Research ResourceNeuroscience

Long-term potentiation modulates synaptic phosphorylation networks and reshapes the structure of the postsynaptic interactome

See allHide authors and affiliations

Sci. Signal.  09 Aug 2016:
Vol. 9, Issue 440, pp. rs8
DOI: 10.1126/scisignal.aaf6716

Defining the PSD risk for psychiatric disease

Li et al. stimulated mouse hippocampi to induce long-term potentiation (LTP; a form of synaptic plasticity), isolated the CA1 region, and then performed both a proteomic analysis and a protein-protein interaction analysis of the postsynaptic density (PSD), a cellular compartment containing more than 1500 proteins. Bioinformatics analysis of this data set indicated that risk factors associated with autism spectrum disorder and schizophrenia were contained within the phosphoproteins that were differentially regulated by the induction of LTP, such that excluding these proteins from the analysis completely removed any association with the risk of having either of these two psychiatric diseases. The data provided, along with this medical association of disease risk, should guide researchers and clinicians toward a better understanding of both the molecular changes that enable learning and memory as well as the complex neurological diseases involving proteins that are part of the PSD.


The postsynaptic site of neurons is composed of more than 1500 proteins arranged in protein-protein interaction complexes, the composition of which is modulated by protein phosphorylation through the actions of complex signaling networks. Components of these networks function as key regulators of synaptic plasticity, in particular hippocampal long-term potentiation (LTP). The postsynaptic density (PSD) is a complex multicomponent structure that includes receptors, enzymes, scaffold proteins, and structural proteins. We triggered LTP in the mouse hippocampus CA1 region and then performed large-scale analyses to identify phosphorylation-mediated events in the PSD and changes in the protein-protein interactome of the PSD that were associated with LTP induction. Our data indicated LTP-induced reorganization of the PSD. The dynamic reorganization of the PSD links glutamate receptor signaling to kinases (writers) and phosphatases (erasers), as well as the target proteins that are modulated by protein phosphorylation and the proteins that recognize the phosphorylation status of their binding partners (readers). Protein phosphorylation and protein interaction networks converged at highly connected nodes within the PSD network. Furthermore, the LTP-regulated phosphoproteins, which included the scaffold proteins Shank3, Syngap1, Dlgap1, and Dlg4, represented the “PSD risk” for schizophrenia and autism spectrum disorder, such that without these proteins in the analysis, the association with the PSD and these two psychiatric diseases was not present. These data are a rich resource for future studies of LTP and suggest that the PSD holds the keys to understanding the molecular events that contribute to complex neurological disorders that affect synaptic plasticity.


Long-term potentiation (LTP) and long-term depression (LTD) are two of the most studied forms of synaptic plasticity. These activity-dependent changes in the efficacy of synaptic transmission of excitatory synapses are believed to represent cellular processes underlying learning and memory (1, 2). The best characterized form of plasticity is the LTP that requires the glutamate receptors NMDAR and AMPAR and occurs at the Schaffer collateral pathway between CA3 and CA1 pyramidal neurons in the hippocampus (2, 3). Changes in the postsynaptic side of the synapse are essential for LTP (4). Since the discovery of LTP more than 40 years ago (5), numerous postsynaptic molecules and mechanisms that affect glutamate receptor function have been associated with LTP (3, 6).

Although changes in the activity and trafficking of AMPA-type glutamate receptors (AMPARs) have a crucial role in the enhancement of synaptic strength that occurs in LTP, the induction of LTP is also associated with many other changes in the postsynaptic neuron (7). In pyramidal cells, LTP involves changes in gene expression and protein synthesis (8), changes in dendritic spine morphology (9, 10), and altered activity of the voltage-dependent ion channels that regulate the intrinsic excitability (11). Changes in protein phosphorylation have a central role in all of these processes, and advances in mass spectrometry (MS) analysis have provided key insights into how phosphorylation regulates the organization of the postsynaptic density (PSD) of excitatory synapses (1216).

The PSD comprises a collection of more than 1500 proteins, including receptors, enzymes, scaffold proteins, and structural proteins of the cytoskeleton. Large-scale studies have identified hundreds of phosphorylation sites within the PSD, many of which are affected by pharmacological activation of neurotransmitter receptors (13, 1720). Coordinated changes in protein kinase and phosphatase activity, producing distinct patterns of protein phosphorylation in the PSD, may provide a mechanism for the processing, integrating, and storing information at synapses (13). Compared with the phosphorylation changes reported to occur in response to pharmacological receptor activation, few phosphorylation sites have been reported that are modulated by the induction of LTP (21, 22). Thus, it remains unclear whether the induction of LTP at Schaffer collateral fiber synapses in the hippocampal CA1 region triggers widespread changes in protein phosphorylation of PSD proteins and, if it does, how these changes in phosphorylation are distributed within PSD protein complexes and potentially organized into distinct signaling networks composed of specific writers (kinases), erasers (phosphatases), readers (proteins that recognize the phosphorylation status of target proteins), and the targets of these three components of the phosphorylation-dependent regulatory system.

To address these questions, we used liquid chromatography–tandem MS (LC-MS/MS) to identify changes in protein phosphorylation that occurred after the induction of LTP by high-frequency, tetanic stimulation of Schaffer collateral fiber synapse in the CA1 region of the mouse hippocampus. Such high-frequency synaptic stimulation altered the phosphorylation status of 570 sites within 222 PSD proteins. We determined the composition of the PSD kinome and phosphatome and identified protein domains and short sequence motifs surrounding the phosphorylated and dephosphorylated sites that were modulated by the induction of LTP. In addition, we isolated scaffold protein signaling complexes from the mouse hippocampus CA1 area and mapped LTP-regulated sites within the protein interaction networks and determined how these sites changed after the induction of LTP. Our results indicated that proteins modulated by LTP induction are differentially phosphorylated in a manner dependent on their position within the PSD protein-protein interaction network. Proteins in the PSD that were differentially phosphorylated by the induction of LTP were significantly associated with the risk for two psychiatric diseases: schizophrenia and autism spectrum disorder. The PSD did not contribute any significant risk for these psychiatric diseases when these proteins were excluded from the analysis. Thus, these LTP-regulated phosphoproteins represent the “PSD risk” for at least these two psychiatric diseases and likely represent functional subnetworks within the PSD.


Identifying the phosphoproteomic changes associated with the induction of LTP in the CA1 region

We induced LTP in mouse hippocampus CA1 slices by stimulating presynaptic fibers from the stratum radiatum layer of CA1 with two trains of 100-Hz stimulation for 1 s separated by an intertrain interval of 10 s. We then collected the stimulated slices 5 min after the induction of LTP (Fig. 1, A and B). To determine changes in protein phosphorylation in the PSD, we isolated PSD fractions (13) from stimulated and nonstimulated CA1 slices, and then we enriched for phosphopeptides and identified and quantified the peptides by LC-MS/MS (Materials and Methods). To control for changes in protein phosphorylation due to LTP-induced changes in protein abundance in the PSD, we quantified PSD protein abundance by LC-MS/MS in stimulated and nonstimulated samples. We identified 1554 phosphorylation sites, of which 570 (37%) were altered by induction of LTP (Fig. 1C and table S1). Changes in protein phosphorylation and protein content were further validated by Western blotting for kinases, scaffolds, and glutamate receptors that are known components of the PSD (fig. S1A).

Fig. 1 Modulation of protein phosphorylation by induction of LTP in the CA1 region of mouse hippocampus.

(A) Potentiation induced by high-frequency stimulation (HFS) in three separate experiments (20 CA1 mini-slices from three to four mice per experiment). There was no significant difference in the amount of potentiation present in the three experiments [F2,57 = 0.104, P = 0.901, one-way analysis of variance (ANOVA)]. (B) Superimposed postsynaptic responses evoked during baseline and 4 min after HFS. (C) Distribution of phosphorylation sites phosphorylated (increase) and dephosphorylated (decrease) by the induction of LTP in mouse hippocampus CA1 area (left) and the functional classification of proteins modulated by LTP (right). Selected functional groups are noted. Glutamate rec. glutamate receptor.

Pharmacological activation of NMDARs triggers both increases and decreases in protein phosphorylation in hippocampal neurons; however, the number of sites that are dephosphorylated exceeds the number of sites with increased phosphorylation (13). In contrast, we observed that similar numbers of sites exhibit increases (18.5%, 287 sites) and decreases (18%, 278 sites) in phosphorylation after the induction of LTP (Fig. 1C and table S1). These phosphorylation sites were distributed in 129 proteins that exhibited increased phosphorylation (up-regulation) and 135 proteins that exhibited decreased phosphorylation (down-regulation) compared with their status in the PSD from nonstimulated slices. These proteins belong to a wide range of functional classes (Fig. 1C). The most abundant functional groups represented structural proteins (44%), including cytoskeletal (23%), scaffold (16%), and cell adhesion (5%) proteins. We define scaffold proteins as those that contain multiple protein interaction domains and, therefore, have the capacity to form protein complexes and adaptor proteins as those that have a single protein-binding domain and thus facilitate the interaction between pairs of proteins. Proteins that exhibited phosphorylation changes were also those involved in neuronal signaling, including protein kinases (9%), adaptor proteins (6%), proteins involved in Ras-dependent signaling (a pathway that controls cytoskeletal dynamics) and classified as Ras GTPase (guanosine triphosphatase)–activating protein (RasGAP) signaling (2%), and the proteins that function as ion channels (Ca2+ channel) and ligand-activated ion channels (glutamate receptor) (2%). Thus, induction of LTP in the CA1 layer of the mouse hippocampus influenced the phosphorylation status of proteins that control multiple aspects of synaptic and neuronal function.

Identifying the PSD kinome

Because the changes that we detected upon LTP induction involved many increases in phosphorylation events, we sought to identify the kinases in the PSD. We isolated PSD-enriched fractions from nonstimulated mouse CA1 regions (see Materials and Methods and fig. S1B) and identified 1626 PSD proteins by LC-MS/MS (table S2). We then identified serine/threonine kinases and tyrosine kinases using the SMART (23) and Pfam (24) databases to match protein sequences identified by MS in the PSD with sequences from serine/threonine and tyrosine kinase domain–containing proteins. With this approach, we identified 81 protein kinases associated with the PSD (table S2, PSD kinome) from a total of 558 mouse protein kinases (25) and mapped them to the mouse kinome (Fig. 2A) (26). This analysis revealed that the PSD kinases were distributed among all branches of the mouse kinome with larger representation in four families: 17% AGC (a family named after PKA, PKG, and PKC groups of kinases), 17% STE (a family of the homologs of the yeast sterile kinases), 22% CMGC (a family named after CDK, MAPK, GSK3, and CLK groups of kinases), and 21% CAMK (the family of calcium/calmodulin kinases). Moreover, within each family, specific subfamilies were overrepresented (Table 1 and table S2). These data showed that the CA1 PSD contains the necessary protein kinases to target diverse protein phosphorylation motifs.

Fig. 2 Modulation of the PSD kinome and kinase-substrate pairs.

(A) Representation of the kinases in the mouse CA1 PSD on the mouse kinome tree. Circles show kinases present in part of the kinome tree. Main groups are indicated: AGC, CMCG, TK, STE, and CAMK. P values indicate kinase groups that are enriched in the PSD. (B) Kinases modulated by the induction of LTP mapped on the mouse kinome tree. (C) Mapping of known kinase-substrate pairs (top) and kinases predicted to phosphorylate sites that increase upon induction of LTP (bottom). The kinases with most phosphorylated substrates in each set are indicated by name. Colors match the families in the kinome trees in (A) and (B). (D) Distribution of functional groups for substrates of kinases modulated by LTP. (E) Differential distribution of functional groups phosphorylated by the indicated kinase families.

Table 1 Kinases identified in the PSD.

The kinases were identified from unstimulated mouse hippocampus, and the number of phosphorylated sites that exhibited an increase or decrease in response to LTP induction in the MS data is shown. Kinases are grouped by family.

View this table:

Protein kinase activity is often regulated by phosphorylation in the catalytic domain (27, 28); thus, we used changes in phosphorylation of protein kinase activation sites to identify how LTP induction modified the activity of the PSD kinome. We found 47 phosphorylation sites that increased in phosphorylation after induction of LTP and that mapped to the activity-regulating domains of 14 protein kinases (Table 2 and tables S1 and S2). These sites included those that enhance kinase activity, those that reduce activity, and those with an unknown effect on activity (Table 2) and were distributed through the kinome (Fig. 2B).

Table 2 Kinases in the PSD with LTP-induced changes in phosphorylation.

Kinases are grouped by family.

View this table:

To map kinases to their substrates containing LTP-modulated phosphorylation sites, we compared the LTP-regulated phosphorylated sequences with established kinase-substrate pairs in the PhosphoSitePlus database ( (29). Initially, we limited the analysis to only those kinases that exhibited altered phosphorylation in response to induction of LTP. With this limitation, we could only match a small fraction (9%) of the LTP-regulated phosphorylation sites to a protein kinase (table S3). When we limited the analysis to those kinases that we detected in the PSD, we matched 52 phosphorylation sites to a protein kinase (table S3), and when we did not impose any limits, we mapped 77 sites to a kinase (table S3). The first most stringent method for mapping kinases to substrates identified many sites predicted to be targeted by the CMGC family member ERK2 and the CAMK family member Camk2a (Fig. 2C, top).

Using the kinase-substrate matching based on the most stringent criteria, we found that, similar to the function assigned to phosphorylated proteins identified in the PSD, proteins of the cytoskeleton, kinases, and scaffold proteins represented a relatively large proportion of the functional groups that exhibited LTP-induced changes in phosphorylation (compare Fig. 1C and Fig. 2D). We also analyzed whether subfamilies of protein kinases activated by the induction of LTP modulated the phosphorylation of substrates with distinct molecular functions. Cytoskeleton and adhesion proteins were phosphorylated principally by CMGC kinases, which include the proline-directed kinases, whereas glutamate receptors, proteins associated with gap junctions, and RasGAP activity were preferentially phosphorylated by kinases of the basophilic-targeted AGC and CAMK families (Fig. 2E). In contrast, phosphorylation of kinases and scaffold proteins was predicted to involve all three families, which could provide a mechanism for crosstalk among the regulatory pathways mediated by kinases of the CMGC, AGC, or CAMK families (Fig. 2E). The functional grouping of activated kinase–substrate pairs after the induction of LTP suggested a pattern of signaling in which the AGC and CAMK families phosphorylate channels and glutamate receptors, whereas the CMCG kinases target downstream cytoskeletal components.

To expand the substrate-kinase mapping, we analyzed the linear phosphorylation motifs for all LTP-modulated sites (table S4) and identified the consensus motifs for each kinase (table S2, kinases) using PrediKin (30) and NetPhorest (31). Predicted kinases for each regulated site were then determined using NetworKIN (31) and the Eukaryotic Linear Motif (ELM) resource (32). The predicted kinase substrates (table S4) exhibited a similar pattern as that observed for the kinases matched to their reported phosphorylation sites (Fig. 2D and tables S3 and S4). In both, proline-directed kinases of the CMCG family targeted a large proportion of the substrates, and the basophilic kinases of the AGC and CAMK families represented another large proportion. The kinases that phosphorylated (table S3) or were predicted to phosphorylate (table S4) the greatest proportion of substrates were also those that exhibited phosphorylation changes consistent with activation after induction of LTP (Tables 1 and 2). Many of the basophilic kinases that exhibited phosphorylation-induced activation profiles mediate receptor-stimulated second-messenger cascades, such as PKA subfamily, which mediates adenosine 3′,5′-monophosphate (cAMP) signals; PKC subfamily, which mediates Ca2+, diacylglycerol (DAG), and phosphoinositide (PI) signaling; and the CAMKII subfamily, which mediates Ca2+ signals. These kinase families were also the ones that targeted the receptors themselves; thus, as observed for NMDAR signaling networks (13), this dual function of mediating the downstream signal and also targeting the receptors suggests that there is a feedback role of second messenger–coupled kinases. In contrast, the proline-directed kinases propagate the signal to cytoskeletal components of the PSD.

Identifying the PSD phosphatome

Phosphorylation is a dynamic process that can be reversed through the action of phosphatases. LTP induction produced both increased and decreased phosphorylation in the PSD proteome. Analysis of protein phosphatase domains (with SMART and Pfam) and regulatory subunits present within the PSD-enriched proteome revealed 20 protein phosphatases within the PSD (Fig. 3A and table S4). PSD phosphatases were enriched principally in serine/threonine-specific phosphatases, in particular members of the PPM and PPP subfamilies. We only detected two dual-specificity phosphatases (DUSPs) in the PSD (Fig. 3A and table S2, phosphatases). Moreover, analysis of ligand-binding motifs for sites dephosphorylated after LTP induction revealed an enrichment in PP2B docking sites (Fig. 3A and table S4). PP2B is a member of the PPM group; thus, these data are consistent with a suggested role for members of the PPM group in LTP (33, 34).

Fig. 3 PSD phosphatases, phosphorylation readers, and modulation of phosphorylation in disordered regions.

(A) Representation of the protein phosphatases in the mouse CA1 PSD on the mouse phosphatase tree. Phosphatases are indicated with circles. P values indicate enrichment of phosphatase groups that are enriched in the PSD. Tyrosine phosphatases: PTP, dual S/T-T phosphatases (DUSP), and specific S/T phosphatases PPP and PPM. (B) Phosphorylation-dependent binding domains present at the PSD. Domains were identified by SMART/Pfam/InterPro identification (Domain ID). (C) Percentage of phosphorylated and dephosphorylated short sequence motifs that mediate binding to protein domains (domain ligands). Domain ligands were identified by ELM. The most significantly enriched domain ligands that exhibited a change in response to LTP corresponded to the proline-binding domains WW (P < 0.001), SH3 (P < 0.005), and the phosphoserine and phosphothreonine (pSer/pThr)–binding domains: 14-3-3 (P < 0.001) and FHA (P < 0.05). (D) Protein domains phosphorylated after the induction of LTP. DUF3498, domain of unknown function found in RasGAPs; DUF737, domain of unknown function in a family of proteins; L27, a domain in scaffold proteins; GuKc, guanylate kinase domain; ANK, ankyrin repeat found in proteins in many functional classes, including the cytoskeleton; PDZ, domain found in scaffold proteins; PH, pleckstrin homology domain involved in protein localization; S_TKc, serine/threonine protein kinase domain; septin, a domain present in a family of GTPases; stathmin, a domain in a protein family that regulates microtubules; tubulin, the GTPase domain in this microtubule protein; vinculin, a domain in this family of actin interacting proteins; 4_1_CTD, an actin-binding domain.

Identifying the readers of changes in phosphorylation status in the PSD

Many PSD phosphorylation sites are present in short sequences (less than 10 amino acids) within disordered regions (13). These sequences can be recognized by specific protein domains (the readers in the writer, eraser, and reader toolkit), and these phosphorylated sequences function as ligands (domain ligands) for their cognate readers. Therefore, we determined the composition of phosphorylation readers in the PSD. We identified by bioinformatics analysis the protein domains present in the 1626 PSD proteins in the mouse CA1 region and scanned for phosphorylation domain readers using the SMART and Pfam databases (see Materials and Methods). This analysis identified proteins with domains that recognize phosphotyrosine motifs (PTB and SH2) and domains that recognize phosphoserine and phosphothreonine motifs (FHA, WD40, 14-3-3, and WW) (table S4). Of these readers, 14-3-3 domains were significantly enriched at the PSD (Fig. 3B and table S4, spreadsheet 2). This analysis confirmed that the CA1 PSD contains phosphorylation readers that decode changes in phosphorylation ratios induced by LTP.

To match the phosphorylation sites modulated by the induction of LTP to the readers, we used the ELM resource to scan short sequence motifs in phosphorylated regions (13) and identified protein domains within phosphorylated regions using the SMART and Pfam databases. We determined that most (72%) of the phosphorylation sites regulated after induction of LTP map within protein domains or motifs that serve as domain ligands (table S4). Many domain ligands exhibited a change in phosphorylation in response to LTP, including to the proline-binding domains WW and SH3, and the phosphoserine and phosphothreonine (pSer/pThr)–binding domains 14-3-3 and FHA (Fig. 3C and table S4). Moreover, there was a close correlation between domain ligands, activated kinases, and protein function. For example, 39% of cytoskeleton proteins modulated by LTP contained WW or SH3 domain ligands; 75% of the FHA and 14-3-3 domain ligands occurred in scaffolds and channels or glutamate receptors (table S4). Both major groups of ligands, the proline-directed ones and the pSer/pThr-directed ones, reproduced the expected pattern of substrate phosphorylation by proline-directed (CMCG) and basophilic (AGC and CAMK) kinases when matched bioinformatically to kinases: most (76%) of the WW and SH3 domain ligands were phosphorylated by proline-directed kinases, and 66% of 14-3-3 and FHA domain ligands were phosphorylated by kinases of the AGC and CAMK families (fig. S2 and table S4).

Identifying how induction of LTP affects phosphorylation of the core scaffolding complexes

Although most of the LTP-regulated phosphorylation sites were present in short disordered regions (~80%), a small fraction including 75 phosphorylation sites mapped to structured protein domains (Fig. 3D and table S4). Phosphorylation sites within protein domains that had an increase in phosphorylation upon LTP included the domains present in the major components of the scaffold machinery of the PSD. These domains included the PDZ, PH, ankyrin (ANK), and guanylate kinase domains. Domains not typically found in scaffold proteins were also enriched. These included the serine and threonine kinase domain and domains present in various cytoskeletal proteins or proteins that associate with the cytoskeleton, such as the septin, stathmin, tubulin, and vinculin domains (Fig. 3D and table S4). DUF3498 was particularly prevalent in the phosphorylated protein domains. This is a predicted domain of unidentified function present in SynGAP1 and potentially relevant for its RasGAP function. By interacting with multiple proteins, scaffold proteins can serve as protein hubs that control the flow of information.

Because protein domains that showed enhanced phosphorylation were present in the core scaffold components of the PSD, we hypothesized that these scaffold proteins may organize subnetworks that are differentially regulated by LTP when compared to the whole PSD. We considered that there are three principal layers of scaffold proteins: a membrane-proximal “top” layer connected to glutamate receptors, which is composed of members of the family of MAGUKs (membrane-associated guanylate kinases), represented by proteins of the DLG (disk large) family; a “bottom” layer close to the cytoskeleton, which includes members of the SHANK family; and a middle layer connecting the top and bottom layers, which includes members of the DLGAP (disk large–associated guanylate-associated protein) family (Fig. 4A). Within these layers, we identified 118 phosphorylation sites within 11 PSD scaffold proteins: in the DLG layer, PSD95, PSD93, SAP102, and SAP97; in the DLGAP layer, DLGAP1, DLGAP2, DLGAP3, and DLGAP4; and in the SHANK layer, SHANK1, SHANK2, SHANK3, and SynGAP1 (Fig. 4A and tables S1 and S2, PSD). The DLG layer contains proteins with the PDZ, SH3, and guanylate kinase domains; the DLGAP layer contains proteins with the GKAP domains; and the Shank layer contains proteins with the SH3, ANK, and PDZ domains. With a few exceptions, these domains were among those identified as having an increase in phosphorylation upon LTP.

Fig. 4 Modulation of phosphorylation within PSD scaffolds.

(A) Cartoon of the modulation of core components of the core scaffold structure of the PSD, representing three layers organized by families of scaffold proteins: DLGs (top), DLGAPs (middle), and SHANKs (bottom). (B) Distribution of phosphorylated and dephosphorylated sites in PSD proteins in multiple functional groups after the induction of LTP. (C) Network representation of Dlg4, Shank3, Syngap1, and Dlgap1 protein complexes isolated from mouse hippocampus, and identified by LC-MS/MS. The intensity of the color indicates the number of phosphorylated sites that increased (phosphorylated) or decreased (dephosphorylated) at each node. Gray nodes represent nonphosphorylated components of the network. (E and F) Bar charts of distance distributions for phosphorylated and nonphosphorylated nodes in the PSD scaffold protein network. Average shortest path length for phosphorylated nodes, 2.1; average shortest path length for nonphosphorylated nodes, 2.5; P < 0.001, t test. (G) Degree distribution of phosphorylated and nonphosphorylated nodes in the scaffold network. P < 0.001, t test.

Whereas the induction of LTP resulted in a similar amount of phosphorylation and dephosphorylation throughout the entire PSD (~20% each), this core scaffold machinery of the PSD exhibited little LTP-induced dephosphorylation (~8%). Instead, most of the changes were an increase in protein phosphorylation (~59%), and the remaining 33% were unchanged by LTP (Fig. 4B). This threefold increase (from ~20 to ~60%) in increased protein phosphorylation of the core scaffolding components of the PSD compared to the regulated sites in the entire PSD is significant (P < 0.001, t test). Other functional categories, such as kinases, calcium channels, and glutamate receptors, exhibited a similar equal distribution of phosphorylated and dephosphorylated sites as we observed for the entire PSD. In contrast, cytoskeletal proteins exhibited a bias toward dephosphorylation with 23% of sites having decreased phosphorylation and only 11% having an increase after induction of LTP (Fig. 4B). These data indicated that the increased phosphorylation of the core scaffold machinery of the PSD is important for transmitting the signals and mediating the reorganization needed to support LTP induction in mouse CA1 hippocampus. We predicted that the modulation of protein phosphorylation by the induction of LTP modified the patterns of protein-protein interactions within the core scaffold structure of the PSD.

Exploring how LTP induction modifies scaffold interactions in the PSD

To explore whether the induction of LTP modulated the composition of protein interaction networks in the PSD, we determined the composition of scaffold signaling complexes in the mouse hippocampus, generated network maps, and overlayed the phosphorylation sites onto the network. We selected DLG4, DLGAP1, and SHANK3 as typical representatives from each scaffold layer (Fig. 4A), as well as SYNGAP1, which is both a scaffold protein and a RasGAP, to interact with multiple other PSD scaffold proteins (16, 35, 36). We immunoprecipitated Dlg4, Syngap1, Shank3, or Dlgap1 from the mouse hippocampus and used high-performance LC–MS/MS (Materials and Methods) to determine the composition of proteins that interacted with each of the representative scaffold proteins and then assembled scaffold protein complexes with 255 nonredundant protein interactions (table S5). We then clustered the protein interaction network using the algorithim of Newman and Girvan (13, 37) and mapped the phosphorylation sites regulated by the induction of LTP onto the proteins in the network (Fig. 4, C and D). About 50% of the total phosphorylated sites that we detected as regulated by LTP induction in the PSD occurred within components of the network. Furthermore, we found that the 35 most connected nodes (table S5, spreadsheet 2) represented 75% of the total differentially regulated phosphorylated sites within the network.

To assess how these nodes integrate enhanced phosphorylation within the PSD scaffold network, we determined the distance, degree, and phosphorylation ratios within nodes. Because distance represents the number of edges required to connect two separate nodes and the degree represents the number of links connected to a node, measuring the phosphorylation ratios (number of sites with increased phosphorylation/number of sites with decreased phosphorylation) can give a representation of the distribution and function of phosphosites within PSD scaffold networks. Distance analyses indicated that the induction of LTP modulates protein phosphorylation within nodes that are separated by an average of 2.1 edges, whereas the nonphosphorylated nodes are separated by an average of 2.5 edges, indicating that the phosphorylated nodes were more closely clustered than the nonphosphorylated nodes (Fig. 4, E and F) in the PSD scaffold network.

Moreover, we found that phosphorylated nodes were significantly more connected, exhibiting an average of 3.13 connections, than were nonphosphorylated nodes, which had an average of only 1.49 connections (Fig. 4G). The phosphorylated nodes can be grouped into two high-degree, high-phosphorylation clusters that contain the core signaling machinery of the PSD, composed by scaffolds, glutamate receptors, and signaling molecules enriched in protein kinases and activators and inhibitors of GTPases (Fig. 5, clusters a and b, and table S5, spreadsheet 2). The two most highly connected clusters, a and b, which contain the core PSD machinery, are shown with their constituent node identities (Fig. 5). Therefore, although the induction of LTP promoted an equal change in phosphorylation and dephosphorylation at the level of the entire PSD, we found that the most connected components of the protein interaction networks organized by the core scaffold proteins were enriched in sites that become more phosphorylated upon LTP.

Fig. 5 Clustering of phosphorylated and dephosphorylated nodes in PSD scaffold networks.

Clustering of PSD scaffold protein-protein interactions by node degree and node phosphorylation/dephosphorylation ratio. Color code indicates phosphorylation ratio for each node. Cluster connectivity: a→b→c→d.

Analysis of the psychiatric disease risk of the PSD

The function of multidomain scaffolds can be understood by their composition of protein domains and their protein interactors. Our data suggested that at the PSD, multidomain scaffolds, along with their interacting partners, are dynamically regulated by the induction of LTP. To explore biomedical implications of this finding, we analyzed whether the PSD of the hippocampal CA1 was enriched for proteins encoded by genes identified as psychiatric disease risk factors.

Many mutations in PSD components have been linked to various brain disorders (Fig. 6A and table S6). Consistent with human genetic data, we identified PSD proteins linked to autism spectrum disorders and schizophrenia (3843). We asked whether the CA1 PSD was enriched in proteins associated with psychiatric disease risk and, furthermore, whether the phosphorylation status of the risk-associated proteins was regulated by the induction of LTP. We found that the CA1 PSD was significantly enriched in both de novo single nucleotide variants (SNVs) for autism spectrum disorder (41) and recurrent de novo SNVs in autism spectrum disorder (44). We also found significant enrichment in SNVs associated with schizophrenia (38) and common variants associated with schizophrenia (Fig. 6B and table S7) (40). The same pattern was observed when different data sets were analyzed for enrichment of autism spectrum disorder–associated proteins, such as by the TADA (transmission and de novo association) test (45) and the Simons Foundation for Autism (SFARI) database (table S7) (46). When we excluded PSD proteins that exhibited phosphorylation-dependent regulation by LTP, we did not observe significant enrichment for any of the human genetics data sets studied with the exception of schizophrenia common variants, which decreased from P = 5.75 × 10−5 to P = 0.02 (table S7). Therefore, our data suggested that the risk for these two psychiatric diseases is not evenly distributed throughout the PSD but tends to be associated with proteins modulated by the induction of LTP. As expected, when only proteins modulated by LTP were considered, the significance ratios were greatly increased (Fig. 6B and table S7).

Fig. 6 Modulation of psychiatric disease risk factors by induction of LTP.

(A) Distribution of psychiatric disease risk factors for autism spectrum disorder (ASD) and schizophrenia (SCZ), intellectual disability (ID), and neurological disorders (Neurological-OMIM) categories within the PSD core scaffold protein interaction network defined by the four scaffolds analyzed. (B) Bar charts showing the enrichment (−log P value) for autism spectrum disorder and schizophrenia for SNVs (single and recurrent) and genome-wide association studies (GWAS). Charts show the enrichment in total PSD (dark green), in PSD without nodes regulated by the induction of LTP (gray), and in nodes only regulated after the induction of LTP (regulated includes both phosphorylated and dephosphorylated). Note that without the regulated nodes, there is no significance, so the gray bars are barely or not visible. (C) Bar charts showing the enrichment for autism spectrum disorder and schizophrenia with highly connected nodes. Note that without the regulated nodes, there is no significance, so the black bars are not visible.

Because proteins regulated by the induction of LTP showed a differential clustering in the PSD protein-protein interaction network, we asked whether the risk for psychiatric disease also correlated highly connected nodes in the clusters. SNVs for schizophrenia and autism spectrum disorder (total and recurrent) and schizophrenia GWAS categories were all enriched in highly connected nodes (Fig. 6C). Because our data indicated that the clusters were highly phosphorylated after the induction of LTP, we analyzed whether regulated proteins represented the principal component of the risk. As expected, proteins modulated by the induction of LTP were the principal component of psychiatric risk in highly connected nodes (Fig. 6B and table S7); removal of the regulated nodes eliminated significant enrichment. Therefore, not only are the proteins that are regulated by the induction of LTP enriched in PSD risk for these psychiatric diseases, but they also cluster in highly connected protein-protein interaction networks that increase protein phosphorylation after HFS of the CA1 area of the mouse hippocampus.

To address whether the induction of LTP modulated the composition of protein complexes enriched in psychiatric disease risk factors, we determined the composition of Dlgap1 and SynGAP1 protein complexes from the CA1 area of the mouse hippocampus in control and LTP samples by immunoisolation of the target protein and quantitation by LC-MS/MS (Materials and Methods). We selected Dlgap1 as a representative of a core PSD scaffold of the middle layer (Fig. 5A) and SynGAP1, which is a node that interacts with components of all the three layers of PSD scaffolds (35, 36). Quantification of protein complexes showed that the induction of LTP reshaped protein interactions of highly connected and phosphorylated nodes (Fig. 7A). Dlgap1 protein complexes showed an increased binding in kinases (for example, Camk2b, Gsk3b, and Mapk1) and phosphatases (for example, Ppp3ca) upon induction of LTP. Moreover, this scaffold protein showed an increased association to the core scaffold machinery of the PSD composed of Dlgs and Shanks. Moreover, SynGAP1 showed a shift from upstream components, such as DLGs, to downstream components, such as Shanks, and exhibited changes in the association to guanine nucleotide exchange factors, GTPase activators, and components of the MAPK signaling cascade.

Fig. 7 Modulation of PSD protein interactions by induction of LTP.

(A) LTP-induced changes in the protein interactions involving SynGAP1 or Dlgap1 and whether the interacting protein exhibited an LTP-induced change in phosphorylation status (increased or decreased). (B) Cartoon showing the scheme of protein interactions that are modulated by the induction of LTP with increased scaffold associations, redistribution of Syngap1 within all three scaffold layers of PSD, and clustering of signaling molecules (represented by kinases and phosphatases) together with phosphorylated domain ligand readers (represented by 14-3-3 proteins) within the PSD network.

In accordance with the increased phosphorylation of 14-3-3 binding motifs observed after the induction of LTP (table S4), SynGAP1 and Dlgap1 protein complexes exhibited increased association of some 14-3-3 proteins (Ywhab, Ywhae, Ywhag, and Ywhah) (Fig. 7A). Therefore, these results suggested that, whereas the core scaffold machinery of the PSD surrounding Dlgap1 seems to increase its structural association to NMDAR signaling and to cluster kinases and phosphatases, SynGAP1, a signal transmitting molecule, begins to reduce its association to scaffolds to enable downstream signaling (Fig. 7B).


The regulation of synaptic phosphorylation networks by synaptic plasticity requires the fine-tuning of crosstalk signaling between the PSD kinases, PSD phosphatases, and protein interaction domains clustered in the PSD protein interaction network. Here, using large-scale phosphoproteomics and protein interaction assays, we showed that the induction of LTP modulates a highly connected PSD signaling network at the CA1 area of the mouse hippocampus. The regulation of protein phosphorylation within the PSD had distinctive characteristics. First, the induction of LTP modulated the phosphorylation state of 570 phosphorylation sites within 222 proteins at the mouse CA1 PSD, with an equal distribution of protein phosphorylation and protein dephosphorylation. Second, LTP induction regulated a combination of writer-reader-eraser toolkits using kinases distributed throughout the kinome to modulate protein interactions through phosphorylation of protein domains and short-sequence linear motifs. Third, highly connected components within the PSD protein interaction networks tended to exhibit increased phosphorylation relative that of the entire PSD and represented functional modules that were associated with two psychiatric diseases associated with altered synaptic function. Last, these nodes modulated protein interactions within the PSD core protein interaction network, reshaping the network composition after induction of LTP.

We determined that the mouse CA1 PSD kinome contained 81 protein kinases that were distributed throughout the mouse kinome tree. The large number of phosphorylated sites identified (570) combined with the relatively small number (14) of observed changes in phosphorylation state consistent with kinase activation indicated that each activated kinase phosphorylated many sites in the PSD. Modulation of basophilic kinases distributed in the AGC and CAMK families correlated with the increase in phosphorylation of mainly channels and receptors (upstream components of PSD network). Many AGC and basophilic kinases have functional protein domains other than the kinase domain (28). These domains can be used to regulate kinase activity in response to glutamate receptor and channels through the modulation of second-messenger production, including cAMP, Ca2+, DAG, and PI. Moreover, this kinase group can also activate downstream cascades by regulating proline-directed kinases, either through direct kinase or RASGAP phosphorylation (fig. S3). LTP induction also increased the activity of STE kinases. This family of kinases can also function as a link between the upstream components (such as the AGC kinases and channels) and the downstream proline-directed kinases (28). Our data indicated that proline-directed kinases were responsible for the phosphorylation of downstream and structural (cytoskeleton and cell adhesion) components of the PSD; both basophilic (AGC and CAMK families) and proline-directed (the CMGC family) kinases phosphorylated the scaffold architecture of the PSD. Thus, the induction of LTP promotes extensive crosstalk signaling through the PSD kinome, providing feedback loops with the receptor and channel outputs and modulating downstream kinase pathways mediated by different families of protein kinases (fig. S3).

We also found that a similar pattern of phosphorylation occurred with ligand domains. WW and SH3 ligands were enriched in cytoskeletal and scaffold proteins and phosphorylated at sites targeted by proline-directed kinases. However, pSer/pThr ligands (recognized by 14-3-3 and FHA) were phosphorylated at sites targeted by basophilic kinases, and these domains were present in channels, receptors, and scaffold proteins. This suggested a close correlation between individual kinase classes and the types of binding interaction modules that they regulate. Analysis of the effects of chemical stimulation of NMDAR on the CA1 region also yielded a similar pattern of kinase family–substrate function (13). However, the two experimental stimulation conditions, pharmacological NMDAR activation (13) and electrophysiological LTP induction (here), produced different patterns of protein phosphorylation, and the activation of individual kinases and phosphatases within families are input-specific. Therefore, this indicates a model in which the PSD machinery uses a core architecture of kinase-substrate groups to produce different patterns of protein phosphorylation by input-specific regulation of kinase and phosphatase activity.

Although here we show that the induction of LTP regulates the phosphorylation rate of more than 500 PSD phosphorylation sites, we observed an unequal distribution of sites within the core of the PSD protein interaction network. When compared to the total PSD, this core component had an increased amount of protein phosphorylation, which was due to the increased phosphorylation of highly connected components of the PSD, which included the scaffold machinery that connects glutamate receptors to downstream enzymatic signaling. These highly connected components are phosphorylated in domain ligands and protein domains, suggesting that the induction of LTP affected phosphorylation-regulated protein-protein interactions mediated by the core scaffold components. As evidence for this case, immunoisolation of protein complexes from scaffolds (Dlgap1) and scaffold interactors (SynGAP1) shows that the induction of LTP promotes the recruitment of basophilic protein kinases to scaffolds and directs a redistribution of protein interactions within the PSD. Because we focused on only two protein complex clusters, it is likely that this scenario will be extended to other components of the core PSD. In particular, this is likely to occur in processes leading to LTP induction and maintenance, involving cytoskeletal remodeling of the PSD.

A more general question about the role of phosphorylation sites regulated by LTP is whether they represent relevant functional sites. Recent advances in psychiatric genetics started to reveal the genetic architecture of psychiatric disorders. Within this polygenetic disease, a number of psychiatric disease risk factors have been described to be components of the postsynaptic site, supporting a role of the PSD in the etiology of these disorders. Here, we show that the PSD-associated risk for autism spectrum disorder and schizophrenia is not distributed evenly throughout the PSD but is localized to the set of proteins that are regulated by phosphorylation after the induction of LTP. Moreover, the principal component of the psychiatric disease risk was also distributed within the highly connected nodes modulated by the induction of LTP. Therefore, our results suggest that highly connected components that exhibit a change in phosphorylation status in response to the induction of LTP are responsible for the psychiatric disease risk of the PSD.

Here, we provide data related to how the induction of LTP modulates protein phosphorylation and protein interaction networks at the mouse CA1 PSD. It will be important to address temporal phosphorylation and protein interaction networks that are specifically involved in the different phases of LTP: induction, expression, and maintenance. It is likely that a number of the observed regulated sites will mediate the induction of LTP, and they may represent intermediate signaling events that translate initial synaptic inputs into the redistribution of glutamate receptors. Dephosphorylation of cytoskeletal components of the PSD, principally at PP2-targeted sites, also occurred, and some of these events may occur at the interface between induction of LTP and maintenance of LTP, during which cytoskeletal components are likely to be important for mediating the redistribution and membrane insertion of AMPA receptors. These results provide a framework for future studies to begin to determine the combinatorial properties of these protein phosphorylation networks and map them to specific patterns of synaptic activity, receptor activation, and synaptic “memory” or previous activity.


Hippocampal slice preparation and stimulation

Standard techniques approved by the University of California, Los Angeles Institutional Animal Care and Use Committee were used to obtain hippocampal slices from 8- to 12-week-old C57Bl/6 mice. Briefly, animals were deeply anesthetized with isoflurane and sacrificed by cervical dislocation. The brain was then rapidly removed and placed into a cold (4°C), oxygenated (95% O2/5% CO2) artificial cerebrospinal fluid (ACSF) containing 124 mM NaCl, 4.4 mM KCl, 25 mM Na2HCO3, 1 mM NaH2PO4, 1.2 mM MgS04, 2 mM CaCl2, and 10 mM glucose. Hippocampi from both hemispheres were dissected free from the rest of the brain, and a tissue slicer (Stoelting Co.) was used to cut 400-μm-thick slices. The dentate gyrus, CA3 region, and subiculum were then removed from each slice by manual dissection to obtain “mini-slices” containing the CA1 region. A small portion of the molecular layer of the dentate gyrus near the hippocampal fissure was left attached to avoid cutting the apical dendrites of the CA1 pyramidal cells. Mini-slices were then transferred to an interface-type chamber perfused with ACSF (2 to 3 ml/min, 30°C) and allowed to recover for 2 hours. Twenty mini-slices containing the CA1 region were used in each triplicate assay.

For each stimulation, a bipolar stimulating electrode fabricated from Formvar-insulated nichrome wire (A-M Systems Inc.) was placed in stratum radiatum near the center of the slice, and an extracellular recording electrode filled with ACSF (resistance, 5 to 10 megohms) was placed in the pyramidal cell layer near one of the ends of the mini-slice. The stimulation intensity was adjusted to evoke about a 5-mV population spike (measured as the difference between the first positive peak of the excitatory postsynaptic potential and the peak amplitude of the negative-going population spike). After recording four to five baseline responses (stimulation rate, 0.1 Hz), slices received two trains of HFS (100 Hz for 1 s) with an intertrain interval of 10 s. Single pulses of presynaptic fiber stimulation delivered 5 s, 2 min, and 4 min after HFS were then used to monitor the effects of HFS on synaptic transmission. (Slices exhibiting a less than twofold potentiation of the population spike 5 s after HFS were discarded.) Slices were collected 5 min after HFS and snap-frozen by transferring them into a prefrozen 1.5-ml microcentrifuge tube maintained in a bed of crushed dry ice. After each slice stimulation, an unstimulated mini-slice from the same chamber was also collected to provide control tissue.

Biochemical analysis of the PSD preparations

Mini-CA1 hippocampus slices from adult (14 to 18 weeks old) mice were pooled from post-HFS or control samples, and PSD fractions were prepared as described (12, 13). PSD-enriched fractions were prepared using a three-step protocol. Hippocampal slices were homogenized in 10 mM Hepes buffer (pH 7.4), containing 2 mM EDTA, 5 mM sodium orthovanadate, 30 mM NaF, 20 mM β-glycerol phosphate, and Roche cOmplete as protease inhibitor cocktail. Homogenates were spun for 4 min at 500g, supernatant was collected and centrifuged at 10,000g, and the membrane fraction was solubilized in 50 mM Hepes (pH 7.4), containing 2 mM EGTA, 2 mM EDTA, 30 mM NaF, 5 mM sodium orthovanadate, 20 mM β-glycerol phosphate, Roche cOmplete, and 1% Triton X-100. Solubilized membranes were centrifuged at 30,0000 rpm in a Beckman Optima Max rotor MLA-130 for 40 min, and the pellet was collected and solubilized in 50 mM tris (pH 9), 30 mM NaF, 5 mM sodium orthovanadate, 20 mM β-glycerol phosphate, 20 μM ZnCl2, Roche cOmplete, and 1% sodium deoxycholate. Before proceeding to MS analysis, a fraction of the samples were analyzed for protein phosphorylation status, using GluR1 pS831, Erk2 pT202/PY204, and CamkII pT286 as positive controls (fig. S1). Validation for phosphorylation sites included phosphorylation of PSD kinases, glutamate receptors subunits, and core PSD scaffold proteins, which were assayed at least in triplicate samples (fig. S1A). Total protein content within stimulated and nonstimulated samples was included to determine changes in the content of proteins within the PSD after induction of LTP. Enrichment and quality of PSD fractions were monitored by analyzing several protein controls, including glutamate receptor subunits, presynaptic markers, cytoplasmic proteins, nuclear proteins, and core PSD scaffolds (fig. S1B).

Antibody validation and immunoprecipitation

Antibodies for immunoprecipitation were screened for specificity by analyzing samples from mutant mice as negative controls (fig. S4). Antibody concentrations were standardized for optimum protein recovery within ranges 0.4 to 0.8 μg/μl. Immunoprecipitations were performed as described in Coba et al. (12) with minor modifications. Protein interactions identified in the immunoisolates analyzed by MS were considered positive if at least two unique peptides were present in triplicate assays and absent in controls.

MS analysis

PSD samples were separated by NuPAGE Novex 4-12% Bis-Tris Gels and subsequently fixed and then stained with InstantBlue. Lanes were cut and placed into 96-well plates for destaining and digested by tripsin at 37°C for 1 hour. Peptides were then extracted with acetonitrile. Peptide desalting and reverse-phase separation of peptides were performed using the Nano/Capillary LC System UltiMate 3000 (Thermo/Dionex).

For phosphopeptide enrichment, 40 μl of freshly prepared TiO2 slurry {10.0 mg/ml in loading buffer [65.0% acetonitrile (ACN)/2.0% trifluoroacetic acid (TFA)]} was used for each of the 12 fractions from in-gel digestion and was added into the sample tube, which contained the desalted and dried peptide sample. One hundred sixty microliters of loading buffer was then added, and the sample was incubated at room temperature for 1 hour with shaking. After a brief centrifugation, the supernatant was collected in a separate low-binding tube, and 200 μl of loading buffer was added into the original tube. After a 30-min incubation and brief centrifugation, the supernatant was again collected in the same tube with the previous supernatant. Then, the beads were washed twice with 200 μl of wash buffer 1 (65.0% ACN/0.5% TFA) and twice with wash buffer 2 (50.0% ACN/0.1% TFA). After each centrifugation, the supernatant was collected in the same collecting tube. We keep these fractions in case the enrichment is not successful. The peptide sample was eluted successively with 200 μl of elution buffer 1 (50.0% ACN/0.3M NH4OH) and 200 μl of elution buffer 2 (5.0%ACN/0.3M NH4OH) (each incubation was at 45˚C with vigorous shaking for 1 hour), and the two supernatant fractions were collected and pooled in a new low-binding tube. The eluate was partially dried in a SpeedVac at 35˚C until the sample volume was less than 10 μl. It was then reconstituted with 5.0% ACN/0.1% FA to 50-μl total volume and was ready for LC-MS/MS analysis.

MS data were processed using Proteome Discoverer (PD) v1.4 (Thermo Scientific) and searched by both Sequest and Mascot v2.4 (Matrix Science) against a modified mouse database, which was downloaded from UniProt and was combined with its decoy database. The mass tolerance used for searching was set as 10 parts per million for precursor ions and 0.8 dalton for fragment ions. No more than two missed cleavage sites were allowed. Static modification was set as cysteine carboxyamidation, and dynamic modification was set as methionine oxidation. False discovery rates (FDRs) were automatically calculated by the Percolator Node of PD based on decoy database hits. A peptide FDR of 0.01 was used for cutoffs. Peptides with high confidence were considered as true hits, and proteins with at least two different peptides were accepted.

For semiquantitative analysis, in-house–written Perl scripts were used to extract and analyze data from the output files of PD database search results. Normalized spectral abundance factor (NSAF) was calculated for each protein and was used for comparison between different samples. For protein interactions assays, a ratio ≥1.5 or ≤0.67 in triplicate samples was considered a biological change.

For phosphorylation analysis, MS data were processed as described for general data analysis, and in-house–written Perl scripts (available upon request) were used to extract and analyze data from the output files of PD database search results. NSAF was calculated for each protein and was used for comparison within triplicates and between different samples. A ratio ≥2 or ≤0.5 was considered a biological change.

For phosphoproteomic analysis, the phosphoRS 3.1 node of PD was used for phosphopeptide identification and phosphorylation site localization. A phosphoRS site probability of 0.75 was used as the cutoff for site assignment. The search results were further exported and analyzed by in-house–written Perl scripts, which organizes data, extracts the same phosphorylation sites from different peptides, and calculates their spectra counts for semiquantitation and comparison.

Testing for enrichment within data sets

For a given domain, we tested whether the domain or site or motif is enriched in the data set. The statistical significance, P value, is calculated as the conditional probability that the domain or site or motif occurs at least the observed number of times given that it occurs at least once. More precisely, let N be the total number of proteins and M be the number of proteins containing the query. Let n be the number of proteins in the data set. Under the null hypothesis that the query is not enriched in the data set, the number of proteins containing the domain or site or motif follows a hypergeometric distribution H(M, N, n). Let m be the number of proteins in the data set that contain the domain or site or motif. Then, the P value is calculated by


We used Bonferroni procedure to adjust for multiple testing. That is, only domains or sites or motifs with a P value less than α/T are considered significant, where T is the number of different domains in the complex. A domain or protein is considered enriched for P < 0.05.

Statistical overlap between sets of molecules

The statistical significance of an overlap between two sets of molecules was calculated as described in Coba et al. (13). Suppose that of N molecules, na and nb belong to sets {a} and {b}, respectively. If a and b are randomly distributed throughout N, then the probability of finding nab molecules belonging to both sets is given by the function

h(nab,na,N,nb)=na!(N-na)!nb!(N-nb)!/[N!(na-nab)!nab!(N-na-nb + nab)!(nb-nab)!].

Given the actual number of molecules μab belonging to both annotations, we estimate its significance by calculating the probability Pab) of an overlap as or less likely under the random distribution


This test was used to analyze which kinases or kinase families catalyze the phosphorylation at motifs identified as phosphorylated in the data or which kinases or kinase families catalyze the phosphorylation of substrates classified by the functional group that were identified as phosphorylated in the data. The approach was applied to both sites that have been reported in the literature as phosphorylated by a particular kinase or kinase family and those motifs that are predicted to be recognized by a specific kinase or kinase family based on the presence of the phosphorylated residue occurring within a consensus motif for the kinase or kinase family.

For analysis of short sequence motifs, the significance was established by random permutations of amino acid sequences as described in Coba et al. (13). Amino acids surrounding phosphorylated sites (constant) were randomly permutated, using regular expression matching to count sites present in each motif. Random permutations were used to estimate the probability of a motif to be present m times as Nm/(number of permutations), for m greater or less than the expected number of occurrences. Nm is the number of random permutations in which the motif occurred m or more (less) times. The same method was used to analyze the number of modulated peptides containing predicted or known phosphorylation sites within functional motifs.

Network clustering was performed with the Newman and Girvan algorithm (39), using edge betweenness, as described in Coba et al. (13), to represent the nodes’ shortest path and degree. A shortest path is defined as a path between two nodes such that the number of its constituent edges is minimized. Cytoscape was used for network visualization (47). To determine the network parameters, we excluded from the analysis the targets (Dlg4, Syngap1, Dlgap1, and Shank3).

For gene set enrichment, the following lists of genes were used: (i) supplementary table 16 from Turner et al. (44), which encompasses a list of 57 genes found to have recurrent de novo likely gene disrupting (LGD) mutations in probands with autism. This data set includes previous exome sequencing studies by Iossifov et al. (41), O’Roak et al. (48), and De Rubeis et al. (45). (ii) Supplementary table 7 from Iossifov et al. (43), which lists 353 genes harboring validated de novo LGD mutations in probands with autism. (iii) A subset of supplementary table 7 from Iossifov et al. (41), which lists 27 recurrent de novo LGD mutations. (iv) Supplementary table 3 from De Rubeis et al. (45), which includes 107 genes with an FDR < 0.3 implicated in contributing to autism through the TADA statistical model. (v) A list of genes implicated in autism as curated by SFARI gene database (; as of February 2016). This list was further restricted to genes with SFARI gene rankings 1 to 4 (strong evidence to minimal evidence) and S (syndromic) to account for genes with high confidence. (vi) Supplementary table 1 from Fromer et al. (38), containing de novo SNVs identified in probands with schizophrenia, excluding silent mutations. (vii) Supplementary table 3 from Ripke et al. (40), containing loci reaching genome-wide significance in patients with schizophrenia. This table was split into three groups (genes contained in all loci reaching genome-wide significance, loci implicating a single gene, and all genes falling under loci that do not implicate a single gene). Lists of proteins analyzed for enrichment include proteins present at the PSD (table S2); proteins present at the PSD without those found regulated by phosphorylation in response to LTP induction (the data in table S2 minus the data in table S1); clusters of proteins generated from protein-protein interaction data from the immunoprecipitation of Dlgap1, Dlg4, Shank3, and Syngap1 (table S5); and these clusters without proteins phosphorylated upon induction of LTP. For all tests, the number of protein-encoding genes in the human genome served as the background, measured by the two-tailed Fisher’s exact test using the (fisher.test) package in R.


Fig. S1. Validation of the LTP-induced phosphoprotein changes and enrichment of the PSD preparation.

Fig. S2. Kinase enrichment for short sequence motifs that mediate phosphorylation-dependent protein interactions.

Fig. S3. Model of the regulation of protein kinase and phosphatase signaling after induction of LTP in the CA1 area of the mouse hippocampus.

Fig. S4. Validation of the specificity of the antibodies recognizing the scaffold proteins.

Table S1. Total phosphorylation sites detected in the CA1 area of the mouse hippocampus.

Table S2. Proteins detected in PSD preparations of unstimulated mouse hippocampus CA1 region.

Table S3. Kinases matched to the phosphorylation sites differentially regulated after induction of LTP.

Table S4. Protein domains and predicted kinases and predicted domain ligand motifs present in short disordered regions with phosphorylated sites altered by the induction of LTP.

Table S5. Protein interactions and network clustering of the proteins that immunoprecipitated with four different PSD scaffold proteins.

Table S6. Disease association of proteins regulated by induction of LTP.

Table S7. Enrichment of proteins in the PSD and in scaffold complexes for mutations associated with autism spectrum disorder and schizophrenia.


Acknowledgments: We thank K. Wang for statistical analysis comments and discussion. Funding: This work was supported by grants from the National Institute of Child Health and Human Development (MH104603-01A1; to M.P.C.) and National Institute of Mental Health (MH060919; to T.J.O.). Author contributions: J.L., B.W., V.A.C., and T.J.O. performed experiments; J.L. and M.P.C. analyzed MS data; J.H. developed software tools for MS analysis; and J.L., M.P.C., and T.J.O. designed the project and wrote the article. Competing interests: The authors declare that they have no competing interests. Data and materials availability: Scripts and data have been deposited (1-20160721-20534) to the ProteomeXchange Consortium through the PRoteomics IDEntifications partner repository.
View Abstract

Navigate This Article