Calcium threshold shift enables frequency-independent control of plasticity by an instructive signal

Significance

Instructive signals play an important role in synaptic plasticity and learning. For example, at cerebellar parallel fiber (PF)-Purkinje cell synapses, climbing fiber (CF) coactivation provides an instructive signal that promotes long-term depression (LTD) by amplifying spine calcium transients above a threshold level that, at these synapses, is higher than for LTP induction. Here, we show that the CF instructive signal maintains its control over PF plasticity regardless of the PF synaptic activation frequency, which, on its own, alters spine calcium signaling. We demonstrate that high-frequency stimulation reduces the calcium sensitivity of LTD, resulting from inhibitory calcium/calmodulin-dependent kinase II autophosphorylation at Thr305/306. We propose that this regulatory mechanism causes a horizontal shift of the long-term potentiation/LTD cross-over point, making plasticity independent from absolute calcium amplitudes.

Abstract

At glutamatergic synapses, both long-term potentiation (LTP) and long-term depression (LTD) can be induced at the same synaptic activation frequency. Instructive signals determine whether LTP or LTD is induced, by modulating local calcium transients. Synapses maintain the ability to potentiate or depress over a wide frequency range, but it remains unknown how calcium-controlled plasticity operates when frequency variations alone cause differences in calcium amplitudes. We addressed this problem at cerebellar parallel fiber-Purkinje cell synapses, which can undergo LTD or LTP in response to 1-Hz and 100-Hz stimulation. We observed that high-frequency activation elicits larger spine calcium transients than low-frequency stimulation under all stimulus conditions, but, regardless of activation frequency, climbing fiber (CF) coactivation provides an instructive signal that further enhances calcium transients and promotes LTD. At both frequencies, buffering calcium prevents LTD induction and LTP results instead, identifying the enhanced calcium signal amplitude as the critical parameter contributed by the instructive CF signal. These observations show that it is not absolute calcium amplitudes that determine whether LTD or LTP is evoked but, instead, the LTD threshold slides, thus preserving the requirement for relatively larger calcium transients for LTD than for LTP induction at any given stimulus frequency. Cerebellar LTD depends on the activation of calcium/calmodulin-dependent kinase II (CaMKII). Using genetically modified (TT305/6VA and T305D) mice, we identified α-CaMKII inhibition upon autophosphorylation at Thr305/306 as a molecular event underlying the threshold shift. This mechanism enables frequency-independent plasticity control by the instructive CF signal based on relative, not absolute, calcium thresholds.

Synaptic activation frequency is an important factor in the induction of long-term potentiation (LTP) and long-term depression (LTD). For example, it has been shown at Schaffer collateral-CA1 pyramidal cell synapses that application of 900 pulses at 1–3 Hz causes LTD, whereas the same number of pulses applied at 50 Hz causes LTP (1). However, LTP and LTD can also be induced at the same stimulus frequency. This phenomenon has been demonstrated at hippocampal, neocortical, and cerebellar synapses, where potentiation and depression mechanisms operate over a wide range of activation frequencies (2–7). In the neocortex and hippocampus, the level of postsynaptic depolarization determines whether LTP or LTD results from stimulation at a given frequency (2, 5). These voltage-dependent thresholds for LTP and LTD induction reflect thresholds in calcium signal amplitudes (3, 4, 8–11) that, when maintained for sufficiently long time periods (12), control synaptic plasticity in concert with distinct calcium sensors that are restricted to local microenvironments (13, 14).

At cerebellar parallel fiber (PF)-Purkinje cell synapses, both LTP and LTD can be induced using 1-Hz and 100-Hz PF stimulation protocols, and at both frequencies, climbing fiber (CF) coactivation promotes LTD, whereas LTP results from PF stimulation alone (6, 7, 15, 16). CF coactivity leads to supralinear spine calcium signaling (17), which helps to reach the calcium threshold for LTD, which is higher than the threshold for LTP at these synapses (6). It seems that for central synapses, there is a computational advantage to be able to undergo potentiation or depression regardless of activation frequency but under the control of instructive signals, such as CF coactivity in cerebellar plasticity (18–20). Here, we address a fundamental problem that arises from LTP and LTD induction under the control of instructive signals over a wide frequency range: How is it possible to maintain the essential role of calcium thresholds, and the role of an instructive signal that helps to reach the higher threshold, when the calcium levels resulting from high-frequency stimulation are likely in a higher range than the calcium levels resulting from low-frequency stimulation? The relevance of this problem is illustrated by the observation that at PF synapses, LTP-inducing 100-Hz PF bursts evoke larger calcium transients than paired single-pulse CF activation, which promotes LTD (21). We thus investigated spine calcium signaling at PF synapses under low- and high-frequency LTD- and LTP-inducing conditions, as well as mechanisms that may shift the calcium sensitivity of the LTD pathway.

Results

To monitor spine calcium transients, we used confocal imaging in slices obtained from P21-75 mice. For the comparison of calcium levels that are reached under low- and high-frequency LTD- and LTP-inducing conditions, we applied defined patterns of PF and CF activation, respectively, that were derived from the plasticity protocols described below. LTD results from PF + CF activation at 1 Hz for 5 min, whereas LTP is evoked when the same PF activation pattern is applied in isolation (6, 16). LTD can also be induced with a train of eight PF stimuli (100 Hz) followed 120 ms after stimulus onset by single-pulse CF stimulation. This activation pattern is applied at 1 Hz for 5 min. With this protocol, too, omission of CF activation causes LTP instead (7). All stimulus protocols cause postsynaptically expressed forms of LTD/LTP (6, 7). In the following, we will use the terms “low-frequency protocol” and “high-frequency protocol” to refer to the highest lead frequency within the PF stimulus patterns. Thus, 100-Hz PF burst stimulation is referred to as the high-frequency protocol, whereas single-pulse PF stimulation at 1 Hz is referred to as the low-frequency protocol. For the comparison of calcium levels that are reached under the four activation conditions, we chose to break down the protocols into the stimulus patterns that are repeated at 1 Hz for 5 min. For the high-frequency protocol, this stimulus pattern is a 100-Hz PF burst (eight pulses), and for the low-frequency protocol, it is a single PF pulse, which are paired/not paired with CF activation. Calcium transients evoked in spines on secondary or higher order branches of Purkinje cell dendrites were calculated as ΔG/R = (G(t) − G₀)/R, where G is the calcium-sensitive fluorescence of Oregon Green 1,2-Bis(2-aminophenoxy)ethane-N,N,N′,N′-tetraacetic acid (BAPTA)-2 (OGB-2; 200 μM), G₀ is the baseline signal, and R is the calcium-insensitive fluorescence of Alexa 633 (30 μM) (22). Calcium transients were monitored under all four stimulation conditions (applied two to three times in randomized order, with 30-s intervals between stimuli) from the same spine that was defined as the region of interest (ROI). The ROI was selected based on the strength of calcium responses to synaptic activation (maximally responding spine). To obtain a calcium transient measure, ΔG/R values were averaged over a period of 200 ms after stimulus onset (17). Single-pulse PF stimulation resulted in small calcium signals (0.02 ± 0.02 ΔG/R; n = 11; Fig. 1 A–D). Paired PF + CF activation resulted in significantly larger calcium transients (0.06 ± 0.01 ΔG/R; P = 0.001). Application of the 100-Hz PF burst evoked calcium signals that were about threefold as large as the calcium signals obtained with paired PF + CF single-pulse stimulation (0.18 ± 0.02 ΔG/R; P = 0.00005). When this 100-Hz PF burst was followed after 120 ms by a single CF pulse, the resulting calcium transient was further enhanced (0.24 ± 0.04 ΔG/R; P = 0.041). OGB-2 is a high-affinity calcium indicator (K_d = 485 nM), allowing for accurate measurement of calcium in the low-amplitude range. To obtain a better dynamic range at high calcium amplitudes, we repeated these measurements using the low-affinity (K_d = 1.8 μM) indicator Fluo-5F (300 μM). Single-pulse PF stimulation resulted in ΔG/R values in the noise range (−0.01 ± 0.01 ΔG/R; n = 11; Fig. 1 E–H). However, with CF coactivation, the signal was significantly enhanced (0.04 ± 0.01 ΔG/R; P = 0.035). A 100-Hz PF burst stimulation resulted in calcium transients that were about 12-fold larger than the calcium transients obtained with single-pulse PF + CF activation (0.46 ± 0.15 ΔG/R; P = 0.013). When this 100-Hz burst was followed after 120 ms by a single CF pulse, the calcium signal amplitude was further enhanced (0.59 ± 0.19 ΔG/R; P = 0.039). With both OGB-2 and Fluo-5F, calcium transients monitored in adjacent shaft areas showed the same amplitude relationships as calcium transients measured in the spines (Fig. S1). The results obtained with both indicators show that 100-Hz PF bursts evoke larger calcium transients than single pulses, and that CF coactivation further enhances calcium signals at both frequencies. To compare spine calcium transients evoked by 100-Hz PF bursts with and without CF coactivation under conditions that are as remote as possible from dye saturation, we went a step further and used the ultralow calcium affinity (K_d = 22.0 μM) indicator OGB-5N (300 μM). These recordings confirmed our previous observations: paired 100-Hz PF burst + CF activation caused spine calcium transients (0.32 ± 0.09 ΔG/R) that were significantly larger than spine calcium transients observed with 100-Hz PF burst stimulation alone (0.21 ± 0.05 ΔG/R; n = 9; P = 0.04338; Fig. S2).

Fig. 1.

Spine calcium transients evoked by LTP- and LTD-inducing PF and CF activation patterns. (A–D) OGB-2 measurements. (A, Left) Purkinje cell filled with Alexa 633. (Scale bar: 20 μm.) (A, Center) Enlarged view. The circle outlines the ROI. (Scale bar: 1 μm.) (A, Right) Green fluorescence of OGB-2. The circle outlines the ROI. (Scale bar: 1 μm.) (B, Top) Electrophysiological responses to the following stimuli: 100-Hz PF burst + CF, PF burst alone, single-pulse PF + CF, and PF pulse alone. (Scale bars: vertical, 20 mV; horizontal, 100 ms.) (B, Bottom) Simultaneously recorded calcium transients. (Scale bars: vertical, 0.1 δG/R; horizontal, 500 ms.) (C) Calcium transients averaged from all Purkinje cell recordings (n = 11). Calcium signals are expressed as the percentage of the peak amplitude in each recording. (D) Bar graph summarizing calcium signal amplitudes (ΔG/R; average over a 200-ms period starting at stimulus onset, n = 11). (E–H) Fluo-5F measurements. (E, Left) Purkinje cell filled with Alexa 633. (Scale bar: 20 μm.) (E, Center) Enlarged view. The circle outlines the ROI. (Scale bar: 1 μm.) (E, Right) Green fluorescence of Fluo-5F. The circle outlines the ROI. (Scale bar: 1 μm.) (F) Electrophysiological responses and calcium transients arranged as in B. (Scale bars: Top, vertical, 20 mV; Top, horizontal, 100 ms; Bottom, vertical, 0.5 δG/R; Bottom, horizontal, 500 ms.) (G) Averaged calcium transients (n = 11). (H) Bar graph summarizing signal amplitudes (n = 11). Error bars indicate SEM. **P < 0.01; *P < 0.05.

Fig. S1.

Calcium transients evoked in spines and adjacent shaft areas by LTP- and LTD-inducing PF and CF activation patterns. (A–D) OGB-2 measurements. (A, Left) Purkinje cell filled with Alexa 633. (Scale bar: 20 μm.) (A, Center) Enlarged view. The circles outline the spine and shaft ROIs. (Scale bar: 1 μm.) (A, Right) Green fluorescence of OGB-2. The circles outline the spine and shaft ROIs. (Scale bar: 1 μm.) (B, Top) Electrophysiological responses to the following stimuli: 100-Hz PF burst + CF, PF burst alone, single-pulse PF + CF, PF pulse alone. (Scale bars: vertical, 20 mV; horizontal, 100 ms.) (B, Middle) Simultaneously recorded spine calcium transients. (B, Bottom) Shaft calcium transients. (Scale bars: vertical, 0.1 δG/R; horizontal, 500 ms.) (C) Bar graph summarizing calcium signal amplitudes in spines (ΔG/R; average over a 200-ms period starting at stimulus onset, n = 11). PF pulse alone: 0.02 ± 0.02 ΔG/R. PF + CF pulse: 0.06 ± 0.01 ΔG/R (P = 0.001; here and in the following, the P values refer to the statistical comparison with the previously stated stimulus condition/calcium signal amplitude); 100-Hz PF burst: 0.18 ± 0.02 ΔG/R (P = 0.00005); 100-Hz PF burst + CF: 0.24 ± 0.04 ΔG/R (P = 0.041). (D) Bar graph summarizing calcium signal amplitudes in shafts (n = 11). PF pulse alone: 0.02 ± 0.01 ΔG/R. PF + CF pulse: 0.03 ± 0.01 ΔG/R (P = 0.588); 100-Hz PF burst: 0.19 ± 0.03 ΔG/R (P = 0.0002); 100-Hz PF burst + CF: 0.24 ± 0.05 ΔG/R (P = 0.083). (E–H) Fluo-5F measurements. (E, Left) Purkinje cell filled with Alexa 633. (Scale bar: 20 μm.) (E, Center) Enlarged view. The circles outline the spine and shaft ROIs. (Scale bar: 1 μm.) (E, Right) Green fluorescence of Fluo-5F. The circles outline the spine and shaft ROIs. (Scale bar: 1 μm.) (F) Electrophysiological responses and spine/shaft calcium transients arranged as in B. (Scale bars: Top, vertical, 20 mV; Top, horizontal, 100 ms; Middle and Bottom, vertical, 0.5 δG/R; Middle and Bottom, horizontal, 500 ms.) (G) Bar graph summarizing calcium signal amplitudes in spines (n = 11). PF pulse alone: −0.01 ± 0.01 ΔG/R. PF + CF pulse: 0.04 ± 0.01 ΔG/R (P = 0.035); 100-Hz PF burst: 0.46 ± 0.15 ΔG/R (P = 0.013); 100-Hz PF burst + CF: 0.59 ± 0.19 ΔG/R (P = 0.039). (H) Bar graph summarizing calcium signal amplitudes in shafts (n = 11). PF pulse alone: 0.01 ± 0.004 ΔG/R. PF + CF pulse: 0.034 ± 0.012 ΔG/R (P = 0.07); 100-Hz PF burst: 0.36 ± 0.1 ΔG/R (P = 0.005); 100-Hz PF burst + CF: 0.46 ± 0.11 ΔG/R (P = 0.008). Values are shown as mean ± SEM. *P < 0.05; **P < 0.01.

Fig. S2.

Calcium transients monitored with the low-affinity calcium indicator OGB-5N. (A, Left) Purkinje cell filled with Alexa 633. (Scale bar: 20 μm.) (A, Right) Enlarged view. (Top) Green fluorescence of OGB-5N. The circle outlines the ROI. (Scale bar: 2 μm.) (Bottom) Corresponding Alexa 633 image. The circle outlines the ROI. (Scale bar: 2 μm.) (B) Electrophysiological responses to 100-Hz PF burst + CF stimulation (Left) and PF burst stimulation alone (Right). (C) Calcium transients recorded simultaneously with the electrical responses shown above. (D) Bar graph summarizing calcium signal amplitudes (ΔG/R; average over a 200-ms period starting at stimulus onset, n = 9). (E) Calcium transients averaged from all Purkinje cell recordings (n = 9). (Left) Calcium transients expressed as ΔG/R. (Right) ΔG/R values expressed as the percentage of the peak amplitude in each recording. Error bars indicate SEM. *P < 0.05.

To examine whether larger calcium transients are required for LTD than for LTP induction in the low-frequency range as well as the high-frequency range, we applied both LTD protocols when the calcium chelator BAPTA was added to the pipette saline. Under control conditions, paired PF + CF activation at 1 Hz for 5 min caused PF-LTD (65.7 ± 6.3%; t = 31–35 min; n = 6; P = 0.0029; Fig. 2). In line with previous observations (6) application of the 1-Hz LTD protocol instead resulted in LTP (149.1 ± 14.5%; n = 6; P = 0.0198; Fig. 2) when BAPTA (20 mM) was present in the pipette saline. Excitatory postsynaptic current (EPSC) amplitude changes observed in these groups differed significantly (P = 0.0051). Application of the 100-Hz PF burst-LTD protocol resulted under control conditions in PF-LTD (77.9 ± 3.8%; n = 6; P = 0.0008; Fig. 2). For these high-frequency stimulus experiments, we selected a lower BAPTA concentration (5 mM) than the concentrations that are typically used to block 1-Hz LTD (10–40 mM) (6, 23, 24). The exact BAPTA concentration was arbitrarily chosen, but the underlying idea of selecting a lower concentration was that moderate manipulation of the higher calcium signals observed with the high-frequency LTD protocol (Fig. 1) might be sufficient to evoke a switch from LTD toward LTP, allowing for a qualitative assessment of the relative calcium dependencies of LTP and LTD. We indeed observed that LTP was induced instead of LTD when the 100-Hz burst protocol was applied in the presence of BAPTA (134.3 ± 12.2%; n = 7; P = 0.031; Fig. 2). EPSC changes observed under these two conditions differed significantly (P = 0.0043).

Fig. 2.

In the low- and high-frequency activation range, BAPTA reverses LTD toward LTP. (A) Application of the 1-Hz LTD protocol results in LTP when BAPTA (20 mM) is added to the pipette saline. (Left) Time graph showing that LTD is induced under control conditions (n = 6), but LTP is induced in the presence of BAPTA (n = 6). (Center) Traces show EPSCs before and after tetanization. (Right) Individual cell data (t = 31–35 min). (B) Application of the 100-Hz LTD protocol results in LTP when BAPTA (5 mM) is added to the pipette saline. (Left) Time graph showing LTD under control conditions (n = 6) and LTP in the presence of BAPTA (n = 7). (Center) Typical traces. (Right) Individual cell data. Arrows indicate tetanization. Error bars indicate SEM. **P < 0.01.

Calcium transients evoked by 100-Hz PF burst + CF activation in the presence of BAPTA (5 mM) were significantly lower than in the absence of BAPTA (+BAPTA: 0.16 ± 0.06 ΔG/R, n = 8; −BAPTA: 0.59 ± 0.19 ΔG/R, n = 11; P = 0.0349; Fig. S3), while they did not differ statistically from calcium transients evoked by LTP-inducing PF burst stimulation (−BAPTA: 0.46 ± 0.15 ΔG/R; n = 11; P = 0.1471). Moreover, these transients were significantly higher than calcium signals evoked by the LTD-inducing single-pulse PF + CF activation (−BAPTA: 0.04 ± 0.01 ΔG/R; n = 11; P = 0.0093; Fig. S3). Similarly, in the presence of 20 mM BAPTA, single-pulse PF + CF activation caused significantly lower calcium transients than without BAPTA (+BAPTA: 0.01 ± 0.002 ΔG/R, n = 8; −BAPTA: 0.04 ± 0.01 ΔG/R, n = 11; P = 0.0349; Fig. S3). These calcium transients were significantly higher than the calcium transients reached with LTP-inducing single-pulse PF stimulation (−BAPTA; −0.01 ± 0.01 ΔG/R; n = 11; P = 0.0147), but it should be noted that as a result of the high K_d of Fluo-5F, these latter stimulus conditions do not reliably evoke transients above noise levels (traces in Fig. S3). To obtain an approximation of calcium amplitudes reached in spines during these stimulus conditions, we calculated [Ca²⁺]_i values based on the Fluo-5F G/R measures. The calcium signals reached peak values of about 0.4 μM for single-pulse PF + CF stimulation, 2.8 μM for 100-Hz PF burst stimulation, and 7.1 μM for 100-Hz PF burst + CF activation (consistent measurements of calcium levels; see ref. 17). In the presence of BAPTA (5 mM), peak calcium levels evoked by 100-Hz PF burst + CF activation were reduced to about 2.2 μM. No attempt was made to calculate [Ca²⁺]_i for single-pulse PF stimulation or single-pulse PF + CF stimulation in the presence of BAPTA because these signals were too close to noise levels in the Fluo-5F recordings.

Fig. S3.

BAPTA prevents LTD induction and reduces calcium transient amplitudes. (A) Application of the 1-Hz LTD protocol results in LTP when BAPTA (20 mM) is added to the pipette saline. (Left) Time graph showing that LTD is induced under control conditions (n = 6), but LTP is induced in the presence of BAPTA (n = 6). (Center Left) Traces show EPSCs before and after tetanization. (Center Right) Plot of individual cell data (t = 31–35 min). (Right Upper) Calcium transients (ΔG/R) evoked by single-pulse PF + CF activation (green trace), single-pulse PF activation (blue trace), and single-pulse PF + CF activation in the presence of BAPTA (purple trace). (Right Lower) Bar graphs show the averaged ΔG/R values determined for the low-frequency stimulus condition (Left) and the high-frequency stimulus condition (Right). (B) Application of the 100-Hz LTD protocol also results in LTP when BAPTA (5 mM) is added to the pipette saline. (Left) Time graph showing LTD under control conditions (n = 6) and LTP in the presence of BAPTA (n = 7). (Center Left) Typical traces. (Center Right) Plot of individual cell data (t = 31–35 min). (Right) Calcium transients (ΔG/R) evoked by 100-Hz PF burst + CF activation (red trace), PF burst activation (orange), and PF burst + CF activation in the presence of BAPTA (brown). Error bars indicate SEM. Note that in the bar graphs shown, all statistical comparisons between non-BAPTA conditions are done using the paired Student’s t test (because these recordings were obtained from the same cells), whereas comparisons between BAPTA and non-BAPTA conditions are done using the Mann–Whitney U test (recordings from different cells). *P < 0.05; **P < 0.01. Note that the panels showing electrophysiological data correspond to Fig. 2.

Together, these observations suggest that regardless of synaptic activation frequency, additional calcium influx contributed by the instructive CF signal is required to induce LTD. Hence, while the absolute calcium levels reached in the high-frequency range are significantly higher than in the low-frequency range, the relative calcium thresholds are preserved at both stimulus frequencies. The transition from absolute to relative calcium thresholds requires that high-frequency stimulation lower the calcium sensitivity of the LTD induction process, in effect shifting the LTD threshold toward a higher value.

To identify the mechanism behind this threshold shift, we postulated two requirements that a candidate signaling pathway needs to fulfill: (i) It requires a calcium sensor to detect the higher calcium levels resulting from high-frequency activation, and (ii) it needs to tip the balance from a kinase-dominated LTD pathway toward a phosphatase-activated LTP pathway at PF synapses (25, reviewed in ref. 26). This description profile matches the inhibitory autophosphorylation that has been described for the alpha subunit of calcium/calmodulin-dependent kinase II (α-CaMKII) at Thr305 and Thr306, which lowers the affinity of CaMKII for CaM, and thus reduces Ca/CaM-mediated activation (27). β-CaMKII is expressed in Purkinje cells as well (28), but the inhibitory autophosphorylation has only been examined in α-CaMKII and has been shown to have a dominant negative effect on the holoenzyme (27). Both α-CaMKII and β-CaMKII are essential for proper LTD induction (28, 29). CaMKII promotes LTD by suppressing the activity of protein phosphatase 2A through negative regulation of phosphodiesterase 1 and subsequent disinhibition of a cGMP/protein kinase G pathway (30). Negative regulation of CaMKII by Thr305/306 autophosphorylation requires prior calcium/calmodulin-mediated activation of CaMKII and subsequent calmodulin dissociation (31). Moreover, it has been shown that Thr305/306 phosphorylation, triggered by 10-Hz priming, reduces hippocampal LTP (32), whereas genetic manipulation of the Thr305/306 phosphorylation site lowers the LTP threshold and prevents priming (27, 32). Thus, CaMKII inhibitory autophosphorylation depends on previous calcium signaling and directly affects the kinase/phosphatase balance, thus fulfilling the two requirements outlined above. To test whether Thr305/306 phosphorylation is indeed involved in reducing the probability for LTD induction at high stimulus frequencies, we examined whether LTP is still induced by the 100-Hz PF burst protocol, or whether LTD is restored, in α-CaMKII mutant mice in which Thr305 and Thr306 are substituted by the nonphosphorylatable amino acids valine and alanine, respectively [referred to as TT305/6VA mice (27)]. In wild-type (WT) littermates, application of the 100-Hz LTP protocol potentiated EPSC amplitudes (131.8 ± 12.4%; t = 36–40 min; n = 12; P = 0.026; Fig. 3B). In contrast, application of the same protocol resulted in LTD in TT305/6VA mice (81.5 ± 8.4%; n = 15; P = 0.044; Fig. 3B). The difference between these groups was significant (P = 0.0021). We next examined whether the same switch toward LTD occurs when the 1-Hz LTP protocol is applied. In WT mice, single-pulse PF activation potentiated EPSC amplitudes (122.3 ± 4.3%; t = 36–40 min; n = 7; P = 0.002; Fig. 3A). In TT305/6VA mice, LTP was observed as well (124.9 ± 7.7%; n = 8; P = 0.007; group comparison: P = 0.42; Fig. 3A). Likewise, LTD was unaffected in TT305/6VA mice, whether it was triggered by the 1-Hz protocol (WT: 72.8 ± 4.4%, t = 36–40 min, n = 8, P = 0.0005; TT305/6VA: 67.6 ± 3.0%, n = 5, P = 0.0004; group comparison: P = 0.38; Fig. S4A) or the 100-Hz protocol (WT: 71.5 ± 5.0%, n = 6, P = 0.002; TT305/6VA: 70.2 ± 5.3%, n = 5, P = 0.005; group comparison: P = 0.93; Fig. S4B). We also tested LTD/LTP protocols in T305D mice, in which Thr305 is replaced by a negatively charged Asp, which serves as a phosphomimetic resembling persistent Thr305 phosphorylation and preventing Ca/CaM binding (27). In slices prepared from T305D mice, application of the 1-Hz LTD protocol and the 100-Hz LTD protocol, respectively, resulted in LTP instead (1 Hz: 127.9 ± 10.2%, t = 36–40 min, n = 6, P = 0.041; 100 Hz: 138.3 ± 12.4%, n = 7, P = 0.021; Fig. 4). These results are significantly different from the respective LTD measures in WT controls (same as reported for the TT305/6VA mice; 1-Hz group: P = 0.0019; 100-Hz group: P = 0.0027). LTP induction was not affected in T305D mice [1-Hz LTP: 136.3 ± 12.3%, n = 5, P = 0.018 compared with WT (P = 0.123); 100-Hz LTP: 134.0 ± 9.6, n = 6, P = 0.017 compared with WT (P = 0.93); Fig. S5]. These results suggest that high-frequency PF activation causes CaMKII phosphorylation at Thr305/306, which blocks LTD induction despite calcium levels above the threshold for 1-Hz LTD. CF coactivation prevents inhibitory autophosphorylation and restores LTD. However, LTD is not induced when CaMKII is inactivated in a phosphomimetic mutant, which demonstrates the existence of one (rather than multiple) threshold(s) for LTD induction that is regulated by the availability of CaMKII for activation.

Fig. 3.

LTP induced by 100-Hz PF burst stimulation, but not 1-Hz PF stimulation, is prevented in mice that express CaMKII and cannot undergo Thr305/306 phosphorylation. (A, Left) Time graph showing that 1-Hz PF stimulation induces LTP in WT mice (n = 7) and TT305/6VA mice (n = 8). The arrow indicates time of tetanization. (A, Center) Typical traces show EPSCs before and after tetanization in WT mice (Top) and TT305/6VA mice (Bottom). (A, Right) Individual cell data (t = 36–40 min). (B, Left) Time graph showing that 100-Hz PF burst stimulation induces LTP in WT mice (n = 12), but LTD in TT305/6VA mice (n = 15). (B, Center) Typical traces show EPSCs before and after tetanization in WT (Top) and TT305/6VA mice (Bottom). (B, Right) Individual cell data. Error bars indicate SEM. **P < 0.01.

Fig. S4.

LTD is not impaired in TT305/6VA mice. (A, Left) Time graph showing that 1-Hz PF + CF activation induces LTD in WT mice (n = 8) and TT305/6VA mice (n = 5). (A, Center) Typical traces show EPSCs before and after application of the 1-Hz LTD protocol in WT mice (Top) and TT305/6VA mice (Bottom). (A, Right) Plot of individual cell data (t = 36–40 min). (B, Left) Time graph showing that 100-Hz PF burst stimulation paired with CF coactivation induces LTD in WT mice (n = 6) and TT305/6VA mice (n = 5). (B, Center) Typical traces show EPSCs before and after application of the 100-Hz LTD protocol in WT mice (Top) and TT305/6VA mice (Bottom). (B, Right) Plot of individual cell data (t = 36–40 min). Error bars indicate SEM.

Fig. 4.

LTD is prevented in T305D mice, in which Thr305 replacement by Asp mimics constitutive inhibitory CaMKII autophosphorylation. (A, Left) Time graph showing that 1-Hz PF + CF activation induces LTD in WT mice (n = 8), but LTP in T305D mice (n = 6). (A, Center) Typical traces show EPSCs before and after application of the 1-Hz LTD protocol in WT mice (Top) and T305D mice (Bottom). (A, Right) Individual cell data (t = 36–40 min). (B, Left) Time graph showing that 100-Hz PF burst + CF activation induces LTD in WT mice (n = 6), but LTP in T305D mice (n = 7). (B, Center) Typical traces show EPSCs before and after application of the 100-Hz LTD protocol in WT mice (Top) and T305D mice (Bottom). (B, Right) Individual cell data. Error bars indicate SEM. **P < 0.01.

Fig. S5.

LTP is not impaired in T305D mice. (A, Left) Time graph showing that 1-Hz PF activation induces LTP in both WT mice (n = 7) and T305D mice (n = 5). (A, Center) Typical traces show EPSCs before and after application of the 1-Hz LTP protocol in WT mice (Top) and T305D mice (Bottom). (A, Right) Plot of individual cell data (t = 36–40 min). (B, Left) Time graph showing that 100-Hz PF burst stimulation induces LTP in WT mice (n = 12) and T305D mice (n = 6). (B, Center) Typical traces show EPSCs before and after application of the 100-Hz LTP protocol in WT mice (Top) and T305D mice (Bottom). (B, Right) Plot of individual cell data (t = 36–40 min). Error bars indicate SEM.

Discussion

The observation that CF signaling promotes LTD regardless of large variations in calcium signaling at different activation frequencies sheds light on the importance of instructive signals in bidirectional plasticity and the way they operate. Our data show that the LTD threshold slides, enabling the instructive CF signal to operate over a wide frequency range by providing, at a given frequency, a spine calcium transient that is higher than in the absence of this signal (Fig. 5). The observation that the difference in the calcium signal amplitude caused by the switch in activation frequency is larger than the difference resulting from CF coactivation (Fig. 1) leads to the question of whether CF activity indeed promotes LTD by amplifying calcium transients, or whether it contributes other signaling factors that trigger the switch toward LTD. For example, CF activity results in the release of corticotropin-releasing factor, which promotes LTD at PF and CF synapses by activation of PKC (33, 34). In addition, in the mature cerebellum, CF stimulation recruits a specific calcium source that is not available at PF synapses, postsynaptically located NMDA receptors (7, 35, 36). It is conceivable that these CF-specific signaling components contribute to LTD. However, the BAPTA experiments demonstrate that a critical factor in the switch from LTP to LTD is the local calcium signal amplitude, which is significantly enhanced upon CF coactivation. A critical prediction of this conclusion is that strong PF activation can induce LTD in the absence of CF coactivation. This replacement effect has indeed been demonstrated (37, 38, also ref. 39), suggesting that strong calcium influx, regardless of synaptic origin, promotes LTD but that CF activity facilitates the induction process by amplifying calcium transients.

Fig. 5.

Sliding plasticity thresholds and role of CaMKII inhibitory autophosphorylation. (A) Schematic presents a model of the relationship between calcium amplitudes and LTD/LTP as assessed in this study (the dashed lines indicate that possible LTP thresholds were not investigated). The numbers below show approximations of peak [Ca²⁺]_i values, which were calculated from ΔG/R measures recorded at each stimulus condition. Note that these peak values were determined from individual protocol-typical stimuli, and not from complete stimulus trains (Fig. S6). [Ca²⁺]_i values are not presented for the low-amplitude signals (N.D.), because no reliable measures above noise could be obtained with the low-affinity indicator Fluo-5F. (B) Diagram showing the role of CaMKII in LTD induction. CaMKII indirectly promotes LTD (dashed arrow) by negative regulation of phosphodiesterase 1 (PDE1), and the resulting facilitation of a cGMP/protein kinase G (PKG) cascade, which ultimately removes a blockade of LTD induction pathways by protein phosphatase 2A (PP2A) (also ref. 30). Inhibitory autophosphorylation of CaMKII at Thr305/306 may disable this negative regulation of PP2A.

The concept of synaptic modification under control of postsynaptic activity thresholds was introduced in the Bienenstock–Cooper–Munro (BCM) theory, which was originally developed as a model for developmental plasticity in the visual cortex (8). A key assumption of the BCM theory is that the modification threshold θ_M, which marks the level of postsynaptic activity at which the polarity of synaptic plasticity changes from depression to potentiation, is not fixed but, instead, may slide as a function of synaptic activation history. Our data show that in the low-frequency stimulus range, the [Ca²⁺]_i threshold for LTD induction is ≤0.4 μM, whereas in the high-frequency range, it is >3 μM. These values fall into the range of [Ca²⁺]_i amplitudes that have been reported as triggering LTD induction in the absence of synaptic activation (40, 41). Thus, these data demonstrate that the shift in the LTD threshold upon high-frequency synaptic activation is a process that prevents the LTD that would otherwise take place at the same [Ca²⁺]_i levels. Previous reports have assigned a key role for such threshold shifts to CaMKII. It has been shown that CaMKII autophosphorylation at Thr286 shifts the threshold for hippocampal LTP to the right, thus facilitating LTD at potentiated synapses (42, also ref. 43). Moreover, it has been demonstrated that priming-induced Thr305/306 phosphorylation prevents subsequent LTP induction (32). In these examples, threshold shifts follow prior synaptic activity; thus, they constitute forms of metaplasticity as predicted by the BCM rule. The threshold shift described here does not constitute a classic form of metaplasticity [i.e., “plasticity of synaptic plasticity” (44)], because the LTD threshold does not shift as a consequence of activity that occurred before stimulus protocol application and separated by an activity-free interval from it but, instead, results exclusively from sensing enhanced calcium levels during stimulus application. Our finding that application of the high-frequency LTP protocol in TT305/6VA mice induces LTD instead, whereas the same polarity switch is not observed when using the low-frequency LTP protocol, suggests that inhibitory autophosphorylation of CaMKII selectively occurs upon high-frequency PF stimulation but is overcome by CF coactivation. Not much is known about the biochemical requirements for Thr305/306 autophosphorylation or its prevention. It has been shown that Thr305/306 autophosphorylation requires Ca/CaM binding to CaMKII and subsequent dissociation (31). A possible, but not experimentally tested, scenario is that repeated 100-Hz PF bursts provide the calcium influx required for Ca/CaM binding to CaMKII, but that during the pauses between the bursts, which last almost 1 s, CaM dissociates from CaMKII, allowing inhibitory phosphorylation to take place. Single PF pulses (1-Hz LTP protocol) do not seem to provide sufficient calcium to trigger Ca/CaM binding to CaMKII. Conversely, the larger calcium transient that results from paired 100-Hz PF burst + CF activation might provide sufficient calcium to prevent CaM dissociation, thus allowing CaMKII to remain in the activated state.

The frequency-dependent calcium threshold shift described here might also explain why it has been difficult to confirm the existence of distinct calcium thresholds for LTD and LTP induction using photolysis of caged calcium compounds (40, 45). In the absence of synaptic activity, the thresholds might assume different, possibly overlapping, values, and additional parameters, such as the involvement of specific calcium sources/sensors, might gain importance. The latter possibility has been demonstrated in cortical pyramidal cells, in which larger calcium transients are needed for LTP than for LTD induction, but activation of metabotropic glutamate receptors provides a switch for LTD. Blockade of this switch allows for LTP induction if the calcium signals reach the higher LTP threshold (14). An important aspect of calcium amplitude thresholds is that they can vary with the duration of calcium exposure. Using calcium uncaging experiments, for example, it has been shown that lower calcium amplitudes are sufficient for cerebellar LTD induction if they are presented for longer periods of time (40). Note that our data do not contradict this “leaky integrator” model. Rather, we kept the overall duration of the tetanization period constant (all stimulus patterns were applied at 1 Hz for 5 min) to study the shift in LTP/LTD induction probabilities that specifically results from a variation in the synaptic activation frequency alone. A consequence of calcium thresholds that shift depending on stimulus conditions, such as variations in duration or frequency, is that threshold values do not generalize, but only apply to very specific activation conditions. For this reason, we only report peak [Ca²⁺]_i values for protocol-typical individual stimuli (Fig. 5), but otherwise describe relations between threshold amplitudes. Our findings have important implications for the control of activity-dependent synaptic plasticity by dendritic calcium transients. The study identifies a fundamental problem that synapses have to solve when enabling LTD and LTP under control of an instructive signal, but independent of the induction frequency: Spine calcium transients reach different amplitudes at different activation frequencies. At PF-Purkinje cell synapses, this problem is solved by a shift of the LTD threshold upon Thr305/306 phosphorylation of CaMKII, allowing for LTD/LTP induction independent of absolute calcium levels, but under control of relative calcium thresholds.

Materials and Methods

Animals.

All procedures were performed in accordance with the guidelines of the University of Chicago’s Animal Care and Use Committee. Experiments were performed using P21-75 mice (C57BL/6). In some experiments, we used P21-75 TT305/6VA or T305D mutant mice and WT littermate controls, which are also in a congenic C57BL/6 background (27).

Slice Preparation.

Animals were anesthetized with isoflurane and decapitated. The cerebellar vermis was removed and cooled in artificial cerebrospinal fluid (ACSF) containing 124 mM NaCl, 5 mM KCl, 1.25 mM Na₂HPO₄, 2 mM CaCl₂, 2 mM MgSO₄, 26 mM NaHCO₃, and 10 mM d-glucose, bubbled with 95% (vol%) O₂ and 5% (vol%) CO₂. Parasagittal slices of the cerebellar vermis (200 μm) were prepared using a Leica VT-1000S vibratome, and were subsequently kept for at least 1 h at room temperature in oxygenated ACSF. Throughout recording, slices were perfused with ACSF that was supplemented with picrotoxin (100 μM) to block GABA_A receptors.

Whole-Cell Patch-Clamp Recordings.

Patch-clamp recordings from the Purkinje cell soma were performed at room temperature using an EPC-10 amplifier (HEKA Electronics). Currents were filtered at 3 kHz, digitized at 25 kHz, and acquired using Patchmaster software (HEKA Electronics). Patch pipettes (2–5 MΩ) were filled with a solution containing 9 mM KCl, 10 mM KOH, 120 mM K-gluconate, 3.48 mM MgCl₂, 10 mM Hepes, 4 mM NaCl, 4 mM Na₂ATP, 0.4 mM Na₃GTP, and 17.5 mM sucrose (osmolarity: 295–305 mmol/kg, pH 7.25). In two groups of experiments, 5 and 20 mM BAPTA (tetrapotassium salt) was added to the internal saline. In the BAPTA experiments, CaCl₂ was added (2.5 and 10 mM) to maintain the resting calcium concentration (46). To evoke synaptic responses, PFs and CFs were activated using glass electrodes filled with ACSF. In the LTD and LTP experiments (Figs. 2–4 and Figs. S3–S5), test responses were recorded in voltage-clamp mode before and after application of the induction protocol at a frequency of 0.05 Hz. Tetanization was applied in current-clamp mode. Series and input resistances were monitored by applying hyperpolarizing voltage steps (−10 mV) at the end of each sweep. Recordings were excluded if series or input resistances varied by >15% over the course of the experiments.

Confocal Calcium Imaging.

Calcium transients were monitored using a Zeiss LSM 5 Exciter confocal microscope equipped with a 63× Apochromat objective (Carl Zeiss MicroImaging). Calcium signals were calculated as ΔG/R = (G(t) − G₀)/R (22), where G is the calcium-sensitive fluorescence (G₀ = baseline signal) of either OGB-2 (200 μM), Fluo-5F (300 μM), or OGB-5N (300 μM) and R is the calcium-insensitive fluorescence of Alexa 633 (30 μM). Using G/R values from the Fluo-5F measurements, peak [Ca²⁺]_i was estimated by calculating (22)

$${\left[{\text{Ca}}^{2+}\right]}_{\text{i}}=K_{\text{d}}\left(\text{G}/\text{R}\mathrm{-}\text{G}/{\text{R}}_{\min }\right)/\left(\text{G}/{\text{R}}_{\max }\mathrm{-}\text{G}/\text{R}\right),$$

which is analogous to the equation (47)

$${\left[{\text{Ca}}^{2+}\right]}_{\text{i}}=K_{\text{d}}\left(\text{R}\mathrm{-}{\text{R}}_{\min }\right)/\left({\text{R}}_{\max }\mathrm{-}\text{R}\right),$$

where R is the ratio of the fluorescence intensities F₁ and F₂ of a dual-wavelength indicator (e.g., fura-2) at excitation wavelengths λ₁ and λ₂, respectively. For Fluo-5F, G/R_max (1.57) was measured in dye-filled Purkinje cells. To obtain maximal calcium signal amplitudes, the CF input was tetanized at 100 Hz in ACSF containing 4 mM CaCl₂. G/R_min (0.0026) was determined in cuvettes using a solution containing 0 mM [Ca²⁺], supplemented with 10 mM K₂EGTA. The K_d values for all dyes were determined using cuvette measurements of solutions prepared from a calcium calibration buffer kit (Molecular Probes). The green fluorescence G resulted from excitation at 488 nm using an argon laser. The red fluorescence R resulted from excitation at 633 nm using a HeNe laser (both from Lasos Lasertechnik). Purkinje cells were loaded with the dyes through diffusion from the patch pipette. The experiments were performed at room temperature and were initiated after the fluorescence at the selected dendritic ROI reached a steady-state level, which typically required ≥30 min. In each recording, the ROI was the spine on a secondary (or higher order) dendritic branch in the field of view close to the stimulus electrode that responded maximally to synaptic activation. The maximally responding spine was selected to ensure comparability between the calcium measures in different neurons. The strategy to focus on protocol-typical stimulus units, rather than measuring calcium levels during the entire tetanization period (5 min), was chosen to be able to test all four stimulus conditions in the same Purkinje cell. This strategy would not be possible with the application of complete protocols, which trigger either LTD or LTP, and would thus affect subsequent measures. This approach is validated in control experiments (Fluo-5F) that show there is no calcium build-up between stimuli over the course of 5-min periods of tetanization, not even when the high-frequency LTD protocol is applied (Fig. S6). Image acquisition was restricted to the first 2 s of each beginning minute of ongoing stimulation to reduce the amount of phototoxicity. This strategy allowed us to monitor baseline and peak calcium levels during tetanization while minimizing transient calcium build-up that may result from prolonged light exposure [compare with studies that focused on capturing the complete calcium profile during ongoing tetanization (48, 49)]. Our recordings show that under these imaging conditions, baseline calcium levels do not plateau.

Fig. S6.

Absence of calcium build-up during prolonged tetanization. (A) Calcium transients during application of the high-frequency LTD protocol (100-Hz PF burst, followed 120 ms after stimulus onset by a CF pulse; this stimulus pattern is applied at 1 Hz for 5 min). Fluorescence was monitored once per minute of tetanization, for a period of 2 s (2 stimuli). In addition, the fluorescence was measured immediately after completion of the tetanization protocol. The traces show the average calcium transients recorded from six Purkinje cells. The calcium signals are expressed as ΔG/R (percentage increase from the baseline preceding tetanization). Note that we selected the strongest tetanization protocol for a proof-of-principle demonstration that calcium levels do not significantly plateau in-between stimuli during the entire tetanization period. (B) Bar graph showing the peak amplitude levels for responses 1(red bars) and 2 (purple bars) during each 2-s sweep (200-ms average starting from stimulus onset) and the baseline calcium accumulation recorded before stimulus onset (average over a 200-ms baseline period; gray bars), as well as following the end of tetanization (end). The corresponding imaging periods are shown in A for the first 2 s of image acquisition. At no time point did the calcium accumulation reach significance compared with the 200-ms baseline period preceding the onset of the first stimulus (n = 6; P > 0.05). Values are shown as mean ± SEM.

Data Analysis.

Data were analyzed using Patchmaster software (HEKA Electronics) and Igor Pro software (Wavemetrics). Imaging data were analyzed using ZEN software (Carl Zeiss MicroImaging). Statistical significance was determined by using the paired Student’s t test (within-group comparison of paired events) and the Mann–Whitney U test (between-group comparison), when appropriate. All data are shown as mean ± SEM.