The human and mouse synaptome architecture of excitatory synapses show conserved features

Large-scale mapping of the location of synapses and their molecular properties in the mouse has shown that diverse synapse types are spatially distributed across the brain. The diversity of synapses is known as the synaptome and the spatial distribution as the synaptome architecture. Synaptome maps in the mouse show each brain region has a characteristic compositional signature. The signature can store behavioral representations and is modified in mouse genetic models of human disease. The human synaptome remains unexplored and whether it has any conserved features with the mouse synaptome is unknown. As a first step toward creating a human synaptome atlas, we have labelled and imaged synapses expressing the excitatory synapse protein PSD95 in twenty human brain regions in four phenotypically normal individuals. We quantified the number, size and intensity of approximately a billion individual synaptic puncta and compared their regional distributions. We found that each region showed a distinct signature of synaptic puncta parameters. Comparison of brain regions showed the synaptome of cortical and hippocampal structures were similar but distinct to the synaptome of cerebellum and brainstem. Comparison of human and mouse synaptome revealed conservation of synaptic puncta parameters, hierarchical organization of brain regions and network architecture. These data show that the synaptome of humans and mouse share conserved features despite the 1000-fold difference in brain size and 90 million years since a common ancestor. This first draft human synaptome atlas illustrates the feasibility of generating a systematic atlas of the human synaptome in health and disease.


Introduction
For over a century synapses have been recognised as a hallmark of brain anatomy, yet there are no reference resources that report the number and distribution of synapse types at the single synapse resolution across human brain regions. The importance of understanding the molecular anatomy of synapses is underscored by the remarkable growth in knowledge about the importance of synaptic proteins in disease. In 2000 there were three human genetic disorders known that disrupted proteins in excitatory synapses (Husi et al., 2000) and by 2011 this had been extended to 133 common and rare brain disorders (Bayes et al., 2011). Schizophrenia, depression, autism spectrum disorders and intellectual disability are all caused by gene mutations targeting synapse proteins (Husi et al., 2000;Fernandez et al., 2009;Bayes et al., 2011;Nithianantharajah et al., 2013;Fromer et al., 2014;Purcell et al., 2014;Fernandez et al., 2017;Howard et al., 2018;Satterstrom et al., 2019). The expression of synapse proteins is also altered in neurodegenerative disorders, brain injury (including stroke and traumatic brain injury), neuroinflammatory conditions (including multiple sclerosis and autoimmune encephalopathies) and addiction (including alcohol and cocaine). It is now accepted that synapse molecular pathology is a feature of most brain disorders.
Mammalian synapses contain thousands of evolutionarily conserved proteins (Bayés et al., 2011;Bayes et al., 2012;Bayés et al., 2017) and these are differentially distributed across the brain (Emes et al., 2008;Hawrylycz et al., 2012;Roy et al., 2018a;Roy et al., 2018b). A transformative concept, that has emerged from the capacity to image synaptic proteins in individual synapses, is that there is a far greater diversity than could be anticipated from historical studies of synapse morphology, neurotransmitters and physiology (O'Rourke et al., 2012;Zhu et al., 2018a). Driven by the ability to systematically characterise the molecular composition of individual synapses across the whole brain, the term 'synaptome' has been introduced to describe the catalogue of synapses in the brain or part thereof (Zhu et al., 2018a).
To map the mouse synaptome, we developed a suite of technologies into a pipeline called SYNMAP: endogenous synaptic proteins are labelled (with genetic tags or immunolabelling) and tissue sections are imaged with high-speed spinning disc confocal microscopy, then sophisticated image analysis tools exploiting computer vision and 3 machine-learning detect individual synapses and measure their molecular and morphological properties, then classify the synapses and map their spatial locations in brain atlases. Mapping the spatial location of synapse types and subtypes in mice reveals there is synaptome architecture of the brain (Zhu et al., 2018a). Synapse diversity is observable at all scales -individual dendrites, cell types, brain regions to the global systems level (Zhu et al., 2018a). Studies of genetic disorders in mice show that synapse diversity is modified and the synaptome architecture, suggesting that this could be a common feature of synaptic diseases.
The synaptome architecture of human brains is poorly understood and as a first step toward creating a human synaptome atlas, we recognized the need to overcome several challenges. First, the need to develop robust molecular labelling of excitatory synapses in post-mortem brain tissue using immunohistochemistry (IHC) suitable for analysis using the SYNMAP pipeline. We chose to label excitatory synapses using antibodies to PSD95, which is an abundant postsynaptic protein known to play a key role in physiology, behaviour and disease (Migaud et al., 1998;Bayés et al., 2011). To optimize and validate our IHC labelling of PSD95, we took advantage of two lines of genetically modified mice that we had previously characterized: mice lacking PSD95 (PSD95-/-) (Migaud et al., 1998) and mice expressing PSD95 fused to green fluorescent protein (PSD95eGFP) (Zhu et al., 2018a). PSD95-/-and PSD95eGFP mice serve as negative and positive controls, respectively, for antibody staining. Second, given that the human brain is 1000-fold larger than that of the mouse, we needed to explore the feasibility of acquiring single-synapse resolution data across the major brain regions.
Although PSD95 immunostaining has been reported in specific human brain regions, including frontal cortex, caudate nucleus, putamen and hippocampus (Glantz et al., 2007;Fourie et al., 2014;Morigaki & Goto, 2015), there is no single comprehensive study of PSD95 distribution across a number of human brain areas at single-synapse resolution using a high-throughput quantitative analysis.
Here we report a detailed qualitative and quantitative survey of the distribution and diversity of PSD95-labelled synapses throughout 20 brain areas from four phenotypically normal individuals. We describe the synaptome architecture of human brain regions and analyze regional similarity within the human brain and between human and mouse brain.

4
We discuss the utility of this dataset for studies of the normal and diseased brain and the logistical considerations for large-scale human synaptome mapping.

Consent
Post-mortem human brain tissue was obtained from the Medical Research Council UK (MRC) funded University of Edinburgh Brain Bank (EBB). All procedures involving post-mortem human brain tissue were approved by East of Scotland Research Ethics Service (16/ES/0084). Informed written consent was obtained in releation to each subject.

Human subjects
Human brains from four control subjects (three males, one female; mean age 51 ± 6.9 (standard deviation) years) were used to image PSD95 using laser scanning confocal microscopy (LSCM). Details of the human subjects are listed in Table 1. None of the control subjects had a history of dementia, neurological or psychiatric disorders at the time of death. Human brain tissue donated to the EBB was processed according to departmental protocols. Briefly, formalin-fixed paraffin-embedded (FFPE) 4 μm tissue sections were stained with hematoxylin and eosin (H&E) for routine neuroanatomical and neuropathological surveys. Selected sections were stained with a panel of antibodies, which included phosphorylated tau for neuronal and glial inclusions, beta-amyloid for vascular and parenchymal amyloid deposition, ubiquitin for ubiquitinated aggregates and/or alpha-synuclein for Lewy body pathology. Gross anatomical and microscopic examinations of the brain tissue revealed only low age-related neuropathological changes in these control brains (Table S1).

Post-mortem procedures
At post-mortem examination brains were macroscopically dissected into 1 cm-thick coronal slices and sampled according to the EBB protocol (Samarasekera et al., 2013).
Small tissue blocks were immediately fixed in 10% formaldehyde for 24-72 h prior to further tissue processing and embedding in paraffin wax. The blocks were sampled from 13 neocortical areas (six frontal, four temporal, one parietal, two occipital), one allocortical 5 area (hippocampus), two subcortical grey matter areas (thalamus and caudate nucleus) and four infratentorial regions (midbrain, pons, medulla and cerebellum) ( Table 2). The 13 neocortical areas are defined as Brodmann areas, and include sensory visual regions (BA17 and BA19), sensory auditory region (BA41/42), motor region (BA4), premotor region (BA6/8), associative frontal regions (dorsolateral BA9 and BA46, ventrolateral BA44/45, and orbital BA11/12), associative temporal regions (BA20/21 and BA38), associative parietal region (BA39) and associative occipitotemporal region (BA37). The pathological changes in neurodegenerative diseases are present throughout the brain parenchyma; however, careful selection of representative areas allows efficient screening of tissue for any potential disease entities (King et al., 2013). The choice of the anatomical regions examined was additionally guided by the availability of all 20 blocks for the four control cases. Tissue blocks were macroscopically dissected by experienced pathologists and the selection of blocks closely followed topographical brain anatomy using the patterns of sulci and gyri. Regions of interest were further identified using an atlas of human brain (Mai et al., 2015). Nissl-stained sections from blocks immediately adjacent (where possible) to those used for IHC were examined to assess and confirm the distinctive cytoarchitectural features of various areas. The absence of significant neuropathological findings was further confirmed by routine examination of immunohistochemical and special stains available from the Department of Academic Neuropathology, Edinburgh University.

Fluorescence immunohistochemistry
FFPE human brain sections were dewaxed and dehydrated as per the EBB neuropathology laboratory protocol. Briefly, sections were dewaxed in two 3-min xylene washes followed by four 3-min graded alcohol washes, comprising two 74°OP (IMS99%) alcohol washes and two 70% alcohol washes. This was followed by a 15-min wash in saturated alcoholic picric acid to remove any formalin sedimentation and a final wash in running tap water for 15 min. Antigen and heat retrieval techniques were used to enhance cell surface staining, antigen detection from antigen-masked FFPE sections and reduction of non-specific background staining. The slides were placed in 250 ml freshly made antigen retrieval solution (0.1 M sodium citrate buffer pH 6.0, Fisher Scientific) and 6 pressure cooked (A. Menarini Diagnostics) at 125°C for 30 s. After cooling in tap water, sections were washed once with PBS and blocked with 5% bovine serum albumin (BSA, Sigma-Aldrich) in 1x Tris-buffered saline (TBS, Fisher BioReagents) containing 0.2% Triton X-100 (Sigma-Aldrich) for 1 h at room temperature (RT). Sections were then incubated overnight at 4°C with 16 ng/ml mouse anti-PSD95 monoclonal IgG2a (NeuroMab) primary antibody diluted (1:250 of 4 μg/ml stock) in TBS containing 3% BSA and 0.2% Triton X-100. After three 5-min washes with TBS and 0.2% Triton X-100, secondary antibodies Alexa Fluor 546 goat anti-mouse IgG2a (γ2a) (Molecular Probes) were added at 1:500 and sections incubated for 2 h at RT. Each IHC staining round was accompanied by nuclear cell counterstaining with 4',6-diamidino-2-phenylindole (1μg/ml DAPI (Sigma) diluted 1:1000 in PBS). Finally, the slides were washed three times with TBS and 0.2% Triton X-100. For elimination or reduction of lipofuscin-like autofluorescence, sections were subsequently treated with Autofluorescence Eliminator Reagent (AER, Millipore Chemicon International). Lipofuscin pigment accumulates in the cytoplasm of brain cells and can hamper fluorescence microscopy owing to its broad excitation and emission spectra. AER is a commercially available Sudan Black (SB)based reagent that has been successfully used in studies using human tissue (Blazquez-Llorca et al., 2010). SB has been reported to provide the best compromise between the reduction of lipofuscin-like fluorescence and maintenance of specific fluorescent labels (Schnell et al., 1999). Briefly, the sections were first immersed in 70% ethanol for 5 min, followed by the AER for 30 s. Slides were washed in 70% ethanol for a further 3 min before being mounted with mounting medium (Mowiol with 1,4-diazabicyclo[2.2.2]octane (DABCO), Sigma-Aldrich) and left to dry overnight.

Validation of PSD95 IHC with PSD95eGFP mouse
Qualitative and quantitative comparisons with a selection of images from PSD95eGFP mice were made to validate the human PSD95 staining. At first, the representative IHC images of coronal sections from several mouse brain areas were acquired using the same LSCM and with the same parameter set-up. Then, a direct quantitative comparison of PSD95eGFP mouse versus human results was made for three wild-type (WT) male mice (aged 18 months) and three human male control subjects from 7 ten brain regions. The PSD95eGFP mouse data were acquired with a spinning disk microscope (SDM) and analysed using the Ensemble analysis as previously described (Zhu et al., 2018a).

Image acquisition
To resolve individual synaptic puncta, antibody-labelled brain tissue sections were captured using a Zeiss laser scanning confocal microscope (LSCM510) using a Zeiss Plan-Apochromat 63x oil-immersion objective lens (NA 1.4) with a frame size of 1024 x 1024 pixels, a pixel dimension of 46 x 46 nm and a depth of 8 bit. All raw images were obtained with x3.1 zoom as stacks of six image planes with z-step of 130 nm pixels and a total length of axial plane 780 nm pixels. The total area used for synaptic density determination varied depending on region, but for cortical areas 18 images randomly distributed per Brodmann area (three per cortical layer) per subject were acquired. A total of 1510 confocal images were acquired, of which 21 (1.4%) were excluded due to technical problems, including too low signal-to-noise ratio precluding adequate image analysis, a lack of alignment in z-stacks, and images with overexposed synaptic puncta.
It is estimated that more than half a billion individual synaptic puncta were identified and counted in this study. Images and figure montages were prepared in Adobe Photoshop and Adobe Illustrator. Where necessary, images were adjusted for brightness for correct display, but no other corrections were made.
For the visualisation of the whole-mounted brain sections, a Zeiss Axio Scan.Z1 Slide Scanner (Carl Zeiss) with a 20x lens (NA 0.8) was used. Images were obtained using sections stained with secondary antibodies conjugated to Alexa Fluor 546, DAPI and Alexa Fluor 633. Pixel resolution was 0.325 x 0.325 μm and image brightness is in a depth of 16 bits. The ZEN lite 2012 software (Zeiss) was used with the scanner for image acquisition and adjustment. Coronal brain section images were background and contrast adjusted to provide consistent comparison of PSD95 expression. All control images were directly comparable with their negative controls using the same settings. The slide scanner-captured images were used only for qualitative comparisons. The slide scanner was also used to capture Nissl-stained material for direct assessment of tissue integrity and cytoarchitecture of sections used in this study.

Image analysis
Image analysis of the human synapse images includes image puncta detection, segmentation and parameter quantifications. The detection and segmentation of individual puncta were done by ensemble learning, a machine learning approach developed in house (Zhu et al., 2018a). Ensemble learning of image detection requires an annotated image set from the LSCM dataset for algorithm training. The raw tile images were firstly selected from the whole data set using random bootstrapping. PSD95-labelled individual synaptic puncta in each of the images were then manually located using the Fiji plugin 'Cell Counter' (Schindelin et al., 2012) by two independent experts. Half of the annotated data were used for training and the other half for testing of thee trained detector. A K folder strategy was also used for validation of the training.
The second step in the ensemble image analysis involves algorithm training of synaptic puncta detector, which is based on a multi-resolution image feature detection and supervised machine-learning technique. A multi-resolution and multi-orientation version of 2nd-order nonlocal derivative (NLD) to calculate intensity differences, or image features, was developed for each of all individual puncta. For instance, for an image of PSD95 IHC, a group of 109 feature images were calculated per punctum. These intensity differences were assembled as feature vectors of each individual punctum. Details of the algorithm training can be referred to our previous work (Zhu et al., 2018a). Once training was finished, the trained ensemble detector was applied to automatically detect puncta in the LSCM images. Once being detected, the object is then segmented by thresholding the pixel intensity of the puncta. The threshold was calculated adaptively for each punctum as the 18% of the maximal intensity of the punctum after subtracting the background intensity.
Different puncta parameters were measured after the detection and segmentation.
In particular, the mean punctum intensity is a measurement of the relative amount of PSD95 protein within the PSD, since the absolute number of PSD95 molecules cannot be resolved using confocal imaging. Nevertheless, the mean punctum intensity provides a good insight into the differences in average packing of PSD95 between areas or individuals. Besides, the puncta size is a measurement of the relative area of PSD95-positive synapses. It is calculated as the number of pixels within the segmented puncta and converted in μm2. The final measurement provided by ensemble image analysis is the puncta density, i.e. the number of detected puncta per 100 μm2.

Statistical analysis
Three PSD95 punctum parameters, synaptic puncta density (100 μm2), puncta intensity (A.U.) and size (μm2), were used for the statistical study. The punctum size is only an approximation when the square root of the results was taken, as majority of puncta are not perfectly spherical. The parameters were plotted with the Tukey-style as box and whisker plots (Krzywinski & Altman, 2014). Normality of LSCM and SDM data distribution was analyzed using D'Agostino-Pearson omnibus normality test. For a heatmap generation, the data were reorganized using an agglomerative hierarchical clustering algorithm and standardized using the z-score normalization. A similarity matrix in Figure   13 was calculated to quantify the synaptome network architecture in the human brain.
Each row/column in the matrix represents a brain region/subregion. The rows/columns are resorted to indicate the brain anatomy: anatomically close regions are placed in the adjacent rows/column in the matrix. Elements of the matrix are the similarity and calculated using the Euclidean distances of the three synaptic parameters between the two regions. It ranges from zero to one, the latter of which indicates identical synaptome of the two regions. Similarity matrix was plotted as a heatmap where the elements were colour coded.

Results
The study is divided into four phases: i) the optimization of synaptic labelling with antibodies to PSD95, ii) the systematic acquisition of image data from 20 brain regions with a detailed description of the synaptic anatomy, iii) an analysis of the populations of synapses in the brain areas, and iv) a comparison of the synaptome architecture of human and mouse.

Optimization of PSD95 immunostaining
To identify optimal IHC conditions we used four complementary approaches. First, we used the standard IHC controls including the presence and absence of primary and secondary antibodies on human tissue (Figure 1). Strong punctate staining was observed with PSD95 antibodies and abolished when either primary or secondary antibodies were absent. Second, to confirm that the PSD95+ puncta observed were synaptic, we labelled the excitatory presynaptic terminals with synapsin 1 and synaptophysin (Figure 2) (Koffie et al., 2009) and detected the characteristic juxtaposed labelling in 91.5% and 87.8% of synapses, respectively. Third, we performed PSD95 IHC on brain sections from Psd95 knockout mice and confirmed the absence of positive signal (Figure 3). Fourth, we compared the puncta distribution of multiple human brain regions with the corresponding regions in mice expressing PSD95eGFP, which revealed similar synaptic puncta distributions in these areas (Figure 4). We screened eight commercially available antibodies and found best results with anti-PSD95 monoclonal IgG2a (NeuroMab) at 16 ng/ml (a 250-fold dilution of the 4 μg/ml stock solution).

Human brain regions show different distributions of excitatory synapses
In order to describe the distribution patterns of the PSD95+ synapses in the human brain, we performed IHC on post-mortem material from a selection of 20 regions.
Representative images from these selected brain areas are shown in Figure 5. PSD95 expression was detected as fluorescent puncta (< 1 m in size) in all regions. In general, the PSD95 puncta appeared intense and densely packed in the neocortex and the hippocampus. However, even within the neocortex, different areas displayed unique combinations of bright, dim, large and small puncta. In the subcortical structures, the thalamus and caudate nucleus, the puncta tended to show moderate densities and intensities. A high density of intense PSD95 puncta was found in some parts of the cerebellum, such as the granular cell layer, but not others, such as the cerebellar white matter. Both the hippocampus and cerebellum showed distinct subregional distribution patterns of PSD95 puncta, as described below. Although present within the brainstem, the synaptic puncta density was low in the midbrain, pons and medulla. Synaptic puncta were almost absent from the cortical and the cerebellar white matter (not shown). Thus, human brain regions show diversity in the densities, intensities and size of PSD95 puncta.
In the following sections, we describe some of the diversity observed within the hippocampal formation, the deep grey nuclei (caudate and thalamus), cerebellum and brainstem.
In the human hippocampal formation, there was intense PSD95 staining within CA1-4 subregions as well as the dentate gyrus (DG) (Figure 6A-C). The highest intensity of staining was seen in CA1 ( Figure 6D-I). Bright PSD95 staining was mostly observed within the dendritic layers, including stratum radiatum and stratum lacunosum-moleculare ( Figure 6J) as well as the molecular layer of the DG ( Figure 6K, S). Staining was observed on the cell bodies of the pyramidal cells of the CA1-4 subregions ( Figure 6L, M), in contrast to the granule cells of the dentate gyrus that lacked staining ( Figure 6N, O). CA1 contained small bright PSD95 puncta ( Figure 6P), whereas CA2 puncta were dimmer ( Figure 6Q). The puncta observed in the neuropil of the pyramidal cell layer of the CA3 (not shown) and CA4 subregions ( Figure 6R) appeared very similar in size, intensity and pattern of staining: they were large, moderately bright, and formed a rosette-like pattern of staining.
The caudate nucleus showed strong PSD95 labelling ( Figure 7A-C), with small patches ( Figure 7B asterisk) that we confirmed as striosomes by IHC with Calbindin-D28K (Ito et al., 1992) (Figure 7D-F), which are embedded within the matrix. In addition to dendritic staining, somatic staining was observed, presumably in medium spiny neurons (MSNs). In the thalamus, moderate numbers of PSD95 positive puncta were detected ( Figure 7G-I). PSD95 puncta were often found lined up along neuronal dendrites ( Figure   7J), although a more diffuse distribution of puncta was also seen.
In the cerebellum PSD95 staining was restricted to the cerebellar folia, absent from the white matter and present at low levels in the dentate nucleus ( Figure 8A

Quantification of PSD95 puncta in human brain regions
To quantify synapses and describe their diversity in the brain regions we used the SYNMAP image analysis pipeline created to map the mouse synaptome (Zhu et al., 2018b). Each PSD95 punctum was quantified in terms of its intensity (a.u.) and size (μm2).
In addition, the density of PSD95 puncta per 100 μm2 was calculated as a mean of the number of puncta per region. The mean values for four individuals are shown in Figure   12A-C and the data from individual subjects are shown in Figure S1. Figure 12D shows a summary of the mean values of each of the puncta parameters on maps of the human brain and highlights the diversity of these regions.
Ranking the 20 brain regions shows that the highest synapse density is in the neocortex and hippocampal formation and the lowest density in pons, midbrain and brainstem. There was a 7.3-fold difference between the highest and lowest mean density in the examined areas. Cortical area BA19, a region of the occipital cortex involved with feature extraction of visual images, had the highest density, intensity and size of synapses ( Figure 12A).
Quantification of PSD95 puncta size across the 20 human brain areas is shown in Figure 12B. Size reflects PSD size within an individual synapse (Broadhead et al., 2016;Zhu et al., 2018a). There was a 0.5-fold change between the largest and smallest mean PSD95 puncta size. The secondary visual cortex, area BA19, had the largest PSD95 puncta, whereas midbrain had the smallest.
Ranking PSD95 puncta intensity across the 20 human brain areas is shown in Figure 12C. Intensity reflects the amount of PSD95 protein in an individual synapse (Broadhead et al., 2016;Zhu et al., 2018a). There was a 0.9-fold change between the highest and the lowest mean intensity in the examined areas. The secondary visual cortex, area BA19, showed the highest intensity of PSD95 puncta, whereas medulla showed the lowest. The deep grey nuclei (thalamus and caudate nucleus) showed greatest variations between individuals. Together these data show that PSD95 puncta parameters vary between the delineated 20 human brain regions ( Figure 12D), with each brain area having a characteristic 'signature' of these parameters.

The human synaptome architecture is hierarchically organized and similar to mouse
With the quantitative synaptic data in hand we could now make comparisons between human brain regions. We first examined the synaptome similarity of the human brain regions by generating a similarity matrix ( Figure 13A), which revealed a number of interesting patterns. All neocortical areas and hippocampus showed high similarity (with the exception of BA19) and these regions were distinct to cerebellum and brainstem structures. BA19 is the visual association area and it has the highest PSD95 density, 14 intensity and size. Interestingly, a very similar organization was observed in the mouse brain (Zhu et al., 2018a). To test if the synaptome architecture of the two species was conserved we correlated the synaptic puncta parameters between the homologous brain regions of the two species and found significant correlations (puncta density, r=0.90, p<0.0001; intensity, r=0.51, p=0.05; size, r=0.56, p=0.03)( Figure 13B).
The synaptome architecture can also be analyzed using network methods and in the mouse the synaptome shows a small-world network architecture (Zhu et al., 2018a).
We analyzed the human synaptome network properties of the 20 human regions: each node in the network represented one region and edges that link nodes were scored for similarity of PSD95 parameters. Three network topological properties, clustering coefficient (CC), modularity and small-worldness ( Figure 13C-E) were calculated and showed that the PSD95 network has a higher CC and modularity than the random network with equivalent network complexity. Thus, the PSD95 synaptome is considered to contain a small-world network structure with high small-worldness topology. Together these findings indicate that the human brain shares a synaptome architecture with similarity to mouse.

Discussion
Using immunolabelling of PSD95 with high-resolution confocal microscopy together with the SYNMAP image analysis software, we have examined the distribution of excitatory synapses across 20 regions of the human brain in four individuals. We found that there are populations of PSD95+ synaptic puncta with varying size and intensity distributed in each brain region. Comparisons of the mean values of the synaptic parameters of these populations revealed a hierarchical organisation of the regions in which cortical and hippocampal regions shared greatest similarities, and cerebellar and brainstems had distinct synaptome signatures. This organisation is remarkably similar to the synaptome organisation in the mouse brain (Zhu et al., 2018a). Moreover, we found the synaptic puncta density, size and intensity correlated between homologous brain regions in mouse and human, showing evidence of a conserved synaptome architecture.
Furthermore, the synaptome architecture of both species showed small world network properties. These data indicate that the synaptome architecture retained invariant features during the 90 million years since humans and mice shared a common ancestor and despite the fact that brain size is 1000-fold greater in humans. Previous studies have shown that the synapse proteome composition is also highly conserved between mouse and human (Bayés et al., 2011;Bayes et al., 2012) and regional diversity in the postsynaptic proteome of the human cortex (Roy et al., 2018a) and mouse brain shows similar features (Roy et al., 2018b).Thus, the molecular anatomy of synapse protein composition, synapse diversity and its spatial organization has conserved features between humans and mice.
The present study has been restricted to a single synaptic marker of excitatory synapses. Because high synapse diversity arises from the differential distribution of synaptic proteins and their cognate multiprotein complexes (Grant, 2018;Zhu et al., 2018a;Grant, 2019), it is likely that a detailed characterization of the synaptome of diverse molecular types of synapses in humans and mice will reveal species-specific synapses and synaptome maps. Understanding these differences could be extremely important for translational studies from the mouse. Mice carrying mutations in human orthologs that cause schizophrenia and autism have been shown to have altered synapse diversity and synaptome maps (Zhu et al., 2018a;Grant, 2019). The availability of synaptomic methods in humans now makes it possible to compare the synaptomes of the two species carrying the same mutations.
A limitation of this study is that we have not created a fine-scaled synaptome map of all the subregions within the 20 regions we have studied. This will require more extensive microscopy, scanning larger areas of brain tissue. We are currently designing experiments and developing technical approaches suitable for mapping all grey matter regions in whole coronal sections of human brain, which is a step toward a brain-wide synaptome atlas of the human. The availability of a human synaptome atlas for multiple synapse types and comparable atlases in other mammalian species will inform on the function of the human brain and the synaptomic basis of behaviour. We thank the patients and families who kindly donated the brains for this work and the staff at the EBB.

Conflict of interests
The authors have no conflicts of interest to disclose.

Data accessibility
The authors confirm that all data underlying the finding are available and will be shared with the research community upon request.

Author contributions
OEC -immunofluorescence, microscopy platforms, image capture, data analysis; ZQimage puncta analysis (detection, segmentation, quantification) and network analysis; OEC, ZQ, CS, SGNG writing; CS, SGNG -direction. Insets (C, F) demonstrate the typical immunofluorescent 'snowman' appearance of colocalised pre-and postsynaptic markers. Overall, the majority of PSD95 puncta are juxtaposed to presynaptic terminals. Scale bars: 10 μm in C, F; 1 μm in insets. (G, H) Quantification of PSD95-positive puncta colocalisation with synapsin 1 (G) and synaptophysin (H). The rotated control shows the degree of random colocalisation obtained by clockwise rotation of the synapsin 1 (G) or synaptophysin (H) images. The colocalisation with either presynaptic marker is significantly higher than expected by chance. Data are mean ± SD; *P ≤ 0.05, Mann-Whitney U-test.  FIGURE 4 Comparison of PSD95 expression between human and mouse brain. Similarity of PSD95 immunofluorescence patterns between human and mouse. Panels in green: tissue from mutant knock-in mouse examined using direct fluorescence microscopy of PSD-EGFP fusion protein in the adult Psd95 EGFP/+ heterozygote. 27   (B) Correlation of PSD95 positive puncta parameters quantification between human and mouse. Scatterplots showing correlations of PSD95 immunofluorescent puncta parameters (density, intensity and size) quantified from 15 brain areas of knock-in mouse and human tissue stained using PSD95 antibody. There was a high, positive linear and significant correlation between the human and mouse PSD95 puncta counts. A moderate positive, linear and significant correlation was found for PSD95 puncta intensity. PSD95 puncta sizes were moderately and significantly correlated between the species. Pearson's product-moment correlation (r) and p values are provided. Values are based on three male WT mice aged 12 months and three male human controls.

FIGURE S1
FIGURE S1 Comparison of PSD95 synaptic puncta parameter distribution from different brain areas for the four subjects studied.