NIH Blueprint grant (S.F.H.). We thank Nelson Spruston, Massimo Scanziani, and Martha Bagnall for their generous advice and critical reading of this manuscript; Nelson Spruston for financial support; and Jeffry Isaacson and Gabrielle Edgerton for comments.
- Correspondence should be addressed to Edward B. Han at his present address: Northwestern University, 2205 Tech Drive, Hogan 2160, Evanston, IL 60208. edbhan{at}gmail.com