The proteomic landscape of synaptic diversity across brain regions and cell types

1The compartmentalization of biological processes in space and time within the single cell enables parallel processing of many reactions. Neurons, highly polarized cells, represent an extreme example of parallel processing—at their synapses, they compartmentalize information processing to communicate with thousands of other neurons. Synapses are plastic in structure and function, depending on their developmental and experiential history. This plasticity provides a molecular basis for learning and memory formation.

2The identity, copy number, post-translational modifications, and interactions of individual proteins largely determine the physiological properties of a given synapse. A growing body of evidence has found substantial structural and functional diversity of synapses. The complement of proteins that comprise average synaptic and sub-synaptic structures, such as the postsynaptic density (PSD) or synaptic vesicles, have been studied, but the proteome diversity that underlies different synapse types and states remains largely unexplored.,

3The traditional biochemical isolation of synaptic terminals or synaptic elements,,,, yields synapse-enriched fractions (synaptosomes) that can be analyzed by mass spectrometry (MS)-based proteomics., These preparations, however, are often of low purity and contain a heterogeneous mixture of many synapse types along with contaminants. Immunoisolation and proximity-labeling strategies have been used to identify proteins in selected synaptic sub-compartments of some synapse types.,,, In order to isolate a defined synapse population containing all synaptic elements from in vivo brain structures, Biesemann et al. introduced fluorescence-activated synaptosome sorting (FASS).,,,, This strategy utilizes fluorescent labeling of a synaptic protein to sort synaptosomes with high purity using a fluorescence-activated cell sorter and has been used to identify proteins that are enriched at growth cones or selected synapse types.,,

4Here, we used Cre-inducible knockin mice and FASS coupled to MS to investigate the diversity of the synaptic proteome across genetically defined synapse types and brain areas. We identified >1,800 unique synapse-enriched proteins that give rise to 18 different synapse-type-specific proteomes. This resource (https://syndive.org) reveals commonly shared synaptic protein modules as well as proteins that customize the proteomes of synapse populations related to their function, for example, at dopaminergic synapses in the striatum (STR) and subtypes of cortical interneurons.

5We developed a streamlined workflow to quantify the proteomes of synapses formed by different cell types in different brain areas. We used various Cre-inducible knockin mice expressing a fluorescently labeled presynaptic protein, synaptophysin-tdTomato (SypTOM), to target synapses that arise from cell types that use different neurotransmitters. We prepared and purified synaptosomes using FASS and coupled the output to a mass spectrometer to reproducibly quantify thousands of proteins (Figure 1A). We first tested the pipeline by comparing well-characterized Gad2-cre (cortical inhibitory) and Camk2a-cre (cortical excitatory) Cre-driver lines, crossed with SypTOM mice (Figure 1A). We prepared synaptosomes from the cerebral cortex (CX), and verified that synaptosome fractions were enriched in pre- and postsynaptic proteins and depleted for non-synaptic contaminants (Figures S1A and S1B). We optimized the FASS,,, strategy to define a population (P3) with a high tdTomato signal in combination with high membrane content, visualized by an FM dye (Figures 1B and S1C). Although control synaptosomes prepared from wild-type mice showed no detectable particles in the P3 gate, Camk2a::SypTOM synaptosomes constituted approximately 40% of the total particles (Figure 1B). Re-analysis of sorted synaptosomes by flow cytometry routinely led to a purity of ∼80%–95% (Figure 1B). Immunolabeling of individually spotted sorted synaptosomes revealed that most synaptosomes contained a postsynaptic element, and ∼65% of Camk2a::SypTOM synaptosomes were labeled for the excitatory postsynaptic marker PSD95, whereas ∼50% of the Gad2::SypTOM synaptosomes were labeled for inhibitory postsynaptic marker gephyrin (Figures S1D–S1G). In this proof-of-principle experiment, we sorted ∼20 million cortical synaptosomes from 8 Camk2a::SypTOM and 8 Gad2::SypTOM mice and quantified their proteomes., For each mouse, we additionally processed control samples consisting of the same number of particles from the precursor synaptosome population (the input to sorting). To achieve this, we purified all particles with a membrane dye signal intensity above the noise threshold but independent of tdTomato signal, resulting in an unsorted control sample. Overall, we quantified >2,300 protein groups. We defined the individual synaptic proteomes as the complement of proteins that were significantly enriched relative to their respective control fractions (Tables S1A and S1B). We found that the proteomes of Camk2a::SypTOM and Gad2::SypTOM synaptosomes were enriched in synaptic proteins and clearly separated from unsorted synaptosomes in a principal-component analysis (PCA) (Figures 1C, 1D, S1H, and S1I). We examined the proteins with the highest enrichment in the direct quantitative comparison of Camk2a::SypTOM and Gad2::SypTOM proteomes and observed that they represent almost exclusively established marker proteins for cortical excitatory and inhibitory synapses (Figures 1D and 1E; Table S1C). We compared the 20 most enriched proteins (top ∼10%) for Camk2a::SypTOM with the SynGO synaptic protein database and found 19 with prior synaptic annotation. These data demonstrate that this workflow can identify synaptic proteomes from a small number of purified synapses.

6We applied the above pipeline to investigate the proteomic diversity of 15 different major synapse subtypes using four Cre-driver lines representing different cell types and microdissection of five different brain areas (CX, hippocampus [HC], STR, olfactory bulb [Bulb], and cerebellum [CER]) (Figure 2A). Besides Camk2a- and Gad2-cre, we included Syn1-cre (synapsin 1), a line that is independent of neurotransmitter type, and Dat-cre, representing the synapses that use the modulatory neurotransmitter dopamine. In control experiments, we first comprehensively assessed the degree to which a given Cre-line labeled the expected synaptic population in a given brain area (Figures S2A and S2B). We found that in the CX and HC, Camk2a::SypTOM and Gad2::SypTOM mice showed specificity for excitatory and inhibitory synapses, respectively. In the STR and Bulb, however, Camk2a::SypTOM and Gad2::SypTOM labeled both distinct and overlapping synaptic populations (Figures S2A and S2B). Finally, two-dimensional (2D) electron microscopy (EM) showed that synaptosome preparations from different brain regions did not differ in their synaptosome abundance, in the size or presence of synaptic vesicles, or in synaptic mitochondria or postsynaptic structures between brain regions (Figures S2C–S2L).

7The fluorescently labeled synaptosome populations that we studied originally comprised ∼5% to ∼50% of a given brain region’s total crude synaptosome population (Figure 2B). We purified all synapse types (but one, Dat::SypTOM) to greater than 75% purity, on average (Figure 2C). For each synapse subtype, we obtained ∼10 Mio sorted (P3 gate) particles and the same number of matched control particles from at least 5 mice. We processed the sorted synaptosomes for label-free quantitative proteome analysis. Overall, we quantified >2,800 protein groups with a high reproducibility (median coefficient of variation [CV] ∼20%) among biological replicates (Figures S3A and S3B). In order to define the individual synaptic proteomes, we determined the complement of proteins that were quantitatively enriched in each synapse type (Table S2A), compared with its own precursor (unsorted) synaptosome population from the same brain region. In a PCA, the different synaptic proteomes were distinguished by both cell types (Figure 2D) and brain regions (Figure 2E). Which feature, cell type or brain region, exerts the greatest influence on the synaptic proteomes? Overall, the cell types explained more of the total variance than the brain regions (Figure 2F), and other factors like the age or sex of the animals had a negligible influence on the observed variance (Figure 2F). In total, we allocated >10,000 protein groups to the 15 synaptic proteomes (Figure 2G) and identified >1,800 unique protein groups that were enriched in at least one of the 15 synapse subtype proteomes (Figure 2G; Table S2A). We created an interactive web tool that allows one to query the abundance and localization of individual proteins in individual synapse types (The Synaptic Diversity Hub, https://syndive.org/). We compared the enrichment of the vesicular GABA transporter (vGat) and vesicular glutamate transporter 1 (VGlut1) proteins from MS with those obtained by immunofluorescence on brain slices (Figure S2B) and found high correlation coefficients of 0.84 and 0.91 for VGlut1 and vGat, respectively (Figure 2H), validating the specificity of these synaptic proteomes.

8To what extent do the proteins identified here represent previously known synaptic molecules? To address this, we compared the union of all the unique synapse-type-enriched proteins with the synaptic protein database SynGO. Overall, there was very good agreement between the two datasets: ∼80% of the identified SynGO-annotated proteins were classified as synapse-enriched in our analysis (Figures S3C–S3E). Furthermore, we identified >1,000 new synapse-enriched protein groups (Table S2B). These novel synapse-enriched proteins are significantly associated with various disease-related pathways and underlie different cellular functions such as G-protein signaling, protein degradation, or tRNA aminoacylation (Figures S3F–S3H). When each synaptic proteome was analyzed individually, each one was significantly enriched in SynGO terms (Table S2C). As might be expected from the localization of the SypTOM protein, we found significantly more proteins annotated as presynaptic than postsynaptic across the 15 synapse types (Figure S3D). We also identified significantly more novel synaptic proteins at Gad2::SypTOM synapse types as compared to Camk2a::SypTOM (Figure S3E) and the most novel proteins at Dat::SypTOM synapses, presumably reflecting a bias of the published synaptic protein literature toward excitatory synapses.

9Are there protein modules that are shared among the different synapse types? To address this, we selected proteins that were detected in at least 14 of the 15 synapse types and performed an enrichment analysis using the SynGO database. We identified a significant enrichment of synaptic vesicle vesicular ATPases (vATPases) as well as proteins mediating synaptic vesicle endocytosis (Figure S4A). Both terms were also significantly enriched when selecting proteins that were detected in minimally 10, 11, 12, or 13 synapse-type proteomes (Figure S4B). The enriched proteins include, for example, clathrin (Clta, Cltb, and Cltc) and proteins of the associated adapter protein complex (AP-2) or the vATPase Atp6v1f, which were enriched at every synapse type. Next, we quantitatively compared Gad2::SypTOM with Camk2a::SypTOM synaptosomes and defined Gad2-enriched, Camk2a-enriched, and shared protein groups in each brain region (Figure 3A). This analysis revealed proteins that were quantitatively enriched at either synapse type over the other (Camk2a-enriched or Gad2-enriched) or enriched in both types over controls and not significantly different between the types (shared). In general, we observed very little overlap (∼1% of all synapse-enriched proteins) between Gad2::SypTOM-enriched and Camk2a::SypTOM-enriched proteins, indicating that across all brain regions, excitatory and inhibitory synapses have subsets of mutually exclusive synaptic proteins (Figure 3B; see further analysis below). However, since Gad2-cre and Camk2a-cre label different neuron populations in different brain regions (Figure S2B), there were shared proteins in some brain regions available for analysis (Figure 3B). To examine in more detail the shared proteins, we focused on the shared proteins of the CX (where Gad2::SypTOM and Camk2a::SypTOM label specifically excitatory and inhibitory synapses, respectively; Figures S2A and S2B). Within the shared proteins, we again observed an overrepresentation of synaptic vesicle endocytosis and synaptic vesicle proton loading, whereas exocytosis and presynaptic active zone proteins were absent or minimally represented in the shared fraction but overrepresented in the Gad2-enriched and/or the Camk2a-enriched fraction (Figures 3C and 3D).

10We constructed a protein-protein interaction network using the proteins associated with the enriched terms and revealed that protein modules that represent different steps of the synaptic vesicle cycle were enriched in either shared or subtype-specific synaptic proteins (Figure 3D). We observed intriguing patterns among certain protein families, such as the Rab, Vamp, Stx, or Syt groups, where related family members displayed distinct type-specific enrichment at excitatory or inhibitory synapses. For a subset of these proteins, we conducted immunoblotting experiments that confirmed these data (Figures S4C–S4E), and notably, we observed an excellent correlation between immunoblot and MS analyses (Pearson’s r = 0.95). This diversity suggests that these proteins may differentially modulate the same synaptic molecular process, potentially influencing synapse-type-specific characteristics.

11Cell-adhesion molecules (CAMs) are known to contribute to the specificity in synaptic connections. We retrieved a curated list of transsynaptic CAMs and asked whether these proteins show preferential enrichment at excitatory or inhibitory synapse types. Indeed, in the CX and HC, we found >95% of the CAMs were exclusively enriched at either excitatory or inhibitory synaptic proteomes (Figures 3E and S4F), indicating a highly cell-type-specific localization. Taken together, the above analyses highlight endocytosis and vATPases as generic synaptic protein modules that are used by synapses independent of neurotransmitter type, whereas the exocytosis machinery, the presynaptic active zone, transsynaptic adhesion molecules, and postsynaptic protein modules make use of different sets of proteins for different synapse types.

12Are there specific protein modules that are associated with synapses that use different neurotransmitters or are associated with different cell types or brain regions? To address this, we used a protein-protein weighted correlation network analysis (WGCNA) to identify protein modules that show correlated abundance patterns across the 15 synapse types. We first asked a simpler question: do proteins of the same complex or functional unit exhibit correlated expression levels? We found a significantly higher median correlation for proteins that are subunits of the same protein complex (Pearson’s r = 0.65) as compared with random protein pairs (Pearson’s r = 0.03) (Figure 4A). For example, the proteasome subunits Psma1 and Psma7 exhibited highly correlated abundance (r = 0.97); similarly, vGat and Gad2 (the essential synaptic GABA synthesis enzyme) were correlated with a near-perfect coefficient of 0.98. We next constructed a protein-protein correlation network using all synapse-enriched proteins and identified 14 protein modules using WGCNA (Figure 4B; Table S3). We discovered that the resulting protein network featured two main opposing clusters, defining two highly connected protein communities. To identify the nature of these protein communities, we correlated all protein modules with traits of the synapse subtypes, including immunofluorescence for vGat and VGlut1 (Figure S2B), cell types, and brain regions. Although most brain regions and the cell type Syn1 showed no correlation with any protein module, we identified three inhibitory protein modules that were significantly correlated with vGat and three excitatory protein modules that were significantly correlated with VGlut1 (Figure 4C). We found that both the excitatory and inhibitory modules were located at the center of each protein community and that each community was characterized by a high correlation or anti-correlation with vGat or VGlut1 (Figure 4D; Table S3). In summary, we constructed a synaptic protein-protein correlation network that revealed two distinct protein communities representing the proteomes associated with excitatory or inhibitory neurotransmitters across distinct cell types and brain regions.

13Can we identify the key proteins for the synaptic proteomes of glutamatergic and GABAergic synaptic proteomes based on the network topology? We identified 188 and 315 proteins that were significantly correlated with vGat or VGlut1 expression, respectively, with correlation coefficients ranging from moderate (∼0.5) to very high (∼0.95) (Table S3). Proteins with high correlation coefficients were also among the most connected nodes within each community (Figure 4D). We inspected the network topology of the protein modules with a significant correlation for vGat and found that the core module (module 10) contained established inhibitory marker proteins, including Gad-1, Gad-2, Gaba transporter-1, Neuroligin-2, multiple GABA-A receptor subunits, and the inhibitory postsynaptic scaffolding protein gephyrin (Figure 4E). Notably, module 10 contained proteins from both the pre- and postsynaptic compartments, indicating a tight co-regulation of synaptic architecture across the synapse. Module 9, by contrast, was characterized by a moderate but significant correlation with vGat. In this module, we detected proteins that have been previously associated with subtypes of GABAergic neurons or synapses, for example, Syt-2, Lamp-5, cannabinoid receptor 1 (Cnr1), or GABA-A receptor subunit alpha-2 (Figure 4E). In total, we identified 130 novel synaptic proteins that were significantly correlated with vGat expression and not previously recognized as synaptic by SynGO, for example, IgLON5, a cell-adhesion protein implicated in a specific neurodegenerative autoimmune disease.

14In order to validate the network topology of the GABAergic proteome, we constructed a protein-protein interaction network of the three inhibitory protein modules using a protein interaction database and found that the proteins from the core module (module 10) accounted for the majority of hubs at the center of the network (Figures S5A and S5B). In conclusion, we present a protein-protein correlation network that represents the diversity of the GABAergic synaptic proteome, revealing many novel synaptic proteins, and highlights core proteins that are strongly associated with vGat across synapses from different brain regions and cell types, as well as moderately associated proteins that may distinguish specific synapse subtypes.

15The excitatory synapse protein community was well-correlated with VGlut1 fluorescence (Figure 4D); correspondingly, we found the VGlut1 protein was among the most connected nodes and there was no significant correlation with the VGlut2 protein. Overall, we detected many more proteins in the VGlut1+ network when compared to the vGat+ network. Presumably, this is due to the elaborate PSD complex and spine architecture present at excitatory synapses. Accordingly, we detected many proteins from established excitatory postsynaptic protein families within the VGlut1 protein community, including the Shank family (Shank1/2/3), the Camk family (Camk4/1d/2d/2b/v/2a), the Dlg family (Dlg1/2/3/4, Mpp2/3, and Dlgap1/2/3/4), the glutamate receptor subunits (Gria1/2/3/4, Grin1/2b, Grik2), the Lrr family (Lrrc7/57/4) and protein phosphatases (Pp2r5a/3ca/1cb/5c/1ca/3cb/3ra). We also identified many proteins not previously associated with glutamatergic synapses. For example, we found Fbxl16, an F-box protein of an E3-ubiquitin ligase with unclear synaptic function,, exhibited the strongest correlation with VGlut1 expression across brain regions. In total, we identified 158 novel synaptic proteins that correlated significantly with VGlut1 expression and were not previously recognized as synaptic by SynGO.

16We next asked whether there are pathways or biological functions enriched specifically at vGat vs. VGlut1 protein communities. To address this, we performed gene set enrichment analysis (GSEA) using the ranked protein correlations with vGat and VGlut1 immunofluorescence, respectively. As expected, we found significant enrichment for proteins associated with postsynaptic signaling pathways and dendritic spines within the VGlut1 community (Figures 4F and S5C). In the vGat protein community, we found significant enrichment for proteins from the GABA receptor complex, but also proteins with functions not previously recognized as enriched at inhibitory (vs. excitatory) synapses: aminoacyl-tRNA synthetases, 19S proteasome subunits, and mitochondrial proteins (Figures 4F and S5C).

17We previously observed an opposite enrichment pattern of some proteins from the same family at excitatory and inhibitory synapses. To investigate this further, we conducted a comprehensive analysis and identified 22 protein pairs that exhibited significant negative correlations across the 15 synapse types (Figures 4G, 4H, and S5D; Table S3). The identified protein pairs spanned diverse functional categories, including exocytosis (Rab3a/b, Syt2/12, Sv2b/c), voltage-gated calcium channels (Cacna2d1/2), protein folding (Fkbp1a/p4), and ATPase activity (Atp1a1/3). Notably, our analysis also revealed proteins, such as the GTPases Diras1 (inhibitory) and Diras2 (excitatory), without prior synaptic annotation. These findings highlight their potential involvement in the same synaptic molecular processes to modulate synapse-type-specific characteristics. Taken together, the above analyses define a roadmap for how molecular systems-biology analysis of synaptic proteomes can be used to identify the key protein modules that underlie synaptic traits.

18Dopaminergic synapses from midbrain dopaminergic neurons that project to the STR are critically important for reward processing and movement control, and their degeneration is a hallmark of Parkinson’s disease pathophysiology. We conducted an in-depth analysis of the synaptic proteome of striatal dopaminergic terminals. To verify the expression of the presynaptic fluorophore in the Dat::SypTOM mice, we immunostained brain sections (Figure 5A) and found a prominent tdTomato signal in the STR and a high correlation with tyrosine hydroxylase (Th) immunoreactivity (Figures 5B and 5C). Comparing striatal Dat::SypTOM synaptosomes with unsorted control synaptosomes, we identified 267 significantly enriched proteins in the striatal dopaminergic proteome (Table S2A). To address specific vs. shared dopaminergic synaptic proteins, we identified proteins that were either enriched or de-enriched in striatal dopaminergic synapses as compared with the other 14 synapse types. Among the differentially enriched proteins were dopaminergic marker proteins, for example, Maoa and Aldh1ha1, but also proteins that have not been previously associated with dopaminergic terminals (Figures 5D and S6A). For example, we identified oxidation resistance protein 1 (Oxr1), as ubiquitously present in Camk2a, Gad2, and Syn1 synapse types but depleted from dopaminergic terminals (Figure 5D). Oxr1 controls sensitivity to oxidative stress, and as such, the absence of Oxr1 in dopaminergic synapses might confer susceptibility to oxidative damage (e.g., during Parkinson’s disease). Another example is the mitogen-activated protein kinase Erk1 (Mapk3), which was specifically enriched at dopaminergic synapses (Figure S6A) and has been linked to Parkinson’s disease via multiple cellular processes. We validated these findings and the findings below with immunoblotting (Figure S6B). Next, we compared the striatal Dat::SypTOM proteome to the striatal Syn1::SypTOM proteome (Figure 5E) and performed GSEA comparing mutually enriched with Dat::SypTOM specific proteins (Figure 5F). Relative to striatal Syn1::SypTOM synapses, the Dat::SypTOM synaptic proteome was enriched in key proteins involved in dopamine biosynthesis, trafficking, and degradation (Figures 5E and 5G). The proteins that were significantly enriched in both groups included vATPases, proteasome subunits, and endocytic proteins (Figures 5F and S6C). Although we identified four vATPases within the shared group (Atp6v1a/h/f/e1), Atp6v1g1 was significantly enriched at dopaminergic synapses, indicating an association with dopaminergic synaptic vesicles (Figure 5G). We identified six proteasome subunits as shared and two proteasome subunits that constitute the modulatory immunoproteasome-associated PA28 complex (Psme1/2) as enriched at dopaminergic synapses. We validated the presence of the Psme1 protein and proteasome activity in striatal synaptosomes but found no evidence for the presence of immunoproteasome subunits, rather, Psme1 was associated with standard proteasome subunits (Figures S6D–S6G). The above data identify the synaptic proteome of striatal dopaminergic terminals and highlight similarities and specializations in synaptic proteome architecture.

19We next asked whether our approach can be used to study synapse types originating from small subfields or rare cell types. To analyze subfields, we microdissected the main hippocampal regions CA1, CA3, and the dentate gyrus (DG) from a single Gad2::SypTOM mouse (per replicate) and quantified the inhibitory synaptic proteomes. Despite the small amount of starting material, we quantified more than 2,000 protein groups and discovered 264 (DG), 328 (CA1), and 493 (CA3) Gad2-enriched proteins in the subfield-specific inhibitory proteomes (Figure S7A; Table S4). Although the unsorted control samples showed subfield-specific clustering and enrichment of known marker proteins, the inhibitory proteomes were highly correlated and had only few differentially enriched proteins (Figures S7B–S7E).

20The neurons that use the neurotransmitter GABA exhibit morphological diversity and differ in the location of their cell bodies and synapses on other cells, as well as their transcriptomic profiles, connectivity patterns, and firing properties.,, Using Cre-driver lines for the main subclasses, parvalbumin (PV), somatostatin (SST), and vasoactive intestinal peptide (VIP) neurons, we targeted synapses from these cortical GABAergic neuron subclasses (Figure 6A) and compared them to the cortical Gad2::SypTOM synaptic proteome. We verified the co-localization of tdTomato signal with subtype-specific markers (Figures S8A–S8C) and validated the specificity of SST- and VIP-Cre-driver lines (Figures S8D and S8E). We note that cortical interneurons are very sparse, making their synaptic proteomes very challenging to study. Indeed, we found that VIP+ synaptosomes represented just ∼0.5% of all particles in a cortical synaptosome fraction, whereas PV+ and SST+ synaptosomes represented approximately 3% and 5% (Figure 6B). After sorting, we achieved greater than 80% purity for all 3 cortical inhibitory synapse types (Figure 6C). With an optimized sample processing protocol, we quantified >2,500 protein groups in total and identified almost 600 unique proteins enriched in at least one cortical interneuron subtype. We found distinct synaptic proteomes for the three subtypes with 325, 234, and 369 significantly enriched proteins detected, respectively, for PV+, SST+, and VIP+ synapses (Figure 6D; Table S5A). All proteomes were significantly enriched with GABAergic synaptic proteins (Figure S8F). The three inhibitory synaptic proteomes were clearly separated in a PCA (Figure 6E) and showed a type-specific signature in their proteome composition (Figure 6F). The union Gad2 proteome was closest to the most abundant type SST+ in PCA space. We searched for canonical markers of each inhibitory synapse type and found, as predicted, the expected enrichment for VIP, PV, and Calbindin proteins in the VIP+, PV+, and SST+ synaptic proteomes (Figure 6G). The remaining markers, Lamp5 and Scng, distinguished a 4th and 5th type of inhibitory neuron, and, appropriately, were most enriched in the Gad2+ cortical synaptosomes (Figure 6G).

21We next asked whether the developmental origin of the cell type is reflected in the synaptic proteome. Although PV and SST neurons arise from the medial ganglionic eminence (MGE), VIP neurons originate from the caudal ganglionic eminence (CGE). Previous studies showed that transcriptomes of cortical inhibitory neurons,, cluster according to their progenitor domain. By contrast, a PCA revealed that the SST+ and VIP+ synaptic proteomes cluster closer together than the SST+ and PV+ proteomes (Figure 6E). Consistent with this, we identified only 33 proteins that were significantly different between the SST+ and VIP+, but 128 and 318 between PV+ and SST+ or VIP+, respectively (Tables S5B–S5D). These analyses indicate that cortical inhibitory synaptic proteomes are predominantly shaped by other factors than the developmental origin of their presynaptic cell type.

22Can we identify proteins that relate to the functional differences observed in these three inhibitory synapse types? PV+ neurons are characterized by fast spiking and corresponding high energy demands. We found that the voltage-gated potassium channels Kcnc1, Kcnc2, and Kcnc3 (Kv3.1/2/3) were among the most enriched PV+ synaptic proteins (Figure 6G), potentially explaining the ability of PV cells to accommodate high firing rates. We also found Hcn1 and Hcn2 were highly enriched in the PV+ synaptic proteome (Figure 6G). The PV+ proteome was also significantly enriched in mitochondrial proteins involved in oxidative phosphorylation (Figure 6F). Additionally, we detected a number of synaptic vesicle-associated proteins (such as Syt2, Vamp1, or Cplx1) and the cell-adhesion proteins Cntnap4 and Hapln4 as specifically enriched at PV+ synaptic proteomes (Figures 6G and S8G; Table S5).

23Cortical VIP neurons are recognized for their inhibitory control over SST and PV interneurons, forming a crucial component of cortical disinhibitory circuits. In comparison to the cortical PV+ proteome, the VIP+ synaptic proteome was enriched for neuroactive ligand-receptor interactions and downstream signaling molecules (Figures 6F and 6G). One of the most enriched proteins was Cnr1, and many proteins involved in downstream G-protein signaling were enriched as well, including Rgs6, Kcd12, Adcy2, Gng2, and Gnai2 (Figure 6G; Table S5). We also detected a strong enrichment of glutamate G-protein coupled receptors, the metabotropic glutamate receptors (excitatory Grm1 and inhibitory Grm7), and, to a lesser extent, the ionotropic glutamate receptor subunits Grik2 and Gria1/4 but not Gria2/3 (Figures 6G and S8G; Table S5). Furthermore, we found a number of cell-adhesion proteins specifically enriched at cortical VIP+ synapses (Nrxn1, Nrxn2, Nlgn3, Dag1, and Igsf8), as well as calcium-binding proteins (Calb2 and Necab2), calcium channels (Cacna1b and Cacna2d3), and synaptic-vesicle-associated proteins (Sh3gl3, Rph3a, Rab3c, Synpr, and Syn3) (Figures 6G and S8G; Table S5). In contrast to PV+ and VIP+, there were few proteins (Calb1, Rab3b, Icam5, Nipsnap3b, Atp2b4, and Nos1) that distinguished SST+ synaptic proteomes (Figure 6G). This reflects the current view that SST neurons represent a diverse group with substantial differences in morphology and physiology. Finally, we identified a few proteins (Ppp1r1b and Cplx3) that were enriched in the overall Gad2 synaptic proteome over the three inhibitory subtypes, presumably because they are specific for one of the other main cortical interneuron subtypes characterized by Lamp5 or Scng expression (Figure 6G). Overall, we identified ∼600 unique proteins that define the cortical interneuron synaptic proteome, allocated them to the three main subclasses, and highlighted specialized groups of proteins that relate to their established functional properties.

24Although there is evidence for substantial structural and functional diversity of synapses,, the underlying diversity in synaptic molecular architecture is much less understood. A detailed understanding of synapse proteome diversity allows us to link the molecular architecture of the synapse to structure and function. Here, we optimized FASS in combination with MS for system-wide analysis of the proteomic landscape of synaptic diversity across 18 distinct synapse types defined by cell type and brain region. We found that the cell type from which the synapses originate explains more of the observed variance than the brain region. Across the brain areas, the Bulb and CER were the most distinct compared with the other brain areas. These regions may exhibit unique molecular profiles that reflect their evolutionary context and developmental origin and potentially their distinct roles in sensory processing (Bulb) and motor coordination (CER). With these data, we now define a common core of synaptic proteins, shared across synapse types, that includes vATPases and synaptic vesicle endocytosis proteins; by contrast, we identified the presynaptic active zone, exocytosis machinery, transsynaptic, and postsynaptic elements as hotspots for synaptic proteome specialization. Our analysis also revealed the polarized enrichment of individual proteins from the same family across synapse types, suggesting a means by which functional differences in these protein family members can modulate the same synaptic molecular process to establish synapse-type-specific characteristics.

25Using a “guilt-by-association approach” (the WGCNA), we found that synaptic proteins form communities that correlate with vGat and VGlut1 expression. Intriguingly, we discovered many proteins at the core of the vGat and VGlut1 protein communities that were not previously recognized as synaptic. The vGat protein community further revealed an intriguing enrichment of unanticipated functional protein groups: we identified proteasome subunits, mitochondrial proteins as well as tRNA synthetases preferentially enriched over the VGlut1 synaptic community. The presence of mitochondrial proteins in the vGat community presumably relates to increased energy demands. Although moonlighting functions have been ascribed to tRNA synthetases, it is possible that the tRNA synthetases scavenge and charge amino acids that originate from local proteasomal degradation.

26We also provide an in-depth analysis of striatal dopaminergic synapses, revealing 267 proteins significantly enriched at dopaminergic terminals, which is an almost 5-fold greater depth compared with a recent proteome analysis of dopaminergic terminals. We compared the dopaminergic proteome to the other 14 synaptic proteomes and identified the absence of Oxr1, a protein that protects from oxidative damage and might render dopaminergic neurons particularly susceptible to oxidative stress., Finally, we identified ∼600 unique proteins that define the synaptic proteomes of the main cortical interneuron subclasses. In contrast to the transcriptomes of cortical interneurons, we find that the synaptic proteomes do not cluster according to their progenitor domain, indicating that synaptic proteomes are shaped by factors other than the developmental origin of their presynaptic cell type. We reveal type-specific signatures in the cortical interneuron proteomes that relate to their established functional properties. For PV+ synapses, characterized by high firing frequencies, we identified a specific enrichment of mitochondrial proteins, voltage-gated potassium channels, and hyperpolarization-activated cyclic nucleotide-gated ion (HCN) channels. We detected the differential distribution of individual CAMs (e.g., Nrxn1, Nrxn2, Cntnap4, Hapln4, and Icam5), suggesting a role in the distinct connectivity patterns observed for the interneuron subtypes.

27Our resource of synaptic proteomes departs from previous studies in many respects. First, we used FASS because it enables the analysis of synaptic proteomes originating from in vivo brain structures and covers all synaptic compartments, including pre-, post-, and transsynaptic proteins. Second, we identified the proteome of relatively scarce synapse types not previously amenable to purification (e.g., the VIP+ synaptic proteome, representing <1% of the total cortical synapse pool). Third, we included many biological replicates per synapse type (>5), increasing the proteome depth to define synapse-enriched proteins in a purely data-driven fashion using linear-mixed effects models, without the requirement for external lists or prior knowledge. Importantly, we used a paired-sample experimental design where every sorted sample was accompanied by an unsorted control, and synapse-enriched proteins were determined by quantitative enrichment. Fourth, we compared proteome diversity across 15 synapse types, whereas all previous studies were limited to one or two synapse types.,,,,,,,,,, Finally, our resource enables a direct comparison of different synapse types, in contrast to comparisons between synapses and other subcellular compartments like the soma.

28We anticipate that multi-omic analysis of synapses will be conducted using an analogous experimental strategy, investigating the synaptic diversity of other biomolecules like glycans, lipids, or RNA. In addition, the future integration of synaptic proteomes with local transcriptomes can delineate the roles of RNA localization and local translation in synaptic proteome diversity.,,, On the cellular level, experiments that use methods such as the Cre/lox system to target a defined cell type to investigate function can now be combined with synapse subtype-specific proteome information and thereby connect different levels of organization. For example, one could probe how a particular phenotype, disease model, behavioral paradigm, or cellular manipulation differentially affects the synaptic proteomes of the synapse subtypes of interest.

29This study combines transgenic reporter mice with FASS and MS-based proteomics, and each aspect has inherent limitations. The reporter animals limit the cell-type-specific synaptic proteomes that can be targeted and dictate the resulting homogeneity. FASS is based on biochemical synaptosome purification and may be biased toward synapses that readily form synaptosomes of the described size and density while limiting the analysis of synapses with anomalous sizes or shapes. Furthermore, synaptosomes are sorted using a presynaptic marker protein, and not all sorted synaptosomes contain an intact postsynaptic compartment; as such, there may be a bias against postsynaptic proteins. MS is biased in its ability to quantify peptides by the individual peptides' biophysical properties that influence ionization efficiency and transfer to the gas phase. We identified enriched synaptic proteins from fluorescently labeled synaptosomes by comparing them to their unsorted precursor synaptosome population, which includes the labeled synapse population of interest as well as unlabeled synapses. As such, proteins that we identify as enriched at synapse type A over type B or correlated with vGat or VGlut1 might be present in neuronal dendrites, axons, or somata to varying degrees. The observed enrichments or de-enrichments are thus a function of protein abundance and specific subcellular targeting or exclusion of proteins from synapses. For example, proteins that showed cell-type-specific expression but broad localization throughout the cell, like the PV protein, were identified as specifically enriched in PV+ synapses when compared with other synapse types. By contrast, specific recruitment or synaptic exclusion could lead to synaptic enrichment or de-enrichment despite comparable average protein abundance in different neuron types. For example, Oxr1 is specifically enriched at many synapse types but depleted from dopaminergic terminals; however, the corresponding mRNA was found at comparable levels throughout many neuron types, including midbrain dopaminergic neurons and GABAergic neurons in the STR.

30Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Erin Schuman (erin.schuman@brain.mpg.de).

31This study did not generate new unique reagents.

32The housing and sacrificing procedures involving animal treatment and care were conducted in conformity with the institutional guidelines that are in compliance with national and international laws and policies (DIRECTIVE 2010/63/EU; German animal welfare law; FELASA guidelines). The animals were euthanized according to annex 2 of § 2 Abs. 2 Tierschutz-Versuchstier-Verordnung. Animal numbers were reported to the local authority (Regierungspräsidium Darmstadt). The following Cre-driver lines were crossed with the SypTOM (JAX strain #: 012570, Ai34D) mouse line: Camk2a-cre (JAX strain #: 005359), Gad2-cre (JAX strain #: 010802), Syn1-cre (JAX strain #: 003966), Dat-cre (JAX strain #: 006660), PV-cre (JAX strain #: 008069), SST-cre (JAX strain #: 013044) and VIP-cre (JAX strain #: 010908). Animals were housed under a 12 h light/dark cycle, and provided with food and water ad libitum. Analyses were performed on adult animals aged 8-13 weeks with a median age of 10 weeks. The sex of the animals was mixed and balanced within conditions.

33Mice were transcardially perfused with 4% paraformaldehyde (PFA) in PBS and brains were post-fixed overnight (ON) in 4% PFA at 4°C. Sagittal sections (50μm) were cut using a Leica vibratome (VT1200S) and washed in PBS. Vibratome sections were blocked and permeabilized by incubation in 5% goat serum 0.5% Triton-x in PBS (blocking buffer) at RT for 4 hours. Primary antibody incubation was performed ON at 4°C on a rocker in the blocking buffer. The following day, sections were washed in PBS and incubated with a fluorescently labeled secondary antibody ON at 4 °C. Slices were washed in PBS, rinsed in ddH2O and air dried on SuperFrost Plus glass slides (Fisher Scientific). Aqua-Poly/Mount (Polysciences) was used for mounting.

34For single molecule fluorescent in situ hybridization experiments (viewRNA, Thermo Fisher), mice were perfused and postfixed with 4% PFA, 4% sucrose in PBS for 1 hour at room temperature (RT). Cryoprotection was performed by incubating ON in 15% and subsequently 30% sucrose in RNAse free PBS at 4 °C. Sagittal sections (40μm) were cut using a Zeiss HYRAX S50 (at -25°C) and washed in PBS. Sections were post-fixed at RT in a 4% PFA solution (4% paraformaldehyde, 5.4% Glucose, 0.01 M sodium metaperiodate, in lysine-phosphate buffer) for 10 minutes, washed in RNAse free PBS and permeabilized using a detergent solution (viewRNA) for 20 minutes at RT. Probe hybridization was performed in the hybridization buffer at 40°C ON. The following day, sections were rinsed in the wash buffer and subjected sequentially to pre-amp DNA, amp DNA and label probe oligonucleotides in their respective buffers for 1 hour at 40°C with washes in the wash buffer at RT in between. After washing in PBS, sections were blocked in blocking buffer (4% goat serum 0.5% Triton-X in PBS). Primary antibody staining was performed ON in blocking buffer at 4°C. After washing in PBS, secondary antibody incubation was carried out at RT for 2 hours in blocking buffer. Sections were washed, counterstained with DAPI (1 μg/ml, Thermo Fisher) in PBS for 3 minutes, washed in PBS and mounted as described above. The following primary antibodies and corresponding dilution factors were used: anti-vGAT (SYSY, 131004 – gp, 1:5000), anti-vGAT (SYSY, 131002, rb, 1:5000), anti-vGlut1 (SYSY, 135304, gp, 1:5000), anti-vGlut1 (SYSY, 135303, rb, 1:5000), anti-TH (SYSY, 213104, gp 1:500), anti-VIP (Thermo Fisher, PA5-78224, rb, 1:500), anti-SST (Thermo Fisher, PA5-85759, rb, 1:500), anti-Syt2 (SYSY, 105223, rb, 1:500), anti-NeuN (Abcam, Ab177487, rb, 1:1000). The following FISH probes were used at 1:100 dilution: TdTomato (viewRNA, VF1-14985 or VF6-13925, Alexa Fluor 647 or Alexa Fluor 750), Sst (viewRNA,VB1-14821-VC or VB4-3112424-VC, Alexa Fluor 750 and Alex Fluor 647). Images were acquired using a Zeiss confocal microscope (LSM-880 or LSM-980) using 63X or 40X oil-immersion objectives (NA 1.4).

35Synaptosomes were prepared as described in Westmark et al. Briefly, animals were sacrificed by decapitation, the brain regions of interest were dissected on ice and subsequently homogenized in gradient medium (GM; 0.25 M sucrose, 5mM Tris-HCl, 0.1mM EDTA supplemented with Calbiochem Protease inhibitor cocktail III) using a glass dounce homogenizer. The homogenate was centrifuged for ten minutes at 4°C at 1’000g. The supernatant (S1) was layered onto a Percoll density gradient with 23%, 10% and 3% Percoll in the GM buffer. The gradient was centrifuged for 5min at 32’500g at 4°C with maximum acceleration and minimum deceleration using a Beckman Coulter JA-25.50 rotor in an Avanti J-26S XPI centrifuge (both from Beckman Coulter). The resulting bands are labeled, from top to bottom, F0, F1, F2/3, F4. Bands were retrieved and directly processed (for EM analysis) or stored at -20°C. For proteasome activity measurements, F2/3 fractions were pipetted onto glass fiber filters using a syringe and stored at -80 °C until further processing. For immunoblotting, synaptosomes were lysed in a detergent buffer (8M urea, 10% SDS, 10% Sodium deoxycholate 5% Triton-X in water; 1 part detergent buffer and 5 parts synaptosome fraction) at 75°C for 5 minutes. BCA assay (Thermo Fisher) was used to approximate protein concentrations of the different fractions.

36For synaptosomes of hippocampal subregions mouse brains were dissected on ice and coronal sections of 300 μm thickness were cut on a vibratome (Leica) in oxygenated slush ice prepared from frozen ACSF-sucrose (87 mM NaCl, 25 mM NaHCO3, 1.25 mM NaH2PO4, 2.5 mM KCl, 10 mM Glucose, 75 mM Sucrose, 0.5 mM CaCl2, 7 mM MgCl2). Hippocampal subregions were microdissected manually under a microscope with cooled stage from individually transferred sections and immediately collected in a Dounce homogenizer with 1 ml of GM buffer on ice. The tissue samples from each brain were processed as described above.

37After fraction collection on filters (see above and below) proteasome activity was assayed by incubation of filters with HR buffer (50 mM Tris–HCl pH 7.4, 5 mM MgCl2, 250 mM sucrose) freshly supplemented with 1 mM DTT, 2 mM ATP and 1 μM Me4BodipyFL-Ahx3Leu3VS (Bio-Techne GmbH, I-190-050) for 1h at 37°C. To assay proteasome activity of 20S (LifeSensors, PS020) and 20Si (Enzo, BML-PW9645-0050) purified samples, 1 μg of protein was incubated in HR buffer complete with 1 μM Me4BodipyFL-Ahx3Leu3VS for 1h at 37°C. Sample protein concentration was measured with the Precision Red advanced protein assay (Cytoskeleton, Inc., ADV02-A).

38For each sample, a volume corresponding to a normalized protein amount was supplemented with 10X SDS sample buffer (500 mM Tris pH 6.8, 25% SDS and 2% bromophenol blue in 70% glycerol-30% dH2O), NuPAGE Sample Reducing Agent (10X) (Thermo Fisher, NP0004) and distilled water to an equal final volume. Samples were denatured and reduced at 90°C for 5 minutes and run on Novex 4-20% Tris-Glycine and Novex 12% Bis-Tris mini gels and 4-20% BioRad midi gels. Activity-based probe fluorescence was measured either in gel or after transfer on the membranes on an Azure Sapphire biomolecular imager. For silver staining the gels were processed as described in the kit’s technical bulletin (Thermo Fisher, 24612). For immunoblot analyses, the gels were wet-transferred onto Immobilon-FL PVDF membranes (Sigma-Aldrich, 05317-10EA) or semi-dry transferred onto 0.2 μm nitrocellulose membranes (BioRad, 1704159). Equal loading and transfer were then assessed by Revert 700 total protein stain (LI-COR, 926-11011) or Ponceau stain (CST, 59803). The membranes were destained, blocked for 1h at room temperature in Intercept (TBS or PBS) blocking buffer (LI-COR, 927-60001, 927-70001) and probed with primary antibodies overnight at 4°C. The next day the membranes were developed with fluorescently labeled secondary antibodies on a LI-COR Odyssey Classic system. The following primary antibodies and corresponding dilution factors were used: anti-PSMA3 (Enzo, BML-PW8110-0100 – ms, 1:1000), anti-PSMB8 (Enzo, BML-PW8845-0100 – ms, 1:1000), anti-PSMB5 (CST, 12919 – rb, 1:1000), anti-PSME1 (Abcam, ab155091 – rb, 1:1000), anti-Histone H3 (Abcam, ab1791 – rb, 1:1000), anti-TH (SYSY, 213104 – gp, 1:1000), anti-PSD95 (Abcam, ab2723 – ms, 1:1000, used for Proteasome validation), anti-PSD95 (Thermo Fisher, MA1-046, ms, 1:1000, used for synaptosome prep analysis), anti-Mbp (Abcam, Ab62631– ms, 1:1000), anti-GFAP (Abcam, Ab7260 - rb, 1:1000), anti-Syn (SYSY, 106002, rb, 1:1000), anti-SYPH (Sigma, S5768, ms, 1:5000), anti-Gad1/2 (Enzo, ADI-MSA-225-E, ms, 1:1000), anti-Syt12 (SYSY, 299 003, rb, 1:1000), anti-Proton ATPase (SYSY, 109 003, rb, 1:1000), anti-Complexin-1/2 (SYSY, 122 102, rb, 1:1000), anti-Stx1a (SYSY, 110 111, ms, 1:1000), anti-Stx1b (SYSY, 110 403, rb 1:1000), anti-Syt2 (SYSY, 105 223, rb, 1:1000), anti-Rab3c (SYSY, 107 203, rb, 1:1000), anti-Vamp1 (SYSY, 104 002, rb, 1:1000), anti-Vamp2 (SYSY, 104 202, rb, 1:1000), anti-TH (SYSY, 213 104, gp, 1:1000), anti-Oxr1 (Abcam, ab251774, rb 1:1000), anti-Mapk3 (CST, 9101, rb, 1:1000) and anti-Atp6v1g1 (ProteinTech, 16143-1-AP, rb, 1:1000). Normality of the data was assessed using a Shapiro-Wilk test. The homogenate and the synaptosome fraction F2/3 were compared using a paired one tailed t-test or the non parametric Wilcoxon signed rank test. P-values <0.05 were considered significant

39Electron microscopy (EM) analysis of unsorted synaptosomes was conducted as described in Sebring et al. with chemicals from Sigma Aldrich and Plano GmbH (Pioloform). F2/3 fractions were fixed with EM grade glutaraldehyde (final concentration 2.5%) for 30 minutes on ice and, subsequently, diluted in PBS. During the fixation period, 5% agarose pucks were prepared by aspirating with a P10 pipette tip. Next, the pipette tip was lodged onto a stereological pipette tip allowing for a larger volume to be spun down onto the agarose puck. The fixed synaptosomes were spun onto the agarose puck using Labofuge 400R swinging bucket centrifuge at ∼4’000g for 1 hour at 4°C. Then, the liquid was aspirated carefully and the agarose puck was pushed out of the pipette tip and encapsulated in an agarose droplet. This droplet was stored at 4°C in PBS until further processing.

40Sorted synaptosomes were processed as described above with minor alterations. First, ∼10 million synaptosomes were sorted into bovine albumin serum in PBS (final concentration 2%). The pipette tip containing the agarose puck was attached to a 15ml Falcon tube to accommodate the higher volume. Synaptosomes were spun down onto the agarose puck at 4°C and 4000g for 3 hours after which glutaraldehyde was added to a final concentration of 2.5%. The synaptosomes were fixed during centrifugation at 4°C at ∼4000g for 30 minutes. Next, samples were washed in 0,1 M Cacodylate buffer (CD) at room temperature and samples were further stained in 1% OsO4 (in 0.1M CD) for 30 minutes. Samples were washed in 0.1M CD and water, and subsequently stained with 1% uranyl acetate (UA) (in water) in the dark for 10 minutes. Samples were washed with water and dehydrated by incubation in a series of ethanol buffers ranging from 30% to 100% ethanol.

41For epoxy resin embedding, the samples were washed in dehydrated Propylenoxid and incubated in a 1:1 mixture of EPON:dehydrated Propylenoxid for 30 minutes at RT. Then, samples were left in EPON overnight at RT. The following day, excess agarose was trimmed and samples were placed in a silicone mold and EPON was polymerized for at least 48 hours at 60°C. 70 nm sections were generated using a Leica Leica Reichert Ultracut S Ultramicrotome with a DiATOME ultra 45° diamond knife and mounted onto a Pioloform film on copper grids. EM imaging was performed with a Zeiss LEO 912 AB Omega and a Sharp Eye TRS (2x2k) CCD camera. ImageSP was used to control the CCD camera and to make tilescans with 20% overlap at a magnification of 31’500X at 120 kEV for characterisation of F2/3 unsorted synaptosomes. FASS synaptosomes were imaged at 4100X magnification at 120 kEV. Tiles were stitched for visualization using TrakEM2 in ImageJ. For image analysis, ∼30 synaptosomes from each sample were randomly selected, blinded (using ImageJ plugin Blind Analysis Tools), converted to 8-bit and the image contrast was normalized to 0.35 saturation. The blinded images were manually annotated for size, mitochondrial area and the presence of an attached postsynaptic density, while the number of SVs was determined with the automatic detection pipeline. Unblinding and subsequent analysis was performed in R. To estimate the true diameter, overcoming the random cross sectional nature of 2d TEM, the assumption of perfectly spherical synaptosomes was made. This simplification allowed for sampling random cross sections from a sphere with a known radius (R) by solving for Y in X² + Y² = R² given randomly drawn X values from the range 0-R. Then 2Y was used as the diameter of the randomly sampled cross section of the sphere. The above described was performed 30 times for four simulated samples.

42Data was tested for extreme outliers, defined as outside the first quartile minus 3 x the interquartile range or third quartile plus 3 x the interquartile range, and for normality using a Shapiro-Wilk test. An ANOVA was used to compare synaptosome ultrastructural features across brain regions. To compare the cross sectional area of synaptosomes with a postsynaptic density (PSD) and without a PSD, a paired two-tailed T-test was used. P-values <0.05 were considered significant.

43Fluorescence-activated synaptosome sorting (FASS) was performed as previously described.,, F2/3 fractions from the synaptosome preparation were diluted 1:50 in the GM buffer and 1.5μg/ml membrane dye (FM4-64, Thermo Fisher) was added. Synaptosomes were analyzed and sorted on a FACSAria Fusion (BD Biosciences) running FACSDiva, equipped with a 70μm Nozzle and the following settings: 488nm laser (for FM4-64), 561nm laser (for TdTomato), sort precision (0-16-0), FSC (317 V), SSC (488/10 nm, 370V), PE “TdTomato” (586/15 nm, 470V), PerCP “FM4-64” (695/40 nm, 470), thresholds (FSC = 200, FM4-64 = 700). Samples were analyzed and sorted at approx. 20’000 events/s and a flow rate of < 3. Doublet particles were excluded based on SSC-H and SSC-W. We constructed a gate (P3) to obtain the synaptosome population that selects particles that are double-positive for TdTomato and FM4-64 by thresholding against synaptosomes from wt mice (Figure 1B). For control synaptosomes all non-doublet particles above the FM4-64 threshold were sorted (P2). For each sorted sample (P3) as well as the matching control sample (P2) we sorted 20 Mio particles (Experiment described in Figure 1), 10 Mio particles (Experiment described in Figure 2A) or 2 Mio particles (Experiments described in Figures 6 and S9).

44Sorted synaptosomes were centrifuged onto gelatin coated coverslips (Electron Microscopy Sciences) for 20 min at 4’000g and 4 °C in a Labofuge 400R and fixed for 15min in 4% PFA, 4% sucrose in PBS. Synaptosomes were then immunolabeled according to the manufacturer's instructions (Alexa Fluor 488 Tyramid SuperBoost Kit, Invitrogen, B40941) using the following antibodies: anti-PSD95 (Thermo Fisher, MA1-046, ms, 1:1000) and anti-Gephyrin (SYSY, 147011, ms, 1:1000), and subsequently mounted and imaged as described above for brain sections. Synaptosome co-localization analysis was performed in ImageJ using the SynaptosomesMacro published by Paget-Blanc et al.

45Sorted synaptosomes and control particles were filtered onto Whatman glass microfiber filters (GF/F, Cytiva) and stored at -80 °C until further processing. For filtration, a miniaturized custom-built apparatus was constructed. Synaptosomes were sorted directly into a 15ml tube (Protein LoBind, Eppendorf) that was connected through tubing (from an infusion set) to a luer connector unit with a 4mm diameter filter fixed between the male and female luer connector units. The tubing and filtration unit was cooled on ice and negative pressure was applied for filtration. Early experiments revealed the necessity of washing the filters with PBS to remove residual sheath fluid (BD FACSFlow), and as a result many of earlier samples were not analyzed. Samples from the experiment across brain regions were allocated to blocks and batch-processed. Synaptosome samples with 10 Mio or 20 Mio particles were lysed in 20μl lysis buffer (100 mM Tris, 1% sodium deoxycholate, 10 mM TCEP, 15 mM 2-chloroacetamide) by repeated sonication using a VialTweeter (Hielscher Ultrasonics), and heated to 95 °C for 5 min. Proteins were digested with 0.1μg LysC and 0.1μg trypsin (Promega) overnight at 37 °C. Samples were then acidified with 10% formic acid to approximately pH 3, and centrifuged for 10 min at 16’000 g. The supernatant was desalted using ZipTip pipette tips (Merck) and dried in a vacuum centrifuge. Synaptosome samples with 2 Mio particles were solubilized in 20ul TEAB buffer (50mM Triethylammonium bicarbonate, 1mM CaCl2, 0.05μg trypsin LysC and 0.05μg trypsin) and digested overnight at 37 °C. 30μl acetonitrile was added and the samples were centrifuged 10 min at 16’000 g. The supernatant was filtered through ZipTip pipette tips by centrifugation for 1 min at 2’000g and then 50μl of 50% acetonitrile in MS-grade water was added to the filter and both steps were repeated. Samples were dried in a vacuum centrifuge and stored at -20 °C until Liquid chromatography–tandem mass spectrometry (LC-MS/MS) analysis.

46All samples were analyzed using label-free quantification in DIA-mode with targeted feature extraction using sample-specific spectral libraries built by DDA runs of in-house generated synaptosome samples. This approach was chosen to assure high sensitivity and reproducible coverage across a high number of relatively heterogeneous samples., The peptide samples were reconstituted in 5% acetonitrile (ACN) and 0.1% formic acid (FA) supplemented with an iRT peptide standard (Biognosys). Peptide mixtures were analyzed using an UltiMate 3000 nano-LC coupled to a Fusion Lumos mass spectrometer (Thermo Fisher Scientific). Briefly, samples were loaded onto a PepMap 100 C18 trap column (75 μm id × 2 cm length, 3 μm particle size; Thermo Fisher Scientific) at 6 μL/min for 6 min with 2% acetonitrile (v/v) and 0.05% trifluoroacetic acid (v/v) followed by separation on an C18 analytical column (75 μm id × 50 cm length, 1.7 μm particle size; CoAnn Technologies) maintained at 55°C. Peptides were separated by a non-linear 120 min gradient using mobile phase A (100% H2O, 0.1% formic acid) and B (80% acetonitrile, 0.1% formic acid). The Fusion Lumos mass spectrometer was operated in DDA mode for spectral library generation and DIA mode for all samples used in this study, using acquisition methods that are described in Muntel et al. In brief, the 40W DIA-method had the following settings: Full scan; orbitrap resolution = 120k, AGC target = 125%, mass range = 350-1650 m/z and maximum injection time = 100 ms. DIA scan; activation type: HCD, HCD collision energy = 27%, orbitrap resolution = 30k, AGC target = 2000%, maximum injection time = dynamic. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD039946.

47LC-MS/MS samples were analyzed with Spectronaut version 16 (Biognosys) as previously described., Briefly, a spectral library was generated from collected in-house generated DDA FASS synaptosome files. The collected DDA spectra were searched against the UniProtKB/Swiss-Prot database for mus musculus (retrieved in May 2021) using the Sequest HT search engine within Thermo Proteome Discoverer 2.4 (Thermo Fisher Scientific). For samples that did not undergo alkylation by 2-chloroacetamide (samples with 2 Mio synaptosome input, see above), carbamidomethylation at cysteines was set as variable instead of fixed modification. The identified proteins were assessed using Percolator and filtered using the high peptide confidence setting in Proteome Discoverer 2.4. Analysis results were then imported to Spectronaut 16 for the generation of spectral libraries. Targeted data extraction of DIA-MS acquisitions was performed in Spectronaut 16 with default settings. The proteotypicity filter “only protein group specific” was applied. Extracted features were exported from Spectronaut for statistical analysis with MSstats 4 using default settings. Briefly, for each protein, features were log-transformed and fitted to a mixed effect linear regression model for each sample. For significance testing, we required a minimum of 6 features (combination of precursor and fragment ion) and 12 measurements per protein per condition. The model estimated fold change and statistical significance for all compared conditions. For the P2 control conditions, all P2 samples were grouped according to the brain region of origin, independent of the mouse line. The Benjamini–Hochberg method was used to account for multiple testing and p-value adjustment was performed on all proteins that met the fold-change cutoff. Significantly different proteins were determined using a threshold fold-change >1.1 and adjusted p-value < 0.05. For analyses comparing conditions originating from the same brain region, the significance testing result and sample quantification (protein abundance) files from MSstats 4 were used. For analyses comparing multiple conditions across brain regions, we calculated the normalized log2(fold-change) of the matched P3/P2 pairs for every sample and protein, which was then used for further analysis.

48The synaptic proteomes for Camk2a+, Gad2+ and Syn+ proteomes were determined by quantitative enrichment against the P2 (unsorted) control samples and the other synapse types originating from the same brain region. Specifically, for a given brain region, the Camk2a+ the proteomes, for example, were defined as proteins that satisfy at least one of following criteria: (I) they met the fold-change and p-value cutoff comparing P3 (sorted) with P2 (unsorted) samples of the same brain region and were not significantly different between Camk2a+ and Gad2+ or (II) proteins that met the fold-change and p-value cutoff comparing Camk2a+ over Gad+ and had a positive fold-change comparing P3 (sorted) with P2 (unsorted) samples of the same brain region or (III) proteins that met the fold-change and p-value cutoff comparing Camk2a+ over Syn1+ and had a positive fold-change comparing P3 (sorted) with P2 (unsorted) samples of the same brain region. For Gad2+ and Syn1+ the proteomes were determined according to the above description. The Dat+,cortical interneuron and hippocampal subfield proteomes were defined by quantitative enrichment against the P2 control samples using the above mentioned cutoffs. To compare Gad2+ enriched synaptic proteomes between hippocampal subfields, we defined interaction terms in the linear model used in MSstats analogous to: Gad2+SubfieldA vs. Gad2+SubfieldB = (“Subfield A Gad2+” - ”Subfield A control”) - (“Subfield B Gad2+” - “Subfield B control). All further analysis was conducted using quantitative values obtained from MSstats 4 in the R environment (R studio 2023.03 running R 4.2). Variance partitioning was analyzed using the VariancePartition R package. The chord diagram was generated using the circlize R package. Protein to gene name conversion was done using either org.Mm.eg.db or the UniProt API. SynGO analyses were performed using the SynGO web page with default settings. Euler diagrams were constructed using the eulerr R package. STRING networks were generated in Cytoscape 3.9.1 using the Cytoscape StringApp and analyzed using CentiScaPe. Protein complexes were annotated using the CORUM database. Weighted gene correlation network analysis was done using the WGCNA R package using soft power 6, a signed hybrid network type and the bicor correlation function. Heatmaps were visualized using ComplexHeatmap. The adjacency matrix was visualized as a protein-protein correlation network in Cytoscape with a frequency cutoff of 0.3. Gene set enrichment analysis (GSEA) for Gene Ontology terms and KEGG pathways was performed with clusterprofiler. ggplot2 was used for visualization. Cartoons were generated using Adobe Illustrator 2023 or Biorender.com. FACS plots were generated with FlowJo 10.

49The PSME1 co-immunoprecipitation was performed using a commercial immunoprecipitation kit (Abcam, ab206996) following the manufacturer's instructions. Briefly, DIV 28 rat cortical neurons were washed three times in cold PBS and lysed in RIPA buffer supplemented with protease inhibitors (Thermo Fisher, 78430). The lysates were clarified by centrifugation at 104 x g for 10 min at 4C and the protein concentration measured using the Precision Red protein assay (Cytoskeleton Inc., ADV02). For each co-immunoprecipitation reaction 500 μg of protein were used in a total volume of 500 μl. The antibodies used and corresponding dilution factors were: rabbit anti-PSME1 (Abcam, ab155091, 1:100), rabbit anti-PSME1 (Cell Signalling, 2408, 1:100) and rabbit isotype control (Cell Signalling, 3900, 1:200). The reactions were incubated on a wheel overnight at 4C and the next day capture was performed using 30 μl of agarose resin for 2h at room temperature. After five washes in wash buffer and two additional washes in PBS the proteins were eluted by boiling in 2X SDS sample buffer.

50Different appropriate statistical methods were used for different experiments and are indicated in the relevant sections of the STAR Methods section and the figure legends. Statistical analysis of mass spectrometry data was performed with MSstats 4 in the R environment (R studio 2023.03 running R 4.2)

51We created an interactive web-tool that allows one to query the abundance and localization of individual proteins in individual synapse types and also download our data (The Synaptic Diversity Hub, https://syndive.org/).