Differential distribution of endoplasmic reticulum controls metabotropic signaling and plasticity at hippocampal synapses

Genetic Approach To Identify ER-Containing Spines in CA1 Pyramidal Cells.

To visualize the ER in intact hippocampal tissue, we constructed a green ER label by fusing EGFP with ER-targeting and ER-retention sequences (9). Organotypic hippocampal slice cultures were cotransfected biolistically with the ER label and a cytoplasmic red fluorescent protein (RFP) to visualize cell morphology (Fig. 1A). Two-photon microscopy was used to image transfected CA1 pyramidal cells at high resolution and identify ER-containing (ER+) and other (ER−) spines on oblique dendrites close to the soma (Fig. 1B). Spines were considered ER+ if they had a clear GFP signal inside the head (Fig. 1C); spines with traces of ER in the neck region were not analyzed. Analysis of several transfected CA1 pyramidal cells showed that 18.7 ± 2.3% of spines were positive for the ER label (n = 318 spines, 8 cells). A similar fraction of ER-containing spines has been reported in acute slices by using electron microscopic reconstruction of CA1 cell dendrites (6). To test whether spine ER was continuous with dendritic ER, we performed fluorescence recovery after photobleaching (FRAP) experiments on the ER label. A brief laser pulse (930 nm) was used to bleach the green ER label inside the spines (Fig. 1D). On average, ER fluorescence recovered with a time constant of 210 ± 39 ms (n = 33 spines). These data indicate that dendritic spines contain ER structures that are connected to dendritic ER and that ER constituents (such as ions and proteins) can rapidly diffuse inside the ER lumen from the spine to the dendrite and vice versa.

ER-Containing Spines Have Large Heads and Often Contain a Spine Apparatus.

Quantification of spine morphology showed that ER+ spines had significantly larger cytoplasmic volumes (0.058 ± 0.005 μm³, n = 49) than the rest of the population (0.028 ± 0.002 μm³, n = 91; Fig. 1E). Spine heads are separated from their parent dendrite by a thin spine neck, allowing biochemical compartmentalization of second messengers (10). The ER could affect the diffusional coupling between spine head and parent dendrite by physically obstructing the neck. To address this issue, we measured FRAP time constants (τ) of cytoplasmic RFP in ER-containing and other spines. We bleached ≈30% of the cytoplasmic red fluorescence inside the spine and monitored fluorescence recovery. The recovery time constants in the two groups of spines were not significantly different (τ_ER+ = 235 ± 40 ms, n = 26; τ_ER− = 218 ± 20 ms, n = 35; Fig. 1F), demonstrating that the ER did not block cytoplasmic diffusion between spine head and dendrite. In fact, because of the larger head volumes of ER+ spines, these would be expected to have longer time constants, assuming identical neck properties (11, 12). The similar time constants therefore suggest that the necks of ER+ spines provided an even weaker diffusional barrier compared with those of spines lacking ER.

In spines of pyramidal cells, ER often forms a specialized organelle consisting of stacked membrane discs, the spine apparatus (6). To assess which fraction of ER+ spines in our sample contained this organelle, we combined live ER imaging with posthoc immunohistochemistry against synaptopodin, a protein associated with the spine apparatus (13) (Fig. S1A). We found that the majority (78%) of ER+ spines were also positive for synaptopodin (Fig. S1B) and thus very likely contained a spine apparatus.

Synapses on ER-Containing Spines Have a High Potency.

To assess the functional properties of spine synapses, we stimulated individual spines by two-photon glutamate uncaging. First, we identified ER+ spines on oblique dendrites (Fig. 2A). After spine preselection, cells were patch-clamped, and glutamate was uncaged on identified spines. On average, uncaging-evoked excitatory postsynaptic currents (uEPSCs) had amplitudes of 11.4 ± 0.7 pA (n = 74 spines, 23 cells), similar to the amplitudes of miniature EPSCs in these cells (12.4 ± 0.6 pA, n = 8 cells). Stimulation of ER+ spines evoked significantly larger uEPSCs (17.1 ± 1.9 pA, n = 30) than stimulation of other spines (7.5 ± 0.5 pA, n = 44; Fig. 2 B and C). To exclude systematic differences in stimulation intensity, we consistently stimulated one to two ER+ and control spines on the same dendrite (distance <25 μm). Furthermore, we compared the amount of bleaching of a cytoplasmic red dye by the uncaging laser pulse (calibration method used in ref. 14), which was identical in ER-containing and other spines (Fig. S2). This control verified that both groups of spines were stimulated with the same local laser intensity. In our sample of spine synapses, uEPSC amplitude was positively correlated with spine volume (R² = 0.46, n = 62; Fig. 2D), which is in line with previous studies (15, 16). Thus, our data indicate that high-potency synapses are located on large spines and these high-potency synapses are often associated with spine ER. We would like to point out that no causal relationship should be inferred from these correlations between morphology and function.

Similar NMDAR-Mediated Calcium Transients in ER+ and ER− Spines.

The ER potentially could modulate the time course of postsynaptic calcium transients, which are an important trigger for synaptic plasticity (17, 18). To compare calcium transients in spines with and without ER, we filled transfected CA1 pyramidal cells with a mixture of a green calcium-sensitive dye (Fluo5F) and a red calcium-insensitive dye (Alexa Fluor 594). At the excitation wavelength and laser power used in the calcium imaging experiments (810 nm, ≈10 mW), GFP and RFP fluorescence was negligible. Glutamate uncaging evoked spine calcium transients (CaTs) with short latencies (Fig. 3A). The amplitudes of these rapid CaTs have been shown to mainly depend on the activation of different subtypes of NMDARs and other voltage-sensitive channels (19–21). Consistent with these reports, we found that the NMDAR antagonist D-4-[(2E)-3-Phosphono-2-propenyl]-2-piperazinecarboxylic acid (dCPP) (10 μM) strongly reduced peak amplitude (22.8% of control amplitude, n = 4). Peak amplitudes of short-latency CaTs (Fig. 3B) were not significantly different between ER-containing spines (ΔG/R = 26 ± 6%, n = 19) and other spines (ΔG/R = 19 ± 3%, n = 26). The decay time constants (τ_decay) were also very similar (ER+ τ_decay = 164 ± 19 ms; ER− τ_decay = 179 ± 34 ms). Depletion of intracellular calcium stores with thapsigargin (10 μM) did not reduce peak amplitude of these rapid calcium transients (116% of control amplitude, n = 9 ER+ spines), suggesting that calcium release did not significantly boost NMDAR-mediated signals. Our data therefore indicate that spine ER is not a major modulator of NMDAR- mediated CaTs during synaptic activity.

mGluR-Dependent Calcium Release in ER-Containing Spines.

As described above, glutamate uncaging triggered fast rising calcium transients (average time to peak, 47 ms) with a slow decay (τ_decay = 173 ± 21 ms) in all spines tested. In addition, in some spines, we observed a second, delayed rise in calcium concentration (average delay, 470 ± 41 ms; Fig. 3C). This second calcium transient (CaT2) reached much higher amplitudes than the fast, NMDAR-mediated transient (CaT1, ΔG/R = 26 ± 6%; CaT2, ΔG/R = 141 ± 26%; Fig. 3D). Delayed CaTs were observed in seven of 19 ER+ spines but never in ER− spines (n = 26). Our best estimate for the frequency of spines with delayed CaTs is 7/19 = 0.37, so assuming a binomial distribution, the probability of observing no delayed CaTs in ER− spines is 6.6 × 10⁻⁶. Therefore, ER+ and ER− spines represent two distinct populations with respect to their calcium signaling.

Delayed CaTs never preceded uncaging stimulation but appeared to be triggered by the stimulation in a stochastic fashion (Fig. 3C). In the spines that showed delayed CaTs, they occurred in ≈20% of the individual stimulations (range 6–43%). Simultaneous voltage-clamp recordings revealed no somatic depolarization during delayed CaTs, making voltage-gated calcium influx rather unlikely (Fig. S3). In many cases, delayed CaTs also were detected in the dendrite (Fig. S3). However, dendritic CaTs were delayed with respect to spine CaTs (2.8 ms, on average) and had smaller amplitudes (52% of spine CaTs, on average), which points to active calcium release inside the spine rather than passive diffusion of calcium from the dendrite into the spine. Blocking group I mGluRs with a mixture of mGluR1 and mGluR5 antagonists (100 μM LY367385 and 10 μM 2-Methyl-6-(phenylethynyl)pyridine hydrochloride (MPEP), n = 9) or blocking IP₃ receptors with intracellular heparin (4 mg/mL, n = 12) abolished delayed CaTs in ER+ spines (Fig. 3 E and F). Depletion of intracellular stores by bath application of thapsigargin also abolished delayed CaTs in ER+ spines (10 μM, n = 9). Whole-cell perfusion with 100 μM IP₃, however, increased the fraction of ER+ spines showing delayed CaTs (n = 5; Fig. 3F). The amplitude of uEPSCs was not affected by any of these pharmacological manipulations (Fig. S4). In summary, our data suggest that ER+ spines represent a specialized subpopulation of dendritic spines, capable of mGluR- and IP₃-mediated calcium signaling.

Group I mGluR receptors are preferentially located at the periphery of the synapse, and glutamate uncaging could activate a larger fraction of these perisynaptic receptors than vesicular glutamate release (22). To address this issue, we performed spine calcium imaging after electrical stimulation of Schaffer collaterals in acute slices (Fig. S5). With synaptic stimulation, four of 26 stimulated spines displayed delayed CaTs. Corresponding to the pharmacology of the uncaging experiments, thapsigargin (2 μM) completely prevented delayed CaTs after synaptic stimulation (0 of 33 spines). We conclude that at a subset of Schaffer collateral synapses delayed CaTs can be triggered by vesicular glutamate release from presynaptic terminals.

mGluR-Dependent Depression Is Restricted to ER-Containing Spines.

With electrophysiological approaches, metabotropic signaling and IP₃-mediated calcium release have been suggested to be involved in the induction of LTD in several cortical areas (23–25). However, whether this form of plasticity is present at all synapses or restricted to a specialized subpopulation is unclear. mGluR-dependent LTD has been shown to have a strong postsynaptic component (26). The specificity of mGluR- → IP₃-mediated spine calcium signals raised the possibility that synapses on ER+ spines are preferential sites for this form of synaptic depression. We tested this possibility by comparing the effects of low-frequency stimulation, a classical LTD induction protocol, on ER+ spines. To isolate mGluR-dependent effects, NMDARs were blocked with dCPP (10 μM). In a series of control experiments, we verified that delayed CaTs were still present under NMDAR block (n = 4). The potency of synapses on spines of interest was measured before and after low-frequency uncaging (LFU) stimulation (0.2 Hz, 15 min; Fig. 4A). Indeed, low-frequency glutamate uncaging induced depression of the uEPSC in ER+ spines (72.4 ± 4.4% of baseline, n = 20; Fig. 4 B and C). This depression was long-lasting (n = 6; Fig. 4A) and stable (no significant difference between uEPSC amplitudes 15–17 and 27–30 min after start of LFU). Importantly, the same LFU protocol evoked no depression in spines lacking ER (104.7 ± 8.3% of baseline, n = 12; Fig. 4 B and C). Whole-cell perfusion with 1,2-bis(2-aminophenoxy)ethane-N,N,N′,N′-tetraacetate (BAPTA) (1 mM) blocked depression of uEPSC on ER+ spines (113.0 ± 5.1% of baseline, n = 3), indicating a calcium-dependent mechanism underlying depression. To test whether depression of uEPSCs at ER+ spines depended on mGluR-mediated signaling, we pharmacologically blocked group I mGluRs (100 μM LY367385 and 10 μM MPEP) and analyzed the effects of the LFU protocol on synaptic potency. Blocking mGluRs did not affect the baseline response amplitude of ER+ spines (ER+, control, 21.0 ± 1.0 pA, n = 20; ER+, mGluR block, 21.3 ± 1.9, n = 7) but completely prevented LFU-induced depression (111.5 ± 16.9% of baseline, n = 7; Fig. 4C). We conclude that an mGluR-dependent mechanism was responsible for the difference in plasticity between ER+ and ER− spines. Depression at ER+ spines also was blocked completely by intracellular application of the IP₃ receptor antagonist heparin (105.5 ± 11.2% of baseline, n = 9; Fig. 4C). The identical pharmacology of synaptic depression, delayed CaTs, and the specificity for ER+ spines suggest that calcium release from the subsynaptic ER is required for the induction of NMDAR-independent depression at these synapses. Last, we investigated potential spread of synaptic depression by probing pairs of neighboring synapses on the same dendrite. As previously, LFU on ER+ spines produced reliable homosynaptic depression (74.7 ± 5.3% of baseline, n = 8). Remarkably, no significant change was detected at spine synapses 2–10 μm away (106.9 ± 10.5% of baseline; Fig. 4E), demonstrating that mGluR-dependent depression remains highly compartmentalized to individual synapses.

ER Is Associated with Mature Spines.

We found spine ER not randomly distributed but specifically localized to large spines containing high-potency synapses (Fig. 2). Consistent with previous reports, synaptic potency was correlated with the volume of the spine head (15). Because newly formed spines are typically small (27), our data suggest that the acquisition of ER is part of the process of spine maturation. Overexpression of the adapter proteins Shank and Homer has been shown to enhance or accelerate this maturation process, leading to spine enlargement and recruitment of ER and IP₃ receptors into spines (28, 29). The preferential localization of ER to mature spines restricts the potential for mGluR- or IP₃-mediated calcium signaling to this specialized subpopulation. In this context, that the intact animal spines that are stable over days and weeks often contain ER is interesting (30). Altered sensory experience, for example, sensory deprivation, can lead to destabilization and removal of these stable spines (31). The mechanism that we describe here might be critical for the weakening and eventual disappearance of initially stable synaptic connections (27). Although only ≈20% of spines posses ER, they carry the synapses that contribute most excitatory drive and presumably need to be most tightly controlled (Fig. 2C).

mGluR Activation and Synaptic Plasticity.

Ionotropic glutamate receptors, most prominently NMDARs, have been established to regulate the amplitude of spine calcium transients, an important trigger for synaptic plasticity (18, 32). The role of metabotropic receptors in regulating biochemical signaling in pyramidal neurons is less well understood. Strong activation of multiple presynaptic fibers can lead to the generation of mGluR-dependent calcium release events in apical dendrites (33). Here, we show that the activation of a single synapse can be sufficient to trigger mGluR-dependent calcium transients and depression if the activated synapse impinges on an ER+ spine (Fig. 3F). A similar metabotropic mechanism of depression has been described in Purkinje cells of the cerebellum (34, 35). In these cells, however, there is no indication of a regulation of plasticity on the single synapse level, a functional uniformity that is reflected in the homogenous morphology of Purkinje cell spines (36). Furthermore, much stronger stimulation is needed to elicit calcium release events in Purkinje cells, possibly related to the simpler, tube-like morphology of spine ER in these cells. Therefore, although the general mechanism of mGluR-mediated depression seems to be conserved in different cell types and brain areas, the selective targeting of the ER to strong synapses and the low activation threshold might be specific features of hippocampal pyramidal cells.

Our experiments suggest that local IP₃ receptor activation and calcium release are essential steps in the pathway leading from mGluRs to synaptic depression. Calcium release events and mGluR-dependent depression had identical pharmacological profiles: Both were blocked by group I mGluR antagonists and the IP₃ receptor blocker heparin. Furthermore, both processes were restricted to ER+ spines (Figs. 3F and 4C). This tight correlation and the sensitivity of depression to the calcium chelator BAPTA suggest that calcium release from the ER is a key step in the induction of synaptic depression. In our experiments, calcium release events were not restricted to the spine but often spread into the dendrite (Fig. S3). Induction of depression, however, was limited strictly to directly stimulated synapses (Fig. 4D). This specificity could be simply caused by differences in calcium signal amplitudes, but the ER also might serve additional functions during plasticity induction: In addition to regulating spine calcium signals, we show that it forms a continuous subcellular conduit for the exchange of proteins between dendrite and spine (Fig. 1D). Furthermore, ER-bound ribosomes provide a substrate for local mRNA translation, a process that is thought to be important for mGluR-mediated depression (37). The impact of ER on synaptic plasticity very likely involves all of the processes mentioned above. In summary, we show that differences in spine microanatomy, which were noted half a century ago (38), play a crucial functional role in determining the competence of individual synapses for long-term plasticity.

Slice Preparation and Plasmid Construction.

Organotypic hippocampal slices were prepared from Wistar rats at postnatal day 5 as described (39). After 3–5 days in vitro, cultures were transfected with synapsin-GFP-ER and synapsin-RFP (tdimer2; R. Y. Tsien, University of California, San Diego), using a Helios gene gun (Bio-Rad). GFP-ER was constructed by fusing EGFP to the ER-targeting sequence of calreticulin and the ER-retention sequence KDEL.

Two-Photon Imaging and Uncaging.

Imaging experiments were performed between 1 and 2 weeks after transfection. We used a custom-built two-photon imaging and uncaging setup (40) controlled by ScanImage (41). Two ultrafast IR lasers were combined for two-photon imaging (930 nm for GFP/RFP imaging, 810 nm for Alexa Fluor 594/Fluo5F imaging) and uncaging of 4-Methoxy-7-nitroindolinyl-caged L-glutamate (725 nm). To measure calcium signals, green and red fluorescence was collected during 250-Hz line scans across the spine head and parent dendrite (7–15 trials per spine). Fluorescence changes were quantified as increases in green fluorescence from baseline normalized to the red fluorescence (ΔG/R) (42). For each imaging trial, photomultiplier dark noise was measured before shutter opening and subtracted from the dataset. Fluorescence was monitored 640 ms before the stimulus and 640 ms after the stimulus.

Glutamate uncaging was achieved by using a 0.5-ms laser pulse for experiments analyzing synaptic properties and a 1-ms pulse for experiments looking at synaptic depression. Mushroom-shaped spines well separated from neighboring spines were used for the experiments. The standard uncaging location was ≈0.5 μm from the spine center, in the direction away from the parent dendrite. These uncaging parameters minimized the possibility of activation of dendritic glutamate receptors. Laser intensity was ≈50 mW for 0.5-ms pulses and ≈40 mW for 1.0-ms pulses, measured in the back focal plane of the objective. The same laser pulse, if directed to the center of the spine, bleached 30% of the Alexa Fluor 594 fluorescence. We used this control to ensure equal stimulation strength for ER+ and ER− spines (14) (Fig. S2).

Electrophysiology.

Slice cultures were superfused with artificial cerebro-spinal fluid (ACSF) at 30–32 °C containing 127 mM NaCl, 25 mM NaHCO₃, 25 mM d-glucose, 2.5 mM KCl, 1 mM MgCl₂, 2 mM CaCl₂, 1.25 mM NaH₂PO₄, 0.05 mM chloroadenosine, and 0.01 mM serine. For uncaging experiments, 5 mM MNI-glutamate (Tocris) was added to the ACSF. For plasticity experiments, serine was replaced by 0.01 mM dCPP in the ACSF. To block mGluR1 and mGluR5 receptors, we added LY367385 (0.1 mM; Tocris) and MPEP (0.01 mM, kindly provided by Novartis AG). To deplete internal calcium stores we used thapsigargin (0.01). Whole-cell recordings were made with an Axopatch 200B amplifier (Axon Instruments), using 4–6 MΩ electrodes filled with 135 mM potassium gluconate, 10 mM Hepes, 10 mM sodium phosphocreatine, 3 mM sodium ascorbate, 4 mM MgCl₂, 4 mM Na₂-ATP, 0.4 mM Na-GTP, and 0.015 mM Alexa Fluor 594 (pH 7.2). For calcium imaging experiments, 0.3 mM Fluo5F (Molecular Probes) was added. Calcium imaging experiments were performed under current clamp conditions, and measurements of postsynaptic currents were performed in voltage clamp. The uEPSC peak amplitude was extracted by fitting the difference of two exponentials to the average electrical response for each spine (five trials per spine). To allow for dye and drug diffusion, we started the stimulation 15–20 min after break-in.

Induction of Synaptic Depression at Single Spines.

We attempted to mimic a conventional low-frequency LTD stimulation protocol (15 min, 1 Hz, 900 pulses). Because, in contrast to the reliable uncaging stimulus, the average release probability of Schaffer collateral synapses is only ≈0.2, we used a five times lower frequency for the LFU stimulation (15 min, 0.2 Hz, 180 pulses). To optimize the signal-to-noise ratio of the electrical recording, only spines with an average uEPSC >7 pA were selected for the experiment. To quantify synaptic response amplitude, spines were stimulated before and after LFU with five test pulses spaced 30 s apart. To quantify depression, the average response after LFU was divided by the average response before LFU. For long-term recordings after LFU, spines were stimulated with test pulses at 0.03 Hz. Cells were held in voltage clamp at −65 mV.

Statistical Analysis.

Data are reported as mean ± SEM. To test for significance we used the Mann–Whitney rank sum test (unpaired data) or the Wilcoxon signed-rank test (paired data) at a significance level of P = 0.05.