DRACC1, a major postsynaptic protein, regulates the condensation of postsynaptic proteins via liquid-liquid phase separation

Numerous proteome analyses have been conducted on the postsynaptic density (PSD), a protein condensate beneath the postsynaptic membrane of excitatory synapses. Each has identified several hundred to thousands of proteins. While proteins with predictable functions have been well studied, functionally uncharacterized proteins are mostly overlooked. In this study, we perform a meta-analysis of the 35 PSD proteome datasets, including 5,869 proteins, identifying 97 uncharacterized proteins that appeared in multiple datasets. We focus on the top-ranked protein, FAM81A, renamed DRACC1. DRACC1 is expressed in forebrain neurons and enriched at the synapse. DRACC1 interacts with PSD proteins, including PSD-95, SynGAP, and NMDA receptors, and promotes liquid-liquid phase separation of those proteins. Consistently, the downregulation of DRACC1 in neurons causes a decrease in the size of PSD-95 puncta and the frequency of neuronal firing. Our results characterize DRACC1 as a novel synaptic protein facilitating the assembly of proteins within PSD. It also indicates the effectiveness of a meta-analytic approach of existing proteome datasets in identifying uncharacterized proteins.

As the technologies of proteome analysis advance, the number of proteins detected in the PSD fraction increases and reaches the order of thousands. Whereas many are bona fide PSD proteins from their known function and distribution, others are apparent contamination (such as proteins in glial cells or presynaptic compartments). In addition, multiple proteins without known functions or localization have been overlooked and left out of the in-depth analysis because it is difficult to determine whether they are contaminants or unreproducible.
Here, we performed meta-analyses of the multiple published datasets obtained under different experimental settings, assuming that those proteins identified in multiple datasets are authentic PSD proteins. To prove the validity of this approach, we identified and performed in-depth characterization of the top-ranked protein, Fam81, renamed as Disordered Region And Coiled-Coil Domain 1 (DRACC1). Our data confirmed that DRACC1 is indeed a functionally crucial postsynaptic molecule modulating the condensation of other PSD proteins. This approach of meta-analysis of multiple proteome datasets will help identify bona fide proteins in a sample that is otherwise overlooked from a single dataset.

Meta-analysis of PSD proteome datasets
To evaluate the published PSD proteome datasets, we analyzed 35 datasets having at least 30 proteins (Table 1), of which 20 are from biochemically isolated PSD fractions followed by mass-spectrometric analysis (unbiased, Table S1, Fig. 1A) and 15 are by immunoprecipitation, pull-down, or proximal labeling of known PSD components (candidate-based, Fig. 1B). 5,869 proteins were detected at least in one dataset, where 5,800 proteins are in biochemical fractionation studies and 995 proteins in other studies (Fig. 1C). Overall, the datasets from biochemical fractionation studies showed a high overlap of identified proteins, although they are derived from various samples different from the species, brain region, and purification protocol. In contrast, immunoprecipitation, pull-down, or proximal labeling datasets showed relatively low overlap. The low overlap is observed even if the same starting protein (such as GluN2B or PSD-95) was used, suggesting the stochasticity of these approaches.
These proteins may include contaminants from non-PSD proteins, as only 1.2% of them are known PSD proteins based on GO annotation (Fig. 1E). In contrast, proteins detected reproducibly in multiple datasets contain a higher fraction of known PSD proteins. When we analyzed 123 proteins detected in more than 20 datasets, we found that about 40% of them are PSD proteins based on GO annotation, suggesting that PSD proteins are highly enriched in this group (Fig. 1E). As expected, proteins detected in even higher (>25) number of datasets are composed of well-known core PSD proteins, including MAGUK family proteins and glutamate receptor subunits (Fig. S1). Our meta-analysis of PSD proteome datasets for the proteins recurrently detected in the multiple datasets successfully identifies known core PSD proteins.

Identification of uncharacterized proteins from PSD proteome datasets
We asked whether this approach can detect proteins that have never been studied as PSD proteins. We first extracted protein names based on cloning ID or chromosome region (FamXX and XXXX…Rik) (Team et al., 2001;Wain et al., 2002). In addition, we also searched for proteins named after specific domains; Tmem for a transmembrane domain, Ccdc for a coiled-coil domain, Cctm for both of them, and Zfp for a zinc finger domain (Brayer and Segal, 2008;Marx et al., 2020;Priyanka and Yenugu, 2021;Schapira et al., 2017;Sohn et al., 2016). As a result, 177 proteins were identified from a total of 5,869 proteins (Table S2). Among them, 97 proteins were detected in at least 2 datasets, and 21 proteins appeared in more than 8 datasets ( Fig. 2A). We focused on the top-ranked protein, FAM81A, which was detected in 21 datasets, including 15 PSD fractionations and 6 other studies ( Fig. 2A and 2B).
Although FAM81A is reported to localize to the PSD (Dosemeci et al., 2019), no further characterization has been reported about this protein. We, therefore, decided to focus on FAM81A to demonstrate the usefulness of the meta-analytic approach of multiple proteome databases.
Analysis of domain architecture using the Simple Modular Architecture Research Tool (SMART) revealed that human FAM81A and FAM81B possess 2 and 4 coiled-coil domains, respectively (Fig. S2). In addition, we found that they contain regions predicted to be intrinsically disordered (Fig. S2). Considering these predictions, we named FAM81A and FAM81B as Disordered Region And Coiled-Coil Domain 1 (DRACC1) and DRACC2, respectively, and the unique orthologs in amphibians, fish, and invertebrates as DRACC.
Human DRACC1 has more than 90% sequence identity with their mouse ortholog, which is comparable to other major PSD proteins (Fig. 2D). By contrast, human DRACC1 and DRACC2 have only 34% and 37% sequence identity with zebrafish DRACC, respectively, whereas other PSD proteins have 60-80% identity with their zebrafish orthologs (Fig. 2D). These data suggest that DRACC1 evolved rapidly in the vertebrate lineage compared to other PSD proteins. Taken together, DRACC1 is a PSD protein that evolved specifically in higher vertebrates.

DRACC1 is heterogeneously distributed in PSD within different brain regions
DRACC1 and DRACC2 are expressed explicitly in the brain and testis in humans and mice, respectively ( Fig. 2E and S3A). DRACC in amphibians is expressed in a broad range of tissue, suggesting that the role of DRACC1 and 2 are differentiated during evolution (Fig. S3B). We then asked whether the abundance of DRACC1 is different across brain regions. In the mouse brain, expression of Dracc1 is limited to the forebrain and hardly detectable in the cerebellum and brainstem (Fig. S3C). In addition, the expression of Dracc1 in the cortex, hippocampus, and olfactory area is about 5-times higher than that of the striatum, pallidum, thalamus, and hypothalamus. This is in marked contrast to the global gene expression of PSD-95 (Dlg4) (Fig. S3C). Our PSD proteome data of the marmoset brain also showed region-specificity of DRACC1 expression on the PSD. PSD in the neocortex and hippocampus showed a high abundance of DRACC1 compared to other regions (Fig.   S4A, S4B). PSD proteome dataset of the human neocortex shows that DRACC1 is heterogeneous across Brodmann areas (Roy et al., 2018b). The abundance of DRACC1 is relatively high in the frontal and temporal cortex compared to the parietal and occipital cortex (Fig. S4C, S4D). These data indicate that DRACC1 is heterogeneously expressed in PSD across brain regions.

DRACC1 is enriched in dendritic spines in neurons
We then tested the subcellular localization of DRACC1. The endogenous DRACC1 was enriched in the detergent-resistant PSD fraction, similarly to the bona fide PSD protein, PSD-95, whereas a presynaptic protein, synaptophysin, was depleted from the fraction (Fig. 3A). In hippocampal neurons, exogenous DRACC1-GFP was colocalized with PSD-95-mCherry at dendritic spines (Fig. 3B). These results confirmed that DRACC1 is enriched in PSD, consistent with the meta-analysis of the proteome datasets as well as an earlier immunoelectron microscopic study (Dosemeci et al., 2019).
To observe the dynamics of DRACC1 in neurons, we performed time-lapse imaging of DRACC1-GFP expressed in hippocampal neurons from DIV (day in vitro) 16 to 18 (Fig. 3C, Movie S1). As the neuron matures and gains mushroom-shaped dendritic spines, DRACC1 was condensed in the structure (Fig. 3D, 3E). In addition to accumulation at dendritic spines, we found the condensates of DRACC1 in both soma and dendritic shafts (Fig. 3F, 3G). The time-lapse observation shows that the condensations are moving rapidly compared with those in PSD, which is rather stable (Fig. 3F,   3G).

Domain structure of DRACC1 required for condensation and synaptic accumulation
We attempted to reproduce the condensate in a heterologous system and found that it can also be reproduced in HEK293T heterologous expression systems, where synaptic proteins are hardly expressed. This suggests that DRACC1 has a propensity to condensate without other synaptic molecules ( Fig. 4A, Movie S2). To examine the molecular architecture of DRACC1 important for the condensation, we asked for the sequence motif required for condensation by generating a series of deletion mutants and expressing them in HEK293T cells (Fig. 4B, Fig. S5). C-terminal half (C-half; 188-364) completely abolished the puncta formation whereas N-terminal half (N-half; 1-187) still formed puncta, indicating that the N-terminal half is essential for condensation. We then made smaller deletion mutants of the N-terminal half and found that all of these mutants resulted in decreased puncta formation ( Fig. 4A, 4C). In particular, Δ1-36 and Δ107-157 showed an approximately 90% decrease in the number of condensates. The size and/or maximum size of puncta of the Δ1-36 and Δ37-74 mutants were larger and more amorphous than that of full-length DRACC1, suggesting that these mutants form aggregates (Fig. 4A, 4D, 4E).Δ75-106, Δ107-157, and Δ158-187 showed a defect in the formation of enlarged structures and the size of Δ75-106and Δ158-187-positive structures were smaller than that of full-length DRACC1, suggesting that these regions contribute to the formation of DRACC1 droplets. These results indicate that the wide-ranged sequence of DRACC1 is important for condensation.
We then used the condensation-deficient mutant, Δ107-157, to test if condensation is required for synaptic accumulation in the hippocampal neuron. It showed diffused cytosolic localization and accumulation at PSD was significantly impaired compared with the full-length DRACC1 (Fig. 4F).
Together, these results suggest that condensation of DRACC1 is essential for its PSD localization.

DRACC1 forms condensates by liquid-liquid phase separation
Intriguingly, these puncta showed flexible shapes and underwent fusion and fission ( Fig. 5A-5C), indicating that these punctate DRACC1 structures have liquid-like properties rather than solid aggregates. We also observed larger structures, often with complicated shapes (Fig. 5D). The shape of such structures was stable over time, suggesting that they are rigid protein aggregates (Fig. 5D, Movie S3). From these observations, we came to the idea that DRACC1 forms these droplets through the mechanism of liquid-liquid phase separation (LLPS). To test this idea, we examined the effect of 1,6hexanediol that disrupts LLPS by interfering with hydrophobic interactions (Kroschwald et al., 2017) on DRACC1 droplets in HEK293T cells. Although large aggregate-like structures remained, most of the puncta disappeared after 10 min incubation, consistent with the idea that DRACC1 undergoes LLPS (Fig. 5E). We then tested whether the multimerization of DRACC1 is involved in LLPS. We found that GFP-DRACC1 was co-precipitated by DRACC1-FLAG expressed in HEK293T cells, suggesting that DRACC1 undergoes LLPS through multimerization (Fig. 5F).

DRACC1 interacts and forms condensate with core synaptic proteins
We then tested if DRACC1 can directly interact with the major PSD proteins. For this purpose, we first tested if DRACC1 can be co-immunoprecipitated with PSD-95, SynGAP, and GluN2B in HEK293T cells and found that these proteins indeed co-immunoprecipitate with DRACC1, indicating that they interact with DRACC1 (Fig. 5G). The interaction of DRACC1 with these proteins in nonneuronal cells, which hardly expresses synaptic molecules, suggests that DRACC1 directly binds to PSD-95, SynGAP, or GluN2B.
It has been shown that PSD-95 and SynGAP undergo LLPS when they are combined (Araki et al., 2020;Zeng et al., 2016). Consistently, we reproduced the condensation of GFP-SynGAP and PSD-95-mCherry in HEK293T cells upon co-expression (Fig. S6). Given the interaction of DRACC1 with both PSD-95 and SynGAP, we tested whether DRACC1 can undergo LLPS with SynGAP or PSD-95 in HEK293T cells. As a result, GFP-SynGAP and DRACC1-FLAG condensed together (Fig.   5H). On the other hand, PSD-95-mCherry co-expressed with DRACC1-GFP showed diffuse distribution, hardly localized with the punctate distribution of DRACC1-GFP (Fig. 5I). However, when all three were co-expressed, they formed condensate together, suggesting that PSD-95 needs SynGAP for condensation with DRACC1 (Fig. 5J). These results indicate that DRACC1 interacts with core synaptic proteins and co-localizes with SynGAP positive droplets.
To test if DRACC1 can undergo LLPS along with PSD-95, GluN2B, and SynGAP in vitro, we bacterially expressed and purified these proteins, combined, and observed them under a microscope.
As a result, DRACC1 formed condensate in combination with SynGAP, GluN2B, and PSD-95 (3 µM each). To examine the role of DRACC1 in forming condensate, we next decreased the concentration of DRACC1 to 1 or 0 µM while maintaining the concentration of other proteins (Fig. 6). Upon reducing the concentration of DRACC1, we found a decrease in the condensate size, as visualized in the PSD-95 channel. This indicates that DRACC1 can facilitate the condensate formation of PSD proteins through the assembling and stabilizing the component proteins.

DRACC1 affects PSD size and neuronal activity
To examine the role of DRACC1 on the formation of PSD in neurons, we performed a knock-down experiment of DRACC1 in the cultured hippocampal neuron. We expressed two different shRNAs against DRACC1 (shDracc1 #1 and #2) by using a lentivirus vector, both of which downregulated the mRNA level of Dracc1 to <10% (Fig. S7A). We then analyzed PSD-95 puncta on neurons, using GFP to visualize the dendrites of the infected neurons (Fig. S7B). The size of PSD-95 puncta is decreased in neurons expressing shRNA for DRACC1 (Fig. 7A, 7B), suggesting that DRACC1 stabilizes PSD-95 at the synapse. To test the physiological role of DRACC1, we next examined whether the neuronal activity is affected by DRACC1 downregulation using a multi-electrode array (MEA). We found a significant decrease in the frequency of neuronal firing in DRACC1 downregulated neurons, indicative of reduced excitatory synaptic transmission (Fig.7C, 7D). These results suggest the structural and functional importance of DRACC1 at the excitatory synapse.

Discussion
In this study, we performed a meta-analysis of the studies that identified PSD proteome and generated a versatile dataset to determine whether a given protein is in the PSD. In-depth, we characterized DRACC1, a PSD protein conserved in terrestrial vertebrates. Considering the recurrent detection of DRACC1 in the PSD proteome datasets in mammals, it is likely abundant in the PSD. Fish, amphibians, and some invertebrates have only one ortholog, DRACC. DRACC in fish is presumably not a PSD protein because DRACC was not detected in PSD proteome in zebrafish (Bayés et al., 2017), which indicates DRACC gained properties as a postsynaptic protein during evolution. We found that DRACC1 facilitates condensate formation in a dose-dependent manner. Consistently, the downregulation of DRACC1 results in decreased size of the PSD (Fig. 6, 7A, and 7B). The interaction of DRACC1 with multiple PSD proteins (Fig. 5G) and the self-interaction of DRACC1 (Fig. 5F) may be involved in the maintenance of PSD through the facilitation of LLPS between PSD molecules. The protein-protein interactions and LLPS of DRACC1 might be involved in the effect of DRACC1 on neuronal activity (Fig. 7C, 7D), as it has been proposed that synaptic activities are modulated through protein-protein interaction and LLPS in PSD (Bayer et al., 2001;Hosokawa et al., 2021;Liu et al., 2021;Saneyoshi, 2021;Saneyoshi et al., 2019). The heterogeneous expression of DRACC1 across brain regions (Fig, S3C, S4) might be involved in brain region-dependent differences in synaptic strength. Considering the limited evolutional conservation, DRACC1 may be involved in the complexity and diversity of synapses in higher vertebrates, which may be involved in the high cognitive function of these species.
DRACC1 has an intrinsically disordered domain (IDR) in its C-terminal end similarly to other proteins that undergo LLPS, such as TDP-43, FMRP, and CTTNBP2 (Shih et al., 2022;Tsang et al., 2019;Zbinden et al., 2020). Dynamics of DRACC1 in neurons observed by time-lapse imaging look similar to that of TLS/FUS , which is among the first identified and best characterized RNA-binding proteins in the field of phase separation (Zbinden et al., 2020). Although we currently have no data that DRACC1 condensates contain RNA granules, including TLS/FUS, DRACC1 may play a role in signaling between synapses and dendrites. Shaft-localized condensates of DRACC1 suggest that they may regulate local translation in response to various stimuli, including synaptic transmission Tsang et al., 2020) like TLS/FUS  or FMRP (Tsang et al., 2019) in addition to synaptic function.
A number of genetic studies point to the genes encoding excitatory synaptic proteins as causative genes for neuropsychiatric disorders such as schizophrenia and autism. We found that common variants of DRACC1 have been registered in GWAS Catalog (Buniello et al., 2019). These variants of DRACC1 (rs28890483 and rs10519005) are located at the 5' side of the DRACC1 gene, which may affect the expression level of DRACC1. They might increase the risk of schizophrenia, bipolar disorder, and alcohol dependence. In addition, DRACC1 is reported as one of the hub genes related to susceptibility to depression (Bagot et al., 2016), suggesting that the expression level of DRACC1, which may affect neuronal activity, is involved in neuropsychiatric disorders, including major depression.
In conclusion, we provided a novel meta-analytical approach to identifying uncharacterized proteins in a given protein fraction. Using this approach, we reported an uncharacterized protein, DRACC1. The same approach will be helpful in the identification of other proteins from several existing proteomic studies on both central and peripheral tissues.

Mice
ICR mice were purchased from Japan SLC, Inc. All protocols for animal experiments were approved by RIKEN Brain Science Institute and Kobe University and performed by institutional guidelines and regulations. After cervical dislocation, the brain was removed from mice, briefly rinsed with ice-cold HBSS (Hanks' Balanced Salt solution), frozen with liquid nitrogen, and stored at -80°C before use.

Preparation of PSD fraction
Preparation of the PSD-I fraction was performed according to the previously described protocol with minor modification (Carlin et al., 1980). Briefly, three brains obtained from adult (12-week-old) ICR mice were homogenized with a glass-Teflon homogenizer in Solution A (0.32 M sucrose, 1 mM NaHCO 3 , 1 mM MgCl 2, 0.5 mM CaCl 2 , and cOmplete EDTA-free Protease Inhibitor Cocktail). The homogenate was centrifuged at 1,400 g for 10 min at 4°C to obtain the pellet and the supernatant.
The pellet was resuspended in Solution A and centrifuged at 700 g for 10 min at 4°C. The supernatant of the first and second centrifugation was pooled as an S1 fraction and subjected to subsequent centrifugation at 13,800 g for 10 min at 4°C. The resulting pellet was resuspended with Solution B (0.32 M Sucrose and 1 mM NaHCO 3 ) and centrifuged in a sucrose density gradient (0.85/1.0/1.2 M) for 2 h at 82,500 g. Synaptosomes were collected from the 1.0/1.2 M border and diluted twice with Solution B. The synaptosome was lysed by adding an equal volume of solution C (1% TX-100, 0.32 M Sucrose, 12 mM Tris-HCl pH 8.1) and rotation at 4°C for 15 min. The sample was centrifuged at 32,800 g for 20 min at 4°C to obtain PSD-I as a resulting pellet.

PSD proteome analysis
Proteome analysis of PSD fraction obtained from mice (Dataset No.19) and common marmosets (Dataset No.20) was performed. For mouse analysis, PSD was prepared from the whole brains of 2, 3, 6, and 12-week-old ICR mice. For marmoset analysis, PSD was prepared from the neocortex, hypothalamus, thalamus, striatum, hippocampus, brainstem, and cerebellum of adult (24-month-old) marmoset. The samples were analyzed with Q Exactive HF-X mass spectrometer (Thermo Fisher Scientific) and Proteome Discoverer version 2.2 (Thermo Fisher Scientific). Proteins detected in all samples were listed and used for the meta-analysis described above.  For 1,6-hexanediol treatment, a regular medium containing 10% 1,6-hexanediol (Sigma-Aldrich, 240117-50G) was used.

Primary culture of cortical and hippocampal neurons
Hippocampi and cortices were dissected from E16.5 mouse embryos and dissociated using Neuron

Live imaging of hippocampal neurons
Cells seeded on 35-mm glass bottom dishes (Matsunami, D11130H) were observed using CellVoyager CV1000 (Yokogawa Electric) equipped with a 60x objective lens. During live imaging, the culture dish was placed in a chamber to maintain incubation conditions at 37°C with 5% CO 2 . Two-color time-lapse images were acquired at 30 min or 5-sec intervals for hippocampal neurons or HEK293T cells, respectively. In the observation of hippocampal neurons, a 20 µm range of Z-stack images (21 slices, 2 µm) were acquired. As for HEK293T cells, a 2 µm range of Z-stack images (3 slices, 1 µm) were acquired. Maximum intensity projection images were shown in the figures and the movies.

Immunoprecipitation and immunoblotting
For immunoprecipitation, HEK293T cells cultured on 10 cm dishes at ~50% confluency were transfected with 7.5 µg of DRACC1-FLAG and 7.5 µg of GFP tagged PSD-95, SynGAP, or GluN2B using PEI Max (Polysciences, 24765-1). 24 h after transfection, cells were washed with ice-cold PBS, collected with centrifugation (5,000 rpm, 2 min 4 °C), and resuspended with lysis buffer (1% Triton X-100, 50 mM Tris-HCl pH7.4, 150 mM NaCl, 1 mM EDTA, 15 mM NaF, 2.5 mM Na 3 VO 4 ) with cOmplete EDTA-free Protease Inhibitor Cocktail). Lysates were kept on ice for 10 min and then centrifuged with 15,000 g for 15min at 4 °C. The supernatant was subjected to immunoprecipitation using 20 µl of ANTI-FLAG M2 Affinity Gel (Sigma-Aldrich, A2220-5ML). After 90 min incubation at 4 °C, the samples were washed five times using lysis buffer, resuspended with sample buffer (62.5 mM Tris-HCl pH 6.8, 4% sodium dodecyl sulfate (SDS), 10% glycerol, 0.008% bromophenol blue, and 25 mM dithiothreitol), and then boiled for 5 min. The samples were subjected to SDSpolyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto Immobilon-FL polyvinylidene difluoride membranes (Millipore, IPFL00010). The membrane was blocked in blocking buffer (Trisbuffered saline with 0.1% Tween 20 (TBST) and 5% skim milk) at room temperature and then incubated with the indicated primary antibody overnight in blocking buffer at 4°C. The membrane was washed three times with TBST, followed by incubation with the respective secondary antibody in a blocking buffer for 1 h at room temperature. The membrane was washed five times with TBST.

Labeling and observation of LLPS of purified proteins
The DRACC1 protein was labeled by iFluor 488-or iFluor 568-succinimidyl ester (AAT Bioquest) as previously described  The PSD proteins were diluted in a phase buffer (50 mM Tris, pH 8.0, 100 mM NaCl, 1 mM TCEP, 0.5 mM EGTA, 5 mM MgCl 2 and 2.5 mM ATP). A protein mixture (5 μl) was injected into a homemade imaging chamber and observed by confocal microscopy (FLUOVIEW FV1200, Olympus). The number and size of the iFluor488-positive droplets were analyzed using Analyze Particles function of Fiji software.

Lentivirus production and infection
Lenti-X 293T cells were co-transfected with pLKO.1 lentiviral plasmid, psPAX2, and pMD2.G using Lipofectamine LTX and PLUS Reagents (Thermo Fisher Scientific, 15338-100). After overnight incubation, the medium was changed to a fresh medium. 60-72 h after transfection, the supernatant was collected and filtrated through a 0.45 µm filter (Millipore, SLHV033RS

Quantitative image analysis
1024x1024 pixel images obtained with confocal microscopy were analyzed using ImageJ. For quantification of DRACC1-GFP-positive structures in HEK293T cells, the signals were extracted by binarization using Find Maxima (noise tolerance: 120). The number and size of the signals in each cell were analyzed using Analyze Particles after selecting individual cells. Unhealthy (shrank) cells, cells with too high (saturated) or too low (invisible) signal intensity, and aggregated cells with unclear borders were avoided from the analysis. Images were first shuffled for blinded analysis to quantify PSD-95-positive structures in neurons expressing GFP. Then, images with single neurons were selected. PSD-95 positive structures were extracted by binarization using Find Maxima (noise tolerance: 120). Structures with more than 50 pixels were eliminated to avoid the detention of nonsynaptic structures. To assess dendrite length, signals of GFP were traced using NeuronJ (Pool et al., 2008). The PSD-95 puncta along the dendrites (within 25-pixel diameter) were extracted using SynapCountJ (Mata et al., 2016). The mean puncta size and the number of puncta per dendrite length were calculated for individual neurons. The result was visualized as a boxplot using R software.
Student t-test was used for statistical analysis.

Multi-electrode array (MEA)
The medium was changed with a fresh neuron culture medium without D,L-APV. 30 min after the medium change, neuronal activity was analyzed using MEA2100 (Multi Channel Systems) and MC_Rack Version 4.6.2. Input voltage range and sampling frequency were set as ±19.5 mV and 20000 Hz, respectively. Neuronal activity was recorded for 2 min. Voltage over 5 standard deviations was used as a threshold to detect spikes.

Declaration of interests
The authors declare no competing interests.

Table 1. List of the 35 PSD proteome datasets referred to in this study
No.1-20 summarise the proteome data of biochemically purified PSD fraction (Fraction) and no. [21][22][23][24][25][26][27][28][29][30][31][32][33][34][35] are that of the PSD protein complex (Complex). The original article, information on the sample and method, and the number of detected proteins of each dataset are described. Note that number of proteins can be fewer than that shown in the original article because proteins that failed ID conversion were eliminated.    antibody. Representative dendrite images (A) and the mean size of PSD-95 puncta (B) are described.
(C-D) Neurons were subjected to extracellular electrophysiological recording using MEA.
Representative signals (C) and spike frequency (D) are described. Scale bars: 50 m (A, white), 10 m (A, yellow), 10 sec (C, x-axis), and 100 V (C, y-axis). *P < 0.05       Scale bars: 100 m. Table S1. List of 5,869 proteins included in PSD proteome datasets Proteins were described as mouse Entrez Gene ID. 0 and 1 indicate undetected and detected, respectively. Table S2. List of 177 proteins in PSD proteome datasets that have not been fully characterized

Supplementary Movie 1. DRACC1 droplets in hippocampal neuron
Mouse primary hippocampal neurons were transfected with DRACC1-GFP and DsRed. Images were obtained every 30 min.