The PICK1 Ca2+ sensor modulates N-methyl-d-aspartate (NMDA) receptor-dependent microRNA-mediated translational repression in neurons

MicroRNAs (miRNAs) are important regulators of localized mRNA translation in neuronal dendrites. The presence of RNA-induced silencing complex proteins in these compartments and the dynamic miRNA expression changes that occur in response to neuronal stimulation highlight their importance in synaptic plasticity. Previously, we demonstrated a novel interaction between the major RNA-induced silencing complex component Argounaute-2 (Ago2) and the BAR (bin/amphiphysin/rvs) domain protein PICK1. PICK1 recruits Ago2 to recycling endosomes in dendrites, where it inhibits miRNA-mediated translational repression. Chemical induction of long-term depression via NMDA receptor activation causes the dissociation of Ago2 from PICK1 and a consequent increase in dendritic miRNA-mediated gene silencing. The mechanism that underlies the regulation of PICK1-Ago2 binding is unknown. In this study, we demonstrate that the PICK1-Ago2 interaction is directly sensitive to Ca2+ ions so that high [Ca2+]free reduces PICK1 binding to Ago2. Mutating a stretch of C-terminal Ca2+-binding residues in PICK1 results in a complete block of NMDA-induced PICK1-Ago2 disassociation in cortical neurons. Furthermore, the same mutant also blocks NMDA-stimulated miRNA-mediated gene silencing. This study defines a novel mechanism whereby elevated [Ca2+] induced by NMDA receptor activation modulates Ago2 and miRNA activity via PICK1. Our work suggests a Ca2+-dependent process to regulate miRNA activity in neurons in response to the induction of long-term depression.

The brain expresses numerous different miRNAs that play important roles in development and the regulation of synaptic plasticity (3). Mice lacking Dicer or DGCR8, two proteins involved in miRNA biogenesis, display altered synaptic protein expression, synaptic transmission, dendritic spines, and defects in memory and learning (4,5). Long-term potentiation (LTP) and long-term depression (LTD) are two forms of long-lasting plasticity that have been widely studied as mechanisms that underlie memory and learning (6,7). LTP results in a persistent enhancement of synaptic transmission and enlargement of dendritic spines, whereas LTD is a contrasting process in which the efficacy of synaptic transmission is reduced and dendritic spines undergo shrinkage or complete elimination. Both forms of long-term plasticity are induced upon NMDA receptor (NMDAR) activation, which stimulates Ca 2ϩ influx and trigger relevant signaling cascades (6).
In neurons, many miRNAs localize to dendrites and are regulated by synaptic activity. Some of these miRNAs have been shown to play regulatory roles in spine morphology by locally targeting actin-regulatory pathways and the expression of AMPA receptor (AMPAR) subunits (8 -10). For instance, the chemical induction of LTD by NMDAR activation results in rapid and wide-ranging changes in the dendritic miRNAome that are required for long-lasting remodeling of dendritic spines (8). miR-191 expression is reduced in dendrites, resulting in up-regulation of tropomodulin-2 and stimulation of actin depolymerization, followed by shrinkage of dendritic spines (8).
In contrast, expression of miR-501-3p increases in dendrites, where it represses GluA1 expression and is required for longlasting remodeling of dendritic spines (9). As miRNAs play fundamental roles in synaptic plasticity, it is important to understand the molecular changes that occur within dendrites and spines that facilitate miRNA-mediated translational repression. Apart from changes in the expression level of specific miRNAs, the mechanisms for the transduction of NMDARmediated Ca 2ϩ influx into changes in miRNA activity in neurons during synaptic plasticity are unknown.
PICK1 is a BAR (bin/amphiphysin/rvs) and PDZ domain protein that regulates the trafficking of AMPARs in LTD and LTP (11)(12)(13). Furthermore, PICK1 is a Ca 2ϩ -sensing protein that responds to NMDAR-mediated Ca 2ϩ influx to regulate its interaction with the AMPAR subunit GluA2. The increased Ca 2ϩ binding to PICK1 upon NMDAR activation results in an enhanced interaction between PICK1 and GluA2, which facilitates AMPAR internalization (14,15). Previously, we reported a novel interaction between the core RISC regulator Ago2 and PICK1 on recycling endosomes in neuronal dendrites. We demonstrated a direct interaction between the C-terminal tail of PICK1 and the PIWI domain of Ago2. During NMDARmediated chemically induced LTD (cLTD), Ago2 is displaced from PICK1-containing recycling endosomes, resulting in enhanced miRNA activity of dendritically localized miRNAs (16). Because cLTD causes a reduction in Ago2-PICK1 complexes, we were interested in identifying the molecular mechanisms that underlie this response. As PICK1 binding to GluA2 is Ca 2ϩ -sensitive, we hypothesized that Ca 2ϩ binding to PICK1 may also regulate its interaction with Ago2.
In this study, we demonstrate that the Ca 2ϩ -sensing property of PICK1 is required for regulating its interaction with Ago2. We show that, in high [Ca 2ϩ ] free , PICK1-Ago2 complexes are disrupted. Furthermore, we show that, when the PICK1 Ca 2ϩbinding site is mutated, the cLTD-induced increase in the activity of dendritically localized miRNAs in neurons is blocked. This study describes a novel mechanism for the regulation of Ago2 activity by Ca 2ϩ in response to NMDAR activation in neurons, which we propose might be involved in initiating changes in protein expression required for LTD.

The PICK1 C-terminal acidic region regulates Ago2 binding
PICK1 contains two acidic regions that have been demonstrated previously to bind Ca 2ϩ (14,15). One is located at the N terminus between amino acids 3-11, and the second site is positioned between residues 380 -390, toward the C terminus of PICK1 and within the region of PICK1 required for Ago2 binding (16). To identify which site was required for regulating the Ca 2ϩ -sensitive interaction with Ago2, we examined the association of Ago2 with PICK1 Ca 2ϩ -insensitive mutants. It has been shown previously that mutating the first three acidic amino acids in the N-terminal acidic stretch to asparagine (D3N) renders the N-terminal site insensitive to Ca 2ϩ (15). The specific residues required for Ca 2ϩ binding between 380 - 390 have not yet been determined. Therefore, we mutated all 10 acidic residues to alanine (Aϫ10) to prevent the C-terminal region from binding Ca 2ϩ (Fig. 3A).
To investigate the role of PICK1 residues 380 -390 in Ca 2ϩdependent Ago2 binding, we performed GST pulldown experiments on cortical neuronal extracts in defined [Ca 2ϩ ] free using the C-terminal tail of PICK1 (GST-⌬354) as bait (Fig. 4A). Ago2 binding to GST-⌬354 displayed a biphasic dependence on [Ca 2ϩ ] free , consistent with our pulldown and co-IP experiments shown in Fig. 1. The strongest interaction was observed at 4 M [Ca 2ϩ ] free and the weakest at 16 M [Ca 2ϩ ] free . Interestingly, Ago2 binding to GST-⌬354 Aϫ10 did not respond to changes in [Ca 2ϩ ] free , indicating that the Ca 2ϩϪ sensitive Ago2-PICK1 interaction requires amino acids 380 -390 of PICK1 (Fig. 4B). Together, these experiments demonstrate that the PICK1-Ago2 interaction is regulated by Ca 2ϩ binding to the C-terminal acidic region of PICK1 located within the Ago2 binding domain.

The PICK1 C-terminal acidic region mediates Ago2 dissociation in response to NMDAR-mediated cLTD
Previously, we demonstrated that the PICK1-Ago2 interaction is significantly reduced after the induction of cLTD (16). During cLTD, Ca 2ϩ entry through NMDARs increases localized intracellular [Ca 2ϩ ]. Therefore, to determine whether the Ca 2ϩ -sensing property of PICK1 is required for mediating the disassociation of Ago2 during cLTD, we performed molecular replacement experiments in cortical neurons, where endogenous PICK1 was knocked down with a previously characterized shRNA for PICK1 and replaced with sh-resistant GFP-PICK1 WT or GFP-PICK1 Aϫ10 (15,16). shPICK1 robustly reduced the expression of endogenous PICK1 by ϳ80% in neuronal dendrites, whereas shRNA-resistant GFP-PICK1 WT and GFP-PICK1 Aϫ10 restored PICK1 expression to endogenous levels (supplemental Fig. S1). To investigate whether these constructs

Ago2-PICK1 regulation by Ca 2؉
altered the neuronal architecture, we performed a Sholl analysis to measure dendritic branching. PICK1 knockdown decreased the number of dendritic intersections 70 -100 M away from the cell body in a similar manner to a previous report (19). This phenotype was restored to control levels by either GFP-PICK1 WT or Aϫ10 (supplemental Fig. S2, A and B). Furthermore, neither PICK1 knockdown nor molecular replacement altered total dendritic length (supplemental Fig. S2C). Together, these experiments suggest that the C-terminal Ca 2ϩ -binding properties of PICK1 do not regulate dendritic branching or dendritic length.
We performed GFP-trap experiments to measure the interaction between GFP-PICK1 and endogenous Ago2 under basal and cLTD conditions. Under basal conditions, similar levels of Ago2 bound to PICK1 WT and Aϫ10 (Fig. 5, A and B). 10 min after cLTD induction, Ago2 binding to PICK1 WT was significantly reduced, as seen previously (16). However, the interaction between PICK1 Aϫ10 and Ago2 was not disrupted by cLTD (Fig. 5, A and B). In contrast, cLTD caused a similar increase in GluA2 interaction with both PICK1 WT and the Aϫ10 mutant (Fig. 5, A and C). These data indicate that the Ca 2ϩ -sensing property of the PICK1 C-terminal acidic region underlies the release of Ago2 from PICK1 following cLTD induction but that it is not involved in regulating the interaction with AMPARs. PICK1 localizes to endosomal compartments (20), and we showed previously that Ago2 is present with PICK1 on recycling endosomes and that Ago2 dissociates from PICK1-containing compartments in response to cLTD induction (16). We therefore hypothesized that Ca 2ϩ sensing by PICK1 triggers Ago2 release from PICK1-positive endosomes. To test this, we examined Ago2 co-localization with GFP-PICK1 WT or Aϫ10 under basal or cLTD conditions. Although we did not detect any difference in the degree of co-localization of Ago2 with GFP-PICK1 WT or Aϫ10 under basal conditions, cLTD caused a significant reduction in co-localization between GFP-PICK1 WT and Ago2, which was completely blocked by the Aϫ10 mutation (Fig. 6, A and B). Together, these experiments demonstrate that the C-terminal acidic region of PICK1 is essential for regulating the disassociation of PICK1 from Ago2 in response to cLTD.
To investigate whether the dissociation of Ago2 from PICK1 caused by cLTD induction required extracellular Ca 2ϩ , we examined endogenous PICK1-Ago2 binding in cortical neurons stimulated with NMDA in the absence of extracellular Ca 2ϩ . NMDAR activation significantly reduced PICK1-Ago2 binding by ϳ50% in the presence of normal (1.8 mM) Ca 2ϩ . In contrast, NMDAR stimulation in the absence of extracellular Ca 2ϩ had no effect on PICK1-Ago2 binding, indicating that extracellular Ca 2ϩ is the primary source of Ca 2ϩ that disrupts PICK1-Ago2 interactions during cLTD (Fig. 7A). We also confirmed that our stimulation protocol induced cLTD by analyzing GluA1 phosphorylation at serine 845. Dephosphorylation at this site by the Ca 2ϩ -dependent phosphatase calcineurin is a hallmark of LTD (21,22). GluA1 phosphorylation at serine 845 decreased in response to NMDAR stimulation in the presence of extracellular Ca 2ϩ but was unchanged in the absence of extracellular Ca 2ϩ (Fig. 7B). This demonstrated successful induction of cLTD in the presence of extracellular Ca 2ϩ .
To further define the dissociation of Ago2 from PICK1 in response to cLTD induction, we performed a time course experiment to analyze endogenous PICK1-Ago2 complexes at 0, 5, 10, and 30 min after NMDAR stimulation. Interestingly, a significant reduction in PICK1-Ago2 binding was observed immediately after stimulation with no further reduction being observed at the later time points (Fig. 8A). Serine 845 dephos- phorylation followed a similar time course (Fig. 8B). Together, these experiments suggest that extracellular Ca 2ϩ entry through NMDARs disrupts PICK1-Ago2 complexes very rapidly after stimulation.

PICK1 Ca 2؉ sensing is required for miRNA-mediated gene silencing caused by cLTD
We reported previously that cLTD enhances miRNA-mediated gene silencing of the dendritically localized miR134 in a PICK1-dependent manner (16), suggesting that PICK1 is an inhibitor of Ago2 function under basal conditions. To investigate whether the C-terminal Ca 2ϩ binding region of PICK1 is involved in the cLTD-induced increase in Ago2 function, we measured the activity of a number of miRNAs using luciferase reporter assays. The 3Ј UTRs of LIMK1, APT1, and LIN41 contain miRNA binding sites for miR134, miR138, and Let7, respectively (23)(24)(25). When these UTRs are fused to DNA encoding luciferase, its expression can be regulated in a miRNA-dependent manner. MiR134 and miR138 are expressed primarily in neuronal dendrites, whereas Let7 is ubiquitously expressed throughout neurons (23-26). PICK1 knockdown reduced luciferase expression from the LIMK1 reporter under basal conditions, which was rescued by co-expression of GFP-PICK1 WT and also by GFP-PICK1 Aϫ10 (Fig. 9A), indicating that the Aϫ10 mutation has no effect on miRNA activity under basal conditions. In response to cLTD induction, we observed a ϳ50% reduction in luciferase activity 10 min after stimulation, as reported previously (16). Gene silencing increased further at later time points, with ϳ70% reduction in luciferase activity at 30 min and ϳ90% reduction at 60 min after NMDAR stimulation in both control and GFP-PICK1 WT-expressing cells. The cLTD-induced increase in gene silencing was completely blocked by expression of GFP-PICK1 Aϫ10. The same findings were observed with the APT1 3Ј UTR luciferase reporter, indicating that miR138 activity is regulated by the C-terminal PICK1 Ca 2ϩ -binding region in a similar manner (Fig. 9C). Luciferase activity was not significantly affected under basal or cLTD conditions with any PICK1 knockdown or replacement background when the miR134-and miR138-binding sites were mutated in the LIMK1 and APT1 reporter constructs so that they could no longer bind miRNA (Fig. 9, B and D). PICK1 knockdown or molecular replacement A, GFP-trap was used to capture GFP-PICK1 WT and GFP-PICK1 Aϫ10 complexes before and after NMDAR-induced cLTD, and bound proteins were detected by Western blotting. B, cLTD induction caused a significant reduction in Ago2 binding to GFP-PICK WT, which was blocked by the Aϫ10 mutation. C, cLTD induction caused a significant increase in GluA2 binding to both GFP-PICK1 WT and GFP-PICK1 Aϫ10. n ϭ 3; ns, not significant; **, p Ͻ 0.01; ***, p Ͻ 0.001; two-way ANOVA, Bonferroni post hoc test.

Ago2-PICK1 regulation by Ca 2؉
with GFP-PICK1 A10 did not alter Let7 activity, as measured using the LIN41 luciferase reporter under basal or cLTD conditions (Fig. 9, E and F). Taken together, these experiments suggest that the C-terminal Ca 2ϩ -binding region of PICK1 regulates dendritic miRNA-mediated gene silencing events in response to cLTD induction.

Discussion
Here we have identified a novel molecular mechanism that regulates the interaction between PICK1 and the miRISC protein Ago2. We showed previously that PICK1 inhibits Ago2 function in neuronal dendrites and that, during cLTD, Ago2 is released from PICK1, causing an increase in miRNA-mediated translational repression (16). In this study, we show that the PICK1-Ago2 interaction is regulated directly by the ability of PICK1 to bind Ca 2ϩ ions. We demonstrate that amino acids 380 -390 in PICK1 are responsible for regulating the Ago2-PICK1 interaction in a Ca 2ϩ dependent manner and that mutating this C-terminal Ca 2ϩ -binding region of PICK1 blocks miR134-and miR138-dependent translational repression in response to cLTD induction.

Ca 2؉ sensitivity of the PICK1-Ago2 interaction
Our results show that the PICK1-Ago2 interaction is significantly reduced at high [ (14). This difference is likely to reflect the distinct Ca 2ϩ -binding sites involved in regulating GluA2 and Ago2 interactions and suggests that PICK1 mediates a variety of complex biochemical responses to changes in [Ca 2ϩ ] free . The intracellular [Ca 2ϩ ] reached during LTD induction is unclear. Our results show that PICK1 binding to Ago2 decreased and binding to GluA2 increased during cLTD induction in neurons. In binding experiments using purified components, we could Figure 6. The PICK1 A؋10 mutation blocks the cLTD-induced decrease in Ago2-PICK1 co-localization in neuronal dendrites. A, cLTD induction caused a reduction in co-localization between endogenous Ago2 and GFP-PICK1 WT, which was blocked by the Aϫ10 mutation. Arrows indicate overlapping foci. B, quantification of co-localization of GFP-PICK1 and Ago2 was determined by Mander's coefficients. n ϭ 5; ns, not significant; **, p Ͻ 0.01, two-way ANOVA, Bonferroni post hoc test.

Ago2-PICK1 regulation by Ca 2؉
mimic these effects at 16 M [Ca 2ϩ ] free , suggesting that the biochemical changes observed in vitro at 16 M [Ca 2ϩ ] free may occur in neurons during NMDAR-mediated cLTD. However, it is important to note that Ca 2ϩ buffering in our in vitro setting presumably does not mimic Ca 2ϩ buffering in dendritic spines. PICK1 contains two acidic regions that are capable of binding Ca 2ϩ ions (14). We show that, although the N-terminal acidic region is not involved in Ago2 regulation, mutating the C-terminal acidic region of PICK1 (380 -390) blocks dissociation from Ago2. A possible explanation for the effect of Ca 2ϩ is

Ago2-PICK1 regulation by Ca 2؉
that Ca 2ϩ neutralizes the negative charge of the PICK1 acidic region. This predicts that the neutral Aϫ10 mutant would mimic the Ca 2ϩ -bound state and have reduced affinity for PICK1, as observed for PICK1 WT, at high [Ca 2ϩ ] free . However, we suggest that, at 16 M [Ca 2ϩ ] free , the acidic motif still retains some negative charge, which may explain why Ago2-PICK1 WT binding at 16 M does not coincide with the PICK1 Aϫ10-Ago2 at zero [Ca 2ϩ ]. Moreover, Ago2 binding to Aϫ10 PICK1 was similar to PICK1 WT in neurons under basal conditions, and, in heterologous cells, we observed an enhanced association of Aϫ10 PICK1 with Ago2 compared with PICK1 WT. We speculate that this difference between heterologous cells and neurons under basal conditions is due to neurons expressing numerous PICK1 interactors such as GluA2 and other neuronspecific channels, transporters, etc. (27,28) that are not expressed in HEK293 cells. These proteins may influence the manner in which Ca 2ϩ regulates PICK1 binding to Ago2. Hence we propose that the optimal [Ca 2ϩ ] for promoting Ago2-PICK1 binding is slightly different in neurons compared with the heterologous system.

Ca 2؉ -sensitive PICK1-Ago2 binding in synaptic plasticity
We demonstrate a Ca 2ϩ -sensitive mechanism for regulating miRNA function in neurons. In response to cLTD, the activity of dendritically enriched miR134 is enhanced in neuronal dendrites and spines to promote silencing of LIMK1 (16,23). PICK1 Aϫ10 blocks the cLTD-induced increase in miR134 activity, suggesting that Ca 2ϩ releases the inhibitory effects of PICK1 on Ago2 during cLTD.
During NMDAR-mediated cLTD, a Ca 2ϩ gradient is likely to exist within dendritic spines, where the [Ca 2ϩ ] free is highest toward the surface of spines in microdomains close to Ca 2ϩ channels (29). This is where PICK1 is likely to have the strongest binding properties for GluA2 to facilitate its internalization (14). Recycling endosomes closely associated with dendritic spines close to synapses (30,31) might also be where Ago2 is released by PICK1 (Fig. 10). Furthermore, NMDAR activation at the cell surface has been demonstrated to provoke the release of Ca 2ϩ from Ryanodine receptor-associated Ca 2ϩ stores such as the endoplasmic reticulum, which are present within dendrites and spines (32,33) and are likely to contribute to increases in cytosolic [Ca 2ϩ ] to attenuate PICK1-Ago2 interactions and, consequently, elevate Ago2 activity. This type of Ca 2ϩ gradient-controlled miRNA-mediated gene silencing could allow neurons to control localized Ago2 activation upon stimulation and provide neurons with a unique mechanism to control localized protein synthesis. In this study, we were able to show that the original source of Ca 2ϩ was extracellular because removal of Ca 2ϩ during cLTD induction blocked the dissociation of PICK1 from Ago2.
Our results show that PICK1-mediated Ca 2ϩ signaling during cLTD is responsible for increasing miRNA-mediated gene silencing of miR134 and miR138 but not Let7. This suggests the Ago2 bound to PICK1 only regulates a subset of miRNA-mediated translational repression events within neurons. miR134 and miR138 are enriched within dendrites, whereas Let7 is ubiquitously expressed throughout neurons (23)(24)(25)(26). We propose that PICK1 specifically regulates the activity of dendritically localized miRNAs to mediate changes in protein expression required for synaptic plasticity. Measuring the activity of a wider range of miRNAs that are expressed in dendrites and/or the cell body will provide a better understanding of the role of PICK1 in miRNA-mediated gene silencing.
PICK1 knockdown mimics and occludes the cLTD-induced increase in miRNA activity at 10 min after stimulation, which is consistent with our observation that cLTD causes a very rapid dissociation of PICK1 from Ago2. However, our results show that cLTD causes a further increase in miRNA activity at later Figure 9. The PICK1 A؋10 mutation blocks the cLTD-induced increase in gene silencing. A, miR134-Limk1 3Ј UTR. A Dual-Luciferase assay was performed in neurons expressing Renilla and firefly luciferase reporter containing the Limk1 3Ј UTR. Replacement of endogenous PICK1 with GFP-PICK1 WT or GFP-PICK1 Aϫ10 has no effect on miR134 activity under basal conditions; however, GFP-PICK1 Ax10 blocks the cLTD-induced increase in miR134 activity at 10, 30, and 60 min after stimulation. B, manipulations have no effect on luciferase reporter expression when the miR134-binding site on the LIMK1 3Ј UTR is abolished by mutation. C, miR138-APT1 3Ј UTR. A Dual-Luciferase assay was performed in neurons expressing Renilla and firefly luciferase reporter containing APT1 3Ј UTR. Replacement of endogenous PICK1 with GFP-PICK1 WT or GFP-PICK1 Aϫ10 has no effect on miR138 activity under basal conditions; however, GFP-PICK1 Aϫ10 blocks the cLTD-induced increase in miR138 activity at 10, 30, and 60 min after stimulation in a similar manner as shown in A. D, manipulations have no effect on luciferase reporter expression when the miR138 binding site on the APT1 3Ј UTR is abolished by mutation. E and F, Let7-LIN41 3Ј UTR. A Dual-Luciferase assay was performed in neurons expressing Renilla and firefly luciferase reporter containing the LIN41 3Ј UTR. Let7 activity is unaffected by cLTD induction or manipulation of PICK1. n ϭ 3; ****, p Ͻ 0.0001, two-way ANOVA, Bonferroni post hoc test. ] is low in dendritic spines, Ago2 is recruited to recycling endosomes by the C-terminal domain of PICK1. Upon cLTD induction, localized intracellular [Ca 2ϩ ] increases, and Ca 2ϩ binding to PICK1 releases Ago2 and increases its interaction with GluA2-containing AMPARs to promote their internalization. The released Ago2 causes increased miRISC activity of miR134 on LIMK1 and miR138 on APT1.

Ago2-PICK1 regulation by Ca 2؉
time points, indicating that additional mechanisms independent of PICK1 are involved in increasing miRNA activity in response to cLTD induction. It is interesting to note that, although we observed an ϳ50% reduction in luciferase activity within 10 min of NMDAR activation, it has been suggested previously that the half-life of luciferase in HEK293 cells is ϳ1-2 h (34). Our results indicate that luciferase has a shorter half-life (in the order of 10 min) in neurons, highlighting luciferase assays as an ideal tool for measuring rapid, activity-dependent changes in expression.
Whether the modulation of miRNA activity by PICK1 is required for LTD expression remains an unanswered question. AMPAR endocytosis occurs a few minutes after NMDAR stimulation (35), too soon to be affected by miRNA activity. Later events in LTD expression include dendritic spine shrinkage, which may involve down-regulation of LIMK1 (36). AMPAR internalization in response to LTD induction depends on an NMDAR-dependent increase in PICK1-GluA2 binding (37), and our results show that this is unaffected by the Aϫ10 mutation. Therefore, in neurons expressing Aϫ10 PICK1, the direct effect of PICK1 on NMDAR-dependent AMPAR trafficking should be the same as in control or WT PICK1-expressing neurons. Further experiments are necessary to determine which aspects of LTD are affected by the PICK1-dependent regulation of miRNA activity.
In conclusion, we have identified a novel mechanism for the regulation of miRNA activity by PICK1 in neurons. To our knowledge, this is the first mechanism to explain how an increase in [Ca 2ϩ ] free following NMDAR activation during the induction of synaptic plasticity causes an increase in RISC activity. Future work will determine the role of this mechanism in various kinds of learning and memory processes and, possibly, also in neurological diseases that involve changes in synaptic function.

HEK293 cell and cortical neuronal cultures
HEK293 cells were cultured in complete DMEM (Gibco) and passaged at ϳ70 -80% confluency. HEK cells were transfected with Jetpei (PolyPlus) transfection reagent according to the instructions of the manufacturer. Rat embryonic cortical neuronal cultures were prepared from embryonic day 18 Wistar rats using standard procedures. The culture medium was Neurobasal medium (Gibco) supplemented with B27 (Gibco) and 2 mM Glutamax. Neurons were plated at densities of 100, 000 cells/well of a 24-well plate, 500,000 cells/well of a 6-well plate, and 900,000 cells/6-cm dish. Neurons were transfected with plasmid DNA at days in vitro (DIV) 10 -13 (unless otherwise stated) using Lipofectamine 2000 (Invitrogen) and used for experiments 2-5 days post-transfection. Chemical LTD was induced by bath application of 50 M NMDA plus 20 M glycine in HBS (140 mM NaCl, 5 mM KCl, 25 mM Hepes, 1.8 mM CaCl 2 , 0.8 mM MgCl 2 , and 10 mM glucose (pH 7.4)) for 3 min at 37°C and harvested or fixed 10, 30, or 60 min later for biochemistry, luciferase assays, or imaging. Where cLTD was induced in the absence of Ca 2ϩ , cortical cultures were washed twice with HBS containing zero CaCl 2 before bath application of 50 M NMDA in the same buffer.

Confocal microscopy and image analysis
Cells grown on coverslips were fixed in 4% paraformaldehyde (Thermo Fisher) in PBS (Sigma) supplemented with 2% sucrose at room temperature. Cells were permeabilized in 0.5% Nonidet P-40 for 2 min. Coverslips were blocked in 3% BSA (Sigma) for 1 h and incubated with anti-PICK1 (Abcam, ab3420; dilution, 1:100) in 3% BSA for 1 h at room temperature, followed by goat anti-rabbit Alexa Fluor 568 for 45 min (Thermo Fisher, A-11011; dilution, 1:1000) and mounted in DAPI-free medium (Sigma) onto slides. Coverslips were imaged on a Leica SP5 confocal system under a ϫ63/1.4 numerical aperture (NA) oil immersion or ϫ40/1.25 numerical aperture (NA) oil immersion objective using filters set up to image GFP, mRUBY, or Alexa Fluor 568. The Leica application suite software was used to acquire 0.37-m stepped Z stacks throughout the depth of the cells, and maximal intensity projections generated by ImageJ Fiji are presented in this work. Image processing and co-localization analyses were performed using ImageJ and Coloc2. Three randomly selected dendrites were analyzed per neuron. Mander's fractions were measured using thresholding, values were normalized for each of at least three independent experiments, and statistical significance was determined using two-way ANOVA. All the error bars on graphs represent standard deviation of the mean. Sholl analysis was performed by thresholding the mRUBY channels and using the Sholl plugin (ImageJ Fiji) to measure the number of dendritic intersections every 10 m over a total radius of 150 m from the cell body. The numbers of intersections were measured from three independent neuronal cultures. Total dendritic length was measured using the Simple Neurite Tracer plugin (ImageJ Fiji) to calculate the total dendrite length of a neuron based on filling neurons with mRUBY. Traces were taken from three independent experiments, and statistical significance was determined using one-way ANOVA.
Levels of PICK1 knockdown and molecular replacement were quantified by measuring integrated densities of PICK1 clusters detected using anti-PICK1 in cortical neuronal dendrites using ImageJ (Fiji).

Co-immunoprecipitations
Cells were lysed in IP buffer (0.5% Triton X-100, 150 mM NaCl, and 20 mM HEPES (pH 7.4)), and 1% of extract was removed for input. The remaining lysate was precleared with protein G-Sepharose beads (GE Healthcare) at 4°C for 1 h. 400 g of cell lysates was incubated with 2 g of anti-PICK1 (Neu-roMab clone L20/8), anti-FLAG (Sigma, F3165), or control IgG (Thermo Fisher, 31903) antibodies and pulled down with protein G-Sepharose beads (GE Healthcare) at 4°C for 1 h. Beads were washed three times (1 min each) in 1 ml of IP buffer at 4°C. Bound proteins were detected by Western blotting. For co-IPs carried out under defined [Ca 2ϩ ] free , 8 -10 6-cm dishes of cortical neuronal cultures were pooled together in 1 ml of co-IP buffer with zero Ca 2ϩ . 1% of this pool was taken as input, and the remaining extract was equally divided and diluted into 5 ml of co-IP buffer with the appropriate [Ca 2ϩ ] total and [EGTA] to give the stated [Ca 2ϩ ] free . Therefore, these experiments share the same input. Defined [Ca 2ϩ ] free were determined using the Max chelator software (Stanford).

GST and His 6 pulldowns
Recombinant proteins were expressed and purified from BL21 bacterial cultures. 1 liter of bacterial culture was induced for 2 h using 0.2 mM isopropyl 1-thio-␤-D-galactopyranoside at 30°C. The bacteria were then pelleted and resuspended in 50 ml of HTG buffer (50 mM HEPES, 150 mM NaCl, and 10% glycerol). Bacteria were sonicated on ice, and debris was removed by centrifugation. The resulting supernatant was incubated with glutathione-Sepharose 4B beads (GE Healthcare) or nickel-nitrilotriacetic acid beads (Qiagen) for 1 h at 4°C. The beads were washed with 1 ml of HTG buffer three times (1 min each). 8 -10 6-cm dishes of cortical neuronal cultures or four T75 flasks of transfected HEK293 cells were pooled together in 1 ml of co-IP buffer with zero Ca 2ϩ . 1% of this pool was taken as input, and the remaining extract was equally divided and diluted into 5 ml of co-IP buffer with the appropriate [Ca 2ϩ ] total and [EGTA] to give the stated [Ca 2ϩ ] free . This extract was then incubated with recombinant protein immobilized on beads for 1 h at 4°C. This was followed by washing the beads three times (1 min each) in 1 ml of the appropriate [Ca 2ϩ ] free co-IP buffers. Bound proteins were detected by Western blotting.

GFP-trap
For GFP-trap (Chromotek) pulldowns, six wells of a 6-well plate of DIV10 cortical neurons were transfected with pXLG3-GFP PICK1 constructs and harvested 5 days post-transfection. GFP-trap was performed according to the instructions of the manufacturer. Briefly, cells were lysed in 500 l of lysis buffer (10 mM Tris (pH 7.5), 150 mM NaCl, 0.5 mM EDTA, and 0.5% Nonidet P-40) on ice. 1% was taken as input, and the remaining extract was incubated with 50 l of GFP-trap beads for 1 h at 4°C. The beads were then washed with 500 l of wash buffer (10 mM Tris (pH 7.5), 150 mM NaCl, and 0.5 mM EDTA) three times (1 min each) at 4°C, and bound proteins were detected by Western blotting.

Luciferase assays
DIV12 cortical cultures were co-transfected with the appropriate luciferase constructs and pXLG3-GFP vectors and prepared for luciferase recordings 48 h post-transfection. The Dual-Luciferase reporter assay system (Promega) was used to perform the assays according to the instructions of the manufacturer. Values were normalized for each of at least three independent experiments, and the appropriate statistical analysis was performed. All error bars on graphs represent standard deviation of the mean.