A positive feedback signal transduction loop determines timing of cerebellar long-term depression

PKC, MAPK and PLA2 work downstream of Ca²⁺ during LTD

We used experiments analogous to genetic epistasis analysis (Lev-Ram et al., 1997) to determine whether PKC, MAPK, and PLA2 are required downstream of Ca²⁺. Specifically, we asked whether the LTD produced by direct elevation of [Ca²⁺]_i could be blocked by inhibitors of PKC, MAPK, and PLA2. If the enzymes work downstream of Ca²⁺, then these inhibitors should prevent Ca²⁺ from producing LTD; conversely, if the enzymes work upstream from Ca²⁺, then the inhibitors should not block LTD.

For this purpose, we took advantage of previous observations that elevating [Ca²⁺]_i via photolysis of a caged Ca²⁺ compound is sufficient to induce LTD (Miyata et al., 2000; Tanaka et al., 2007). This Ca²⁺-induced LTD shares a common signal transduction pathway with LTD induced by synaptic activity (Tanaka et al., 2007), justifying the use of this paradigm for analysis of the role of the signaling enzymes in LTD. [Ca²⁺]_i was elevated by using a spot of UV light, approximately 10 µm in diameter, to photolyze a caged Ca²⁺ compound (DMNPE-4) within Purkinje cell dendrites at the site where PFs were stimulated. LTD was evident as a reduction in the amplitude of excitatory postsynaptic currents evoked by electrical stimulation of PF axons (PF-EPSCs; Figure 1B, left). On average, PF-EPSCs were depressed by 37.8% ± 6.0% (n = 7) at 20–40 min after the elevation of [Ca²⁺]_i (Figure 1C). Introducing a PKC inhibitory peptide (PKCI; House and Kemp, 1987) into Purkinje cells caused LTD to be significantly smaller (13.0% ± 5.5%; p = 0.01; n = 6). Ca²⁺-triggered LTD also was blocked by two small-molecule inhibitors of the MAPK pathway (Figure 1D). PD98059, which blocks this pathway by binding to the inactive form of the kinase upstream of MAPK, MEK (Alessi et al., 1995), reduced LTD to 10.7% ± 9.8% (n = 5). Likewise, U0126, which is a noncompetitive inhibitor of MEK (Favata et al., 1998), also reduced LTD to 1.2% ± 6.9% (n = 4). The reduction in LTD produced by each of these inhibitors was significantly different (p < 0.05) from the 36.3% ± 3.7% LTD (n = 7) measured in the presence of DMSO, the vehicle used to dissolve these inhibitors. Finally, OBAA, an inhibitor of PLA2 (Köhler et al., 1992), caused a reduction in LTD (9.7% ± 4.9%, n = 5) that also was significantly (p = 0.004) different from the LTD measured in the presence of the DMSO solvent (Figure 1E). Thus, inhibitors of all three enzymes blocked LTD induced by direct elevation of [Ca²⁺]_i, indicating that these enzymes are required for LTD induced by Ca²⁺ and that they work downstream of Ca²⁺, as predicted by the positive feedback loop model.

Reciprocal activation of PKC and MAPK during LTD

The essence of the model is that PKC and MAPK mutually activate each other to create a positive feedback loop (Figure 1A). We tested this prediction by performing additional epistasis experiments that determined (1) whether PKC and MAPK activate each other during conditions that induce LTD; and (2) whether such reciprocal activation of PKC and MAPK is required for LTD.

Positive feedback loop sustains PKC activation during LTD

Kinetic simulations predict that one of the main functions of the positive feedback loop is to translate the brief signals generated by synaptic activity into a sustained activation of PKC (Kuroda et al., 2001). We next tested this prediction by defining the time window when the PKC inhibitor, BIM, could prevent LTD. BIM alone did not affect PF-EPSCs (Figure 6A): PF-EPSCs measured 10–20 min after BIM application were 97% ± 7% (n = 4) of their amplitude measured prior to BIM application, which was not a significant difference (p = 0.69). Applying PF&ΔV stimulation without BIM produced LTD (Figure 6B; see also Figure 5D). But if BIM was applied as late as 20 min after the PF&ΔV stimulus, LTD was reduced or eliminated (Figures 6C–6F). Even when BIM was applied at 20 min after the PF&ΔV stimulus, the amount of depression calculated 40–50 min after this stimulus was significantly (p = 0.03) smaller than in control experiments where BIM was not added. Applying BIM soon after LTD began (eg. 10 min after the stimulus, Figure 6E), caused EPSC amplitude to partially recover, suggesting that the action of PKC was reversible at this time. Similar results were obtained by using another PKC inhibitor, Gö6976, demonstrating that the effects of BIM result from inhibition of PKC (Figure S1). However, applying BIM 30 min or later after the stimulus did not affect LTD (Figures 6G and 6H). The relationship between the time of BIM application and the degree of LTD impairment indicated that PKC remains active for at least 20–30 min after the end of the stimulus that induces LTD (Figure 8A). Thus, activation of PKC during LTD is prolonged, as predicted by the model.

To determine whether this prolonged activation of PKC requires the positive feedback loop, we disrupted this loop pharmacologically. For this purpose, we used the PLA2 inhibitor, OBAA, because the MAPK pathway inhibitor U0126 affected basal synaptic transmission (164% ± 7%; of control; n = 4) while OBAA did not (Figure 7A; 95% ± 6%, n = 7). As was the case for BIM, OBAA was effective in reducing LTD even when applied 20–30 min after the end of the LTD-inducing stimulus (Figures 7C–7F). The relationship between the time of OBAA application and the degree of LTD impairment (Figure 8B) closely paralleled the relationship shown for BIM application in Figure 8A. We therefore conclude that the positive feedback loop is responsible for sustaining activation of PKC for at least the first 20–30 min after synaptic activity. This result also provides yet another indication that the loop is involved in LTD.

Signaling molecules involved in the positive feedback loop

Although experiments from several different experimental preparations previously implicated MAPK, PKC and PLA2 in LTD (Linden and Connor, 1991; Hartell, 1994; Linden, 1995; De Zeeuw et al., 1998; Kawasaki et al., 1999; Reynolds and Hartell, 2001; Leitges et al., 2004), the targets of these enzymes during LTD induction had not been established aside from the phosphorylation of AMPA-type glutamate receptors (AMPARs) by PKC (Chung et al., 2003; Steinberg et al., 2006). Our experimental results indicate that these enzymes serve to reciprocally activate each other to form a positive feedback loop. The pair-wise interactions of these enzymes have been established by in vitro experimentation or by studies of signal transduction pathways in several cell types: MAPK directly phosphorylates and activates PLA2 (Lin et al., 1993), PLA2 produces AA (Clark et al., 1991), and AA binds to and activates PKC, alone or in conjunction with Ca²⁺ or diacylglycerol (Murakami and Routtenberg, 1985; Shinomura et al., 1991; Shirai, 1998; López-Nicolás et al., 2006). Although the mechanism involved is controversial, PKC-dependent activation of MAPK also has been observed in many types of cells (Marais et al., 1998; Corbit et al., 2003; Zheng et al., 2005; Wen-Sheng, 2006), including in Purkinje cells (Endo and Launey, 2003; Ito-Ishida et al., 2006). These interactions served as the basis for a theoretical study proposing that these enzymes can form a positive feedback loop (Bhalla and Iyengar, 1999). Our results provide direct experimental support for the notion that such a positive feedback loop can serve as a general mechanism for cellular signaling and specifically demonstrate that such interactions are important for LTD in Purkinje cells.

Kinetic modeling suggests that the main purpose of the positive feedback loop during LTD is to sustain activation of PKC (Kuroda et al., 2001). Our results provide several lines of experiment evidence supporting this suggestion. First, histochemical data indicated that PKC activity is sustained during conditions that evoke “chemical LTD” (Figure 3C). Second, biochemical data indicated that activation of MAPK also is sustained under such conditions (Figure 2C). Although our histological results (Figure 2E) suggest that MAPK activation in Purkinje cells may be more transient than indicated by the biochemical measurements, this discrepancy might be due to a sustained activation of MAPK that is too small and/or too localized to be detected in our histological experiments. (Even if MAPK activation were short-lived, the resultant phosphorylation-dependent activation of PLA2 should be sustained because subsequent dephosphorylation of PLA2 is very slow (Gordon et al., 1996)). Finally, timed application of pharmacological inhibitors indicated that activation of both PKC (Figure 8A) and PLA2 (Figure 8B) is sustained for at least 20 minutes after the end of the PF&ΔV stimulus that was used to induce LTD. These results are important because they demonstrate that the sustained activation of PKC by the positive feedback loop is required for LTD induction.

Time course of signal transduction during LTD

Our studies provide novel information about the timing of reactions involved in LTD induction and expression. The 20–30 min time window we observed for the activity of the positive feedback loop during LTD is briefer than the 60–90 min predicted by the computational model of Kuroda et al. (2001). This indicates that mechanisms downstream of this loop sustain LTD earlier than expected. Thus, while our data largely confirm the computational model, they yield a new view of the signal transduction mechanisms underlying LTD induction, expression and maintenance (Figure 9). We propose that synaptic activity transiently increases [Ca²⁺]_i (blue in Figure 9). Although the increase in [Ca²⁺]_i disappears within 1 s after synaptic activity ceases, activation of PKC and/or PLA2 by Ca²⁺ triggers the positive feedback loop. Once this loop is triggered, it causes PKC activity to be self-sustaining for at least 20–30 min (red in Figure 9). This time frame coincides with the time that LTD is gradually expressed (eg. Figure 6B), indicating the cause of such slow development of LTD. Specifically, phosphorylation of PKC substrates, which presumably include AMPAR (Chung et al., 2003; Steinberg et al., 2006), during this time leads to internalization of AMPAR via clathrin-dependent endocytosis (Wang and Linden, 2000; Tanaka et al., 2007) and yields a gradual expression of LTD. At later times, more than 60–90 min after stimulation (green in Figure 9), synthesis of unknown new proteins is required to maintain LTD (Linden, 1996; Ahn et al., 1999). It is not yet clear what role, if any, the positive feedback loop plays in this late phase of LTD.

Between the time that the positive feedback loop is active (20–30 min) and the time that protein synthesis-dependent mechanisms are involved (60–90 min), transmission at the PF synapse is maintained at a constant, depressed level (white in Figure 9). We presume that this phase of LTD arises from sustained changes in AMPAR trafficking. In Purkinje cells, the AMPARs involved in LTD are mainly composed of GluR2/3 subunits (Ito, 2001; Chung et al., 2003), which are constitutively inserted and removed at many glutamatergic synapses (Bredt and Nicoll, 2003). Basal transmission at PF synapses is kept constant by maintaining a balance between insertion and removal of these AMPARs, because PF-EPSC amplitude is reduced when insertion of these AMPARs is blocked (Tatsukawa et al., 2006). Thus, the constant, reduced level of PF transmission produced between 20–60 min after stimulation probably is caused by AMPAR trafficking being maintained at a new equilibrium level with fewer surface AMPARs. The signaling mechanisms involved in establishing this new equilibrium remain to be determined.

When inhibitors of PKC or PLA2 were applied soon after LTD began, LTD was interrupted and EPSC amplitude recovered (Figure 6E, Figure 7D, and Figure S1B). These results indicate that LTD induction consists of at least two components that can be distinguished by the response to inhibitors of the positive feedback loop: one component causes a reversible depression of synaptic transmission, while the second component causes an irreversible depression. There seems to be a time-dependent progression from the reversible to irreversible components: under our experimental conditions, the reversible component was most evident when such inhibitors were applied 10 minutes after the LTD-inducing stimulus ended, while the irreversible component was more evident at later times. The molecular underpinnings of these two components remain to be determined; it is possible that the reversible component reflects reversible phosphorylation of AMPAR by PKC, with subsequent endocytosis of AMPAR, while the irreversible component may be some other reaction that results from positive feedback loop activity and causes LTD expression to be locked into an irreversible state. Our observation that inhibition of MAPK activation by U0126 completely blocked the LTD produced by TPA application (Figure 2F) suggests that MAPK has an additional role beyond activating PKC. Thus, perhaps the reversible component reflects internalization of AMPAR caused by PKC-dependent phosphorylation, while MAPK activity is responsible for subsequent conversion of AMPAR trafficking into an irreversible state. Such an arrangement may account for the observation that PLA2 activation prolongs the duration of LTD, converting a short-term synaptic depression into LTD (Linden, 1995).

Although LTD is expressed progressively over time, the precise time course of LTD expression varies according to the type of stimulus and stimulus intensity (Ito, 2001). Even in our experiments there were some stimulus-dependent differences in LTD time course: LTD induced by uncaging Ca²⁺ often reached a steady state within 5–10 min (Figure 1), while LTD induced by PF&ΔV (Figure 7 and Figure 8) or K-glu treatment (Figure 2 and Figure 3) was more gradual and reached a maximum over 20–30 min. Kinetic simulations predict that such stimulus-dependent variability arises from the efficacy of the stimulus in activating the positive feedback loop (Kuroda et al., 2001). Our refinement of the positive feedback model postulates that PKC activity need not be sustained after LTD expression is maximal; this may account for the observation of Tsuruno and Hirano (2007) that sustained PKC translocation cannot be observed in cultured Purkinje cells after the time that LTD is maximal.

Significance of the positive feedback loop

The positive feedback loop was first postulated to serve as a bistable switch that allows brief stimuli to produce sustained activation of protein kinases (Bhalla and Iyengar, 1999). In Purkinje cells, rises in [Ca²⁺]_i associated with synaptic activity are short-lived (Konnerth et al., 1992; Eilers et al., 1997; Finch and Augustine, 1998; Wang et al., 2000a) and are limited in time by the activity of highly expressed Ca²⁺ binding proteins and Ca²⁺ pumps (Baba-Aissa et al., 1998; Hartmann and Konnerth, 2005). Thus, the bistable character of this loop effectively serves as a temporal integrator that translates brief Ca²⁺ signals into the long-lasting activation of PKC that is required for LTD. Bistabilitiy arising from positive feedback loops has also been proposed as a mechanism for regenerative spatial propagation of signals from the plasma membrane to the nucleus (Markevich et al., 2006). Although LTD is known to spread from active PF synapses to inactive ones (Wang et al., 2000b; Reynolds and Hartell, 2000), it is unlikely that the bistable properties of the PKC/MAPK/PLA2 positive feedback loop are responsible for this spread, because the positive feedback loop works downstream of Ca²⁺ and it is known that LTD cannot spread beyond the site of [Ca²⁺]_i elevation (Wang and Linden, 2000; Tanaka et al., 2007). However, it remains possible that the positive feedback loop could be involved in signaling between active synapses and the nucleus during LTD (Linden, 1996).

Similar regenerative mechanisms may also serve as temporal integrators to maintain other forms of long-lasting synaptic plasticity. One prominent example has been well-established for long-term potentiation of the Schaffer collateral synapse in the hippocampus. Here, a brief Ca²⁺ signal in postsynaptic pyramidal cells (Malenka et al., 1992; Petrozzino et al., 1995; Yang et al., 1999) leads to autophosphorylation of Ca²⁺/calmodulin-dependent protein kinase II (CaMKII), a regenerative mechanism that yields the sustained activation of this kinase that is required for induction and maintenance of LTP (Lisman and Goldring, 1988; Giese et al., 1998; Lisman et al., 2002; Atkins et al., 2005; Sanhueza et al., 2007). Although CaMKII is also known to participate in cerebellar LTD (Hansel et al., 2006), it is not yet known whether it serves a regenerative function as in pyramidal cells. A second example can be found in a form of long-term synaptic facilitation that is important for long-term memory in Aplysia. This synaptic facilitation requires the self-perpetuating activation of a translational regulator, cytoplasmic polyadenylation element binding protein, that stabilizes a late phase of facilitation (Si et al., 2003; Bailey et al., 2004). Given that most forms of long-term synaptic plasticity are initiated by brief Ca²⁺ signals (Malenka et al., 1992; Yang et al., 1999; Franks and Sejnowski, 2002; Bonsi et al., 2004; Gall et al., 2005; Nevian and Sakmann, 2006; Humeau and Luthi, 2007), it is possible that regenerative mechanisms could be a common means of sustaining long-term synaptic plasticity.

Patch clamp recording

Whole-cell patch clamp recordings were made from Purkinje cells as described previously (Miyata et al., 2000; Wang et al., 2000b). Sagittal slices (200 µm) of cerebella from 14- to 21-day-old mice were bathed in extracellular solution containing (in mM): 125 NaCl, 2.5 KCl, 1.3 Mg₂Cl, 2 CaCl₂, 1.25 NaH₂PO₄. 26 NaHCO₃, 20 glucose, and 0.01 bicuculline methochloride (Tocris). TPA, PD98059, U0126, BIM (all from Calbiochem), or OBAA (Tocris) were added into the extracellular solution as indicated. When inducing “chemical LTD”, an extracellular solution containing high K⁺ (50 mM) and glutamate (10 µM) was applied under current-clamp conditions. Patch pipettes (resistance 3–5 MΩ) were filled with (in mM): 130 potassium gluconate, 2 NaCl, 4 MgCl₂, 4 Na₂-ATP, 0.4 Na-GTP, 20 HEPES (pH 7.2), and 0.25 Ca²⁺ indicator (Oregon Green 488 BAPTA-1, Molecular Probes) or EGTA. When LTD was induced by uncaging Ca²⁺, 10 mM caged Ca²⁺ compound (DMNPE-4, gift from Dr. Graham Ellis-Davies), and 5 mM CaCl₂ were added into the internal solution (Tanaka et al., 2007). For the experiments using caged MEK proteins, caged MEK-CA or MEK-KD (1–2 mg/ml) were added into the internal solution along with dextran-conjugated fluorescein (Molecular Probes), to monitor the dialysis of intracellular solutions into Purkinje cells. In some experiments, PKCI (Sigma) was added into the internal solution.

EPSCs were evoked in Purkinje cells (holding potential of −70 mV) by activating PFs with a glass stimulating electrode on the surface of the molecular layer. PF-EPSCs were acquired and analyzed using LTP software (W.W. Anderson, Univ. Bristol, UK; Anderson and Collingridge, 2001). To evoke LTD by electrical stimulation (PF&ΔV), PF stimuli were paired with Purkinje cell depolarization (0 mV, 200 ms) 300 times at 1 Hz. Data were accepted if the series resistance changed less than 20%, input resistance was greater than 80 MΩ, holding current changed less than 5%, and Ca²⁺ levels returned to baseline following photolysis of caged Ca²⁺. Uncaging neither Ca²⁺ nor MEK proteins changed the input resistance or holding current, and UV illumination in the absence of caged compound did not affect PF synaptic transmission (Tanaka et al., 2007).

To evaluate the effect of applying BIM or OBAA at different times on the LTD induced by PF&ΔV (Figure 6 and Figure 7), we measured the mean amount of depression at 45–55 min after LTD induction. The values used in Figure 8 were calculated as the percentage of depression in the presence of inhibitors divided by the amount of depression measured at this same time period in control conditions.

Live cell imaging and photolysis of caged compounds

The procedures described previously (Tanaka et al., 2007) were used to image fluorescent dyes and to photolyze caged compound. Cells were dialyzed with a solution containing Oregon Green 488 BAPTA-1 and DMNPE-4 or a solution containing fluorescein dextran and caged MEK protein. The set-up described by Wang and Augustine (1995) was used to uncage Ca²⁺ or MEK. UV light (351–364 nm) from an argon ion laser (Coherent model 305) was delivered to the slice, via an optical fiber and an Olympus 40X water-immersion objective, to make a light spot with a half-width of 5–10 µm that was focused on the primary or secondary dendrites of a Purkinje cell. Photolysis was accomplished via a train of brief UV light flashes (5 ms duration, 1Hz) that was controlled by an electronic shutter. For uncaging Ca²⁺, the train was 90 sec in duration and the total UV energy applied was 300 µJ. For uncaging MEK, the duration and total energy were 5 min and 900 µJ, respectively.

Immunoblotting and immunohistochemistry

Primary antibodies used for immunoblotting and immunohistochemistry were monoclonal anti-MAPK, monoclonal anti-phosphorylated MAPK (Cell Signaling Technology), monoclonal PKCα (BD transduction Laboratories), and polyclonal anti-calbindin (Sigma). Secondary antibodies were horseradish peroxidase (HRP)-conjugated anti-mouse IgG, Cy5-conjugated anti-mouse IgG (Jackson ImmunoResearch Laboratories) and FITC-conjugated anti-rabbit IgG (Invitrogen).

Cerebellar slices were incubated for 30 min in normal ACSF, transferred to K-glu solution for 5 min, and then washed in normal ASCF for 0–10 min. Inhibitors were added for the entire duration of the experiment. For immunoblotting, slices were lysed in a RIPA buffer containing (in mM): 150 NaCl, 50 Tris (pH8), 1% NP- 40, 0.5% deoxycholate, 0.1% SDS, 1% aprotinin, 1 AEBSF, 5 NaF, 1 orthovanadate, 12 glycerophosphate, 1 pyrophosphate tetrabasic decahydrate. After homogenization and sonication, samples were centrifuged to remove cell debris. A part of supernatant was directly applied into SDS-PAGE to detect total amount of MAPK. The rest of the supernatant was used to immunoprecipitate phosphorylated MAPK by using an immobilized phosphorylated MAPK antibody (Cell Signaling Technology). The immunoprecipitate was suspended in SDS-PAGE sample buffer and applied into SDS-PAGE to detect phosphorylated MAPK by immunoblotting. To quantify MAPK phosphorylation, band intensities were measured by Igor Pro (Wave Metrics, Inc) and the ratio of phosphorylated MAPK to total MAPK was then calculated.

For immunohistochemistry, slices were fixed with 4% paraformaldehyde at 4°C, blocked in 5% normal goat serum in phosphate buffered saline, and then incubated in primary antibodies overnight at 4°C. After several washes, slices were incubated in secondary antibodies for 3 hours at room temperature. Images were acquired by a Leica TCS SP2 laser scanning confocal microscope (Leica). Quantification of phosphorylated MAPK in Purkinje cells and PKCα membrane translocation was performed by Photoshop (Adobe) and Igor Pro. To compare phosphorylated MAPK, a threshold level of fluorescence was subtracted from all images of phosphorylated MAPK and calbindin. In these subtracted images, the number of pixels stained by both antibodies (P_pMAPK&cal) and those stained by calbindin (P_cal) were counted, allowing the percentage (P_pMAPK&cal/P_cal × 100) to be calculated for each image. To quantify PKCα membrane translocation, intensity of PKCα staining was measured along a line 2 µm in width across the Purkinje cell soma, as shown in Figure 3B.

Construction and characterization of caged, constitutively active MEK

Recombinant glutathione-S-transferase-fused MEK1 was purified from E. coli by using glutathione sepharose (Amersham Biosciences). MEK-CA was a double mutation of serine residues 218 and 222 to glutamic acid, while MEK-KD was a single mutation of asparatic acid 208 to alanine (Cowley et al., 1994; Okazaki and Sagata, 1995). Purified MEK protein was caged according to a protocol described previously (Marriott et al., 2003) to cage actin-binding proteins. In brief, purified MEK protein, which was dialyzed in PBS and adjusted to approximately pH 9.0 via the addition of NaHCO₃, was incubated with a 10-fold molar excess of the amino-directed protein-caging compound, 4,5-Dimethoxy-2-nitrobenzyl chloroformate (NVOC-Cl, Sigma), in the dark for 30 min at 4°C. The unbound NVOC was removed by centrifugation at 15,000 rpm for 5 min followed by desalting through a PD-10 column (Amercham Biosciences). Final concentrations of MEK and NVOC were determined by a commercial protein assay (Bio-Rad) and OD₃₆₀ measurements (OD360 = 5000 M⁻¹ cm⁻¹), respectively. The labeling ratio typically ranged from 5 to 7 NVOC-CI per MEK molecule. MEK binds to MAPK via its MAPK binding domain, which includes three lysine residues important for this interaction (Tanoue and Nishida, 2003). We suspect that these lysine residues are masked by NVOC labeling and unmasked by UV photolysis.

To examine the activity of caged or uncaged MEK, in vitro assays of MAPK phosphorylation were performed. Caged MEK was photolyzed by 360 nm UV light from a transilluminator for 5 min. Caged or uncaged MEK was mixed with recombinant MAPK and ATP in kinase buffer (Cell Signaling Technology), and incubated for 30 min at 30°C. The kinase reaction was terminated by adding SDS-PAGE sample buffer. Samples were applied into SDS-PAGE and phosphorylated MAPK was detected by immunoblotting with a phospho-MAPK antibody.

Statistical analysis

Statistical differences were determined by paired comparisons with the Student’s t-test.