Abstract
Growing evidence suggests a remarkable diversity and complexity in the molecular composition of synapses, forming the basis for the brain to execute complex behaviors. Hence, there is considerable interest in visualizing the spatial distribution of such molecular diversity at individual synapses within intact brain circuits. Yet this task presents significant technical challenges. Expansion microscopy approaches have revolutionized our view of molecular anatomy. However, their use to study synapse-related questions outside of the labs developing them has been limited. Here we independently adapted a version of Magnified Analysis of the Proteome (MAP) and present a step-by-step protocol for visualizing over 40 synaptic proteins in brain circuits. Surprisingly, our findings show that the advantage of MAP over conventional immunolabeling was primarily due to improved antigen recognition and secondarily physical expansion. Furthermore, we demonstrated the versatile use of MAP in brains perfused with paraformaldehyde or fresh-fixed with formalin and in formalin-fixed paraffin-embedded tissue. These tests expand the potential applications of MAP to combinations with slice electrophysiology or clinical pathology specimens. Using male and female mice expressing YFP-ChR2 exclusively in interneurons, we revealed a distinct composition of AMPA and NMDA receptors and Shank family members at synapses on hippocampal interneurons versus on pyramidal neurons. Quantitative single synapse analyses yielded comprehensive cell type distributions of synaptic proteins and their relationships. These findings exemplify the value of the versatile adapted MAP procedure presented here as an accessible tool for the broad neuroscience community to unravel the complexity of the “synaptome” across brain circuits and disease states.
Significance Statement
Expansion microscopy approaches can revolutionize studies of synaptic protein diversity in intact brain circuits but have been minimally applied outside specialized labs developing these techniques. An independently adapted step-by-step Magnified Analysis of the Proteome protocol is reported here for analysis of excitatory synaptic components under a wide range of fixation conditions compatible with multidisciplinary basic science and clinical pathology studies. The main advantage of this approach arises from increased antigen recognition, secondarily from physical expansion. Analyses of cell type-specific distributions in mouse hippocampal CA1 region revealed distinct compositions of glutamate receptors and Shank protein family members. Further applications of this approach may expand the “synaptome” concept by mapping the distributions of a multitude of native synaptic proteins across brain circuits.
Introduction
Synapses, central structures of intercellular communication in the brain, are complex in their molecular composition, with thousands of different proteins involved in their function (Koopmans et al., 2019; Sorokina et al., 2021). It is now accepted that synapses are extremely heterogeneous hubs of communication, far beyond the standard classification based on neurotransmitter identity, related to diversity within the presynaptic compartment and postsynaptic compartment (Micheva et al., 2010; Grant, 2019; Grant and Fransén, 2020; Marcassa et al., 2023). Visualizing the spatial distribution of this diversity at the synaptic level, in their original location in tissues, has always been a challenge. The gold standard technique has long been serial section postembedding immunogold electron microscopy (EM), despite its limitations of labor-intensive slow acquisition, small tissue volume, and difficulty to colabel multiple proteins.
Visualization of synapses by light microscopy is subject to different limitations, a major one being limited resolution. Indeed, the XY resolution of conventional confocal microscopes is ∼250 nm due to the diffraction limit (Paddock, 2000), with Z resolution significantly worse. The mean nearest neighbor distance between synapses in the cortex and hippocampus is on the order of 500 nm (Rusakov et al., 1999; Merchán-Pérez et al., 2014) while the distance between the active zone and the postsynaptic density (PSD) is below 100 nm (Valtschanoff and Weinberg, 2001; Siksou et al., 2007; Limbach et al., 2011; Liu et al., 2019). Super-resolution approaches have overcome this limitation to enable imaging with single synapse and subsynaptic resolution (Chen et al., 2022; Fuhrmann et al., 2022) but such technologies are not available to every lab. Moreover, regardless of the imaging system, the immunostaining quality for many native synaptic proteins in brain tissue is very poor using conventional methods. The poor staining is likely due to the high density of molecules at synapses limiting antibody accessibility as well as the presence of tissue components mediating nonspecific binding. Antigen retrieval protocols can improve staining quality for a few synaptic proteins (Fritschy et al., 1998) and along with etching have been very successful in visualizing synaptic proteins in ultrathin sections (Holderith et al., 2020). However, high-quality conventional staining protocols are not available for most excitatory synaptic proteins.
In this context, a hydrogel-tissue hybridization expansion technique was developed in the past decade, in which proteins are anchored to a gel matrix which is then expanded isotropically to compound physical magnification with the optical magnification of microscopes (Chen et al., 2015). This technique includes steps to remove light-scattering material, allowing imaging deep into the transparent tissue–gel hybrid. Many variants of expansion microscopy have been published (Chozinski et al., 2016; Ku et al., 2016; Tillberg et al., 2016; Truckenbrodt et al., 2018; Park et al., 2019; Wassie et al., 2019), some of which allow immunostaining after tissue clearing, which may be considered similar to antigen retrieval. With such techniques, studying synaptic proteins at single synapse resolution using confocal microscopes is now possible. However, few studies have been published using expansion microscopy approaches to answer biological questions related to synapses, especially without the collaboration of the scientists who developed the techniques.
As a lab outside the expansion microscopy field with a history of studying synapses, we sought to adapt an expansion approach for routine study of synapses in brain tissue and to test its parameters. We focused on the Magnified Analysis of the Proteome (MAP) procedure (Ku et al., 2016), largely because it involves immunostaining after tissue clearing and it avoids the use of proteases. Surprisingly, we found the advantage conferred by the physical expansion secondary to that conferred by the increased specificity of antibody binding as a consequence of tissue clearing. We offer a step-by-step protocol for nonspecialists adapted from the initial MAP procedure and a validated list of antibodies targeting synaptic proteins. We assessed this procedure on tissues fixed with various conditions, including formalin-fixed paraffin-embedded (FFPE) MAP for detecting synaptic antigens in pathology specimens. To highlight its utility for circuit-specific studies of synapse diversity, we combined MAP with genetic labeling to reveal a differential molecular identity of excitatory synapses made onto mouse hippocampal CA1 interneurons versus pyramidal neurons.
Materials and Methods
Animals
All animal experiments complied with institutional requirements of the University of British Columbia and conformed to ethical and procedural guidelines of the Canadian Council on Animal Care. The experimental analyses for each group of mice were conducted between 7 and 15 months, males and females. Although underpowered, our analysis did not unveil a significant difference between male and female, thus data from both sexes was mixed. For experiments done on wild-type mice, C57BL6/J were used. To obtain mice expressing ChR2-EYFP in hippocampal interneurons, we crossed the Cre line Dlx5/6-Cre (Tg(dlx5a-cre)1Mekk) with Ai32 Gt(ROSA)26Sortm32(CAG-COP4*H134R/EYFP)Hze (Zerucha et al., 2000; Madisen et al., 2012; Luo et al., 2020). All mice were housed in a 12 h dark/light cycle with unrestricted access to food and water.
Antibodies
The primary antibodies used for the main figures were as follows: anti-parvalbumin (rabbit polyclonal, Abcam, catalog #ab11427), anti-VGAT (guinea pig polyclonal, Synaptic Systems, catalog #131 004), anti-panMAGUK (mouse monoclonal, Antibodies Inc., catalog #75-029, clone K28/86), anti-ELKS (mouse monoclonal, Sigma, catalog #E4531), anti-CaV2.1 (rabbit polyclonal, Synaptic Systems, catalog #152203), anti-PSD-95 (mouse monoclonal, Antibodies Inc., catalog #75-028, clone K28/43), anti-GluN1 (mouse monoclonal, Synaptic Systems, catalog #114011), anti-GluA2 (rabbit polyclonal, Synaptic Systems, catalog #182103), anti-panAMPAR (guinea pig, Frontier Institute, catalog #Anti-pan-AMPAR-GP-Af580, clone N355/1), and anti-GFP (rabbit polyclonal, Thermo Fisher Scientific, catalog #A11122). The secondary antibodies used were all goat antibodies, anti-mouse IgG1 Alexa 568 (Thermo Fisher Life Technologies, catalog #A21124), anti-mouse IgG2a Alexa 488 (Thermo Fisher Life Technologies, catalog #A21131), anti-rabbit Alexa 647 (Invitrogen, catalog #A21245), anti-mouse IgG2a Alexa 568 (Invitrogen, catalog #A21134), anti-mouse IgG2b Alexa 488 (Thermo Fisher Life Technologies, catalog #A21141), anti-guinea pig Alexa 647 (Invitrogen, catalog #A21450), and anti-rabbit Alexa 488 (Invitrogen, catalog #A32731). For the complete list of primary antibodies tested for the MAP procedure, see Extended Data Figure 3-1; RRIDs and other associated information are included.
Adapted MAP protocol (see also Extended Data Fig. 1-1 for a step-by-step detailed protocol)
This protocol was adapted from Ku et al. (2016).
Perfusion
Mice were anesthetized by injecting 0.125 ml/15 g of 20% urethane into the abdominal cavity. Mice were perfused transcardially with 6 ml Ringer's solution containing heparin and then with 25 ml fixative solution (4% PFA in PBS). After the perfusion, the brain was extracted and incubated at 4°C overnight with agitation in fixative solution.
Sectioning and postfixation
Brains were kept in the fixative for 2 h at RT and then washed three times in washing solution (PBS with 0.02% of NaN3) before sectioning. Sections of 170 μm were cut with a vibratome and then incubated in fixative solution at 4°C overnight with agitation and then at 37°C for 2 h with agitation.
AA-integration
Brain sections were washed with washing solution twice at 37°C for 2 h each. After the washing, sections were incubated in 300 nM DAPI in washing solution for 1 h at RT and images acquired to be used for expansion factor calculation (see below). Sections were incubated in low-AA solution [4% acrylamide (Sigma, A3553), 4% PFA in PBS] at 4°C with agitation overnight, followed by a 2 h incubation at 37°C with agitation.
Inactivation
Brain sections were washed with washing solution three times at 37°C for 45 min each. Sections were then incubated in Inactivation solution [1% acetamide (Sigma, A0500), 1% glycine (Sigma, G7126), 0.02% NaN3 (Sigma, S2002) with the pH titrated to 9.0 with NaOH] at 37°C for 4 h with agitation in the EasyClear device (LifeCanvas Technologies). From this step onward, samples were protected from light.
Monomer incubation
Brain sections were washed with washing solution three times at 37°C for 30 min each. MAP solution was made by freshly adding V-50 initiator (Sigma; 440914) to a solution of 30% acrylamide, 0.1% bisacrylamide (Sigma, M7279), and 10% sodium acrylate (Sigma, 408220) in PBS; the brain sections were incubated in MAP solution overnight at 4°C.
Mounting and gel embedding
Brain slices were placed in MAP solution with freshly added V-50 initiator in a gelation chamber composed of a regular slide with cut high precision #1.5 coverslip glued on as spacers and topped with a coverslip. Gelation chambers were placed in a humid gelation box which was sealed and the air purged with nitrogen gas. The gelation box was put in the oven at 45°C for 2 h, following which the hybrid gel–tissue was removed and excess gel around the tissue cut away.
Denaturation, clearing, and expansion
The tissue–gel hybrid sections were immersed in denaturation solution (in mM, 200 SDS, 200 NaCl, 50 Tris, pH 9) and incubated at 37°C for 2 h and then at 95°C for 60 min with gentle shaking using an EasyClear device. Sections were washed twice for 1 h each with 0.001× PBS and then incubated in 0.001× PBS at RT overnight with gentle shaking. To prepare the sections for immunostaining, the 0.001× PBS was replaced with PBST (0.1% Triton X-100 in PBS) for 1 h before staining.
Immunostaining
The mix of primary antibodies was prepared in PBST with the dilutions summarized in Extended Data Figure 3-1. Sections were incubated in primary antibodies in sealed wells at 4°C for 2 d with gentle shaking. Sections were washed with PBST at 37°C three times for 1 h each and then incubated in secondary antibodies (1:200) in PBST at 4°C for 2 d with gentle shaking. Sections were washed with PBST at RT three times for 1 h each, with the addition of 300 nM DAPI during the second wash. After the last washing, the sections were incubated in 0.001× PBS overnight at RT to allow expansion.
Different conditions of fixation
Five fixation conditions were tested: the PFA MAP corresponded to the regular adapted MAP procedure with the perfusion described above. For the Orig MAP, the fixative solution was replaced by the MAP fixative solution (30% acrylamide, 0.1% bisacrylamide, 10% sodium acrylate, 4% PFA in 1× PBS) for the perfusion and postfixation. The Fresh MAP corresponded to the direct extraction of the brain after the sacrifice of the animal followed by incubation in MAP fixative solution for 2–3 d at 4°C. The Form1w MAP and Form1m MAP corresponded to the direct extraction of the brain after the sacrifice of the animal followed by the incubation in formalin for 1 week or 1 month, respectively, at 4°C. These conditions mimic the conditions in which human brain tissue is often fixed and stored after a surgery or autopsy. The Postmort MAP was processed as for Form1w MAP, except that between the extraction and the incubation in formalin, the brain was stored for 3 h at 4°C.
FFPE-MAP (see also Extended Data Fig. 6-1 for a step-by-step detailed protocol)
The FFPE sections, some generously provided by Dr. Huaiyu Hu from SUNY Upstate Medical University and some prepared by Wax-it Histology Services, were all prepared as follows: PFA-perfused mouse brains were incubated in 70, 80, and 95% alcohol, 45 min each, followed by three incubations in 100% alcohol, two in xylene and three in paraffin, 1 h each. The resulting brain tissues were embedded in a paraffin block and sectioned at 5 or 10 μm thickness by a microtome and melted onto glass slides. For FFPE-MAP, sections were deparaffinized and rehydrated by washing them with xylene three times for 5 min each and then two times for 10 min each in the following solutions: 100% ethanol 95% ethanol, 70% ethanol, and 50% ethanol. Finally, the samples were rinsed two times with deionized water for 5 min each. After this step, the tissues were incubated in similar solutions as described above from the postfixation step, but for shorter durations given the thinness of the tissue, as detailed in Extended Data Figure 6-1. Moreover, the gelation was done directly on the original slide to which the sections were attached (using similar spacers) as was the denaturation. The FFPE-MAP tissue–gel hybrids generally floated free of the slides after denaturation. The side that had been facing the original slide, thus which had the tissue at the surface, was mounted next to the coverslip for imaging.
Conventional immunostaining
The PFA-perfused mouse brain (see MAP protocol above) was cut into two hemispheres and half processed for regular immunostaining as follows. The hemisphere was sectioned at 50 μm thickness using a vibratome and then incubated into a blocking buffer (1× PBS, 5% normal goat serum, 3% BSA, 0.5% Triton X-100 at pH 7.4) for 1 h at 37°C. Sections were then incubated in primary antibodies in 30 mM phosphate buffer, pH 7.4, containing 0.2% gelatin, 0.2% BSA, 0.5% Triton X-100, and 0.2 M NaCl at 4°C for 2 d. Sections were washed with PBS containing 0.1% Triton X-100 at 37°C three times for 1 h each and then incubated in secondary antibodies (1:200) in the same buffer at 4°C for 2 d with gentle shaking. Sections were washed with PBST at RT three times for 1 h each, including one washing with 300 nM of DAPI.
Mounting and imaging
MAP-processed expanded brain sections were mounted onto #1.5 high precision coverslips that had been washed with 70% ethanol, dried, coated with 0.1% poly-L-lysine overnight, rinsed with water, and dried. The sections on coverslips were immersed in 0.001× PBS and mounted on custom-designed chambers (3D printing instructions available at https://osf.io/w6c9u/). Slides containing sections processed by conventional staining were mounted with ProLong Glass Antifade Mountant (Thermo Fisher Scientific, P36980) and a #1.5 high precision coverslip. Transparent nail polish was used to seal it.
For the imaging for Figures 1 and 2 conventional immunostaining, an LSM800 Zeiss confocal microscope was used with the objective Plan-Apochromat 63×/NA1.4 Oil. The MAP-processed sections that were imaged for comparison were also imaged on a LSM800 Zeiss confocal microscope with the water objective C-Apochromat 63×/NA1.2. MAP samples for Figures 3–9 were imaged using a LSM700 Zeiss confocal microscope, a LSM880 Zeiss confocal microscope with Airyscan, a LSM980 Zeiss confocal microscope with Airyscan, or a Leica TCS SP8 laser confocal with Stimulated Emission Depletion system (STED). For the images taken with the Airyscan (on the LSM800, LSM880, and LSM980 systems), the super-resolution mode was used, and these images were directly processed using the Airyscan processing from Zeiss. The image stacks taken with the LSM980 microscope of the ChR-YFP sections were processed for chromatic aberration using the chromatic aberration correction function from Zeiss before the Airyscan processing. For the images taken with the Zeiss systems, the objective lens LD LCI Plan-Apochromat 40×/1.2 Imm Corr DIC M27 was used, with the immersion medium “Immersol” W 2010. For the images taken with the Leica STED system, the objective lens HC PL APO 86×/1,20 W motCORR STED white was used.
Both MAP and conventional immunohistostaining allow the visualization of VGAT and parvalbumin. Confocal images of mouse hippocampal CA1 pyramidal layer, one processed by a classical immunohistostaining procedure (IHC; A), the other processed by the adapted MAP procedure (B), labeled for VGAT and parvalbumin. We quantified the fraction of VGAT touching or overlapping with parvalbumin for the true images and with one channel flipped relative to the other to estimate background colocalization (C). ANOVA (p < 0.0001) and post hoc Holm–Sidak's multiple comparison revealed a significant difference between MAP and MAP flipped (p < 0.0001) and IHC and IHC flipped (p < 0.0001) but not between MAP and IHC (ns, not significant; n = 3 mice after averaging data from 3 images per mouse). Images were acquired using Zeiss automated tiling. Scale bars (biological scale), 10 µm. See Extended Data Figure 1-1 for step-by-step adapted MAP protocol.
Figure 1-1
MAP Protocol. Detailed step-by-step protocol for the adapted MAP procedure written for non-specialists. Download Figure 1-1, DOCX file.
MAP improves the visualization of multiple synaptic proteins compared with conventional methods. Confocal images of mouse hippocampal CA1 stratum radiatum, one processed by a classical immunostaining procedure (IHC), the other processed by the adapted MAP procedure, labeled for CaV2.1, ELKS, and panMAGUK (A, C) or GluA2, GluN1, and PSD-95 (B, D). The enlarged regions are from the white rectangles located in the merged images. We quantified the proportion of synaptic proteins colocalizing with another synaptic protein for every antigen labeled (E, F). In the two-way ANOVA (procedure p < 0.0001, proteins p = 0.357, interaction p = 0.0035; n = 3 mice after averaging data from 3 images per mouse) and Tukey's post hoc comparisons, the MAP versus MAP flipped comparison p values were p < 0.0001 for all of the synaptic antigens. In contrast, the IHC versus IHC flipped comparisons were not significant (ns). Images were acquired using Zeiss automated tiling. Scale bars (biological scale), 10 µm (A–D) and 1 µm (enlarged regions).
MAP images for over 40 proteins. Confocal images of mouse brain processed by MAP for the indicated antigens; numbers correspond to the list in Extended Data Figure 3-1. Images are from hippocampal CA1 stratum radiatum (A), hippocampal CA1 stratum oriens (B), hippocampal CA1 stratum pyramidale (C), hippocampal CA3 (D), cortex (E), cerebellum (F,J), hippocampal CA1 (G), Dlx5/6-Cre Ai32 mouse CA1 interneurons expressing ChR2-YFP (H), and dentate gyrus (I). Scale bars (biological scale): 1 µm (A, E), 2 µm (B, F), 4 µm (C), 20 µm (D), 5 µm (G, H, I), and 10 µm (J). See Extended Data Figure 3-1 for antibody details.
Figure 3-1
Antibodies Tested. Excel document listing all the antibodies tested. One worksheet lists antibodies validated to give appropriate signals in the adapted MAP and/or FFPE-MAP protocols, with *** indicating the best signal to noise. Images for the validated antibodies are shown in Figure 3 with full original images available at https://osf.io/w6c9u/. The images are typically from mouse hippocampal CA1 region unless indicated otherwise. The other worksheet lists antibodies that did not work for MAP in our hands, but it remains possible these might work under slightly different conditions. Download Figure 3-1, XLSX file.
MAP tissue clearing increases antigen recognition; the expansion increases the spatial resolution but dilutes the fluorescence signal. Confocal images of mouse hippocampal CA1 stratum radiatum, processed by the MAP procedure, one for CaV2.1, ELKS and panMAGUK (A–C) and the other for GluA2, GluN1, and PSD-95 (D–F). Images were taken of the processed sections reshrunk in buffer to ∼1.4× the original volume (Shrunk; A, D) or expanded in water ∼4-fold in each dimension according to the typical MAP procedure. Expanded section images were taken with imaging settings identical to the ones used for the shrunk tissue (Expanded shrunk settings; B, E) or with imaging settings optimized for the expanded sections (Expanded optimal settings; C, F). Scale bar (biological scale), 1 µm.
MAP applied on brains fixed with a range of conditions yields single synapse resolution. Confocal images of expanded mouse hippocampal CA1 stratum radiatum, stained for CaV2.1, ELKS, and panMAGUK (A) or GluA2, GluN1, and PSD-95 (B). Brains were fixed with different conditions before their processing with the MAP procedure: original MAP fixative solution (Orig MAP), PFA (PFA MAP, as in our standard adapted MAP protocol), extracted without fixation and then fixed with the MAP fixative solution (Fresh MAP), extracted without fixation and then stored in formalin for 1 week (Form1w MAP) or 1 month (Form1m MAP), or extracted without fixation, stored 3 h at 4°C without solution and then stored in formalin 1 week (PostMort MAP). Although the images appeared qualitatively similar, two-way ANOVAs (n = 3 mice after averaging data from 6 images per mouse) revealed significant differences in signal-to-background ratio (mean intensity of the signal/mean intensity of the background, C), fixation p < 0.0001, antigen p < 0.0001, and interaction p = 0.3991; signal coefficient of variation (D), fixation p < 0.05, antigen p < 0.0001, and interaction p < 0.05; and background coefficient of variation (E), fixation p = 0.4491, antigen p < 0.0001 and interaction p = 0.901). Scale bar (biological scale), 0.5 µm.
MAP can be successfully applied to FFPE tissue. Confocal images of expanded mouse hippocampal CA1 stratum radiatum processed by FFPE-MAP, stained for CaV2.1, ELKS, and panMAGUK (A, B), or GluA2, GluN1, and PSD-95 (C, D). Large field images (A, C) were acquired using Zeiss automated tiling. Scale bars (biological scale): 10 µm (A, C), 0.5 µm (B, D). See Figure 6-1 for step-by-step FFPE-MAP protocol.
Figure 6-1
FFPE-MAP Protocol. Detailed step-by-step protocol for the FFPE-MAP procedure written for non-specialists. Download Figure 6-1, DOCX file.
Super-resolution imaging of brain sections processed for adapted MAP enhances the final resolution. Images of PSD-95 in expanded mouse hippocampal CA1 stratum radiatum, acquired with a Zeiss LSM700 confocal, a Zeiss LSM880 Airyscan, a Zeiss LSM980 Airyscan, or a Leica STED microscope (A, single PSD-95 clusters enlarged in B). 2D magnitude spectrum obtained for each microscope using images of expanded mouse hippocampus sections, MAP processed for PSD-95 (C). Spatial frequency graph for the different microscopes tested, using expanded mouse hippocampus sections stained for PSD-95 (D). The blue lines represent the resolution achievable by the microscopes, at the physical and biological scale (4-fold expansion factor considered here). The number of images used for measuring was 25 (LSM 700), 15 (LSM 880), 7 (LSM 980), and 34 (STED). Scale bars (biological scale): 0.5 µm (A) and 0.1 µm (B).
MAP reveals cell type-specific differences in postsynaptic glutamate receptor content. YFP-ChR2 is observed in a pattern consistent with interneuron-specific expression in expanded hippocampal CA1 stratum radiatum from Dlx5/6-Cre Ai32 mice (A, acquired using Zeiss automated tiling). Sections were processed by MAP for YFP, panMAGUK and panAMPAR (B), and for YFP, panMAGUK, and GluN1 (C). We quantified different parameters of the fluorescence signals, analyzing at the single synapse level 3 image stacks containing interneuron dendrites per mouse, for three animals (D, F, H, J, L, O, Q) and reporting mean values per animal (E, G, I, K, M, P, S). PanMAGUK signals were used to generate PSD volumes within which all antigens were measured and which were classified as touching (IN interneurons) or not (PY pyramidal neurons) the YFP-ChR2 signal, used as an indicator of interneuron dendrites. There are significant differences in the volume of postsynaptic densities on and off the interneuron dendrite, from the single synapse analysis (D; Mann–Whitney test, p < 0.001; nIN = 656 clusters; nPY = 11,252) and per animal metric (E; unpaired t test; p < 0.01). For panMAGUK there is no significant difference in the mean intensity (F, Mann–Whitney test, p = 0.52; G, unpaired t test, p = 0.81), but there is a difference in the sum intensity (integrated intensity per cluster; H, Mann–Whitney test, p < 0.001; I, unpaired t test, p < 0.01). For panAMPAR, there are significant differences in the mean intensity (J, Mann–Whitney test, p < 0.001, nIN = 899 clusters, nPY = 11,962 clusters; K, unpaired t test, p < 0.01) and the sum intensity (L, Mann–Whitney test, p < 0.001; M, unpaired t test, p < 0.01). Linear regression of the sum intensity of panAMPAR and panMAGUK volume shows a positive correlation (Spearman correlation r = 0.79 and p value < 0.001). For GluN1, there are significant differences in the mean intensity (O, Mann–Whitney test, p < 0.001, nIN = 1,240 clusters, nPY = 18,396 clusters; P, unpaired t test, p < 0.001) and the sum intensity (Q, Mann–Whitney test, p < 0.001; R, unpaired t test, p < 0.01). Linear regression of the sum intensity of GluN1 and panMAGUK volume shows a slight correlation (Spearman correlation r = 0.15 and p < 0.001). Scale bars (biological scale): 10 µm (A), 1.5 µm (B, C).
MAP reveals cell type-specific differences in postsynaptic Shank family proteins. Sections from Dlx5/6-Cre Ai32 mice were processed by MAP for YFP, panMAGUK, and Shank1 (A), Shank 2 (B), or Shank3 (C). We quantified different parameters of the fluorescence signals in hippocampal CA1 stratum radiatum, analyzing at the single synapse level 3 image stacks containing interneuron dendrites per mouse, from 3 animals (D, F, H, J, L, N) and reporting mean values per animal (E, G, I, K, M, O). PanMAGUK signals were used to generate PSD volumes within in which all antigens were measured and which were classified as touching (IN, interneuron) or not (PY, pyramidal) the YFP-ChR2 signal. For Shank1, there are no significant cell type differences in the mean intensity (D, Mann–Whitney test, p = 0.16, nIN = 743 clusters, nPY = 11,467 clusters; E, unpaired t test, p = 0.34), nor the sum intensity per animal metric (G, unpaired t test p = 0.24), but the sum intensity single synapse analysis showed a difference (F, Mann–Whitney test, p < 0.001). For Shank2, there are significant differences by all measures in the mean intensity (H, Mann–Whitney test, p < 0.001, nIN = 968 clusters, nPY = 10,370; I, unpaired t test, p value <0.01) and the sum intensity (J, Mann–Whitney test, p < 0.001; K, unpaired t test, p < 0.05). For Shank3, there are no significant differences in the mean intensity (L, Mann–Whitney test, p = 0.29, nIN = 1,060 clusters, nPY = 10,231; M, unpaired t test, p = 0.83), nor the sum intensity, per animal metric (N, unpaired t test, p = 0.20), but sum intensity single synapse analysis showed a difference (O, Mann–Whitney test, p value <0.001). Scale bar (biological scale), 1.5 µm.
Signal quantification
For all of the quantitative analyses except for the comparison of imaging systems, three mice were studied and three images or image stacks analyzed from one tissue slice per mouse.
Expansion factor calculation
During one of the washings postfixation, the sections were stained with DAPI only, and the hippocampus of the sections was imaged with a Zeiss Axio Zoom macroscope. After the final immunostaining, in which DAPI was again included, another DAPI image of the hippocampus was taken with the same macroscope. The software Adobe Photoshop was used to manually align these two images and calculate the expansion factor.
Synaptic protein colocalization
Stacks of images were taken for this experiment in hippocampal CA1 stratum radiatum with a LSM800 Zeiss confocal microscope using the Airyscan. The analysis was performed in 3D using Arivis Vision 4D (Zeiss) software as follows: thresholds were determined to generate objects encompassing the signal coming from the different proteins labeled using “Segmentation.” One threshold was used per protein per animal, for the three image stacks. Filters were applied to remove signals from proteins touching the edges of the imaged volume to exclude incomplete synaptic structures and to remove signals below a minimal size presumed to be too small to be synapses. For every protein labeled, objects were classified whether they touch or overlap with another reference object. Colocalization was assessed as the number of overlapping objects divided by the total number of source objects (e.g., VGAT overlapping with parvalbumin divided by total VGAT).
Effect of different conditions of fixation
Images were taken for this experiment with a LSM700 Zeiss confocal microscope. The analysis was performed manually using ImageJ as follows: a threshold was determined to encompass the signal coming from synaptic proteins, and 5–6 synaptic clusters were measured for the mean intensity, the sum intensity, and the standard deviation. After the measurement of the signal, the position of the region of interest was displaced by 30 pixels to a background region, manually ensuring that this did not correspond to another synaptic cluster, and the same measurements were made.
Cell type-specific distributions
Stacks of images were taken for this experiment with a LSM980 Zeiss confocal microscope using the Airyscan. The analysis was performed in 3D using Arivis Vision 4D (Zeiss) software as follows: thresholds were determined to generate objects encompassing the signal coming from the dendrite of interneurons covered by YFP-ChR2 and to generate PSD volumes containing panMAGUK using “Segmentation.” One threshold was used per protein per animal, for the three image stacks. Filters were applied to remove panMAGUK signals touching the edges of the imaged volume to exclude incomplete PSDs and to remove signals below a minimal size presumed to be too small to be synapses. Using the YFP-ChR2 signal as an anchor, two categories were defined: PSDs touching or overlapping with the YFP-ChR2 (on the interneuron dendrite) and PSDs not touching or overlapping with the YFP-ChR2 (off the interneuron dendrite thus on the pyramidal cells). Using these surfaces generated for panMAGUK to define cell type-specific PSDs, different parameters of the fluorescence signal were quantified for panMAGUK and for costained synaptic proteins: the mean intensity per PSD and the sum intensity per PSD.
Spatial frequency analysis of images
To assess the effective resolution provided by the combination of the adapted MAP procedure and various optical microscopes, we performed a similar spatial frequency analysis to Gao et al. (2019). Briefly, we used a Google Colab python notebook to read the images collected on Zeiss (LSM700, 25 images; LSM880, 15 images; and LSM980, 7 images) and Leica (SP8 with STED, 34 images) systems with Bio-Formats (Linkert et al., 2010). We chose to analyze images of PSD-95 labeled with Alexa 568 as this was common across the samples considered. The magnitude of the 2D Fourier transform was used to calculate the spatial frequency content of the images, shown in Figure 7C. We annotated these with colored circles corresponding to physical length scales of 60, 100, 250, and 500 nm to aid interpretation. As we do not expect major differences in the x and y directions, we azimuthally averaged the 2D spatial frequency plots in Figure 7C to make the summary plot shown in Figure 7D. The summary plot is annotated with vertical blue lines indicating the physical length scale (as above) followed by the biological length scale considering the expansion factor of the MAP procedure.
Statistical analysis
Analysis was performed using Microsoft Excel and GraphPad Prism 8. Statistical details are provided in the figure legends. For most of the analyses, data was averaged per animal for comparisons using one-way or two-way ANOVA, depending on the number of variables. For the individual synapse level analyses of excitatory synaptic proteins on and off the hippocampal interneurons, in all cases datasets did not pass the D’Agostino and Pearson’s test for normality so nonparametric tests were used. All data are reported using the mean ± standard error of the mean (SEM) with individual values represented. For the graph representing the volume of panMAGUK surface against the mean intensity of synaptic proteins, individual values are displayed with the linear regression line.
Results
MAP processing greatly improves visualization of synaptic proteins compared with standard immunofluorescence methods
We based our expansion microscopy approach on the report of Ku et al. (2016) and a protocol generously provided by this team at a workshop in 2017. With some modifications, we were able to adapt a MAP protocol which works reproducibly in our hands for many presynaptic and postsynaptic antigens (brief procedure provided in the Materials and Methods; extensive step-by-step protocol written for nonspecialists provided in Extended Data Fig. 1-1). For the remainder of this paper, we use the term MAP to refer to our adapted protocol.
First, we compared MAP with conventional immunohistochemistry (IHC). We processed the same mouse brains split into two halves, one with the adapted MAP protocol and the other with a regular immunostaining procedure (as described in the Materials and Methods) with the same concentrations of primary and secondary antibodies and imaged both with a confocal microscope. Initially, we chose two neuronal components which we have previously visualized using conventional immunohistostaining (Soto et al., 2011; Lipina et al., 2016), parvalbumin which fills a subset of interneurons including their terminals and the inhibitory presynaptic vesicular GABA transporter (VGAT). The conventional staining procedure and the adapted MAP protocol revealed the expected protein distributions (Fig. 1). In the CA1 hippocampal region, VGAT and parvalbumin-positive inhibitory terminals were most concentrated in the pyramidal layer. Omission of either the primary antibodies or the secondary antibodies produced essentially no signal with either conventional immunohistostaining or with the adapted MAP (data not shown; images can be found at https://osf.io/qv6sr). To quantify the colocalization of VGAT with parvalbumin, we measured the fraction of VGAT puncta colocalizing with parvalbumin (touching or overlapping) in the pyramidal layer. To determine a background control level of colocalization with the same puncta in an essentially random distribution between channels, we flipped (i.e., rotated by 180°) the channel for parvalbumin relative to the channel for VGAT and then measured the colocalization the same way. The colocalization measures for VGAT and parvalbumin showed a significant difference between MAP and MAP flipped (p < 0.0001) and between IHC and IHC flipped (p = 0.0001; Fig. 1C), supporting the visual assessment that both MAP and conventional methods yield appropriate localization for these antigens.
To further test the MAP procedure, we chose two antibody combinations representing six synaptic components which are all challenging to visualize using conventional immunohistostaining. The first combination marks the active zone (AZ) scaffold protein ELKS (Held and Kaeser, 2018); the PSD-95, PSD-93, SAP102, and SAP97 excitatory postsynaptic membrane-associated guanylate kinase proteins (panMAGUK; Won et al., 2017); and the P/Q type calcium channel subunit CaV2.1 which functions in transmitter release (Uchitel et al., 1992; Dunlap et al., 1995). ELKS and panMAGUK were chosen to visualize the alignment of the AZ and the PSD, while CaV2.1 on the presynaptic membrane should be located between the AZ and the PSD. The second antibody combination marks the excitatory postsynaptic scaffold protein PSD-95 (Cho et al., 1992), the essential subunit GluN1 of NMDA receptors (Moriyoshi et al., 1991; Furukawa et al., 2005), and the major subunit GluA2 of AMPA receptors (Hollmann and Heinemann, 1994; Gan et al., 2015). These postsynaptic components should largely colocalize although with some separation into subsynaptic nanodomains (Nair et al., 2013; Biederer et al., 2017) and with important differences in relative levels among synapse types. The nuclear stain DAPI was also used to locate the different hippocampal subregions.
In the comparison of MAP with conventional immunostaining, these synaptic components showed differences in staining patterns (Fig. 2). Essentially no signal was observed when primary or secondary antibodies were omitted (data not shown; images can be found at https://osf.io/qv6sr). Qualitatively, the conventional staining procedure yielded poor signal specificity for all of these proteins (Fig. 2A,B). Puncta of roughly the expected size were observed for some antigens, such as CaV2.1, ELKS, and panMAGUK, but these were present not just in the synaptic neuropil of CA1 stratum radium but also in the pyramidal layer localized to nuclei. The signal for other antigens, such as GluN1, GluA2, and PSD-95, was more homogeneous than expected for a synaptic distribution and again showed some overlap with nuclei. In general, none of the conventionally processed tissue displayed convincing signal specificity. Processing the brain sections with the adapted MAP protocol allowed a dramatic improvement of the staining quality for the same proteins (Fig. 2C,D). All antigens were detected in a punctate pattern selectively in the stratum radium, consistent with an excitatory synaptic distribution; additional puncta for ELKS and CaV2.1 in stratum pyramidale likely correspond to inhibitory synapses. Furthermore, the alignment of elongated clusters of panMAGUK and ELKS was apparent, with CaV2.1 located between these PSD and AZ components. We also note a heterogeneous distribution of GluN1 and GluA2, with GluN1 being present at all synapses while GluA2 seems to be more concentrated at bigger synapses. We performed a similar quantification of the colocalization of the synaptic components as above, flipping one channel relative to the other channels to assess background colocalization. In the two-way ANOVA and Tukey's post hoc comparisons, the MAP versus MAP flipped comparison p values were p < 0.0001 for all of the synaptic antigen pairs indicating highly significant colocalization of the true images versus flipped controls (Fig. 2E,F). In contrast, the IHC versus IHC flipped comparison was not significant for any of the synaptic antigen pairs, suggesting an inability to properly detect the synaptic antigens by this conventional method. These results show that processing of brain sections by the adapted MAP protocol generated clean and specific signals for presynaptic and postsynaptic proteins at an individual synapse resolution using confocal microscopy.
We proceeded to test >100 antibodies against 73 synaptic components and found that 50 worked well for MAP (Fig. 3; table in Extended Data Figure 3-1; original images available at https://osf.io/w6c9u/). Among the components that can be visualized are numerous classes of synaptic proteins including AZ proteins, vesicular transporters, ion channels, trans-synaptic organizing proteins, PSD scaffolding and signaling proteins, and neurotransmitter receptors. Working antibody concentrations that yielded reasonable images are listed; however, users may wish to perform a dose–response curve to optimize the concentration for a particular antibody of interest. For some components such as VGluT1, even using high concentrations of antibody we observed a decrease of the fluorescence signal with depth into the tissue–gel hybrid, which we attribute to difficulty to approach saturating concentrations of antibody for highly abundant proteins. It may be noted that we had less success with components of inhibitory than excitatory synapses (scaffolding proteins and receptors) but also that many inhibitory synaptic components can be detected well and individual inhibitory synapses resolved without requiring expansion protocols (Fritschy et al., 1992; Boccalaro et al., 2019).
The improvement in protein visualization with the adapted MAP comes primarily from tissue clearing and secondarily from expansion
In the early studies describing MAP and other tissue expansion microscopy procedures (Chen et al., 2015; Ku et al., 2016; Tillberg et al., 2016), the emphasis was given to the improvement of protein visualization due to the expansion, resulting in an effective increase in resolution. However, with techniques where denaturation is performed before the antibody staining (Ku et al., 2016), there may also be an increase in antigen accessibility and/or signal specificity due to the removal of cellular components. We wanted to investigate if the observed improvement in synaptic protein visualization with MAP comes primarily from the expansion or from the tissue clearing. To investigate this, we looked at brain sections processed for MAP and reshrunk in buffer, which brings them to a size close to nonprocessed sections, ∼1.4× linear expansion (Fig. 4A,D). Even with such minimal expansion, signals were observed for all antigens at an individual synapse resolution. The ELKS signals and the panMAGUK signals aligned, with clusters of CaV2.1 present within these alignments. PSD-95, GluN1, and the bright clusters of GluA2 colocalized, although for some antigens there was considerable background. These results provide evidence that the tissue clearing and resultant increase in signal specificity contribute in a major way to the gain of synaptic protein visualization with MAP. The quality of immunofluorescence in MAP cleared tissue was vastly superior to the quality of standard immunofluorescence of noncleared tissue (compare Figs. 4A,D with Fig. 2).
To investigate the additional gain provided by the final ∼3× linear expansion, we expanded (immersed in 0.001× PBS) the sections previously reshrunk. When the same imaging settings were applied as for the shrunk sections, the fluorescent signals were dramatically fainter (Fig. 4B,E). The physical expansion of the section reduces the fluorescence signals detected, due to the dilution of this signal into a larger volume. Thus, increasing the detection parameters (e.g., laser power, exposure time, gain), as well as adjusting the display intensity of images, was necessary for MAP-processed sections (Fig. 4C,F). Once the MAP fluorescent signal detection was optimized, more details and less background were apparent compared with the shrunk sections. ELKS and CaV2.1 often displayed brighter subclusters within each synapse, and sometimes also panMAGUK and PSD-95, likely corresponding to nanodomains (Tang et al., 2016; Biederer et al., 2017).
Another point worth highlighting is that the concentration of antibodies used to realize these stainings was 2–3 times higher than we typically use for a conventional immunofluorescence procedure on cultured neurons. This increase in antibody concentration may help counteract the effect of the signal dilution due to expansion. Our observation that this higher antibody concentration was necessary for a good signal supports the idea that there is increased antigen accessibility and not just reduced background binding relative to conventional immunofluorescence. Thus, altogether, MAP appears to improve synaptic protein visualization by increased antigen accessibility, reduced nonspecific binding, and increased effective resolution due to expansion.
MAP applied on brains fixed with a range of conditions allows single synapse visualization
The original procedure (Ku et al., 2016) indicated an initial fixation of the brain with a solution specific for MAP composed of acrylamide and paraformaldehyde (see Materials and Methods), which may limit tissue sharing for other purposes. We wanted to investigate the potential of modifying the MAP procedure for the study of synaptic proteins with various initial conditions of fixation, which would increase the possible applications of this technique. As described in the Materials and Methods, five conditions of fixation were tested. The PFA MAP condition would be particularly useful as many techniques use tissue fixed with PFA, such as conventional immunostaining procedures (Evilsizor et al., 2015). The Fresh MAP could be applied to tissue used for electrophysiology experiments. The Form1w and Form1m MAP mimic the conditions of formalin fixation and conservation of human tissue extracted during surgeries (e.g., patients with temporal lobe epilepsy, tumors; the effect of further paraffin embedding is assessed below). Finally, the postmort MAP mimics the conditions of fixation of postmortem tissue, where the brain is extracted several hours or days after death and fixed with formalin (Blair et al., 2016).
The two combinations of synaptic antibodies were used on MAP-processed sections which were fixed with these various conditions. The staining looked qualitatively similar across the different fixation conditions for panMAGUK, ELKS, CaV2.1, and PSD-95, GluN1, and GluA2 (Fig. 5A,B). However, quantitative analyses revealed significant effects of the fixation condition and of the specific antibody on the signal-to-background ratio (Fig. 5C). There were also significant effects of the fixation, the antibody, and their interaction on the coefficient of variation (CV) of the signal (Fig. 5D) and an effect of the antibody on the CV of the background (Fig. 5E). Nonetheless, the signal-to-background ratio was always >6.5 and the signal CV was in a range of ∼0.4–0.6 for all conditions and antibodies. Thus, while there was a detectable effect of the fixation condition used on the quality of the synaptic protein staining, it was not a dramatic effect. Overall, the quality of the staining with all fixation conditions tested allows for resolution and measurement of individual synapses.
A large bank of tissue available for scientists is composed of FFPE specimens (Kokkat et al., 2013). This type of fixation and conservation allows brain blocks to be stored for a long time while limiting degradation of the protein's antigenicity. Using an adapted version of the MAP procedure on FFPE mouse tissue (detailed step-by-step protocol in Extended Data Fig. 6-1), we stained for the two synaptic antibody combinations. FFPE-MAP gave similar staining quality as the regular adapted MAP procedure, showing well-resolved individual synapses with aligned clusters of panMAGUK, ELKS, and CaV2.1 and of PSD-95, GluN1, and GluA2 for the larger synapses (Fig. 6). Most of the antibodies tested on tissue processed with FFPE-MAP worked, targeting proteins from the AZ, the PSD, receptors, and ion channels (Extended Data Fig. 3-1). These findings expand the potential applications of our adapted MAP.
Super-resolution imaging of brain sections processed for MAP enhances final resolution
Various optical microscopes are now available, including super-resolution microscopes which can overcome the Abbe diffraction limit and reach resolution claimed to be as high as 40 nm. To provide an assessment of what resolution is achievable when MAP is combined with a variety of optical microscopes, we tested four different microscopes on expanded samples processed for PSD-95 with the adapted MAP procedure: an older confocal microscope lacking super-resolution, Zeiss LSM700; two newer confocal microscopes with some super-resolution capability, Zeiss LSM880 and LSM980 systems with Airyscan (Huff, 2015; Huff et al., 2017); and a super-resolution STED Leica SP8 microscope (Vicidomini et al., 2018). Qualitatively, all systems could achieve individual synapse resolution (Fig. 7A,B). The PSD-95 signal appeared sharper with enhanced ability to detect nanodomains as the nominal resolution of the microscope increased. The 2D magnitude spectrums confirmed the expected differences in resolution (Fig. 7C). Given that we did not expect differences in resolution in the x and y directions, we azimuthally averaged the 2D magnitude spectrums to obtain the spatial frequency graph in Figure 7D. The accessible biological length scales achieved using the different microscopes combined with the MAP procedure for PSD-95 are shown (Fig. 7D): in contrast to the LSM700, the LSM880 and LSM900 (with Airyscan) were not limited out to spatial scales corresponding to 50 nm but were both outperformed by the STED microscope.
In addition to the differences in resolution, we noticed differences in sensitivity among the microscope systems which may be critical for expanded samples in which the fluorescence is diluted relative to conventional samples (Fig. 4). It was our experience that the older LSM700 confocal was suitable for obtaining single optical sections but not z stacks, as using settings to obtain a minimally acceptable signal resulted in bleaching within small z stacks. This was not a limitation on the other more sensitive systems. The LSM700's curve in Figure 7D flattens between biological length scales of 125 and 62.5 nm indicating that the imaging is dominated by spatial frequency independent noise. This highlights that, in order to exploit the resolution gain of the technique, we still must have sufficient signal-to-noise ratio which may be difficult to achieve on this generation of confocal microscope.
Adapted MAP reveals cell type-specific differences in postsynaptic glutamate receptor content
It is commonly accepted that excitatory synapses, even those having the same neurotransmitter glutamate, are diverse in their molecular composition (Micheva et al., 2010; Grant and Fransén, 2020; Marcassa et al., 2023). What has been less studied is to attribute this diversity of molecular composition to specific cell types. Although many cell types can be genetically defined, the accessibility of synaptic proteins in tissue has been limiting for imaging native cell type-specific synaptic composition. Having now a procedure enabling the study of the protein content in synapses, we wanted to use MAP to investigate this diversity of synapse molecular identity in the hippocampus. To do this, we used a transgenic mouse, Dlx5/6-Cre (Tg(dlx5a-cre)1Mekk) crossed with the Ai32 reporter (B6;129S-Gt(ROSA)26Sor tm32(CAG-COP4*H134R/EYFP)Hze/J), expressing the light-sensitive ion channel channelrhodopsin coupled to the fluorescent protein YFP (YFP-ChR2) in all the hippocampal interneurons (Zerucha et al., 2000; Madisen et al., 2012; Luo et al., 2020). In this Dlx5/6-Cre Ai32 model, the YFP is located at the membrane of the interneurons, allowing one to image dendrite profiles from interneurons in the stratum radiatum of the CA1 subregion of expanded mouse hippocampi (Fig. 8A).
We focused first on the distribution of glutamate receptors. Processing Dlx5/6-Cre Ai32 hippocampal sections using MAP, we stained for GFP (recognizing the genetically encoded YFP-ChR2), panMAGUK, and either all AMPAR subunits (panAMPAR) or the essential NMDAR subunit GluN1. Overall, AMPAR appeared more concentrated in fewer, larger clusters compared with GluN1 (Fig. 8B,C). To investigate this quantitatively, we used the panMAGUK signal to define excitatory PSD volumes within which the fluorescence signal of the different antigens was quantified. These volumes were classified as touching (IN for interneurons) or not (PY for pyramidal neurons) the YFP-ChR2 signal. We found that PSD volumes were significantly larger in interneurons than in pyramidal neurons (Fig. 8D,E). PanMAGUK mean intensity did not differ between cell types, although the sum intensity (integrated intensity of the fluorescence signal per cluster) was higher for the interneuron than pyramidal cell synapses (Fig. 8F–I). For panAMPAR, both the mean and the sum intensities were significantly higher in synapses on interneurons than those on pyramidal neurons (Fig. 8J–M). In direct contrast, for GluN1, both the mean and the sum intensities were significantly lower in synapses on interneurons than those on pyramidal neurons (Fig. 8O–R). These findings directly parallel results from previous immunogold EM studies showing a higher AMPAR content and lower NMDAR content at excitatory synapses on hippocampal CA1 interneurons than on pyramidal neurons (Nusser et al., 1998; Nyíri et al., 2003).
Additionally, immunogold EM studies have reported relations between the size of excitatory synapses and their glutamatergic receptor content. The total number and density of AMPARs strongly correlate with PSD size for pyramidal cells (Nusser et al., 1998; Takumi et al., 1999), while the number of NMDARs weakly correlate and density shows little correlation with PSD size (Takumi et al., 1999; Racca et al., 2000; Nyíri et al., 2003). In agreement with these studies, we found a strong positive correlation between the sum intensity and mean intensity of panAMPAR with PSD volume for pyramidal synapses (Fig. 8N; Table 1). There was a weak correlation of the sum intensity of GluN1 with PSD volume and no correlation for mean intensity of GluN1 (Fig. 8S; Table 1). We also studied the distribution of glutamate receptors among individual synapses. The AMPAR sum intensity and mean intensity per synapse was more variable and the distribution more skewed for pyramidal neurons synapses compared with interneurons (Table 1). Similar findings were also previously reported based on immunogold EM (Racca et al., 2000), supporting the utility of MAP to investigate biological questions traditionally studied by more labor-intensive approaches. Furthermore, we rapidly generated a more comprehensive quantitative analysis of the distributions of multiple synaptic components on both interneuron synapses and pyramidal neuron synapses (Table 1).
Variability and relation to PSD volume of synaptic components.
Adapted MAP reveals cell type-specific differences in postsynaptic Shank family proteins
Extending beyond glutamate receptors, previous research on the molecular diversity of hippocampal synapses found heterogeneity in the postsynaptic scaffolding proteins of the MAGUK family (Zhu et al., 2018; Cizeron et al., 2020). Here, we focus on scaffolding proteins of the SH3 and multiple ankyrin repeat domains (Shank) family (Kursula, 2019), which are strongly linked to neurodevelopmental disorders. SHANK3 copy number variation or pathogenic mutation is a highly expressed single-gene risk factor for autism spectrum disorders and a genetic cause of Phelan-McDermid syndrome (Li et al., 2018; Vitrac et al., 2023). Furthermore, mutations in all three SHANK genes are associated with schizophrenia and intellectual disability as well as autism spectrum disorder (Guilmatre et al., 2014; Monteiro and Feng, 2017). All Shank proteins are expressed in the hippocampus (Lim et al., 1999; Böckers et al., 2004; Peça et al., 2011) although to our knowledge cell type-specific differences in synaptic levels have not been assessed. Considering the relation between Shanks and neurodevelopmental disorders, studying their synaptic distribution on hippocampal interneurons and pyramidal cells could unveil new perspectives. We thus processed Dlx5/6-Cre Ai32 brain sections by MAP staining for each Shank protein with GFP and panMAGUK. Immunofluorescence for all Shank proteins was tightly localized to excitatory synapses colocalizing with panMAGUK (Fig. 9A–C). Shank1 and Shank3 showed no obvious cell type-specific differences, reflected in no difference in the mean intensity per PSD and no difference in sum intensity when analyzed per mouse, although sum intensity was higher at interneuron synapses in the individual synapse analysis (Fig. 9D–G,L–O). In contrast, Shank2 was detected at markedly lower levels at synapses on interneurons compared with synapses on pyramidal cells as reflected in all measures (Fig. 9H–K). Looking at the statistical distribution, the mean and sum intensities for synapses on interneurons are more variable for Shank2 than for Shank1 and Shank3 (Table 1; the high values for Shank2 came from a small subset of interneurons). These results show that Shank1 and Shank3 are present similarly at excitatory synapses made onto interneurons and pyramidal neurons in the hippocampal CA1 subregion, whereas Shank2 seems to be selectively concentrated at synapses on pyramidal neurons, with variable amount at synapses on interneurons.
Discussion
Here, we independently adapted the hydrogel–tissue hybridization-based technique MAP to study the molecular identity of excitatory synapses. We provide a step-by-step protocol, following which any lab with access to a confocal microscope can study the distribution of proteins in tissue with single synapse resolution. This protocol was used here with >40 antibodies and various fixation conditions for basic science, clinical, and pathology applications. Highlighting its utility for studying cell type-specific synapse diversity, we used our adapted MAP on brains from mice with genetically labeled interneurons to reproduce cell type differences in AMPAR and NMDAR distributions previously reported by serial section immunogold EM and to reveal differences in Shank family proteins.
Our study highlights the main reason for the improved synaptic protein labeling with MAP: it is mainly due to the denaturation and tissue clearing of the hybrid gel–tissue. This step removes many cellular components which contribute nonspecific signal and which may block antibody access to the antigen. This step also alters protein conformation which may unmask antigens recognized by commercially available antibodies, many of which are validated for denatured proteins using Western blots. Supporting this idea, replacement of the proteinase step from the standard expansion microscopy with an SDS/heat denaturation step and performing subsequent staining as done here was found to give a good signal with many primary antibodies (Shen et al., 2020). Although the staining in both these cases is “postexpansion,” any increased separation between proteins would be contributing little to the improvement in specific antibody binding in MAP because the tissue is reshrunk in PBS and thus only ∼1.4-fold expanded during the antibody binding. The final physical expansion of the tissue to ∼4-fold provides a sharper signal and reduces the apparent overall background, the latter likely due to the enhanced z separation. A recent study pushed these ideas further, doing gelation, SDS/heat denaturation, expansion, and then a second round of gelation and expansion prior to immunostaining to achieve nanoscale resolution (Sarkar et al., 2022). However, expansion also dilutes the signal, as demonstrated by the reduction in signal we observed with ∼4× MAP compared with ∼1.4× reshrunk tissue. The ∼20× expansion achieved with the iterative approach (Sarkar et al., 2022) or up to ∼10× expansion by modifying the hydrogel (Truckenbrodt et al., 2018; Park et al., 2019; Damstra et al., 2022) may best be combined with sensitive imaging systems. While our study was in progress, another study reported a newer version of MAP, called eMAP, where proteins are not anchored to the hydrogel but rather physically trapped, to increase antibody accessibility (Park et al., 2021). However, in our hands, we did not obtain better results with eMAP than with the protocol described in this paper.
Our protocol can be applied on tissue prepared with various fixatives, expanding the possibilities to combine this method with other approaches for multidisciplinary studies. Our adapted MAP worked well on fresh-fixed or PFA-perfused tissue, with the potential for combining it with slice electrophysiology or traditional histochemistry. Surprisingly, it also worked well on mouse brains fixed a few hours postmortem and those stored for 1 month in formalin, mimicking some clinical fixation conditions. Our FFPE-MAP protocol can be applied to clinical pathology specimens stored in FFPE tissue banks. ExPath, a variation on expansion microscopy using pre-expansion staining and proteinase treatment (Zhao et al., 2017; Bucur et al., 2020), can also be applied to FFPE pathology samples. We suggest that FFPE-MAP may be better suited than ExPath for synaptic proteins because of the postexpansion staining in FFPE-MAP, for the reasons described above.
Our data using the adapted MAP to reveal AMPAR and NMDAR distributions in hippocampal CA1 interneurons and pyramidal cells matched remarkably well with previous data obtained using serial section immunogold EM (Nusser et al., 1998; Takumi et al., 1999; Racca et al., 2000; Nyíri et al., 2003). Parallel findings include the higher density of AMPARs and lower density of NMDARs at interneuron compared with pyramidal synapses and the high variability and relation with synapse size of AMPAR content at pyramidal synapses. It is important to highlight here the relative ease to reproduce these results, using equipment available in most labs. We extend here the analyses of cell type-specific synaptic differences to members of the Shank family. We found that Shank2 is selectively enriched at CA1 pyramidal cells relative to interneuron synapses. SHANK2 has been linked to multiple neurodevelopmental disorders (Monteiro and Feng, 2017), and a mouse model selectively deleting Shank2 in forebrain excitatory neurons showed deficits in social interaction and anxiety-like behaviors (Kim et al., 2018). This was associated with impaired excitatory transmission in hippocampal CA1 pyramidal neurons, consistent with our observation of high Shank2 synaptic content at these synapses. Also consistent with our observations, deletion of Shank2 from all GABAergic interneurons affected synaptic transmission of dorsolateral striatal but not hippocampal CA1 pyramidal neurons, resulting in other behavioral deficits (Kim et al., 2018).
These cell type distributions highlight the utility of our MAP protocol to characterize the synaptic content of defined circuits, limited only by genetic and/or viral labeling techniques. For the optimal cellular label, we recommend plasma membrane proteins such as YFP-ChR2 used here which provide better visualization of synaptic junctions than soluble proteins. While our imaging was limited to small tissue volumes, synaptic distributions could be mapped in large brain regions using lattice light sheet imaging (Gao et al., 2019). Alternately, standard microscopes could be used to image larger regions of MAP-processed reshrunk tissue, at the expense of nanoscale resolution but with vastly improved antigen recognition compared with conventional methods, yielding single synapse resolution. The utility of MAP can be further enhanced by multiplexing, either by antibody stripping and subsequent staining as described in the original MAP study (Ku et al., 2016) or potentially by a DNA-PAINT approach using oligonucleotide conjugated antibodies and transient binding of fluorescently labeled DNA probes (Schueder et al., 2020). As new DNA probes must be applied in buffer, each round of multiplexing would involve tissue shrinking in buffer and re-expansion in water for the ∼4× linear expansion MAP, but a DNA-PAINT version of the reshrunk ∼1.4× expansion MAP might be designed for multiplex imaging in place.
The molecular diversity of synapses has mainly been studied to date using proteomic approaches, involving brain homogenization combined with various fractionation and isolation approaches (Koopmans et al., 2019; Sorokina et al., 2021; Marcassa et al., 2023). While still useful in the study of synaptic protein composition (Paget-Blanc et al., 2022), these approaches lack precise spatial information and often lack individual organism information due to pooling multiple brains to obtain sufficient material. In an interesting exception, one group used an imaging approach to bring new insights into the diversity of excitatory synapses across the mouse brain (Zhu et al., 2018; Cizeron et al., 2020). These studies were limited to imaging two postsynaptic scaffold proteins and were generated with specific transgenic mouse lines. With the adapted MAP protocol described here, the “synaptome” concept can be refined by mapping the distributions of a multitude of native synaptic proteins across brain circuits. Furthermore, using the various banks of FFPE brain tissue, this approach is applicable to studying human brain disorders.
Footnotes
We thank Taeyun Ku and Kwanhung Chung for generously sharing their detailed MAP protocol, Xiling Zhou for excellent technical assistance, as well as Nick Michelson, Pankaj Gupta, Annika Wevers, Carlos Doebeli, Ashli Forbes, Nicole Cheung, Sarah Wang, and Amy Wang for assistance in coding. This work was supported by Canadian Institutes of Health Research Awards FDN-143206 and PJT-183943 (to A.M.C.), Simons Foundation Autism Research Initiative Bridge to Independence Award and National Institutes of Health 1R01MH130476 (to P.Z.), and by resources made available through the Dynamic Brain Circuits cluster and the NeuroImaging and NeuroComputation Centre at the UBC Djavad Mowafaghian Centre for Brain Health (RRID: SCR_019086).
The authors declare no competing financial interests.
S.O.’s present address: Department of Physiology and Pathophysiology, University of Manitoba, Winnipeg, Manitoba R3E OJ9, Canada
- Correspondence should be addressed to Peng Zhang at pxz187{at}case.edu or Ann Marie Craig at acraig{at}mail.ubc.ca.
