Capping of the N-terminus of PSD-95 by calmodulin triggers its postsynaptic release



Postsynaptic density protein-95 (PSD-95) is a central element of the postsynaptic architecture of glutamatergic synapses. PSD-95 mediates postsynaptic localization of AMPA receptors and NMDA receptors and plays an important role in synaptic plasticity. PSD-95 is released from postsynaptic membranes in response to Ca2+ influx via NMDA receptors. Here, we show that Ca2+/calmodulin (CaM) binds at the N-terminus of PSD-95. Our NMR structure reveals that both lobes of CaM collapse onto a helical structure of PSD-95 formed at its N-terminus (residues 1–16). This N-terminal capping of PSD-95 by CaM blocks palmitoylation of C3 and C5, which is required for postsynaptic PSD-95 targeting and the binding of CDKL5, a kinase important for synapse stability. CaM forms extensive hydrophobic contacts with Y12 of PSD-95. The PSD-95 mutant Y12E strongly impairs binding to CaM and Ca2+-induced release of PSD-95 from the postsynaptic membrane in dendritic spines. Our data indicate that CaM binding to PSD-95 serves to block palmitoylation of PSD-95, which in turn promotes Ca2+-induced dissociation of PSD-95 from the postsynaptic membrane.



Ca2+ influx promotes Ca2+/calmodulin binding to the N-terminus of PSD-95, which blocks PSD-95 palmitoylation leading to reduced retention of PSD-95 at synapses. This effect will likely decrease postsynaptic glutamate receptor content and thereby synaptic strength.

  • Ca2+/calmodulin forms a collapsed structure around the N-terminal helix of PSD-95 that sequesters the palmitoylation sites (Cys3 and Cys5) and a key tyrosine (Tyr12).
  • Binding of Ca2+/calmodulin to the N-terminus of PSD-95 decreases its palmitoylation to release PSD-95 from postsynaptic sites
  • Binding of Ca2+/calmodulin also displaces the serine/threonine kinase CDKL5 from PSD-95, which otherwise helps augment synaptic strength
  • A point mutation of PSD-95 that prevents Ca2+/calmodulin binding turns the Ca2+-induced reduction in PSD-95 at synapses into an increase, uncovering the existence of a second mechanism that augments postsynaptic PSD-95 enrichment upon Ca2+ influx.


The vast majority of synapses in the brain use glutamate as neurotransmitter (Micheva et al, 2010). It activates postsynaptic AMPA-type glutamate receptors (AMPARs) that are precisely juxtaposed to its presynaptic release site. Periods of heightened synaptic activity lead to activation of NMDA-type glutamate receptors (NMDARs). NMDARs mediate Ca2+ influx that can permanently increase the postsynaptic strength causing long-term potentiation (LTP; Lisman & Hell, 2008; Nicoll & Roche, 2013), which is most likely the physiological basis of learning and memory (Morris, 2013).

A critical player in postsynaptic AMPAR targeting is PSD-95 (El-Husseini et al, 2000, 2002; Elias et al, 2006; Schluter et al, 2006). PSD-95 consists of three PDZ domains followed by an SH3 domain and a guanylate kinase homology domain (GK). PSD-95 anchors AMPARs at postsynaptic sites by binding with its first and second PDZ domains to the intracellular C-termini of auxiliary subunits known as transmembrane AMPAR regulatory proteins (TARPs), which include stargazin (Stg, γ2) and its homologs γ3, γ4, and γ8 (Chen et al, 2000; Schnell et al, 2002; Opazo et al, 2010). Furthermore, PSD-95 is palmitoylated at two cysteine residues (C3 and C5) at its N-terminus, which is required for its postsynaptic localization and for postsynaptic targeting of AMPARs (El-Husseini et al, 2002). Ca2+ influx through NMDARs causes PSD-95 to be temporarily released from postsynaptic membranes (Sturgill et al, 2009; Nelson et al, 2013). This release seems to reflect an important step for postsynaptic restructuring leading to enlargement of the postsynaptic site and thereby for an increase in postsynaptic AMPARs during LTP (Sturgill et al, 2009). Ultimately, PSD-95 augments postsynaptic AMPAR localization upon Ca2+ influx during LTP in part due to stimulation of phosphorylation of the C-terminus of Stg/γ2 upon NMDAR-mediate Ca2 influx by the Ca2/calmodulin (CaM)-dependent kinase CaMKII. This phosphorylation fosters membrane detachment of the C-terminus of Stg/γ2 and thereby binding of the very C-terminus of Stg/γ2 to PSD-95 (Tomita et al, 2005; Sumioka et al, 2010). The initial mobilization of PSD-95 upon Ca2 influx could either be necessary for structural plasticity to occur or constitute a negative feedback mechanism to curb potentiation.

Ca2+/CaM interacts with PSD-95 and its homologs SAP97 and SAP102 in a Ca2-dependent manner (Masuko et al, 1999; Paarmann et al, 2002; Fukunaga et al, 2005) and stimulates PSD-95 binding to the non-receptor tyrosine kinase Pyk2 to augment NMDAR activity and thereby LTP (Bartos et al, 2010). Previous studies focused on the HOOK region at the C-terminal end of the SH3 domain of PSD-95, SAP97, and SAP102 as the CaM-binding site. We now report that Ca2+/CaM binds not only to the SH3 domain but also to the N-terminal region of PSD-95 as well as the second PDZ domain (PDZ2, residues 155–249). We present the NMR structure of Ca2+/CaM bound to the N-terminal region of a PSD-95 fragment [residues 1–71, called PSD-95(1–71)]. Both CaM lobes collapse around an N-terminal helix of PSD-95 (residues 1–16) and sequester two palmitoylation sites (C3 and C5) that mediate postsynaptic targeting of PSD-95. CaM forms extensive contacts with Y12 in PSD-95, and mutating Y12 to Glu prevents binding of CaM to PSD-95 upon Ca2+ influx. Importantly, although this mutation does not affect PSD-95 palmitoylation under basal conditions, it counteracts the reduction in palmitoylation of PSD-95 following Ca2+ influx and the Ca2+-induced release of PSD-95 from postsynaptic membranes. It also blocks displacement of CDKL5, which binds to the N-terminus of palmitoylated PSD-95 (Zhu et al, 2013) and can strengthen postsynaptic responses (Ricciardi et al, 2012; Zhu et al, 2013). Our results identify the N-terminal CaM-binding site of PSD-95 as a critical regulator of the dynamic postsynaptic localization of PSD-95 and thereby of postsynaptic plasticity.


CaM binds to the N-terminus of PSD-95

To characterize binding of CaM to PSD-95 beyond the established interaction between CaM and the SH3 domain of PSD-95, pull-down assays of CaM were performed with the various GST-tagged PSD-95 fragments (Fig 1A). Accordingly, Ca2+/CaM interacted with the N-terminal region of PSD-95 (residues 1–71 preceding PDZ1), PDZ2, and SH3 but not PDZ3, GK, or GST alone (Fig 1B and D). All samples contained comparable amounts of the respective full-length fusion protein with little to no degradation (Fig 1C). To define the CaM-binding site at the PSD-95 N-terminus more precisely, CaM binding to fluorescein-labeled peptides of residues 1–13, 15–36, 37–52, and 49–65 was monitored by fluorescence polarization (FP; Fig 2A and B). FP of the peptide spanning the first 13 residues showed saturable binding upon adding increasing amounts of CaM. Accordingly, this peptide binds to CaM with a dissociation constant of 18 ± 0.3 μM (Fig 2B, black circles). None of the other peptides (collectively covering residues 15–65) bound CaM at protein concentrations up to 100 μM (Fig 2B). These results indicate that CaM exhibits Ca2+-induced binding to the first 13 residues of PSD-95.

Figure 1.

Figure 1. Ca2+/calmodulin (CaM) binds to the N-terminus, PDZ2, and SH3 of PSD-95

  1. Schematic diagram of PSD-95 domains depicting the N-terminal region (residues 1–64), PDZ1–3, SH3, and GK.
  2. Pull-down of Ca2+-bound CaM with the various PSD-95 domains.
  3. Probing for GST to visualize fusion proteins on a blot run in parallel to the pull-down blot.
  4. Relative band intensities from (B) are shown for each PSD-95 domain as determined by densitometry from 11 independent experiments.
Figure 2.

Figure 2. N-terminal 13 residues of PSD-95 bind to Ca2+/CaM

  1. Amino acid sequence of the N-terminal 65 residues of PSD-95. Rectangular boxes underneath define the length of synthetic peptide fragments used in binding assays with Ca2+/CaM. Palmitoylation sites are highlighted in orange.
  2. Fluorescence polarization (FP; see 'Materials and Methods') from each synthetic peptide covalently labeled with fluorescein is plotted as a function of CaM concentration. The increase in FP from the N-terminal peptide (residues 1–13) as a function of increasing CaM concentration indicates binding. The Kd for binding of Ca2+/CaM to the N-terminal peptide is calculated to be 18 μM. None of the other peptides showed any increase in FP as a function of increasing CaM concentration (up to 100 μM), indicating that these peptides do not bind to Ca2+/CaM.
  3. Peptide alanine scanning spot array of N-terminal PSD-95 peptide (residues 1–13) for Ca2+/CaM binding. Mutagenesis of PSD-95 residues highlighted in orange show the largest effect on CaM binding.
  4. Ca2+/CaM binding to mutant N-terminal PSD-95 peptides (residues 1–13; WT titration is same as in B).
  5. Binding of CaM mutants to N-terminal PSD-95 peptide (residues 1–13; WT titration is same as in B).

CaM forms a collapsed structure around the N-terminal helix in PSD-95

NMR spectroscopy was used to characterize the structural interaction of CaM bound to PSD-95(1–71). The 15N-1H HSQC NMR spectrum of 15N-labeled PSD-95(1–71) in the absence of CaM exhibits poor chemical shift dispersion, indicative of an unstructured and random coil conformation (Fig 3A). The NMR assignments for PSD-95(1–71) were determined as shown in Fig 3A. The addition of saturating CaM causes the PSD-95 NMR peaks assigned to residues 1–16 to broaden significantly, whereas the NMR peaks assigned to residues 17–71 were unaffected by CaM. Thus, the CaM-binding site on PSD-95(1–71) is localized within the first 16 residues from the N-terminus, consistent with the results in Fig 2.

Figure 3.

Figure 3. NMR analysis of Ca2+/CaM binding to PSD-95 (1–71)

  1. Two-dimensional 15N-1H HSQC spectra of 15N-labeled PSD-95(1–71) in the absence (red) and presence (green) of unlabeled Ca2+/CaM. The NMR resonances assigned to the N-terminal 16 residues in PSD-95 show as much as 100-fold decrease in peak height caused by CaM binding. The NMR assignments have been deposited in the BMRB (Accession Number 19238). The decreased NMR intensity is mostly due to chemical shift exchange broadening at these sites caused by CaM binding that has exchange kinetics on the chemical shift time scale. NMR signals assigned to PSD-95 residues 17–71 are unaffected by CaM binding. NMR resonance assigned to non-native residue (S0) upstream of the N-terminal Met is marked by an asterisk.
  2. 15N-1H HSQC spectra of 15N-labeled Ca2+/CaM in the absence (red) and presence (green) of unlabeled PSD-95(1–71). CaM residues that show the largest spectral changes caused by binding to PSD-95(1–71) are indicated by residue labels and are listed in Supplementary Table S1.

The 15N-1H HSQC NMR spectrum of 15N-labeled Ca2+-free CaM does not change upon adding a 10-fold excess of PSD-95(1–71) (not shown), consistent with a lack of PSD-95 binding to apo-CaM. By contrast, the 15N-1H HSQC NMR spectrum of 15N-labeled Ca2+/CaM changes significantly upon adding a stochiometric amount of PSD-95(1–71) (Fig 3B). The spectral changes saturate after adding one equivalent of PSD-95(1–71), indicating a 1:1 binding stoichiometry. CaM residues in the N- and C-lobe exhibit amide NMR peaks that either broaden or change chemical shift upon adding PSD-95(1–71) (Supplementary Table S1), suggesting the respective CaM residues are at or near the PSD-95(1–71) binding site. Most of these residues are clustered in exposed hydrophobic patches on both CaM lobes. However, some CaM residues have NMR signals that are not affected by PSD-95(1–71), including residues in the EF-hand Ca2+-binding loops (G25, G61, G98, G134) and polar surface of the CaM lobes (E7, K13, K30, E114). The lack of chemical shift changes to these residues suggests that the internal main chain structure within each lobe does not change much upon binding to PSD-95(1–71) and target binding is localized mainly to the exposed hydrophobic sites on both CaM lobes.

A detailed protocol for determining the NMR structure of the CaM/PSD-95(1–71) complex is summarized in Supplementary Methods. Residual dipolar couplings (RDCs, Supplementary Fig S1 and Table S2; Tjandra & Bax, 1997) and chemical shift perturbation data (Supplementary Table S1) provided structural restraints for a molecular docking calculation using the software HADDOCK (de Vries et al, 2010). Ten lowest energy structures of the CaM/PSD-95 complex are superimposed in Fig 4A and have a main chain RMSD of 0.69 Ǻ (structural statistics in Table 1). The energy-minimized average main chain structure is shown in Fig 4B.

Table 1. Structure statistics for NMR-derived structures of Ca2+/calmodulin (CaM)/PSD95-NT complex
  1. a

    Q-factor = RMS(Dcalc − Dobs)/RMS(Dobs), where Dcalc and Dobs are calculated and observed residual dipolar couplings (RDC) values, respectively.

NMR restraints
Short-range NOEs12
Dihedral angles46
Hydrogen bonds34
Ambiguous interaction restraints (AIR)69
RDC Q-factora0.11
Ramachandran plot
Most favored region (%)86.3
Allowed region (%)13.7
Disallowed region (%)0.0
RMSD from average structure (Å)
CaM backbone atoms0.59
PSD95 (1–19) backbone atoms0.70
All backbone atoms0.69
Figure 4.

Figure 4. Ca2+/CaM forms a collapsed structure around the N-terminal helix of PSD-95

  • ASuperposition of main chain atoms of 10 lowest energy NMR-derived structures of CaM (red) bound to the N-terminal helix in PSD-95 (cyan). Bound Ca2+ is depicted in yellow. The atomic coordinates have been deposited into the Protein Databank (2MES).
  • BRibbon representation of the energy-minimized average main chain structure of CaM (N-lobe: pink, C-lobe: red) bound to the N-terminal helix of PSD-95 (cyan). The four EF-hands in CaM are labeled EF1, EF2, EF3, and EF4. Side-chain atoms of CaM residues (L39 and V108) show contact between N-lobe (pink) and C-lobe (red).
  • C, DClose-up view of exposed side-chain atoms in CaM (red) interacting with side-chain atoms from PSD-95 residues C3, C5, I6, Y12 and R13 (C) or L4, V7 and K10 (D). Salt bridges to PSD-95 K10 and R13 are depicted by dashed lines.

The NMR-derived structure of the CaM/PSD-95(1–71) complex reveals that both lobes of CaM collapse around the N-terminal helix of PSD-95 (residues 1–16) (Fig 4B). The helical structure of the first 16 residues of PSD-95 in the complex was confirmed by circular dichroism (Supplementary Fig S2). Several hydrophobic residues in PSD-95 including L4, I6, V7, and Y12 interact with exposed hydrophobic residues of CaM (Fig 4C and D). Most striking are the two palmitoylation sites (C3 and C5; El-Husseini et al, 2002, 2000) in PSD-95 that are completely sequestered by hydrophobic residues in the N-lobe of CaM (F19, L32 and L39). Also noteworthy is that PSD-95 Y12 fits into a cavity formed by CaM residues in the C-lobe (M124, E127, V136, F141, and M144). In addition to hydrophobic contacts, PSD-95 K10 forms a salt bridge with CaM E11, which might play a role in specifying the orientation of the PSD-95 helix bound to CaM, and PSD-95 R13 forms salt bridges with E114 and E120 of CaM. The two lobes of CaM in the complex form interdomain contacts between L39 and V108 (Fig 4B, N-lobe: pink and C-lobe: red), which stabilize a collapsed structure of CaM as seen in previous structures of CaM bound to various target proteins (Hoeflich & Ikura, 2002). The EF-hand interhelical angles for CaM bound to PSD-95 are typical for Ca2+-bound EF-hands in the familiar open conformation (87˚ from EF1, 106˚ for EF2, 91˚ for EF3 and 97˚ for EF4).

Our structural model of CaM bound to the PSD-95 N-terminus reveals that the two palmitoylation sites in PSD-95 (C3 and C5) are almost completely buried by CaM and therefore not accessible. To test whether CaM blocks the accessibility of C3 and C5, we monitored whether CaM can inhibit the covalent attachment of a fluorescent dye (fluorescein-5-maleimide) at both C3 and C5 in PSD-95(1–71). This modification should mimic the attachment of palmitate at these sites. Fluorescein-5-maleimide is a thiol-reactive molecular probe that should specifically attach to both C3 and C5 in PSD-95(1–71). The dye was added to PSD-95(1–71) with no CaM, Ca2+/CaM, or apo-CaM. After removing unreacted dye, fluorescence emission spectra were recorded for each sample (Supplementary Fig S3A). Fluorescence was 50% less following incubation in the presence of Ca2+/CaM compared with no CaM or apo-CaM. These results indicate that Ca2+/CaM decreases the efficiency with which the fluorescein dye is attached to PSD-95 likely by blocking the accessibility of the two cysteines. PSD-95 will spend some fraction of time free of CaM even in the presence of stoichiometric amounts of Ca2+/CaM, which would allow fluorescein attachment of PSD-95 to build up over time and could explain partial rather than full block of fluorescein attachment.

To further demonstrate that Ca2+/CaM binding and palmitoylation of C3 and C5 in PSD-95 are mutually exclusive, we examined whether the attachment of fluorescein at C3 and C5 might prevent binding of Ca2+/CaM. Indeed, the 15N-1H HSQC NMR spectrum of 15N-labeled CaM does not substantially change upon adding a stoichiometric amount of fluorescein-labeled PSD-95(1–71) (Supplementary Fig S3B) in contrast to the dramatic change upon addition of a stoichiometric amount of unconjugated PSD-95(1–71) (Fig 3B). These results show that Ca2+/CaM binds preferentially to unconjugated PSD-95(1–71), consistent with binding of Ca2+/CaM to depalmitoylated but not palmitoylated PSD-95.

Mutagenesis of CaM/PSD-95 structural interface weakens binding

To validate the NMR structural model of the CaM/PSD-95 complex, Ala scanning was performed through the peptide that covers the first 13 N-terminal residues of PSD-95. Peptides were spot synthesized on a membrane and probed by overlay with Ca2+/CaM, which was detected with an anti-CaM antibody (Fig 2C). Substitution of PSD-95 residues Y12 and R13 by Ala either individually or combined completely abolished binding to CaM in this assay, consistent with these residues making multiple contacts with CaM as seen in our structure (Fig 4C and D). Substitution of L4 with Ala also decreased the binding affinity to CaM at least when combined with two additional Ala substitutions (Fig 2C), consistent with L4 making hydrophobic contacts with CaM (Fig 4D). However, triple Ala mutations of C5, I6 and V7 that form contacts with CaM did not substantially affect CaM binding (Fig 2C), perhaps because Ala is similar enough to these residues to retain binding. Instead of using Ala mutations, the PSD-95 residues (C3, C5, I6 and V7) were replaced with the more polar Glu for more quantitative analysis by FP. As expected, all of the Glu mutants (C3E, L4E, C5E, I6E and V7E) weaken CaM binding to PSD-95 (Fig 2D; Supplementary Table S3), consistent with each of these residues making hydrophobic contacts with CaM (Fig 4C and D).

The CaM/PSD-95 structure was further validated by mutations in CaM. Mutagenesis of CaM residue L39 to Gly at the lobe interface (L39G) weakens binding to PSD-95 (Fig 2F; Supplementary Table S4). L39 in the CaM N-lobe forms hydrophobic contacts with V108 from the C-lobe that stabilizes interaction between the two lobes in the collapsed CaM/PSD-95 structure (Fig 4B). The L39G mutation abolishes the hydrophobic interaction with V108, which weakens contact between the two lobes and destabilizes the collapsing of the two lobes around the N-terminus of PSD-95. Thus, a cooperative interaction of both CaM lobes is necessary to create proper binding to PSD-95 and explains why residues from both the N-lobe and C-lobe participate in binding to PSD-95 (Fig 4C and D). The cooperativity of the lobes is also consistent with the observation that the individual CaM lobes (N-lobe: residues 1–77 and C-lobe: residues 82–148), which are structured (Zhang et al, 2012), cause no detectable change in the NMR spectrum of PSD-95(1–71) (Supplementary Fig S4), in contrast to the large spectral change induced by full-length CaM.

PSD-95 residue Y12 forms extensive contacts with CaM (Fig 4C). The Y12E mutation disrupts hydrophobic interactions with CaM residues (M124, V136, F141, M144) and weakens the binding affinity by approximately 37-fold (Fig 2D; Supplementary Table S3). In addition to the hydrophobic contacts, the phenolic side chain of Y12 forms a hydrogen bond with the side-chain carboxyl group of CaM residue E127. The E127A mutation in CaM eliminates this hydrogen bond and weakens the binding affinity by 2.5-fold (Fig 2E; Supplementary Table S4), which in terms of free energy is consistent with the loss of a hydrogen bond, further confirming the CaM/PSD-95 structure.

N-terminal Ca2+/CaM binding impairs PSD-95 palmitoylation and CDKL5 binding

To test in intact cells whether Ca2+ influx negatively controls palmitoylation of PSD-95, HEK293 cells were transfected with PSD-95WT and PSD-95Y12E. Y12 makes multiple contacts with Ca2+/CaM, and changing it to Glu strongly impairs binding of Ca2+/CaM (Fig 2D and E). At the same time, this mutation should not affect basal palmitoylation of PSD-95 per se, because Y12 is not located in the palmitoyltransferase binding site (PSD-95 residues 3–9; El-Husseini et al, 2000; Fukata et al, 2004).

Under basal conditions, PSD-95WT and PSD-95Y12E showed comparable levels of palmitoylation in HEK293 cells (Fig 5A and B). Normal basal PSD-95Y12E palmitoylation is also indicated by clustering of the ectopically expressed PSD-95WT and PSD-95Y12E when expressed in COS cells (Supplementary Fig S5), which requires palmitoylation (Hsueh & Sheng, 1999; El-Husseini et al, 2002; Fukata et al, 2013). Ca2+ influx mediated by the Ca2+ ionophore ionomycin induced a strong loss of palmitoylation of PSD-95WT but not PSD-95Y12E (Fig 5A), indicating that binding of Ca2+/CaM to the N-terminus of PSD-95 shifts PSD-95 toward its depalmitoylated state.

Figure 5.

Figure 5. Ca2+/CaM antagonizes palmitoylation of PSD-95 and CDKL5 binding in HEK293 cells

  • A, BHEK293T cells were transfected with RFP-PSD-95WT or RFP-PSD-95Y12E. After 24 h, cells were treated with vehicle (CT) or 5 μM ionomycin for 5 min and extracted. NEM was added to alkylate free SH groups on Cys. Palmitoylated Cys were deacylated with hydroxylamine (NH2OH), and newly available SH groups biotinylated for pull-down with streptavidin, followed by immunoblotting for PSD-95. Omission of NH2OH prevented pull-down (left lane) indicating the specificity of this procedure. Ionomycin-induced Ca2+ influx decreased palmitoylation of PSD-95WT but not PSD-95Y12E. Total levels of PSD-95WT and PSD-95Y12E were comparable as determined just before pull-down (Input).
  • C–EHEK293T cells were transfected with RFP-PSD-95WT or RFP-PSD-95Y12E plus CDKL5 for treatment as above, immunoprecipitation of PSD-95, and immunoblotting as indicated. Ionomycin-induced Ca2+ influx nearly abolished co-immunoprecipitation of CDKL5 with PSD-95WT but not PSD-95Y12E (C, E). Pre-treatment with CMZ (10 min, 30 μM) prevented loss of CDKL5 co-immunoprecipitation. Lack of PSD-95 and CDKL5 immunoprecipitation by control IgG shows specificity of their co-precipitation (C, left lane). Total expression of PSD-95WT and PSD-95Y12E and of CDKL5 in the various conditions was comparable (D; tubulin served as a loading control). Quantification (E) shows the ratio of CDKL5 co-precipitation versus PSD-95 content in each sample.

Data information: In (B) and (E), data are given as means ± s.e.m., ***< 0.001, one-way ANOVA with Dunnett's post-test.

The cyclin-dependent protein kinase-like kinase CDKL5 binds to the first 19 residues of PSD-95 but only if PSD-95 is palmitoylated (Zhu et al, 2013). Because postsynaptic targeting of CDKL5 by this interaction augments postsynaptic strength (Zhu et al, 2013; see also Ricciardi et al, 2012), a functionally important question is whether this interaction is downregulated by Ca2+/CaM. HEK293 cells were co-transfected with PSD-95 and CDKL5 before treatment with ionomycin and co-immunoprecipitation of CDKL5 with PSD-95. As expected from its negative effect on palmitoylation, ionomycin disrupted the co-precipitation of CDKL5 with WT PSD-95 nearly completely (Fig 5C–E). This effect depended on Ca2+/CaM binding to PSD-95 as PSD-95Y12E did not show any loss of co-precipitation upon Ca2+ influx and the CaM antagonist calmidazolium (CMZ) prevented the Ca2+-induced loss of co-precipitation of WT PSD-95.

To test whether Ca2+/CaM affects PSD-95 palmitoylation in intact neurons, acute forebrain slices were treated with 100 μM NMDA for 5 min. This treatment triggers postsynaptic Ca2+ influx that stimulates recruitment of CaMKII to NMDARs and subsequent phosphorylation of the AMPAR GluA1 subunit on S831 (Leonard et al, 1999 #551; Halt et al, 2012 #550). Postsynaptic Ca2+ influx had been shown earlier to reduce PSD-95 palmitoylation (El-Husseini et al, 2002). Accordingly, NMDA treatment reduced PSD-95 palmitoylation (Fig 6A and B). This effect is selective for palmitoylation of PSD-95 as palmitoylation of Gαi3 was unaffected (Fig 6A). NMDA induced in parallel association of Ca2+/CaM with PSD-95 and displacement of CDKL5 from PSD-95 (Fig 6C–E). Importantly, three different CaM antagonists [CMZ, trifluoperazine (TFP), W7] prevented the loss of palmitoylation and of CDKL5 co-precipitation and the increase in CaM co-precipitation. We conclude that NMDAR-mediated Ca2+ influx induces CaM binding to PSD-95, thereby reducing both, palmitoylation of PSD-95 and CDKL5 displacement in neurons.

Figure 6.

Figure 6. Ca2+/CaM antagonizes palmitoylation of PSD-95 and CDKL5 binding in neurons

Acute forebrain slices were treated with vehicle (CT) or 100 μM NMDA for 5 min before lysis.

  • A, BNMDA-induced Ca2+ influx decreased palmitoylation of PSD-95 (reduced pull-down by Streptavidin after biotin switch) without affecting total PSD-95 levels (A, C, middle blots). This decrease was blocked by pre-treatment with the CaM antagonists calmidazolium (CMZ; 30 μM), W7 (20 μM), and trifluoperazine (TFP; 20 μM). Note that palmitoylation of the trimeric G protein subunit Gαi3 was not affected by Ca2+ influx showing specificity of the effect on PSD-95 and serving as an additional internal loading control.
  • C–ECo-immunoprecipitation of CaM with PSD-95 was increased and of CDKL5 decreased upon NMDA-induced Ca2+ influx. Both effects were blocked by pre-treatment with CMZ, W7, and TFP. Quantification shows ratio of CDKL5 and CaM co-precipitation versus PSD-95 content in each precipitation.

Data information: In (B), (D) and (E) data are given as means ± s.e.m., *< 0.05, **< 0.01, ***< 0.001, one-way ANOVA with Dunnett's post-test.

Ca2+ influx stimulates loss of WT but not Y12E PSD-95 from spines

Diffusion of PSD-95 out of spines is limited under basal conditions (~15% over 30 min; Steiner et al, 2008). Ca2+ influx increases the rate of this diffusion of PSD-95 by 2–3-fold (Steiner et al, 2008; Sturgill et al, 2009; Nelson et al, 2013). This activity-triggered increase in PSD-95 mobility is important for LTP (Steiner et al, 2008) and LTD (Nelson et al, 2013). Postsynaptic PSD-95 localization requires palmitoylation of C3 and C5 (El-Husseini et al, 2000, 2002; Sturgill et al, 2009; Fukata et al, 2013). Thus, we hypothesized that Ca2+ triggered binding of CaM to the very N-terminus of PSD-95, and the consequent shift toward depalmitoylated PSD-95 is a major factor in the Ca2+-induced destabilization of postsynaptic PSD-95 localization. We expressed PSD-95WT and PSD-95Y12E with a RFP tag at the C-terminus in cultures of dissociated hippocampal neurons. Brief treatment with glutamate to stimulate Ca2+ influx via NMDARs triggered a reduction in the spine-to-shaft ratio for RFP-PSD-95 by about 20% over a 15-min washout period (Fig 7). This reduction reflected a decrease in PSD-95 concentration as it was normalized to changes in spine volume, which was determined by the spine-to-shaft ratio of GFP. Strikingly, PSD-95Y12E not only showed no such reduction, it actually exhibited a remarkably strong increase after 5 and 15 min of glutamate treatment. Again, this increase reflects an increase in concentration as it was normalized to changes in spine size (which were minimal; Fig 7D). We conclude that CaM binding the N-terminus of PSD-95 curbs its postsynaptic accumulation upon Ca2+ influx.

Figure 7.

Figure 7. Ca2+/CaM binding to the N-terminus of PSD-95 is important for loss of PSD-95 from spines upon Ca2+ influx

  • A, BHippocampal cultures were transfected at 4 DIV with plain GFP as volume marker and PSD-95WT (A) or PSD-95Y12E (B) carrying TagRFP at their C-termini. At 18 DIV, cultures were treated for 2 min with vehicle or glutamate (100 μM) and the NMDA-type glutamate receptor co-agonist glycine (10 μM). Cells were fixed 5, 15, and 30 min after treatment.
  • CCo-localized GFP and RFP puncta were normalized to the respective intensity in the dendritic shaft to account for differences in expression levels. Synaptic enrichment was assessed as ratio of RFP to GFP for every synaptic area (assessed by RFP fluorescence). All values were normalized to unstimulated cells. The time courses of the change in synaptic enrichment of PSD-95WT (t = 0: 1.00 ± 0.06; t = 5: 0.95 ± 0.05; t = 15: 0.81 ± 0.07, *< 0.05; t = 30: 0.90 ± 0.06; one-way ANOVA with Bonferroni's multiple comparison test) and PSD-95Y12E (t = 0: 1.00 ± 0.07; t = 5: 1.47 ± 0.09, **< 0.01; t = 15: 1.26 ± 0.11; t = 30: 0.90 ± 0.11; one-way ANOVA with Bonferroni's multiple comparison test) are significantly different after 5 min (two-way ANOVA, Bonferroni's post-test, F1,265 = 14.62, ***< 0.001) and 15 min (**< 0.01) of washout (values are mean ± s.e.m., n = 17–60 cells from four independent experiments).
  • DAs a proxy for synapse size, the area of individual RFP puncta was compared between PSD-95WT (t = 0: 4.51 ± 0.12 μm; t = 5: 4.40 ± 0.08; t = 15: 4.08 ± 0.11, *< 0.05; t = 30: 4.17 ± 0.17; one-way ANOVA with Bonferroni's multiple comparison test) and PSD-95Y12E (t = 0: 4.40 ± 0.15 μm; t = 5: 4.35 ± 0.08; t = 15: 4.17 ± 0.10; t = 30: 4.07 ± 0.15).


PSD-95 possesses multiple CaM-binding sites

By taking a fresh look at CaM binding to PSD-95, we found that Ca2+/CaM specifically binds not only to the previously described HOOK domain but also to the N-terminus (residues 1–16) and to PDZ2 (Fig 1). The existence of several CaM-binding sites within a protein is not unusual. For instance, the central, pore-forming α11.2 subunit of the L-type Ca2+ channel Cav1.2 interacts with CaM at its N-terminus and three independent sites near the C-terminus to control fast Ca2+-dependent inactivation of Cav1.2 (Pitt et al, 2001; Dick et al, 2008; Evans et al, 2011; see also Johny et al, 2013) as well as its surface removal via displacement of α-actinin (Hall et al, 2013). Ca2+/CaM binding to the HOOK region likely affects interactions of PSD-95 with other binding partners such as the NMDARs (Fukunaga et al, 2005; but see Masuko et al, 1999), SAP102 (Masuko et al, 1999), or Pyk2 (Bartos et al, 2010). A potential function of the newly described association of Ca2+/CaM with PDZ2 on similar or other protein interactions is intriguing and will stimulate future work. Here, we focus on the structural and functional analysis of Ca2+/CaM binding to the N-terminus of PSD-95 because this region and specifically its palmitoylation is critical for postsynaptic targeting of PSD-95 and with it of AMPARs (Chen et al, 2000; El-Husseini et al, 2000, 2002; Schnell et al, 2002; Fukata et al, 2013).

CaM-binding motif at the PSD-95 N-terminus

The CaM-binding site at the N-terminus of PSD-95 (Figs 1 and 2) differs from other CaM target sites (Hoeflich & Ikura, 2002). It does not follow the consensus sequence of other CaM-binding motifs (IQ, 1–10, 1–14 and 1–16) (Rhoads & Friedberg, 1997). However, PSD-95 N-terminal residues 2–16 have some similarity to the CaM-binding site found in Ca2+/CaM-dependent kinase, CaMKK (residues 338–354, 50% homology and 25% identical; Kurokawa et al, 2001). The CaMKK sequence contains an aromatic (F352) and adjacent basic residue (R351), which both form contacts with CaM residues in the C-lobe. The PSD-95 sequence has an aromatic residue (Y14) preceded by R13 that both form contacts with CaM residues in the C-lobe (E114, E120, M124, E127, V136, F141, M144). The CaMKK sequence also has a conserved leucine (L340) flanked by mostly hydrophobic residues (T339, I341, L342, V343) that each form contacts with CaM residues in the N-lobe. The corresponding residues in the PSD-95 sequence (C3, L4, C5, I6, V7) also form contacts with CaM residues in the N-lobe (L18, F19, I27, L32, L39, V55, I63). The structure of CaM bound to PSD-95 has an inter-lobe contact (between L39 and V108) that is also found in the structure of CaM bound to CaMKK but not in CaM bound to other target proteins (Ikura, 1996; Hoeflich & Ikura, 2002). An important structural difference between CaM/PSD-95 and CaM/CaMKK is that PSD-95 lacks the hairpin-loop structure formed by CaMKK residues (352 – 357), which might explain why CaMKK binds CaM with 10-fold higher affinity than PSD-95.

CaM controls palmitoylation and CDKL5 binding of PSD-95

Palmitoylation–depalmitoylation cycles are highly dynamic and can occur within seconds (Vartak et al, 2014). Any interference with palmitoylation is expected to shift this equilibrium toward depalmitoylated protein. Our data show that Ca2+ influx into HEK293 cells reduces palmitoylation of PSD-95WT within 5 min by approximately 80% and, based on loss of CDKL5 coprecipitation, this decrease apparently occurs within 1 min (Fig 5). Loss of Ca2+/CaM binding to PSD-95Y12E completely prevents this shift toward depalmitoylation even over the extended 5-min time period. Similarly, Ca2+ influx through the NMDAR decreases palmitoylation of endogenous PSD-95 in acute forebrain slices (Fig 6). This decrease is blocked by three structurally very different CaM antagonists again implicating Ca2+/CaM in antagonizing PSD-95 palmitoylation in neurons (Fig 6). We conclude that Ca2+/CaM binding to PSD-95 makes C3 and C5 inaccessible for palmitoyl transferase. Ca2+/CaM might physically block the binding of palmitoyl transferase to PSD-95 at residues 3–9 (El-Husseini et al, 2000; Fukata et al, 2004). In this manner, Ca2+/CaM shifts the dynamic equilibrium of the constitutive palmitoylation/depalmitoylation cycle toward the latter (Fig 8).

Figure 8.

Figure 8. Model of Ca2+/CaM-induced postsynaptic release of PSD-95

(Left) Under basal conditions, PSD-95 (yellow) is attached to the postsynaptic site through palmitoylation of C3 and C5. It keeps AMPA-type glutamate receptors (blue) via binding with its first and second PDZ domains to the C-termini of their auxiliary TARP subunits (green) at postsynaptic sites. It also anchors CDKL5 (red) via its palmitoylated N-terminus. PSD-95 frequently cycles between palmitoylation and depalmitoylation. (Right) Ca2+ influx through NMDA-type glutamate receptors induces binding of Ca2+/CaM to depalmitoylated PSD-95 to prevent re-palmitoylation. Consequently, a fraction of PSD-95 leaves the postsynaptic site and CDKL5 anchoring is reduced.

At the same time, Ca2+ influx causes a loss of CDKL5 binding to PSD-95 in HEK293 cells (Fig 5) and in neurons (Fig 6). The CDKL5 displacement is of functional interest because CDKL5 has recently been identified as a positive regulator of postsynaptic strength (Ricciardi et al, 2012), which depends on CDKL5 binding to the N-terminus of PSD-95 (Zhu et al, 2013). Remarkably, this interaction requires palmitoylation of PSD-95 (Zhu et al, 2013). It is thus likely that Ca2+/CaM displaces CDKL5 by impairing palmitoylation. In addition, Ca2+/CaM could also directly compete with CDKL5 for binding to PSD-95 in the Y12-T19 region as T19 is critical for CDKL5 binding to a peptide spanning PSD-95 residues 1–19 (Zhu et al, 2013).

CaM controls postsynaptic localization of PSD-95

Earlier work indicates that postsynaptic Ca2+ influx destabilizes PSD-95 in spines such that a substantially larger fraction (30–50%) than under non-stimulated conditions (15%) of spine PSD-95 exchanges with shaft PSD-95 over a period of 30 min (Steiner et al, 2008; Nelson et al, 2013). Consistently, postsynaptic Ca2+ influx induces a temporary reduction in total PSD-95 content in spines in our cultures (Fig 7). The loss of PSD-95 is prevented and even overcompensated by our PSD-95Y12E mutant (Fig 7). Collectively, our data indicate that N-terminal capping of PSD-95 by Ca2+/CaM blocks the accessibility of palmitoylation sites at C3 and C5 (Supplementary Fig S3), which prevents palmitoylation of PSD-95 and triggers its redistribution away from postsynaptic membranes at high Ca2+ levels (Fig 7). PSD-95 at the postsynaptic membrane is in dynamic equilibrium with palmitoylated and depalmitoylated forms controlled by the relative activity of palmitoyl transferase and esterases (Fukata et al, 2004, 2013). In our model (Fig 8), we suggest that palmitoylation of PSD-95 (and hence membrane anchoring) and CDKL5 binding (Zhu et al, 2013) are both favored under resting basal conditions (absence of glutamate) when cytosolic Ca2+ levels are low. Stimulation of glutamate receptors leads to channel opening that causes Ca2+ influx at glutamatergic synapses. The rise in postsynaptic Ca2+ level then induces Ca2+/CaM to bind preferentially to depalmitoylated PSD-95 at the N-terminus to promote diffusion of PSD-95 away from the postsynaptic membrane. The Y12E mutation blocks Ca2+/CaM binding to PSD-95 without affecting palmitoylation of PSD-95 because Y12 is located outside the consensus sequence for palmitoylation (El-Husseini et al, 2000; Fukata et al, 2004). Consistent with our model (Fig 8), PSD-95Y12E prevents Ca2+-induced dissociation of PSD-95 from both CDKL5 (Figs 5 and 6) and its removal from postsynaptic sites (Fig 7). We posit CaM capping of the PSD-95 N-terminus as an important mechanism for decreasing the abundance of PSD-95 and its binding partners at the postsynaptic site in response to Ca2+ influx associated with LTD.

Materials and Methods

Ethical approval

All animal procedures had been approved by the University of California at Davis and followed NIH guidelines.

Fluorescence polarization assays

Fluorescein-labeled peptides (100 nM; ChinaPeptides, Shanghai, China) were titrated with increasing concentrations of purified CaM in FP buffer (50 mM HEPES, pH 7.4, 100 mM KCl, 1 mM MgCl2, 10 mM CaCl2, 0.05 mM EGTA, 5 mM nitrilotriacetic acid) and FP determined with a Synergy 2 plate reader (BioTek, Winooski, VT, USA) as described (Lim et al, 2002).

Brain slices

Three hundred fifty-micrometer-thick forebrain slices containing hippocampus were prepared from 8- to 14-week-old mice in ice-cold artificial cerebrospinal fluid (ACSF, containing: 127 mM NaCl, 26 mM NaHCO3, 1.2 mM KH2PO4, 1.9 mM KCl, 2.2 mM CaCl2, 1 mM MgSO4 and 10 mM D-glucose, 290–300 mOsm/kg) with a vibratome (Leica VT 1000A) as in Halt et al, 2012 and Hall et al, 2013. CaM antagonists were applied for 30 min before NMDA (100 μM, 5 min). Slices were frozen and later homogenized in Lysis Buffer (1% Triton X-100, 150 mM NaCl, 10 mM EDTA, 10 mM EGTA, 10 mM Tris-HCl, pH 7.4 and protease inhibitors) before ultracentrifugation.

Immunoprecipitation and immunoblotting

Immunoprecipitation and immunoblotting were performed as in Hell et al, 1993. Multiple exposure times ensured that immunosignals were in the linear range (Davare & Hell, 2003; Hall et al, 2006). Signals were quantified through densitometry using Photoshop CS3 software (Adobe).

Primary hippocampal neuronal cultures

Cultures were prepared from E18 rat embryos and grown on coverslips coated with 0.1% (w/v) poly-L-lysine at a cell density of 3 × 104/cm2 in Neurobasal medium (Life Technologies) supplemented with NS21 (Chen et al, 2008) as described (Chen et al, 2008; Halt et al, 2012; Hall et al, 2013).

NMR spectroscopy

Proteins were expressed in E. coli strain BL21 (DE3) in LB medium (unlabeled proteins) or M9 media supplemented with 15NH4Cl or 15NH4Cl/13C-glucose for single- or double-labeled proteins. CaM was prepared as described (Zhang et al, 2012). GST- or His-tagged PSD-95(1–71) were purified from E. coli supernatants by standard affinity purification on GST columns (Hengen, 1996; Smith, 2000) or Ni-NTA columns (Bornhorst & Falke, 2000). Purified GST-fusion protein was digested by thrombin to remove the N-terminal GST tag, producing an N-terminal fragment of PSD-95 [residues 1–71, called PSD-95(1–71)] that was further purified by gel-filtration size-exclusion chromatography (Superdex 75). PSD-95(1–71) containing a C-terminal His-tag from Ni-NTA column was loaded onto a size-exclusion column (Superdex 75) for further purification.

The purified 15N- or 13C/15N-labeled PSD95(1–71) protein was exchanged into NMR buffer (25 mM CD3COONa, 1 mM EDTA-d, 5 mM DTT-d, 8% D2O, pH5.0). The NMR samples of 15N- or 13C/15N-labeled CaM bound to unlabeled PSD-95(1–71) were prepared by adding 1.5-fold excess of unlabeled PSD95(1–71) to labeled CaM and vice versa. Briefly, the pure CaM and PSD-95(1–71) proteins were separately exchanged into NMR buffer (pH7.0) containing 20 mM Tris-d11 with 5 mM CaCl2, 50 mM NaCl, 5 mM DTT-d and 95% H2O/5% D2O. The isotopically labeled Ca2+-bound CaM (0.3 mM) was mixed with an equal volume of PSD-95(1–71) (0.45 mM) and then concentrated to give a final concentration of 0.3 mM for the 1:1 complex (called CaM/PSD95) in a final volume of 0.3 ml.

For the measurement of RDCs of CaM bound to PSD-95(1–71), the filamentous bacteriophage Pf1 (Asla Biotech Ltd., Latvia) was used as an orienting medium. Pf1 (17 mg/ml) was added to 15N-labeled CaM bound to unlabeled PSD-95(1–71) at pH 7.0, to produce weak alignment of the complex.

All NMR experiments were performed with 800 MHz Bruker Avance III spectrometer equipped with TCI cryoprobe. NMR resonance assignments were obtained as described by Zhang et al, 2012. 15N T1, T2, and 15N-{1H} NOE experiments were performed on PSD-95(1–71) at 283K or CaM bound to PSD-95(1–71) at 310K using standard pulse sequences described previously (Farrow et al, 1994).


This work was supported by NIH grants R01 EY012347 (JBA), R01 AG017502 (JWH), R01 NS078792 (JWH), the American Heart Association Postdoctoral Fellowship 11POST7020009 (LM), and the NARSAD Young Investigator Grant 20748 from the Brain & Behavior Research Foundation (LM). The NMR assignments have been deposited in the BMRB (Accession Number 19238). The atomic coordinates have been deposited into the Protein Databank (2MES).

Author contribution

YZ, LM, TP, ZM, DC, AR, JBA, and JWH designed experiments; YZ, LM, TP, ZM, DC, and DKP performed experiments; YZ, LM, TP, ZM, DC, DKP, AR, JBA, and JWH analyzed data; YZ, LM, TP, DC, AR, JBA, and JWH wrote the manuscript.

Conflict of interest

The authors declare that they have no conflict of interest.