A synaptomic analysis reveals dopamine hub synapses in the mouse striatum

Dopamine transmission is involved in reward processing and motor control, and its impairment plays a central role in numerous neurological disorders. Despite its strong pathophysiological relevance, the molecular and structural organization of the dopaminergic synapse remains to be established. Here, we used targeted labelling and fluorescence activated sorting to purify striatal dopaminergic synaptosomes. We provide the proteome of dopaminergic synapses with 57 proteins specifically enriched. Beyond canonical markers of dopamine neurotransmission such as dopamine biosynthetic enzymes and cognate receptors, we validated 6 proteins not previously described as enriched. Moreover, our data reveal the adhesion of dopaminergic synapses to glutamatergic, GABAergic or cholinergic synapses in structures we named “dopamine hub synapses”. At glutamatergic synapses, pre- and postsynaptic markers are significantly increased upon association with dopamine synapses. Dopamine hub synapses may thus support local dopaminergic signalling, complementing volume transmission thought to be the major mechanism by which monoamines modulate network activity. The neurotransmitter dopamine is an important regulator of brain function. Here the authors describe “dopamine hub synapses”, where dopamine transmission may act in synergy with other neurotransmitters.

S ince the 1950s with the first ultrastructural characterization of the synapse in the central nervous system 1 a wide variety of synapse types has been described based on morphological criteria 2 . The archetypal synapse type is the so-called asymmetric excitatory synapse on dendritic spines 1 , which represent the vast majority of synapses (~80%). Its ultrastructure is easily identifiable in the tissue by the presence of a postsynaptic density and a dense cluster of synaptic vesicles, and it has been extensively studied in vitro using primary neuronal cultures 3 . Alternatively, symmetric synapses are predominantly inhibitory or modulatory. They do not display postsynaptic densities and are more difficult to identify in situ 4,5 . Moreover, many types of synaptic organizations are not abundant enough and/or accessible with in vitro models. These limitations hinder our understanding of neuronal network functioning.
While glutamate and GABA (Gamma-Amino Butyric Acid) neurotransmissions drive point to point information locally, modulatory neurotransmitters pace regional activity through volume transmission in the neuropil 6,7 . Dopamine transmission is a major neuro-modulatory system involved in several functions such as movement initiation, reward prediction error and incentive processes, notably by its projections onto spiny projection neurons (SPNs) of the striatum 8 . Dopamine signalling is presumed to modulate glutamate transmission onto SPNs through the release of dopamine mainly from varicosities devoid of synaptic differentiation. Previous investigations assessing the presence of dopaminergic synapses found that only a minority of axon terminals form synapses onto SPN spines, dendrites, or presynapses 4,5,9 . Recent work also challenges the model of volume dopamine transmission by providing evidence for local point-to-point signalling. In particular, optophysiology approaches revealed rapid and local transmission at dopaminergic projections to the striatum [10][11][12] , which is in accordance with the existence of the machinery allowing fast dopamine release at striatal varicosities 13 . Moreover, the distribution of varicosities in the striatal neuropil appears biased toward proximity with the surrounding glutamatergic synapses 5 , and dopamine receptors interact physically and functionally with glutamate and GABA receptors [14][15][16][17] , suggesting a tight coupling between dopamine, glutamate and/or GABA signalling.
In the present work, we unravel the cellular and molecular synaptome of a single projection pathway 18 . This critically complements current connectomic approaches using optophysiology and tracing methods, which are limited in terms of molecular analysis of specific synapses at play in a given circuit 19 .
To that end, we established a workflow combining fluorescence tracing of the dopaminergic pathway, fluorescence-activated synaptosome sorting 20,21 and an array of semi-quantitative analysis methods ranging from conventional immunofluorescence characterization to mass spectrometry-based proteomics. With this approach, we provide a proteome and validate 6 new proteins (Cpne7, Apba1/Mint-1, Cadps2, Cadm2/SynCAM 2, Stx4, Mgll) enriched at dopaminergic synapses from the mouse striatum. Moreover, we show a physical coupling between dopaminergic and other synapses in a tight multipartite complex that we name "dopamine hub synapses".

Results
Fluorescence-activated synaptosome sorting (FASS) of dopaminergic synaptosomes reveals multipartite synaptic hub structures. We labelled the dopaminergic projection onto the striatum through stereotaxic injection of an adeno-associated viral vector carrying Cre-dependent EGFP 22 in the midbrain of Dopamine Transporter promoter (DAT)-Cre transgenic mice 23 (Fig. 1a). We applied a classical synaptosome fractionation miniaturized to 1.5 ml tubes as previously published 24,25 to generate samples labelled with EGFP (Fig. 1a). To validate our labelling and fractionation approach, we performed a complete subcellular fractionation of the dissected striata and measured the amount of two soluble reporter proteins, tyrosine hydroxylase (Th), that catalyses the limiting step for dopamine synthesis 26 , and the fluorescent reporter EGFP. They were probed using a semi-automatic capillary immunoblot system producing electrophoregrams (Fig. 1b) or membrane-like band patterns (Fig. 1c). Quality controls of the fractionation show the enrichment of synaptophysin (Syp) in synaptosomes (SYN) and crude synaptic vesicle (LP2) fractions, while the plasma membrane glutamate transporter GLAST (Slc1a3/GLAST) is enriched in synaptic plasma membranes (SPM). We confirm the high concentration of Th and EGFP signals in synaptosomes (SYN) and derived soluble fractions (LS1 and LS2) while they are weak in nuclear P1 and cytosolic S2 fractions, relative to homogenate (H) (H: Th = 1 ± 0. 26 Fig. 1c, d). Based on these reporters, we can conclude that most of the cytosolic content of dopaminergic axons is present in the synaptosome fraction (SYN) and available for discrimination by the fluorescence-activated synaptosome sorting (FASS) 21,27 procedure (Fig. 1a). Of note, a small fraction of Th seems associated with light membranes of the crude synaptic vesicle fraction, an observation reminiscent of the one made with GABA synthesizing enzymes 28 .
FASS 21,27 applied to this sample allowed recovering up to 35 million fluorescent synaptosomes according to cytometer counts (Fig. 1a). The collection of an equal number of singlet particles regardless of their EGFP fluorescence status served as a control for all comparisons (SYN). Bulk FASS samples were used to perform immunoblot and mass spectrometry assays. In addition, we established the immobilization of particles on glass coverslips to analyze them through quantitative immunofluorescence, super-resolution STED microscopy and electron microscopy ( Fig. 1a). Our gating strategy was adapted from the previous work 27 to avoid sorting aggregated particles, i.e. particles with high forward scatter (FSC) and side scatter (SSC) values and sort specifically singlets, particles with FSC values around 0 ( Supplementary Fig. 1a-c). Among singlets, EGFP + events are specifically detected by setting a fluorescence threshold from the autofluorescence of synaptosomes from non-injected mice ( Supplementary Fig. 1b). Synaptosomes from DAT-Cre-EGFP mice (SYN) contained on average 3.86 ± 0.53% EGFP + synaptosomes (N = 9 sorts; Fig. 1e, f). Upon reanalysis of the DA-FASS sample in the cell sorter, EGFP + events represented around 50% of the total (48.9 ± 2.3%, N = 8 sorts; Fig. 1f) and EGFP − synaptosomes were concomitantly depleted (SYN: EGFP − Singlets = 66.1 ± 4%, N = 9 sorts; DA-FASS: EGFP − Singlets = 30.9 ± 2.8%, SYN-DA-FASS: Šídák's multiple comparison ****p < 0.0001 N = 8 sorts; Fig. 1f). Based on these values, we can expect an enrichment of a specific component of dopaminergic synaptosomes of 48.9/3.9 = 12.5-fold in DA-FASS relative to SYN samples. With the use of mNeonGreen as a fluorescence reporter we increased the yield up to 35 million mNG + synaptosomes, which we compared to an equal number of singlet particles, regardless of their green fluorescence status (SYN; Supplementary Fig. 1e-g).
We further validated these sorts using capillary electrophoresis-based immunoblots. As expected, Th and the dopamine transporter (Slc6a3/DAT) display a strong enrichment after DA-FASS. In contrast, GLAST is strongly depleted while the glutamate receptor (Gria1/GluA1) or the synaptic active zone protein Munc18 (Stxbp1) are reduced (Fig. 1g). We then performed qualitative ultrastructural analysis on DA-FASS samples using transmitted electron microscopy (TEM) (Fig. 1h-l). We identified synaptosome profiles with resealed presynaptic elements (Fig. 1h) and in some cases a clear adhesion with a postsynaptic membrane (Fig. 1i). Surprisingly, we also identified profiles displaying several presynapses organized around possible postsynaptic membranes (Fig. 1j-l). Most of the synaptosomes were cut at an angle that prevented the clear identification of all synaptic elements (Fig. 1l). In another example, we found two distinct presynapses, one electron-dense terminal with a synaptic vesicle cluster adhering to a presynaptic element with few vesicles and to another compartment that could be dendritic (Fig. 1j). Finally, a postsynaptic element displayed adhesion to three different "boutons", one of them displaying a clearer background and fewer vesicles (Fig. 1k). Contrarily to aggregates, these multipartite synapses were preserved even though our procedure exposed them to shearing forces twice, first during tissue homogenization and second, in flight through the nozzle of the cell sorter 21,27,29 (see the workflow in Fig. 1a; Supplemental Figs. 1 and 5). Beyond displaying axo-axonic or axo-spinous synapses, our fractionation isolates multipartite bound synaptic elements that we name "dopamine hub synapses".
Label-free semi-quantitative proteomics reveals 57 proteins highly enriched at DA-FASS synaptosomes. To identify the molecular nature of dopaminergic synaptosomes, we generated a set of six DA-FASS samples and processed them for label-free quantification of proteins through mass spectrometry (MS). We accumulated 35 million mNeonGreen + synaptosomes from three independent DA-FASS experiments. All SYN singlets representing the conventional synaptosome preparation were used as control samples (3 in total). A total of 3824 proteins were identified with one peptide or more, throughout the six samples. Among these, 2653 proteins were identified robustly and quantified with at least 2 distinct peptides. We considered a significant difference between samples for proteins displaying a ratio greater than 1.5 in DA-FASS samples compared to SYN, with an adjusted p-value smaller than 0.05. Based on these criteria, 63 proteins are significantly depleted upon sorting while 57 others appear significantly enriched (Fig. 3g, Supplementary Table 3). The depleted proteins did not have a clear gene ontology signature and Fig. 1 Single projection fluorescence-activated synaptosome sorting (FASS) isolates dopaminergic hub synaptosomes. a Workflow of DAT-cre/AAV-EGFP synaptosome sorting and analysis. DAT-Cre mice injected with a Cre-dependent AAV expressing EGFP or mNeonGreen ( Supplementary Fig. 1) in the Substantia Nigra pars compacta and the Ventral Tegmental Area. Fluorescence guided dissection in the Striatum (Str, red-dashed circle). Subcellular fractionation and FASS. Collection on filters or glass coverslips for FASS sample analysis by mass spectrometry, immunoblot, electron microscopy, or immunofluorescence. b-d Analysis of subcellular fractionation through capillary electrophoresis immunoblot. b EGFP and Th chemiluminescence peaks for H (blue) and LS1 (green) fractions. c Chemiluminescence bands of Synaptophysin1, Slc1a3/GLAST, Th and EGFP. d Relative integrated intensity for Th (grey) and EGFP (green) for each subcellular fraction (H to LP2), (mean ± SEM from 3 independent fractionations; Two-way ANOVA: Interaction F 8,36 = 0.273 p = 0.971, Fraction F 8,36 = 7.062 ****p < 0.0001, Protein F 1,36 = 2.387 p = 0.131). e Flow cytometry analysis of synaptosomes through DA-FASS sorting. The EGFP + gate was set to have 0-0.2% of events in control SYN samples. Before sorting SYN show 2-6% of EGFP + synaptosomes while DA-FASS contains 40-60% of EGFP + synaptosomes. f Averaged percentage of gated particles from SYN (N = 9) and DA-FASS (N = 8) biologically independent synaptosome preparations (mean ± SEM; Two-tailed Mann-Whitney test for all singlets SYN-DA-FASS *p = 0.01 and Two-way ANOVA for EGFP ± singlets: Interaction F 1,30 = 218.3 ****p < 0.0001, Gating F 1,30 = 66.42 ****p < 0.0001, Condition F 1,30 = 3.302 P = 0.079 with Šídák's multiple comparisons test). g Immunoblot against Th, DAT, Munc18, GluA1 and GLAST through DA-FASS (N = 1). h-l Electron micrographs of sorted synaptosomes (N = 2). h, i Typical synaptosomes displaying a bouton (b), synaptic vesicles (SV) and an opened postsynaptic membrane (arrowheads in i). j Example of a multipartite synaptosome displaying a SV-rich bouton (b1) contacting a postsynaptic membrane (arrowheads) and a second bouton (arrows) less populated with SVs (b2). k Multipartite synaptosome displaying 3 distinct presynaptic profiles (b1, b2, and b3) contacting a postsynaptic membrane (arrowheads). l Multipartite synaptosome cut through a plane that is not optimal. Boutons (b), Mitochondria (m). Scale bar, 200 nm; for uncropped immunoblots see supplementary material or original data files in source data. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-30776-9 were not studied further. We thus focused on the enriched proteins.
We first compared the 2653 proteins dataset to the broad survey of mouse brain proteins produced by Sharma and colleagues 32 . Ninety percent of our dataset is common to the global mouse brain proteome. Among 158 proteins significantly enriched in the bulk dissection of the striatum, 89 are represented in the synaptosome samples which is consistent with the selectivity of our subcellular fractionation (Fig. 3b, c, Supplementary Table 4). 403 proteins are common with those previously identified to be specific to a given cell type. A heatmap analysis of these shows the main neuronal origin of our synaptosome samples (Fig. 3d, Supplementary Table 4, Supplementary Fig. 3). We then compared the obtained proteome with the curated database of known synaptic gene ontologies (SynGO) 33 . Among 2653 genes from this proteome, 684 genes are documented in SynGO covering all localizations and functions reported in the second level of SynGO terms. This gene set is associated with synapse organization (184 genes over 306 genes in the category), a process in the presynapse (183/269), process in the postsynapse (131/218), synaptic signaling (106/193), metabolism (25/94) and transport (23/36) (Fig. 3e).
Beyond 57 proteins highly enriched during DA-FASS procedure, we identified a strong enrichment of the reporter protein mNeonGreen (12 unique peptides, 5.12-fold increase, adjusted pvalue = 1.6 × 10 −16 ; Fig. 3f, g, Supplementary Table 3). mNeon-Green enrichment thus represents the target enrichment value for the most specific dopaminergic proteins. In line with this, the major canonical proteins involved in dopamine metabolism (Th; Ddc: DOPA decarboxylase; Slc18a2/VMAT2: Vesicular Monoamine Transporter type 2) show similar enrichment values. Slc6a3/DAT displays a slightly lower enrichment that may be c Analysis of staining as in a and b showing particle proportions per frame. d STED images of EGFP (green) and Th (magenta) labelled synaptosomes. e, f Same as a, b for EGFP and D1 dopamine receptors. g Proportion of differently stained particles per frame. h STED microscopy detects D1 receptor clusters (magenta) apposed to the EGFP + synaptosomes (green). i, j Same as a-b for anti-EGFP and anti-D2 dopamine receptors. k Proportion of differently stained particles per frame. l STED images display D2R (magenta) patches apposed to EGFP (green). All data are mean ± SEM and pulled from N = 2 to N = 3 independent sorts and n = 4 to n = 11 field of view per independent sort. Each independent sort pooled at least 3 animals. Statistical significance was analyzed using Two-way ANOVA; c EGFP/Th: Interaction F 2,96 = 65.04 ****p < 0.0001, Condition F 1,96 = 0.034 p = 0.855, Immunolabelling F 2,96 = 510.3 ****p < 0.0001; g EGFP/D1R Interaction F 2,96 = 208 ****p < 0.0001, Condition F 1,96 = 0.007 p = 0.931, Immunolabelling F 2,96 = 437.9 ****p < 0.0001; k EGFP/D2R: Interaction F 2,180 = 149.4 ****p < 0.0001, Condition  Supplementary Fig. 2) 34 . Of note, 14 proteins quantified with only 2 peptides display enrichment scores much higher than the five-fold increase of our reporter. The extent of enrichment is possibly distorted by a weak detection in MS/MS. We probed the cell type expression pattern of the 57 enriched proteins with DropViz single-cell RNA sequence database. For this analysis we focused on afferent and efferent neurons in the mouse striatum 35 . This meta-analysis provides hints regarding the identity of neurons expressing the enriched markers. It defines four clusters of gene expression, from ubiquitous expression to expression restricted to Th neurons of the midbrain. This analysis further suggests that some of the DA-FASS enriched proteins belong to various partners of the synaptic hubs ( Fig. 3h and Supplementary Fig. 3). Finally, we summarized our proteome data on a model of dopaminergic transmission inspired by the KEGG database (Mmu04728) 36 to represent proteins either enriched, retained or absent from our screen ( Fig. 3i Supplementary Tables 3 and 5).

Validation of 6 new proteins enriched at dopamine synapses.
We identified a set of 12 proteins enriched in our dataset and previously shown to be important for dopamine signaling (Fig. 3f, g marked with a #). We also selected six proteins involved in membrane traffic, cell adhesion and neurotransmission, but not previously described at dopaminergic synapses, to validate   their association with dopaminergic synapses (Fig. 3f, g marked with a *). We monitored their segregation after DA-FASS using immunofluorescence. Copine7 is a C2 domain-containing, calcium-dependent, phospholipid-binding protein (Cpne7; 1.72-fold enrichment measured in MS/MS, adjusted p-value *p = 0.01; see  Table 3), which displays a strong expression in dopaminergic cells of the midbrain, but also a significant expression in cholinergic interneurons (CINs) of the striatum (here labelled Interneuron Chat) and in putative corticostriatal cells 37,38 ( Fig. 4a and Supplementary Fig. 3). We find Copine7 either co-localized or apposed to Th + synaptosomes in 8% of labelled synaptosomes, a percentage that is maintained through DA-FASS (SYN: Th + /Cpne7 + = 8 ± 0.9%; DA-FASS: Th + /Cpne7 + = 8.5 ± 0.4%, N = 3 sorts, n = 31 fields of view each; SYN-DA-FASS Šídák's multiple comparison p = 0.995) (Fig. 4b).
Altogether, we validated six new proteins from our screen for their selective association with dopaminergic synaptosomes. Interestingly, a comparison between MS/MS label-free quantification and immunodetection enrichment ratios reveals a linear correlation between the results (Fig. 4h).
Proteins retained during DA-FASS delineate the association of dopaminergic varicosities in dopamine hub synapses. To further characterize the partners in dopaminergic synaptic hubs, we Total protein content was quantified by silver staining, normalized to 140 ng for each replicate (N = 3) and cleaned onto SDS-PAGE gel before tryptic digestion. Proteins were analyzed by high-resolution tandem MS. Of 2653 quantified proteins, 63 were depleted in DA-FASS while 57 were enriched (Supplementary Table 3). b-d Venn diagrams representing the comparison of DA-FASS proteome with the mouse brain proteome 32 b as a whole, c as enriched in the striatum and d as cell type specific. d Heatmap showing cell type specific protein abundance (from 32 ) among overlapping proteins. e Overlap analysis of the DA-FASS proteome with the second level terms of the SynGO database (mostly glutamatergic and GABAergic synapse related genes) 33 . f Volcano plot of DA-FASS protein quantification. Values of fold changes versus corrected p-value are plotted for each protein (on logarithmic scales). Thresholds are set at ±1.5-fold change and p < 0.05. Proteins are colored by subclass of canonical (green) enriched (red), depleted (cyan) and retained (grey) in the DA-FASS sample. Proteins previously described as playing a role in dopamine signaling (#). Targets selected for further experimental validations (*). Statistical significance was tested using background based two-tailed pairwise t-test adjusted using Benjamini-Hochberg correction for the false discovery rate. g Complete list of depleted and enriched DA-FASS proteins. h Heatmap of mRNA abundance of the enriched DA-FASS proteins in striatal neurons (STR) or afferent cells to the striatum (SN Substantia Nigra, Thalamus, FC Frontal Cortex, PC Posterior Cortex) (DropViz; 35 ). Hierarchical clustering display 4 major clusters relating to the selectivity of mRNA expression (for a more detailed heatmap see Supplementary Fig. 3). i Scheme of the molecular organization of a dopaminergic synapse (Adapted from ref. 36 ). Enriched proteins from our DA-FASS sample are in red, retained in grey, and absent in white. Gene names for each protein class can be found in Supplementary Table 5 with absent ones greyed out. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-022-30776-9 ARTICLE NATURE COMMUNICATIONS | (2022) 13:3102 | https://doi.org/10.1038/s41467-022-30776-9 | www.nature.com/naturecommunications compared our screen with neurotransmission pathways reported in KEGG (Fig. 3i). The pathway of SV and neurotransmitter cycling shares a very high coverage with our proteome (50 protein families present out of 71 listed in the pathway; Fig. 5a grey boxed text, Supplementary Tables 3 and 5). To complete this observation, we probed for the phospho-proteins Synapsin 1&2 that are found at all presynapses 50 (abundance ratio 1.03 for both isoforms in our screen). EGFP + /Synapsin + synaptosomes representation rises from 7 to 45% upon DA-FASS (SYN: EGFP + /Synapsin + = 6.6 ± 1.2%; DA-FASS: EGFP + /Synapsin + = 44.9 ± 3.3%, N = 2 and N = 3 sorts, n = 9 and 10 fields of view respectively; SYN-DA-FASS Šídák's multiple comparison ****p < 0.0001; Fig. 5b) while EGFP − /Synapsin + synaptosomes are reduced from 83 to 34% after sort (SYN: EGFP − /Synapsin + = 83.2 ± 1.1%; DA-FASS: EGFP − /Synapsin + = 33.8 ± 2.4%, N = 2 and N = 3 sorts, n = 9 and 10 field of view, respectively; SYN-DA-FASS Šídák's multiple comparison ****p < 0.0001). (Fig. 5b).
Hence, our proteomic and immunofluorescence data support a very frequent association of dopamine presynapses with all the major synaptic partners operating in striatal neuropil. This finding further validates our earlier electron microscopy observations (Fig. 1j-l). To validate the accuracy and specificity of our results, we performed several controls. A random associations test was applied to our images in order to establish the probability for separate particles to sediment at the same sites by chance (see methods). Indeed, for all our datasets, random associations occur on less than 0.5% of all events while we observe at least 10% for synaptic hub-related associations in sorted samples (see Table 1). As a final control for the specificity of hub-synaptosome adhesion, we performed an additional VGLUT1-FASS experiment in which we selectively sorted aggregates and large events to analyze them with electron microscopy ( Supplementary Fig. 5h-j). Upon reanalysis, sorted aggregates display a strong increase in the representation of small and large aggregates ( Supplementary  Fig. 5h, i). Singlets are still strongly represented in the reanalyzed sample as it is common to break down aggregates into singlets through the shearing forces applied in the nozzle of the sorter ( Supplementary Fig. 5h, i). Electron micrographs display profiles of large particles (3-6 µm in diameter) that are difficult to relate to identifiable features of the tissue and very different from the DA-FASS synaptosomes displayed in Fig. 1 (Supplementary  Fig. 5j).
Altogether, we identified the association of dopaminergic with glutamatergic and GABAergic synapses in synaptic hub structures that may mediate the modulatory influence of dopamine over excitatory and inhibitory synaptic signalling. Cholinergic inputs from CINs may also take part in this association.
From the data gathered in Figs. 2, 4-6 we propose a model of the dopaminergic projection in which most dopaminergic varicosities adhere to postsynaptic elements labelled by either D1R or D2R with the presence of SynCAM 2 and Stx4. In addition, around 80% of dopaminergic synapses are also associated with other presynapses in synaptic hub structures clearly identified in electron and STED microscopy. Dopamine hub synapses are formed with cholinergic (14%), GABAergic (26%) and Glutamatergic synapses (42% VGLUT1 corticostriatal + VGLUT2 Thalamo-striatal; Fig. 6h).
Finally, we propose a model that summarizes our findings regarding dopamine hub synapses involving VGLUT1 (Fig. 7j). Altogether, our results show that a selective set of markers of SV cluster, active zone, postsynaptic density and spine apparatus at VGLUT1 synapses on SPNs display a significant increase upon innervation by dopaminergic varicosities. This observation strengthens the notion that dopamine hub synapses represent a mechanically resilient functional structure.

Discussion
To unravel specific molecular and cellular features of modulatory neurotransmission, we targeted the dopaminergic projection from the substantia nigra and ventral tegmental area to the striatum using FASS 18,27 . Specificity for dopaminergic synaptosomes was validated by the enrichment for presynaptic dopaminergic markers as well as the adhesion of dopaminergic varicosities to postsynaptic elements containing either D1R or D2R. We produced a proteome with proteins significantly enriched through DA-FASS purification and validated the enrichment and localization of 6 of them. We show the association of dopaminergic synapses with glutamatergic, GABAergic and cholinergic synapses in "dopamine hub synapses" that connect tightly several synapses together. Finally, we observed that innervation of glutamatergic synapses by dopaminergic varicosities correlates with a molecular strengthening of the whole synapse.
The cross-analysis of our screen with single-cell RNA sequencing data allowed us to spot the synaptic adhesion protein SynCAM 2/Cadm2 ( Fig. 3 and Supplementary Fig. 3) as a potentially important player of adhesion at dopaminergic varicosities 45,66,67 . The immunofluorescence data confirm the strong expression of SynCAM 2 at dopaminergic varicosities (Figs. 4e, 6a, b, g). SynCAM 2 was also reported to label axons 68 . Therefore, we propose that SynCAM 2 is part of an axonal adhesion complex responsible for the formation of dopaminergic synapses and hub synapses with SPNs. SynCAM 2 is thought to engage in heterophilic interactions with SynCAM 1 or 4 45,46 . Interestingly, SynCAM 1 is involved in cocaine-induced synaptic plasticity in the striatum 69 and SynCAM 2 is implicated in food intake and energy balance 70 , two phenomena directly related to the integrity of the dopaminergic system 71,72 . Besides, SynCAM 1 is thought to be preferentially acting at the postsynapses to induce presynaptic adhesion 73,74 . Hence, SynCAM 1 and 2 are strong candidates to mediate adhesion through heterophilic interaction at dopamine synapses, in addition to their roles at other types of synapses.
A previous contribution suggested that adhesion at dopaminergic synapses occurs through neuroligin 2 (Nlgn2) 75 . Even though we detected all four neuroligins and all three neurexins in our proteomic analysis, none of them displayed a specific enrichment through DA-FASS. Also, Nlgn2 mRNA is not enriched in SPNs 35 . Thus, Nlgn2 did not appear to us as a putative major player in dopamine hub synapses compared to SynCAM 2 35 . Nlgn2 action is preeminent at inhibitory synapses 76,77 . To our knowledge, there is no record for the expression of Nlgn2 on spines or terminals of excitatory synapses which would be required in the context of dopamine hub synapses with excitatory inputs 77 . One possible explanation could be that Nlgn2 plays a role in the association with inhibitory synapses in the context of synaptic hubs. Also, some findings suggest a direct inhibitory function of dopamine projection on SPNs through GABA signalling 78,79 . Nlgn2 may have a function related to this inhibitory phenotype. Finally, a combination of synaptic adhesion molecules is certainly involved and further investigations will be important to clarify the complete machinery responsible for dopaminergic hub synapse formation and maintenance 80 .
Cellular organization of dopaminergic projections to the striatum. The nature of dopaminergic synaptic structures is the topic of a long-standing debate. Previous anatomical investigations identified that the distribution of dopamine varicosities in the neuropil is biased toward proximity to glutamatergic or GABAergic synapses, but only a minority were shown to make synapses with a target structure in the striatum 4,5,8 . However, other authors reported a frequent occurrence of symmetrical synaptic contacts of dopaminergic thin axonal portions with SPNs spines or dendritic shafts 9,81-83 . Our current dataset strongly advocates for specific and frequent adhesion of dopaminergic axonal varicosities with target structures (Figs. 1h-l  and 2). Indeed, around 50% of our EGFP + varicosities displayed apposed D1R, while~80% displayed D2R labelling (Fig. 2). This is in accordance with SPNs being the main target of dopamine terminals in the striatum, with roughly half of the SPNs expressing D1 receptors, while the D2 receptor is expressed by the other half 31,84 as well as by dopaminergic and other presynapses 30,85,86 .
Moreover, our data reveal that adhesion at dopaminergic varicosity extends to synaptic hubs with glutamatergic and GABAergic synapses. We find that around a third of dopaminergic varicosities make hub synapses with putative corticostriatal VGLUT1 synapses, around 12% associate with putative thalamo-striatal VGLUT2 synapses, and more than a quarter is associated with putative VIAAT inhibitory synapses (Figs. 5 and 6h). Additionally, around 14% are also contacted by cholinergic inputs. Conversely, 41% of VGLUT1 striatal synaptosomes are contacted by a Th + input (32/(32 + 45) × 100 = 41%; Fig. 7b, j). VGLUT1 and VGLUT2 synaptosomes display little to no association when probed from a striatal sorting from VGLUT1venus mice ( Supplementary Fig. 5f, g). Also, little documentation exists supporting a GABAergic innervation on spines of SPNs 8,31 . Hence, little overlap may exist between those hub associations, and up to around 80% of dopaminergic varicosities may adhere to hub synapses rather than conventional bipartite symmetric synapses. According to the literature, cholinergic inputs to these hubs may target dopaminergic varicosities 55,57 . Further investigations will be required to characterize this in detail. As we show that most of the cytosolic content of dopaminergic axons is engulfed in synaptosomes (Fig. 1b-d), we propose that most dopaminergic presynapses are associated with hub synapses, while a small minority is not bound to any target cell in the tissue. Such a high occurrence of synaptic hubs may explain previous observations that striatal dopaminergic synaptosomes sediment faster than other synaptosomes in a linear sucrose gradient 87 . At the ultrastructural level, synaptic hubs are composed of electrondense terminals containing many SV profiles associated with clear varicosities much less populated with SVs (Fig. 1f, h). This is in accordance with previous work stating that dopaminergic terminals are less populated with SVs and appear less dense to electrons 5,81,83 .
Further investigations will be necessary to unravel whether synaptic hub formation is a structural invariant common to all sub-divisions of the striatum and whether the proportion of dopamine hub synapses of different kinds can vary depending on subregions and/or physiological states. The existence of hub synapses in other structures and modulatory inputs is an open question.
Dopaminergic input to cortico-striatal synapses correlates with increased synaptic markers. Beyond showing the existence of dopamine hub synapses, we identified that the binding of Th varicosities to cortico-striatal synapses correlates with an increase in VGLUT1, Bassoon, Homer1c, and Synaptopodin and a modest decrease in PSD-95 (Fig. 7). We found that nearly all dopaminergic terminals seem to adhere to a postsynaptic element populated with cognate receptors. Yet, around 2/3 of the dopaminergic varicosities are thought to be "silent" at a given time in the striatum and do not contain the full active zone molecular complement 12,13,88 . In synaptosomes, we found less than 20% of Th varicosities containing Bassoon, a percentage increased to 40% when Th varicosities are involved in dopamine hub synapses. We also found 70% of Th synaptosomes positive for Rim1. The precise combination of active zone scaffolds necessary and sufficient for dopamine release at hub synapses remains to be established. Similarly, it remains unclear whether dopamine release is required to induce the molecular potentiation we describe or whether trans-synaptic signalling of adhesion complexes is sufficient 10 . It will be important to characterize the function of newly identified proteins such a Syntaxin-4 or Syn-CAM 2 in the differentiation process.
The discovery of a molecular differentiation at synaptic hubs provides a unique ex vivo paradigm to study the complex interactions of receptors-through signalling crosstalk or heteromeric interactions-identified in the past decades [14][15][16][17] . Therefore, the question of the co-recruitment of glutamate or GABA receptors with dopamine receptors at synaptic hubs is raised and the plasticity of this recruitment upon reward-based processing and in dopamine-related pathological states remains to be established. Beyond, several metabotropic receptors (to adenosine, cannabinoid, glutamate and acetylcholine) are also important players in the striatal integration of cortical and thalamic inputs. The increase in Homer1c suggests a potential involvement of metabotropic glutamate receptors in the differentiation process 60 . Also, both Homer1 and Synaptopodin were shown to be involved in calcium signalling regulation, a point of interest for future investigations 60,63 . Downstream targets of signalling such as ionic channels may also take part in the critical scaffolds at play 31,89 .
Altogether, our work paves the way for a better understanding of dopaminergic synaptic transmission in physiology and pathology 90 . Future developments will allow a more thorough multi-omics 91 as proposed recently with other techniques 92,93 . More generally, results from our study and the work of Apóstolo and colleagues 94 on mossy fibre terminals of the hippocampus show that FASS synaptomics is a powerful workflow for exploring projection-specific synaptomes 18,95 .

Methods
Animals. We used a transgenic mouse line expressing cre recombinase under the control of the dopamine transporter (DAT-cre + ; MGI:3770172, RRID: MGI:3770172) 23 , WT C57BL/6 N littermates, as well as a VGLUT1 venus knock-in mouse line (MGI Cat# 5297706, RRID:MGI:5297706) 58 . Mice were maintained in C57BL/6 N background and housed in 12/12 LD with ad libitum feeding, 50-70% humidity, and 18-22°C ambient temperature. Every effort was made to minimize the number of animals used and their suffering. The experimental design and all procedures were in accordance with the European guide for the care and use of laboratory animals and approved by the ethics committee of Bordeaux University (CE50) and the French Ministry of Research under the APAFIS n°8944 and #21132.
AAV Vector and stereotaxic injection. Stereotaxic injections were performed in heterozygous DAT-cre + and wild-type (WT) mice of either sex at 8 to 11 weeks of age 96 . An Adeno-Associated Virus (AAV) containing an inverted sequence of EGFP (AAV1 pCAG-FLEX-EGFP-WPRE, University of Pensylvania) 22 or mNeongreen (AAV1 pCAG-FLEX-mNeongreen-WPRE) 97 coding gene flanked by loxP-sites was injected into DAT-cre + mice ( Fig. 1 Panel 1). Saline injected littermates were used as autofluorescence controls. The stereotaxic injections were performed in Isoflurane-anesthetized mice using a 10 μl NanoFil syringe and a 35 G beveled NanoFil needle (World Precision Instruments). Injection coordinates for the Substantia Nigra pars compacta (SNc) were anterior/posterior (A/P)− 3.6 mm, lateral (L)± 1.3 mm, dorsal/ventral (D/V)− 4.2 mm. Injection coordinates for the Ventral Tegmental (VTA) were with a 12°angle A/P− 2.9 mm, L± 1.6 mm; D/V− 4.6 mm. A/P and L coordinates are given with respect to the bregma, whereas D/V coordinates are given with respect to the brain surface ( Fig. 1 Panel 1). The animals were euthanized after 28-35 days at the maximal viral EGFP/ mNeongreen expression. For each fluorescence-activated synaptosome sorting (FASS) experiment, three to six DAT-cre + mice and one WT mouse were used independently of their sex.
Subcellular fractionation of synaptosomes. The preparation of synaptosomes was adapted from a previously published protocol 25 . Briefly, animals were euthanized by cervical dislocation, decapitated and the head was immersed in liquid nitrogen for a few seconds. The striatum of WT and bright fluorescent parts of the striatum of DAT-cre + mice were subsequently dissected under an epi-fluorescence stereomicroscope (Leica Microsystems, Germany, Fig. 1 Panel 2). Non-fluorescent control striata were dissected following anatomical borders. Samples were then homogenized in 1.5 ml of ice-cold Isosmolar buffer (0.32 M sucrose, 4 mM HEPES pH7.4, protease inhibitor cocktail Set 3 EDTA-free (EMD Millipore Corp.)) using a 2 ml-glass-Teflon homogenizer with 12 strokes at 900 rpm. The homogenate (H) was centrifuged at 1000 × g for 5 min at 4°C in a benchtop microcentrifuge. The supernatant (S1) was separated from the pellet (P1) and centrifuged at 12,600 × g for 8 min at 4°C. The crude synaptosomes pellet (P2) was resuspended in 350 µL of isosmolar buffer and layered on a two-step ficoll density gradient (5 mL of 13% Ficoll and 5 mL of 7.5% Ficoll, both in 0.32 M sucrose, 4 mM HEPES). The gradient was centrifuged at 50,000 × g for 1 h and 10 min at 4°C (Thermo Sorvall WX Ultra 90 with a Th 641 rotor). The synaptosome fraction (SYN) was recovered at the 7.5 and 13% ficoll interface using a 0.5 ml syringe. For complete subcellular fractionation 200 µL of the P2 fraction was transferred to a 10 cm 3 ice-cold glass/Teflon potter and quickly homogenized at full speed in 1.8 mL ultrapure water to create an osmotic shock. For synaptic vesicle fractionation the lysate was centrifuged at 25,000 × g for 7 min at 4°C. The lysate supernatant (LS1), was centrifuged at 200,000 × g for 120 min at 4°C. The supernatant LS2 was collected, concentrated to 50 µL and aliquoted. The pellet containing crude synaptic vesicles (LP2) was resuspended in 50 µL of ice-cold isosmolar buffer and aliquoted. The lysate pellet LP1 was centrifuged on the same discontinuous ficoll gradient as for synaptosomes (7.5-13%) at 60,000 × g for 33 min at 4°C and the fraction at the interface of the two gradients containing synaptic plasma membranes (SPM) was collected and aliquoted. For each complete subcellular fractionation independent experiment, three DAT-cre + 15 weeks old mice of either sex were pooled.
Simple Western™ immunoblot. Proteins collected on filters were solubilized in 70 μl of SDS-PAGE loading buffer at 70°C for 3 min and cooled down on ice. Conventional gradient synaptosomes total proteins were titrated using the Bradford assay. FASS samples were titrated using silver staining performed on SDS-PAGE gels against a standard curve of gradient synaptosomes. Detection proteins of interest were determined using an automated capillary electrophoresis-based immunoblot to separate, identify and quantify a protein of interest (WES, Pro-teinSimple, USA). Reagents (Dithiothreitol, DTT; Fluorescent 5× Master Mix, Biotinylated Ladder) were prepared according to the manufacturer's protocol. Samples were diluted with 0.1× Sample Buffer and mixed with 5× Master Mix (4 to 1) to obtain 50 ng/mL and finally denatured 5 min at 70°C. Primary antibodies were diluted to their tested optimal concentration and Luminol-Peroxide (1 to 1) mix was prepared. The plate was filled following the protocol scheme (5 µL of Biotinylated Ladder, 5 µL of Samples, 10 µL of Wes Antibody Diluent, 10 µL of Primary Antibody, 10 µL of Streptavidin-HRP, 10 µL of Secondary Antibody and 15 µL of Luminol-Peroxide Mix). Simple Western™ standard immunodetection protocol was run (separation matrix loading: 200 s, stacking matrix loading: 15 s, sample loading: 9 s, separation: 25 min at 375 V, antibody diluent: 5 min, primary antibody: 30 min, secondary antibody: 30 min, detection: high dynamic range). Capillary chemiluminescent images captured through a charge-coupled device camera were analyzed by the manufacturer's Compass software. Briefly, the protein peak area under the curve (AUC) was fitted using a Gaussian distribution. The fitted protein AUC is expressed either as a ratio to the fitted AUC H fraction for each WES (Fig. 1) or as fitted protein AUC (Fig. 7).
Immunofluorescence. Synaptosomes on coverslips were fixed (4% Paraformaldehyde, 4% sucrose, 1× PBS) for 10 min at room temperature, washed three times with PBS for 5 min and then stored at 4°C until use. Synaptosomes were blocked and permeabilized with PGT buffer (PBS, 2 g/L gelatin, 0.25% Triton X-100 and when needed 5% normal goat serum) and subsequently incubated with primary antibodies in PGT buffer (1 h at room temperature), washed three times with PGT and incubated with secondary antibodies in PGT (1 h at room temperature). Three final washes with PGT buffer were performed prior to a washing step in 1× PBS and a final rinse in ultrapure water. Coverslips were mounted on glass slides with Fluoromount-G mounting solution (Sigma) and stored at 4°C until observation.
Antibodies. All antibodies used and their dilution are reported in Supplementary  Table 1.

Proteomics
Sample preparation and protein digestion. Triplicates of 35*10 6 DA-FASS synaptosomes were accumulated for proteomic analyses and were compared to triplicates of 35*10 6 SYN synaptosomes. Synaptosomes counts were obtained through the cytometer software BD FACSDiva v9.0.1. Both samples were treated in parallel at all steps. Protein samples were solubilized in Laemmli buffer. A small part of each triplicate was analyzed by silver staining using SilverXpress R staining kit (Invitrogen, Cat#LC6100). Protein content was normalized across triplicates to 140 ng (least concentrated sample of the series) and ran onto SDS-PAGE (Sodium Dodecyl Sulfate-Poly Acrilamide Gel Ellectrophoresis) for a short separation. After colloidal blue staining, each lane was cut in 2 bands, subsequently cut in 1 × 1 mm gel pieces. Gel pieces were unstained in 25 mM ammonium bicarbonate 50% Acetonitrile (ACN), rinsed twice in ultrapure water and shrunk in ACN for 10 min. After ACN removal, gel pieces were dried at room temperature, covered with the trypsin solution (10 ng/µl in 50 mM NH 4 HCO 3 ), rehydrated at 4°C for 10 min, and finally incubated overnight at 37°C. Samples were then incubated for 15 min in 50 mM NH 4 HCO 3 at room temperature with rotary shaking. The supernatant was collected, and an H 2 O/ACN/HCOOH (47.5:47.5:5) extraction solution was added to gel slices for 15 min. The extraction step was repeated twice. Supernatants were pooled and dried in a vacuum centrifuge. Digests were finally solubilized in 0.1% HCOOH.
nLC-MS/MS analysis and label-free quantitative data analysis. The peptide mixture was analyzed on a Ultimate 3000 nanoLC system (Dionex, Amsterdam, The Netherlands) coupled with an Electrospray Orbitrap Fusion™ Lumos™ Tribrid™ Mass Spectrometer (Thermo Fisher Scientific, San Jose, CA). Ten microliters of peptide digests were loaded onto a 300 µm-inner diameter × 5 mm C 18 PepMap TM trap column (LC Packings) at a flow rate of 10 µL/min. The peptides were eluted from the trap column onto an analytical 75 mm id × 50 cm C18 Pep-Map column (LC Packings) with a 4-40% linear gradient of solvent B in 105 min (solvent A was 0.1 % formic acid and solvent B was 0.1% formic acid in 80% ACN). The separation flow rate was set at 300 nL/min. The mass spectrometer operated in positive ion mode at a 1.8 kV needle voltage. Data were acquired using Xcalibur 4.3 software in a data-dependent mode. MS scans (m/z 375-1500) were recorded in the Orbitrap at a resolution of R = 120 000 (@ m/z 200) and an AGC target of 4 × 10 5 ions collected within 50 ms. Dynamic exclusion was set to 60 s and top speed fragmentation in HCD mode was performed over a 3 s cycle. MS/MS scans were collected in the Orbitrap with a resolution of 30 000 and a maximum fill time of 54 ms. Only +2 to +7 charged ions were selected for fragmentation. Other settings were as follows: no sheath nor auxiliary gas flow, heated capillary temperature, 275°C; normalized HCD collision energy of 30%, isolation width of 1.6 m/z, AGC target of 5 × 10 4 and normalized AGC target of 100%. Advanced Peak Detection was activated. Monoisotopic precursor selection (MIPS) was set to Peptide and an intensity threshold was set to 2.5 × 10 4 .
Database search and results processing. Data were searched by SEQUEST through Proteome Discoverer 2.5 (Thermo Fisher Scientific Inc.) against the Mus musculus SwissProt protein database (v2021-02-04; 17,050 entries) added with the green fluorescent reporter (mNeonGreen). Spectra from peptides higher than 5000 Daltons (Da) or lower than 350 Da were rejected. Precursor Detector node was included. Search parameters were as follows: mass accuracy of the monoisotopic peptide precursor and peptide fragments was set to 10ppm and 0.02 Da, respectively. Only b-and y-ions were considered for mass calculation. Oxidation of methionines (+16 Da), phosphorylation of serines, threonines and tyrosines (+79), methionine loss (−131 Da), methionine loss with acetylation (−89 Da) and protein N-terminal acetylation (+42 Da) were considered as variable modifications while carbamidomethylation of cysteines (+57 Da) was considered as a fixed modification. Two missed trypsin cleavages were allowed. Peptide validation was performed using Percolator algorithm 98 and only "high confidence" peptides were retained corresponding to a 1% false positive rate at the peptide level. Peaks were detected and integrated using the Minora algorithm embedded in Proteome Discoverer. Proteins were quantified based on unique and razor peptides intensities. Normalization was performed based on the total protein amount. Protein ratios were calculated as the median of all possible pairwise peptide ratios. Background Based pairwise t-test was used to calculate p-values adjusted using Benjamini-Hochberg correction for the false discovery rate. This method assumes that most protein abundances usually do not change in response to any stimulus and automatically determines the range of protein ratios that are essentially constant between conditions and then tests each protein ratio against the median and variance derived from this background population. Quantitative data were considered for proteins quantified by a minimum of two peptides.
The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD027534 99 .
Meta-Analysis with other databases. Meta-analysis was carried out using databases from the mouse brain proteome 32 , SynGO 33 and DropViz 35 . Volcano plots and heatmaps were created using python based bioinfokit 100 .
Epifluorescence microscopy and image processing. Immuno-stained synaptosomes were imaged using either a Nikon Eclipse NiU (with a ×40/NA 0.75 dry objective equipped with a sCMOS ANDOR Zyla 5.5 camera), a Leica DMI8 epifluorescence microscope (with a ×63/NA 1.4 oil immersion objective equipped with a sCMOS Hamamatsu FLASH 4.0v2 camera) or a Leica DM5000 epifluorescence microscope (with a ×40/NA 1.25 immersion objective equipped with a sCMOS Hamamatsu FLASH 4.0 camera) and Metamorph software. Ten to twenty frames were chosen randomly on each coverslip and imaged.
Correlation of synaptosomes' labelling is automated by a homemade macrocommand, using the ImageJ software 101 (SynaptosomesMacro_Randomization, version 10, with functional description and code that can be accessed here https:// github.com/fabricecordelieres/IJ-Toolset_SynaptosomesMacro). The workflow is composed of three steps. First, the images are pre-processed. The original images, trans typed to 32-bits, are centred and reduced: their respective average intensity is subtracted and division by their standard deviation is performed. It is assumed that both signals lay close one from the other: both images are therefore combined into one to serve for synaptosomes' detection. On each pixel, the maximum signal from both channels is retained to produce a new image, subjected to both median filtering and gaussian blurring (3 pixels radius). Each potential synaptosome now appears as a bell-shaped blob, which centre might be determined using a local maximum detection (tolerance to noise: 3). Second, the detections are reviewed and user-validated. Part of the original images is cropped around the local maxima and displayed to the user as a mosaic. Each thumbnail is displayed on a clickable frame, allowing the user to include or reject a signal detection from the analysis. Criteria of rejection included: the presence of competing particles in the quantification area, bad focus on the particle, proximity of the image border preventing proper quantification. Finally, data is extracted, exported, and displayed. A circular region is positioned over the centre of the thumbnail. The centroid's coordinates are retrieved and logged. From the two sets of coordinates (one per channel), the intersignal distance is computed and stored in µm. Signal quantification is performed by placing a circular region of interest (24 pixels radius) around the centroid, and measuring the integrated intensity. A measurement of the local background is performed through a doughnut-shaped ROI surrounding the measurement circle. All values are logged for both channels, for all retained structures and reported in a "_Pooled_CytoFile.csv" file. Further analysis was performed using the FlowJo and GraphPad PRISM softwares. xy-plots of integrated intensity values are displayed with a quadrant analysis of single or double signal detections. Quadrant gate positions were defined from raw images using the distribution of fluorescence intensities in the ROI. In the range of fluorescence intensities between the background peak and the next peak, we determined the quadrant borders at the values where the ROI start to show a clear punctum of the signal. For all analyses, randomly chosen particles were displayed in a gallery to give an overview of the population analyzed.
For association analysis, a randomization plugin was developed and integrated into the "SynaptosomesMacro_Randomization". The source code is available here: https://github.com/flevet/RandomizerColocalization, while the compiled version can be found here: https://github.com/fabricecordelieres/IJ-Toolset_ SynaptosomesMacro/blob/master/Plugins/RandomizerColocalization_.class. Two separated particles were considered associated if d < 2 µm, with d being the Euclidian distance between their centroid. To statistically determine if these associations were significant or happening by chance, we performed randomization tests. Previously stored particles centroids coordinates were retrieved. For each colour channel, we fixed the position of its particle while randomizing all the ones of the other channel using 10,000 Monte-Carlo simulations. Since there is no underlying structure, the probability of having a particle at a certain position is identical for the whole image space. Consequently, randomization was performed by generating a complete spatial random distribution having the same number of points as the number of particles of the channel being randomized. Associations between 2 markers were then computed as explained above. The final random association values reported were defined as the mean of 10,000 randomizations. All randomization values expressed in percentages are logged for both channels, for all retained structures and reported in a "_Pooled_RandomizationResults.csv" file.
Stimulated emission depletion (STED) microscopy. Images were acquired using a Leica TCS SP8/STED3X microscope equipped with a HC PL APO 93×/1.30 GLYC motCORR -STED WHITE objective and Leica Application Suite X (LAS X) software. We used depletion laser lines at 592 nm for Alexa488 and 775 nm for Alexa594 or ATTO647n fluorophores. A 25% 3D-STED effect was applied to increase Z resolution. Metrology measurements were regularly performed using fluorescent beads to test proper laser alignment. Less than 2 pixels (pixel size = 20 nm) shift between channels was measured.
Electron microscopy. Synaptosomes for transmission electron microscopy were fixed right after centrifugation on coverslips with a 1% Glutaraldehyde and 2% PFA (Electron Microscopy Sciences) in 1× PBS solution and kept at 4°C until further treatment. They were then washed with PB and post-fixed in 1% osmium tetroxide and 1% K3Fe(CN)6 in PB for 2 h on ice in the dark. Washed in H 2 O and dehydrated in an ascending series of ethanol dilutions (10 min in 50% ethanol, 10 min in 70% ethanol, twice 15 min in 95% ethanol, twice 20 min in absolute ethanol). After absolute ethanol, coverslips were lifted into Epon 812 resin (Electron Microscopy Sciences) and 50% ethanol for 2 h at room temperature and then left in pure resin overnight at 4°C. Coverslips were then placed on microscope slides, embedded with capsules filled with pure resin and polymerized at 60°C for 48 h. The resin block was then trimmed with razor blades. Sections, 65 nm thick, were then cut using a diamond knife Ultra 35°(Diatome) with an ultra-microtome (Leica UC7) and collected on 150 mesh copper grids (Electron Microscopy Sciences).
The sections were stained with UranyLess® (Chromalys and Deltamicroscopy). Samples were then observed with a Hitachi H7650 transmission electron microscope equipped with a Gatan Orius CCD camera. Synaptosomes were identified by their size (0.5-2 μm), their shape and the presence of intracellular compartments and organelles such as vesicles.
WES' data statistical analysis was performed using Two-way ANOVA and Šídák's multiple comparison testing (Supplementary Table 2, Figs. 1d and 7c).
Proteomics data was tested using a Background Based pairwise t-test used to calculate p-values adjusted using Benjamini-Hochberg correction for the false discovery rate and is described in the proteomics section of the methods. Proteins with an abundance ratio above 1.5 or below 0.75 were considered different to the control providing that data displayed a statistically adjusted p-value lower than 0.05.
No statistical method was used to predetermine sample size. Synaptosomes detected during image analysis were excluded when overlapping biological material prevented an accurate quantification or when the focus was not good enough for accurate quantification. When possible, experimental results were confirmed using a different method (WES immunoblots, different sort criteria like VGLUT1 venus for instance). Each replication attempt has been successful in reproducing the data. The experiments were not randomized. The Investigators were not blinded to allocation during experiments and outcome assessment.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
All data supporting the findings of this study are provided within the paper and its Supplementary Information. DA-FASS Proteomic dataset is publicly available at the ProteomeXchange Consortium http://proteomecentral.proteomexchange.org/cgi/ GetDataset via the PRIDE partner repository with the dataset identifier PXD027534. The mouse brain proteome is available at http://www.mousebrainproteome.com/. SynGO database is available at https://syngoportal.org/index.html. DropViz database is available at http://dropviz.org/. KEGG database is available at https://www.genome.jp/kegg/ pathway.html. Source data are provided with this paper as Source data file. The WES, FACS and immunofluorescence data generated in this study have been deposited in the Zenodo database (https://doi.org/10.5281/zenodo.6482952). Because of their size, the raw microscopy images underlying the results will be made available upon request to the corresponding authors. Requests will be answered within a week. Source data are provided with this paper.

Code availability
The macro code to perform automated correlation of synaptosomes' labelling and association analysis are available on zenodo at the following link: https://doi.org/10.5281/ zenodo.6483605.