Wistar rats were exclusively from a commercial source (Janvier, St Berthevin, France). All animals were handled and treated in accordance with the ARRIVE Guidelines. Animals had free access to water and food. Lightning was controlled as a 12 h light and dark cycle and the temperature maintained at 23 ± 1°C. For brain retrieval, PND14 pups were euthanized by decapitation according to a protocol approved by the Animal Care and Ethics Committee and forebrains were immediately excised and used.

Synaptosomal preparation was adapted from previously published protocols (Loriol et al., 2014). Fresh forebrains of four PND14 rats were homogenized in 26 mL of ice-cold Sucrose Buffer (0.32 M Sucrose, 10 mM Tris–Hcl pH 7.4, complete EDTA-free protease inhibitor cocktail Roche) with 20 mM NEM to prevent proteins from deSUMOylation using a 30 mL glass-Teflon homogenizer with 12 gentle strokes. The homogenate (total fraction, Tot) was centrifuged at 1,000 g for 5 min at room temperature (RT) in a JA4.2 rotor (Beckman). The supernatant (cytoplasmic fraction, Cyt) was then layered by 2 mL on the top of a four-step Percoll-sucrose density gradient (2 mL of 20% Percoll, 2 mL of 10% Percoll, 2 mL of 6% Percoll, and 2 mL of 2% Percoll in Sucrose Buffer) and centrifuged at 18,000 rpm for 10 min at 4°C in a SW41Ti rotor (Beckman). Synaptosomal fraction was recovered at the 10–20% interface Percoll layers. This fraction was then washed in 10 mL of HEPES Buffer (5 mM HEPES, 140 mM NaCl, 3 mM KCl, 1.2 mM MgSO4, 1.2 mM CaCl2, 1 mM NaH2PO4, 5 mM NaHCO3, 10 mM Glucose, and complete EDTA-free protease inhibitor cocktail Roche, 10 mM NEM) and centrifuged at 10,000 g for 10 min in JA25.5 rotor (Beckman). The synaptosomal pellet was resuspended in Denaturing Lysis Buffer (20 mM Sodium Phosphate buffer pH7.4, 150 mM NaCl, 5 mM EDTA, 5 mM EGTA, 1% Triton X-100, 0.5% DOC, 1% SDS, complete EDTA-free protease inhibitor cocktail Roche, 10 mM NEM), supplemented with 50 mM DTT and boiled for 10 min at 95°C. Protein concentration was determined using standard Bradford assay.

For IP beads preparation, Protein-G Sepharose beads (Sigma, P3296) were washed three times and incubated in PBS for 45 min at RT with SUMO-2/3 mouse monoclonal antibody (Clone 12F3, Cytoskeleton) or mouse IgG antibody as a negative control. After two washes in 20 mM Sodium Phosphate buffer (NaPi) pH7 and two washes in Triethanolamine Buffer (200 mM Triethanolamine, pH 8.2), beads were incubated for 1 h at RT in 30 mM dimethylpimelidate (DMP) freshly resuspended in Triethanolamine Buffer to crosslink the antibodies to the beads. The crosslinking step was stopped by resuspension of the beads in 50 mM Tris–HCl pH8. Finally, the beads were washed once with 200 mM Glycine pH2.9 and twice with 20 mM NaPi pH8.

The denaturing immunoprecipitation protocol was adapted from Becker et al., 2013. Briefly, the denaturated synaptosomal fractions were diluted in Dilution Buffer (20 mM NaPi pH7.4, 150 mM NaCl, 5 mM EDTA, 5 mM EGTA, 1% Triton X-100, 0.5% DOC, and complete EDTA-free protease inhibitor cocktail Roche, 10 mM NEM) to obtain a final concentration of 0.1% SDS. The lysate was centrifuged for 10 min at 20,000 g at 4°C. 8 mg of proteins, corresponding to synaptosomes isolated from ten PND14 rat brains, were incubated with 160 μg of SUMO-2/3 mouse monoclonal antibody, or mouse IgG antibody as a negative control immobilized on 60 μL of Protein-G Sepharose beads, overnight at 4°C under soft rotation. Beads were washed three times for 5 min at 4°C in Wash Buffer (20 mM NaPi pH7.4, 150 mM NaCl, 5 mM EDTA, 5 mM EGTA, 1% Triton X-100, 0.5% DOC, 0.1% SDS, and complete EDTA-free protease inhibitor cocktail Roche). Beads were additionally washed in Wash Buffer containing 500 mM NaCl for 30 min at 37°C. Proteins bound to the beads were then eluted three times for 30 min at 37°C with 200 μL of Wash Buffer supplemented with 0.5 mg/mL of SUMO2 peptide (CQIRFRFDGQPINE). Eluates were pooled and subjected to TCA precipitation prior to quality controls and mass spectrometry analysis.

SUMO2/3-ylated proteins immunoprecipitation from total brain extracts were performed as described above with few modifications: Beads were coated after antibody immobilization in Coating Buffer (PBS 1X, BSA 5 mg/mL, Dextran 40 kDa, 5 mg/mL, and Gelatin 1 mg/mL). Proteins bound to the beads were eluted in 2X Laemmli buffer + 5% β-mercaptoethanol.

Samples from the SUMO2/3 immunoprecipitations and the associated IgG negative control were resolved by SDS-PAGE followed by silver staining (Silver Quest Kit; Invitrogen) for quality control or Coomassie staining (Imperial blue) prior to in gel trypsin digestion and mass spectrometry analysis. For immunoblotting, proteins separated by SDS-PAGE were transferred onto nitrocellulose membrane (BioTrace NT, Pall Corporation) and immunoblotted using the following primary antibodies at the indicated dilution: mouse anti-SUMO2/3 (clone 12F3, Cytoskeleton, ASM23) 1/1000, rabbit polyclonal anti-SUMO2/3 (Dr. Guillaume Bossis, IGMM, France) 1/750, mouse anti-SUMO1 (clone 21C7, DSHB) 1.5 μg/mL, mouse anti-Synaptotagmin (Stressgen, SYA-148) 1/2000, rabbit anti-PSD95 (Millipore, AB9708) 1/2000, rabbit anti-Homer (Synaptic System, 160003) 1/1000, mouse anti-CoxIV (Abcam, ab14744) 1/2000, mouse anti-SOD2 (Santa Cruz, sc-137254) 1/2000, mouse anti-Calnexin (Santa Cruz, sc-23954), rabbit anti-GM130 (BD Biosciences, 610823) 1/500, mouse anti-Nopp140 (Santa Cruz, sc-374033) 1/700, mouse anti-Fibrillarin (Invitrogen, MA3-16771) 1/2000, rabbit anti-Histone H4 (Active Motif, 39269) 1/2000, mouse anti-Ubc9 (BD Biosciences, 610749) 1/250, rabbit anti-SynGAP (Synaptic System, 157002) 1/1000; and mouse anti-Flotillin-1 (BD Biosciences, 610820) 1/500. Primary antibodies were revealed using the appropriate horseradish peroxidase (HRP)-conjugated secondary antibodies (GE healthcare) or True Blot (Rockland, Tebu-Bio). Proteins were then identified using Immobilon Western (Millipore), SuperSignal Femto chemiluminescence (Thermo Scientific) or Western Lightning Ultra (Perkin Helmer) chemiluminescent solutions and images acquired on a Fusion FX7 system (Vilber Lourmat).

To compare proteome datasets, protein identifiers were converted to Entrez Gene identifiers using Uniprot, DAVID and db2db conversion tools. When rat datasets were compared to human or mouse datasets, human or mouse homologs to rat proteins were identified using the DRSC Integrative Ortholog Prediction Tool or alternatively Blast. Gene Ontology (GO) terms enrichment analyses were performed using clusterProfiler (ontology level 2) on the web interface ProteoRE (Nguyen et al., 2019). KEGG and Reactome pathways enrichment analysis were performed using DAVID. Enrichment analysis using the synapse specific database SynGO (Koopmans et al., 2019) were performed against the “brain expressed” background, setting medium stringency and second level terms as labels for Cellular Component representation and top-levels terms as labels for Biological Pathways representation. Disease annotations were conducted with the ToppGene portal on the human homologs of the rat protein dataset. Network analysis were performed using the STRING web tool (v. 11.0) combined to the Cytoscape application (v.3.8.2) implemented by the StringApp plug-in. Details for enrichment analyses are available in Supplementary Table 6.

The expression levels of SUMO-modified substrates as well as the components of the SUMOylation machinery are developmentally and spatially regulated in the mammalian brain, with a peak of SUMO2/3-ylated proteins at the post-natal day (PND) 14 (Loriol et al., 2012) during the synaptogenesis period (Semple et al., 2013; Figures 1A,B). PND14 rat brains were thus subjected to biochemical fractionation to purify the synaptosomal fraction (Figure 1C). The quality of the fractionation was assessed by immunoblotting using antibodies against specific markers for different subcellular compartments (Supplementary Figure 1A). As expected, the synaptosomal fraction was highly enriched in synaptic markers such as PSD95, Synaptotagmin and Homer1 compared to the other fractions. Synaptosomes also contain mitochondria and some endoplasmic reticulum. In line with this, the respective markers for mitochondria and endoplasmic reticulum, CoxIV and Calnexin, were also detected in the synaptosomal fractions. Importantly, synaptosomes were totally devoid of nuclear markers confirming the high quality of the fractionation process (Supplementary Figure 1A).

Four independent synaptosomal preparations from PND14 rat brains were subjected to proteomics analysis. We established a list of 4379 proteins detected in at least 3 of the 4 preparations (Supplementary Table 1). This protein list was then used as our synaptic proteome reference. More than 88% of the proteins listed were found in at least three previously published synaptic proteomes and 97% were detected at least once as synaptic components (Supplementary Table 2). Gene Ontology (GO) enrichment analysis showed that among the top 20 for enriched GO Cellular Components terms, ten classes are associated to synapses whereas the other terms are linked to mitochondria and vesicular structures. Consistently, the top 20 for enriched GO Biological Processes terms revealed the involvement of the identified proteins in synapse organization, vesicle organization and trafficking, mitochondria organization as well as nucleotide metabolic processes (Supplementary Figures 1B,C). Altogether, these data confirmed the enrichment of the samples in synaptic components further highlighting the quality of the synaptosomal purification.

Previous proteomic studies on brain SUMOylation were performed using adult and/or whole brain extracts (Tirard et al., 2012; Yang et al., 2014; Matsuzaki et al., 2015; Hendriks et al., 2018; Stankova et al., 2018) and consequently, identified a high proportion of nuclear proteins as neuronal SUMO targets. To elaborate the most efficient protocol to assess synaptic SUMO2/3-ylation, we compared the levels of SUMO2/3-ylated proteins between synaptosomes and other subcellular fractions of the brain (Figure 1C). As expected, the SUMO2/3 immunoreactivity was maximum in the nuclear fraction, whereas synaptic SUMO2/3-ylation levels represented less than 1% of the total brain SUMO2/3-ylation. Considering the dynamic range of the Orbitrap Q-Exactive Plus mass spectrometer used in this study, it was absolutely necessary to enrich the preparations in synaptic material prior to isolate SUMO2/3-ylated proteins by denaturing immunoprecipitation (Becker et al., 2013) and identification by mass spectrometry.

To specifically isolate synaptic SUMO2/3-modified proteins, we performed immunoprecipitation using a commercially available mouse monoclonal SUMO2/3 antibody (Clone 12F3) that binds an epitope within the peptide CQIRFRFDGQPINE in SUMO2/3 (Becker et al., 2013; Supplementary Figure 2). This antibody specifically recognized proteins conjugated to SUMO2 but not conjugated to SUMO1 (Supplementary Figure 2A). Besides, peptide competition experiments monitored by dot blot using recombinant SUMO2 confirmed that free CQIRFRFDGQPINE peptides interfere between SUMO2 and the monoclonal SUMO2/3 antibody (Supplementary Figure 2B). We then showed that SUMO2/3 antibodies are able to specifically immunoprecipitate SUMO2/3-ylated proteins from rat brain in denaturing conditions (Supplementary Figures 2C,D). We thus performed a SUMO2/3 immunoprecipitation combined with a peptidic elution (Becker et al., 2013) to purify SUMO2/3-ylated proteins from denatured rat synaptosomes. Solubilization in 1% SDS efficiently extracted proteins from synaptic membranes, in particular those embedded in detergent resistant compartments such as the post-synaptic density (Jordan et al., 2006). Given the low levels of SUMO2/3-modified proteins at synapses, we performed immunoprecipitation experiments on synaptosomal lysates purified from a total of 10 PND14 rat brains per point. To assess the quality of this process, 10% of the proteins isolated with SUMO2/3 antibodies or mouse IgG as negative control were separated on gradient SDS-PAGE and silver stained (Figure 1D, left panel) or immunoblotted for SUMO2/3 (Figure 1D, right panel). Eluates from SUMO2/3 antibody-mediated immunoprecipitation showed a clear enrichment in SUMOylated proteins compared to eluates from matching IgG controls confirming that this approach allows to efficiently isolate synaptic SUMO2/3-ylated proteins.

Four synaptic SUMO2/3 immunoprecipitation assays with their IgG controls were performed independently. After protein separation by SDS-PAGE and in gel Trypsin digestion, samples were analyzed by liquid chromatography coupled to tandem mass spectrometry (LC-MS/MS). MS data were processed using Proteome Discoverer 2.2 against the Swissprot Rattus Norvegicus database and only Master Proteins with a 1% False Discovery Rate (FDR) for both peptides and proteins were considered. Then, the list of synaptic SUMO2/3-ylated proteins was built using the following selection criteria: (i) for each SUMO2/3 immunoprecipitation experiment, proteins present in the respective IgG control were subtracted from the proteins identified in the SUMO2/3 immunoprecipitates and (ii) the proteins present in at least three of the four filtered lists were selected as candidate SUMO2/3 targets. This finally led to a list of 803 synaptic SUMO2/3-ylated proteins with 291 (36%) present in all four experiments (Supplementary Table 3).

To analyze the relevance of the synaptic SUMO2/3-ylome, we first compared the list of SUMO2/3-ylated target proteins with previously published proteomes obtained from synaptosomes or subspecific components of synaptosomal fractions (Figure 2 and Supplementary Table 4). We showed that more than 97% of the SUMO2/3-ylome were detected in at least five synaptic datasets with about 40% of the candidate SUMO2/3 target proteins associated with the Post-Synaptic Density, on average over the different selected PSD proteomes (Supplementary Table 4). Thus, the list of SUMO2/3-ylated candidate proteins identified largely overlapped with previously characterized synaptosomal protein datasets, demonstrating that the SUMO2/3-ylome is directly connected to the synaptic compartment.

Noteworthy, our dataset presented many proteins found to be SUMOylated in previous target specific studies such as mitochondrial proteins (DJ-1/Park7 (Guerra de Souza et al., 2016), Sirt3 (He et al., 2020), constituents or regulatory factors of the cytoskeleton (β-Actin, α-Tubulin, RhoGDI, and Septins) (Alonso et al., 2015; Feng et al., 2021), signaling molecules (RAS (Choi et al., 2018), β-catenin (Karami et al., 2017), CYLD (Kobayashi et al., 2015)), proteins involved in cellular trafficking (Flotillin-1 (Jang et al., 2019), EDH3 (Cabasso et al., 2015), and hnRNPA2B1 (Villarroya-Beltri et al., 2013)) as well as the well described SUMO-conjugating enzyme Ubc9 (Loriol et al., 2012; Hendriks et al., 2018), the PKC-α and PKC-ε (Zhao et al., 2020; Henley et al., 2021) or SOD1 (Fei et al., 2006).

To go deeper into the validation of the proteomic screen, we next verified the endogenous SUMOylation profile of a selection of candidate SUMO2/3 targets by immunoprecipitation/blotting with specific target antibodies. We first analyzed the modification of the SUMO conjugating enzyme Ubc9 that was previously reported to be SUMOylated in the brain by both SUMO1 and SUMO2/3 (Loriol et al., 2012; Hendriks et al., 2018). In PND14 synaptosomes, free Ubc9 is barely detected as a 18 kDa protein and appears mainly as a 38 kDa modified form (Figure 3A, left panel) that corresponds to a SUMO1-conjugated form (Loriol et al., 2012). Several SUMO-targeted lysines were identified in Ubc9 by proteomics or target specific approaches (Knipscheer et al., 2008; Loriol et al., 2012; Hendriks et al., 2018). Here, we showed at least four bands upper to the unmodified Ubc9 by immunoprecipitation assays on PND14 synaptosomal lysates using SUMO2/3 antibodies followed by anti-Ubc9 immunoblot (Figure 3A, middle panel). Similar results were obtained in converse experiments where Ubc9 immunoprecipitation was combined with SUMO2/3 immunoblotting (Figure 3A, right panel), thus confirming the modification of Ubc9 by SUMO2/3 at synapses.

Using a similar approach based on conversed and complementary immunoprecipitations, we analyzed the SUMO2/3-ylation profile of Flotillin-1/Reggie-2 (Figure 3B). Flotillin-1 is a multifunctional protein involved in signaling, actin cytoskeleton remodeling or receptor endocytosis/trafficking (Kwiatkowska et al., 2020). In the brain, it participates in synapse formation and neurite branching (Kwiatkowska et al., 2020). Flotillin-1 has been shown to be SUMOylated on several target lysines in cell lines (Hendriks et al., 2018) and in a cellular model of prostate cancer (Jang et al., 2019). Our data indicate that Flotillin-1 is also modified by SUMO2/3 at synapses, presenting multiple conjugated forms in line with the previous reports (Figure 3B).

We next examined the SUMO2/3-ylation of the Synaptic Ras-GTPase-activating protein SynGAP (Figure 3C). SynGAP is a downstream effector of NMDA receptors that tunes down the activity of Ras and Rap GTPases and impacts AMPA receptor trafficking within postsynaptic membranes (Jeyabalan and Clement, 2016). SynGAP is predominantly localized at the PSD where it interacts with the scaffolding protein PSD-95. Both the SUMO2/3 immunoprecipitation on synaptosomal extracts followed by SynGAP immunoblotting and its converse experiment allowed the detection of a 140–150 kDa SUMO2/3-ylated form of the protein (Figure 3C). This slight band shift corresponding to the SUMO-modified form of SynGAP may indicate that the protein is only modified by one SUMO2/3 moiety.

In addition to these biochemical validations, we compared our dataset to two previously published rodent brain SUMO2/3-ylomes (Yang et al., 2014; Hendriks et al., 2018). Thirty-six proteins of our synaptic SUMO2/3-ylome (i.e., 4.5%) were formerly shown to be modified by SUMO2/3 in the Central Nervous System (Supplementary Table 5). This narrow overlap is largely explained by the fact that these proteomic analyses were performed on whole adult brain extracts without any synaptic enrichments. Interestingly, an extended comparison with SUMOylomes obtained from 15 previous proteomics analysis in other cellular models indicates that 62% of our identified SUMO targets were also found in these studies. The overlap increased to 74% when alternative approaches were included (Supplementary Table 5). Altogether, these analyses further validated the SUMO immunoprecipitation approach on synaptosomal-enriched fractions from PND14 brains.

We identified 803 SUMO2/3-modified protein candidates, which represents around 18% of our reference synaptic proteome (Figure 2 and Supplementary Table 4). To assess the biological meaning of this synaptic SUMO2/3-ylome, we performed Gene Ontology terms as well as KEGG and Reactome pathways enrichment analyses (Figure 4). Consistent with the clear overlaps with known synaptic proteomes and sub-proteomes (Figure 2 and Supplementary Table 4), the top 25 for enriched GO Cellular Components terms are connected to synapses, mitochondria and vesicles. Regarding GO Biological Processes or KEGG/Reactome pathways (Figures 4C,D), the synaptic SUMO2/3-modified proteome was enriched in proteins associated with neuronal development, organization and function, such as dendritic development (30 genes/proteins found in the SUMO2/3 dataset over 301 genes/proteins in the category), axon guidance (19/218), post-synapse organization (27/205), long-term potentiation (14/65) or depression (13/62), as well as glutamatergic (21/115), dopaminergic (19/129) and GABAergic synapses (14/87). In line with an enrichment in glutamatergic synapses, the synaptic SUMO2/3-ylome was also highlighted for the molecular function of glutamate receptor (12/67) and ionotropic glutamate receptor binding (9/44) (Figure 4B). Besides, vesicle-associated pathways as well as mitochondrial organization and processes, closely related to the synaptic function, clearly emerged in these analyses. Interestingly, the regulation of the actin cytoskeleton, which is involved in the shaping and intracellular trafficking at synapses as well as the proteasome, which participates in synaptic protein homeostasis, were also featured (Figures 4C,D). Cellular Components and Biological Processes enrichment analyses using the synapse specific annotation database SynGO (Koopmans et al., 2019) clearly corroborated the involvement of the synaptic SUMO2/3-ylome in synaptic organization, transmission and plasticity (Figures 4E,F). Last, enrichment analysis of the SUMO2/3-modified proteins dataset in KEGG pathways also highlight classes referring to brain disorders such as addictions and neurodegenerative diseases (Figure 4D).

One of the main roles of SUMOylation is the modulation of protein-protein interactions. Therefore, proteins subjected to SUMOylation are highly interconnected (Hendriks et al., 2014). Besides, synaptic subdomains are organized around large dynamic protein networks. We thus tested the interconnectedness of the synaptic SUMO2/3-ylome using the STRING tool. As illustrated in Figure 5A, the synaptic SUMO2/3 targets assembled in a highly organized network. 74% of all identified proteins (599/803) were connected at high STRING confidence with an average node degree of 5.61, and 92% of these proteins (551/599) belonged to the core cluster. Using the Cytoscape resource, only experimentally validated interactions were extracted and the resulting network was submitted to cluster analysis (Figure 5B). Our settings highlight eight sub-complexes with an average node degree superior or equal to 3. The four prevalent clusters comprised numerous functionally related proteins from the cytoplasmic and mitochondrial ribosomes, the NADH dehydrogenase and the proteasome. Consistent with these findings, modification of the proteasomal subunits was robustly reported in different SUMO proteomes (Hendriks et al., 2014; Im and Chung, 2016; Pfammatter et al., 2018; Supplementary Table 5). In addition to the proteasome, ribosomes were also detected as one of the most dominant clusters in SUMOylome network analysis (Hendriks et al., 2014; Pfammatter et al., 2018) and many mitochondrial ribosomal proteins were detected as SUMO substrates in several proteomics studies (Supplementary Table 5). The other clusters were composed of proteins of the PSD, some aminoacyl t-RNA synthetases (ARS) associated with the ARS-interacting multi-functional protein two or some members of the cytoplasmic dynein complex as well as the retromer complex involved in intracellular trafficking.

In the above sections, we highlighted an enrichment in SUMO2/3-modified proteins dataset for KEGG pathways referring to addiction and neurodegenerative diseases (Figure 4D). To get a broader view of the connections between the synaptic SUMO2/3-ylome and human brain disorders, we interrogated diseases databases using the ToppGene web tool. We found that 318 proteins out of the 803, which corresponds to about 40% of the SUMO2/3-ylome, are connected with at least one brain disease (Supplementary Table 6). In the enrichment analysis, the first 25 hits displayed many neurodevelopmental, psychiatric or neurodegenerative disorders, including, but not restricted to, schizophrenia, autism, epilepsy, Alzheimer’s and Parkinson’s diseases (Figure 6).