Introduction

Precise mapping of individual proteins with subcellular resolution would improve understanding of cellular processes at the molecular level. In many cases, the same protein has several subpopulations with different biological roles within a cell. For example, transmembrane proteins such as receptors or cell adhesion molecules have spatially different subpopulations: cell-surface populations for signal sensing and transduction and intracellular populations for reserve and recycling pools1,2,3,4. Another example is temporally different subpopulations: pre-existing and nascent protein populations often play distinct roles in basal and activity-induced cellular functions5,6,7. Thus, quantitative, subcellular mapping of different subpopulations of endogenous proteins in the brain is needed to delineate the molecular mechanisms underlying neuronal functions.

Neuronal functions rely on complex protein regulation at synapses to respond to distinct activity in the neural network. A neuron receives a wide range of inputs from other neurons via thousands of synapses8. Individual synapses are highly heterogeneous in their protein composition, trafficking and turnover, which underlies the different structures and functions of each synapse9,10,11,12. This synaptic heterogeneity, revealed through brain-wide “synaptome” analysis, is relevant to behavior, cognitive functions, and brain disorders11,12,13, while brain-wide synaptome analysis conflates variation between cells with synapse variation within a single neuron. Single-cell, quantitative mapping of spatially and temporally different subpopulations of endogenous proteins at thousands of synapses would elucidate the spatial organization of synaptic heterogeneity all along the neuronal dendritic tree, providing an accurate and informative representation of the synaptic landscape in single neurons. Thus, this “single-cell synaptome” analysis would allow for a better understanding of the single-cell organization and computations in the brain. However, there is no simple and generalizable method to differentially visualize these subpopulations of endogenous proteins in single neurons in brain tissue.

We previously developed several methods to label endogenous proteins in single cells in the mammalian brain14,15. These methods, named SLENDR and vSLENDR, are based on precise, homology-directed repair (HDR)-mediated genome editing using the clustered regularly interspaced short palindromic repeats (CRISPR)-associated endonuclease Cas9. They provide a high-throughput approach to determine the subcellular localization of endogenous proteins with high resolution, specificity and contrast. In SLENDR and vSLENDR, epitope tags or fluorescent protein tags are genetically fused to specific endogenous proteins. Incorporated epitope tags can be visualized by immunohistochemistry using specific antibodies against the tags. However, the large size of antibodies (~ 150 kDa) hinders their penetration through thick brain tissue, making it challenging to label target proteins within densely-packed protein complexes at the synaptic membrane16. These limitations preclude quantitative spatial measurements of epitope-tagged proteins in brain tissue. Alternatively, fluorescent protein-fused target proteins can be directly imaged without antibody labeling. However, fluorescent proteins’ low brightness and instability provide an insufficient signal-to-noise ratio for quantifying endogenous proteins, especially proteins with low expression levels.

An alternative to epitope tagging and antibody labeling or fluorescent protein tagging is chemical tagging via self-labeling proteins such as HaloTag17, SNAP-tag18, or CLIP-tag19. Chemical tags enable covalent and irreversible labeling with bright and photostable organic fluorophores20,21. The small size (~ 1 kDa) of the organic fluorophore ligands allows fast and efficient tissue penetration to evenly label proteins throughout thick brain tissue22. Notably, the availability of different organic fluorophore ligands in multiple colors and with different cellular permeability permits the detection of spatially restricted subpopulations of tagged proteins. For example, cell-impermeable fluorophore ligands can selectively label cell-surface subpopulations of tagged proteins23. Additionally, pulse-chase labeling with fluorophore ligands of different colors can differentially visualize temporally distinct subpopulations, including pre-existing and nascent proteins7,[24](https://www.nature.com/articles/s41467-025-65813-w#ref-CR24 “Wang, H.Y., Lin, Y.-P., Mitchell, C.K., Ram, S. & O’Brien, J. Two-color fluorescent analysis of connexin 36 turnover: relationship to functional plasticity. J. Cell Sci. jcs.162586. https://doi.org/10.1242/jcs.162586

(2015).“). Thus, chemical tag-mediated protein labeling can provide quantitative and spatiotemporal imaging of tagged proteins in tissue.

In this study, we combined chemical tag labeling with the SLENDR approach to label endogenous proteins in the brain with organic fluorophores. We show that SLENDR and vSLENDR efficiently insert self-labeling tag sequences into a variety of endogenous genes encoding for presynaptic, postsynaptic, and signaling proteins. Multiplexed labeling with different fluorophore ligands enabled single-cell visualization of neuronal protein subpopulations (i.e., surface/intracellular, nascent/pre-existing) in mouse brain tissue. Integrating this approach with a semi-automatic analytical pipeline, we quantified endogenous glutamate receptor subpopulations at the thousands of synapses in single neocortical neurons, providing a whole-cell map of the strength and plasticity of each excitatory synapse (Fig. 1a). Through single-cell synaptome mapping of endogenous protein subpopulations, we reveal the spatial organization of synapse diversity in protein localization, trafficking and turnover in single neurons in vivo.

Fig. 1: Single-cell, chemical tag labeling of endogenous proteins in fixed and living brain tissue.

a Schematic of the workflow for single-cell synaptome mapping in the brain. b**–k, m–r**, Confocal images for HaloTag-fused endogenous proteins (green) including Halo-CaMKIIα (b, c), PSD95-Halo (d, e), Halo-GluN1 (f, g), Halo-GluA1 (h, i), Halo-GluA2 (j, k), Halo-Gephyrin (m, n), and Halo-βCatenin (o**–r**) in single mEGFP-labeled pyramidal cells (magenta) in the primary somatosensory cortex layer 2/3 at P28. PSD95, a major postsynaptic scaffold at excitaotry synapses; GluN1; a NMDAR subunit; GluA1 and GluA2, AMPAR subunits; Gephyrin, a major postsynaptic scaffold at inhibitory synapses; βCatenin, a multifunctional protein associated with cell adhesion and canonical Wnt signaling. Note selective accumulations of Halo-CaMKIIα (c), PSD95-Halo (e), Halo-GluN1 (g), Halo-GluA1 (i), Halo-GluA2 (k), and Halo-βCatenin (p) at glutamatergic postsynapses formed on dendritic spines, Halo-Gephyrin (m, n) and Halo-βCatenin (q) at GABAergic postsynapses localized on the axon initial segment (arrows in m) and head portion of a few dendritic spines (arrow in n) or immunolabeled for vesicular inhibitory amino acid transporter (VIAAT, yellow) on the somatic surface (arrows in q), and Halo-βCatenin (r) at presynaptic boutons, validating the specificity of our HaloTag labeling. l Structured illumination microscope (SIM) image of the same view as in k, resolving nano-clusters for Halo-GluA2 (arrows). Scale bars, 10 μm (b, d, f, h, j, m, o), 2 μm (c, e, g, i, k, l, n, p**–r**).

Results

A generalizable platform for single-cell, chemical tag labeling of endogenous proteins in the brain

Since HaloTag labeling with organic fluorophore ligands provides bright and photostable fluorescence21,25, we set out to build a toolbox to label endogenous proteins with HaloTag via CRISPR-Cas9-mediated HDR using SLENDR14. However, the original SLENDR technique had low efficiency (< 1%) of knock-in for sequences similar in size to HaloTag or SNAP-tag (~ 600–900 bp). In this study, we utilized a pre-assembled Cas9 protein-guide RNA (gRNA) ribonucleoprotein complex (RNP), which efficiently induces DNA double-strand breaks (DSBs) immediately after the delivery into target cells by in utero electroporation (IUE)26. Moreover, as RNPs are rapidly degraded in cells compared with plasmid vectors, this strategy improves genome editing specificity by decreasing off-target DSBs27,28. The homology donor template was designed with the HaloTag and short linker sequences (0.9 kbp) flanked by ~1 kbp of the genomic sequences upstream and downstream of the HaloTag insertion site. The RNPs and homology donor template were introduced together with a plasmid vector encoding mEGFP as a transfection marker into cortical layer 2/3 pyramidal cells by IUE at embryonic day 14 (E14) or E15 in mice. Brains were fixed at postnatal 3–4 weeks, sectioned, and labeled by incubation with Janelia Fluor (JF) dyes conjugated to HaloTag ligands (JF549-HTL or JF646-HTL) for confocal fluorescent imaging20 (Fig. 1a). We confirmed that brief chemical fixation, matching the duration of our perfusion fixation, allows for spatiotemporally-precise localization of HaloTag signals without any significant signal attenuations (Supplementary Fig. 1a, b)

We first targeted HaloTag to the N-terminus of CaMKIIα by using the same homology arms previously validated for the knock-in of mEGFP14,15. HaloTag-labeled CaMKIIα (Halo-CaMKIIα) was found in the somatodendritic compartment of single pyramidal cells with a peak intensity at dendritic spines, as previously described for endogenous mEGFP-CaMKIIα14,15 (Fig. 1b, c). The RNP-based HaloTag knock-in achieved significantly higher efficiency than the plasmid-based approach14 (Supplementary Fig. 1c, d). This improvement was also observed for other genes, such as βActin and GluA2, indicating the generalizability of our strategy (Supplementary Fig. 1c, d). We confirmed the genome editing specificity by replacing the CaMKIIα-targeting gRNA with a GluA2-targeting gRNA, to find no detectable knock-in signal (Supplementary Fig. 1c). Thus, RNP-based SLENDR achieves specific and efficient HaloTag labeling of endogenous proteins in the brain.

We then applied the HaloTag labeling platform to 10 other endogenous proteins such as postsynaptic scaffolds, receptors, presynaptic release machinery, and intracellular signaling proteins (Fig. 1d–r, Supplementary Fig. 1e–m, and Supplementary Data 1, 2). We confirmed a protein species-specific distribution of HaloTag-fused endogenous proteins in single pyramidal cells. Importantly, we successfully visualized excitatory postsynaptic scaffolds (Fig. 1d, e) and ionotropic glutamate receptors (iGluRs, Fig. 1f–l and Supplementary Fig. 1e), which are difficult to detect by conventional immunohistochemistry due to the poor accessibility of antibodies to densely-packed postsynaptic structures16. Their HaloTag-labeled puncta exclusively overlapped with or were closely apposed to immunohistochemical signals for excitatory presynaptic and postsynaptic markers, demonstrating the specificity of HaloTag signals to excitatory synapses (Supplementary Fig. 2). Moreover, HaloTag-labeled endogenous proteins were further compatible with super resolution imaging by structured illumination microscopy (SIM) (Fig. 1l and Supplementary Fig. 1e) and live monitoring of protein translocation by two-photon microscopy (Supplementary Fig. 1j–n). Therefore, our HDR-mediated protein tagging with HaloTag provides a scalable, single-cell labeling approach for a wide variety of endogenous postsynaptic, presynaptic and cell signaling proteins.

Whole-neuron synaptome mapping of endogenous proteins in thick brain tissue

High brightness, photostability, and permeability of the small-sized fluorogenic HTLs are suitable for quantitative labeling of HaloTag-fused proteins in large tissue22. To confirm this, we performed quantitative, volumetric imaging of endogenous CaMKIIα tandemly tagged with HaloTag and mEGFP (Halo-mEGFP-CaMKIIα) by using vSLENDR in transgenic mice that constitutively expressed SpCas9 (Supplementary Fig. 3a, b)15. In thick (200 µm), fixed and SeeDB2G-cleared brain slices29, fluorescently-labeled HTLs uniformly visualized Halo-mEGFP-CaMKIIα proteins throughout the slice, while antibodies against mEGFP limited the labeled area to the slice surface (Supplementary Fig. 3c–e). The signal intensity for HTLs highly correlated with mEGFP fluorescence even at dendritic spines, where CaMKIIα is enriched at the postsynaptic density30 (Supplementary Fig. 3f). These results show that HTLs enable quantitative measurements of endogenous proteins in thick brain tissue that includes entire individual neurons.

We therefore applied this approach to quantitatively map α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs) in pyramidal cells of layer 2/3 somatosensory cortex. AMPARs are the primary transducer of fast excitatory neurotransmission and have been shown to increase or decrease in response to synaptic plasticity4. Thus, their abundance can be an indicator for synaptic strength. We fused HaloTag to endogenous GluA2 using SLENDR and performed whole-brain labeling of Halo-GluA2 with JF646-HTL in cleared brain slices (300 µm in thickness) (Fig. 2a). Large and high-resolution confocal image stacks captured individual Halo-GluA2 puncta from the entirety of whole neurons (Fig. 2b–d), which likely represented a functional postsynaptic site of excitatory synapses mostly formed on dendritic spines31,32. This was also supported by the immunohistochemical validation for Halo-GluA2 puncta at excitatory synapses (Supplementary Fig. 2e, f). For quantification, we developed a deep learning-based algorithm to detect regions of interest (ROIs) for individual Halo-GluA2 puncta from single neurons in a semi-automatic manner (Supplementary Fig. 3g). Our algorithm simultaneously detected 3016, 1013, and 700 ROIs from three different pyramidal cells within a single imaging volume (Fig. 2e). We measured and mapped the signal intensity for Halo-GluA2 at individual ROIs on each pyramidal cell (Fig. 2f). All three pyramidal cells showed variable signal intensities at each ROI (Fig. 2g). We further reconstructed dendritic branches of the single pyramidal cell with 3016 ROIs by the tree-based method using arrays of ROIs (Supplementary Fig. 3h, see Methods). Based on the reconstructed dendritic branches, we calculated and mapped the density, mean intensity, and coefficient of variation (CV) of Halo-GluA2 puncta on dendritic segments between branching points (Fig. 2h–m). The density of Halo-GluA2 puncta peaked at 25–75 µm distant from the soma (Fig. 2h, k). In contrast, the mean intensity and CV were not significantly different among distinct dendritic segments. However, these findings were not the case for other analyzed neurons with 6311 and 2865 ROIs (Supplementary Fig. 4), suggesting cell-to-cell variation in the spatial distribution of Halo-GluA2 puncta within individual neurons. Taken together, our approach provides quantitative, whole-cell synaptome mapping of endogenous proteins in the brain.

Fig. 2: Whole-cell, quantitative imaging of HaloTag-fused endogenous glutamate receptors with synaptic resolution in thick brain tissue.

a Schematic of the labeling and analyzing procedure for Halo-GluA2 in single pyramidal cells in the primary somatosensory cortex layer 2/3. b 3D rendering of three pyramidal cells bearing the soma and dendritic branches labeled for Halo-GluA2 in an imaging volume of 424 µm x 424 µm x 257 µm with the lateral and axial resolutions of 0.26 and 0.5 µm/pixel. c Maximum projection image of a pyramidal cell bearing punctate signals for Halo-GluA2 along dendritic branches. d Enlarged image of the boxed area in (c). e Mapping of 3016 (Cell #1, orange), 1013 (#2, green), and 700 (#3, blue) regions of interest (ROIs) for Halo-GluA2-labeled puncta on three single pyramidal cells. f Heat map of Halo-GluA2 intensity at individual ROIs on the whole pyramidal cell (Cell #1). The color bar indicates % of the averaged intensity of all ROIs. g A scatter plot between the distance from soma and Halo-GluA2 intensity at individual ROIs in three single pyramidal cells (Cell #1-3). h**–m** Heat maps (h**–j**) and proximal-to-distal distributions (k**–m**) of the density (h, k), mean intensity (i, l), and coefficient of variation (j, m) of Halo-GluA2-labeled puncta on dendritic segments (n = 156) of the whole pyramidal cell (Cell #1). Bar graph data are presented as mean ± SEM. **, p < 0.01; *, p < 0.05. p = 0.0013 (k, 25–50 vs 100–125); p = 0.0057 (k, 25–50 vs 150–175); p = 0.021 (k, 50–75 vs 100-125); p = 0.0414 (k, 50–75 vs 150–175). One-way ANOVA with Kruskal-Wallis multiple comparison test (k**–m**). Scale bars, 50 μm (c, e, f, h**–j**), 10 μm (d). AU indicates arbitrary units. Source data are provided as a Source Data file.

Quantitative profiling of spatially-distinct subpopulations of endogenous AMPARs and NMDARs at individual synapses

HTLs with different cell membrane permeability and colors would allow for multicolor labeling of the surface and intracellular subpopulations of HaloTag-fused transmembrane proteins23. We tested this application to label surface and intracellular iGluRs at individual synapses in single neurons. Surface iGluRs generate excitatory synaptic transmission, while intracellular iGluRs mediate their dynamic trafficking required for synaptic plasticity2,4. Thus, quantitative imaging of surface and intracellular iGluRs, especially AMPAR GluA2 and N-methyl-D-aspartate receptor (NMDAR) GluN1 subunits, would provide a good readout of synaptic strength and plasticity. We generated Halo-GluA2 or Halo-GluN1 using SLENDR in single pyramidal cells in the somatosensory cortex layer 2/3. N-terminal HaloTag fusions of the surface or intracellular receptors result in the HaloTag protein arranged outside or inside of the cell, respectively. To label the extracellular HaloTag fraction, we developed a membrane-impermeable ligand, FLAG2-JF646-HTL, in which JF646 fluorophore was engineered to conjugate two FLAG peptides with high charge and biocompatibility (Supplementary Fig. 5a). FLAG2-JF646-HTL required cell-membrane permeabilization to label intracellular HaloTag proteins in heterologous cell cultures, confirming its membrane impermeability in cellular labeling (Supplementary Fig. 5b, c). Using whole brains fixed at postnatal 3–6 weeks, we sequentially labeled the surface subpopulation (sHalo-GluA2 or -GluN1) with an excess amount of a membrane-impermeable FLAG2-JF646-HTL (1 µM), and then the intracellular subpopulation (iHalo-GluA2 or -GluN1) with a membrane-permeable JF549-HTL (50 nM) (Fig. 3a). JF549-HTL, but not FLAG2-JF646-HTL, clearly labeled the cell bodies, consistent with the contrasting distribution pattern of the surface and intracellular iGluRs in soma (Fig. 3b and Supplementary Fig. 5d)31,32. We also confirmed that FLAG2-JF646-HTL occupied the surface receptors in a dose-dependent manner, and the concentration of 1 µM was high enough for the saturated labeling of the surface receptors (Supplementary Fig. 5e–i). Thus, our sequential labeling specifically visualizes the surface and intracellular subpopulations of endogenous iGluRs.

Fig. 3: Dual labeling of the surface and intracellular subpopulations of endogenous glutamate receptors at individual synapses.

a Schematic of the dual-labeling for surface and intracellular subpopulations of Halo-GluA2 (sHalo-GluA2 and iHalo-GluA2) or Halo-GluN1 (sHalo-GluN1 and iHalo-GluN1). b Confocal images for sHalo-GluA2 (green) and iHalo-GluA2 (cyan) in a pyramidal cell in the primary somatosensory layer 2/3 at P28. c, g Confocal images for the surface (green) and intracellular (cyan) subpopulations of Halo-GluA2 (c, sHalo-GluA2 and iHalo-GluA2) or Halo-GluN1 (g, sHalo-GluN1 and iHalo-GluN1) in single dendritic branches labeled for mEGFP (magenta) in layer 2/3 pyramidal cells in the primary somatosensory cortex at P28. d–f, h–j, Correlations between the spine-head volume and fluorescent intensity for sHalo-GluA2 (d) or sHalo-GluN1 (h) at individual dendritic spines. Correlations between the spine-head volume and fluorescent intensity for iHalo-GluA2 (e) or iHalo-GluN1 (i) at individual dendritic spines. Correlations between fluorescent intensities for sHalo-GluA2 and iHalo-GluA2 (f) or sHalo-GluN1 and iHalo-GluN1 (j) at individual dendritic spines. These correlations are measured using the same pool of dendritic spines for Halo-GluA2 (n = 215) or Halo-GluN1 (n = 222) obtained from seven dendritic branches of different pyramidal cells in five mice at postnatal 3-6 weeks. k Heat maps of the sHalo-GluA2-to-iHalo-GluA2 (S/I) ratio at individual ROIs on control (left) and SynGAP1-KO (right) pyramidal cells in the primary somatosensory layer 2/3 at P36-38. l–n Cumulative distributions of sHalo-GluA2 intensity (l), iHalo-GluA2 intensity (m), and S/I ratio (n) at individual ROIs in control (blue) and SynGAP1-KO (orange) pyramidal cells. The analyzed ROIs (n = 44,713 and 42,361 for control and SynGAP1-KO, respectively) are obtained from 12 pyramidal cells in four mice for each condition. ***, p < 0.001; ns, not significant. p = 0.8514 (l); p < 0.0001 (m, n). Two-tailed Kolmogorov–Smirnov test (l–n). Scale bars, 10 μm (b, k), 1 μm (c, g). AU indicates arbitrary units. Source data are provided as a Source Data file.

Halo-GluA2-labeled pyramidal cells showed a selective distribution of punctate sHalo-GluA2 signals at the spine-head portion, but not in dendritic shafts (Fig. 3c). This suggests that the surface population of endogenous GluA2-containing AMPARs is highly concentrated on the postsynaptic membrane at excitatory synapses, while at very low or undetectable levels on the extrasynaptic membrane of dendritic shafts. The integrated signal intensity of sHalo-GluA2 at the spine-head portion varied from very low to high levels among individual spines and was positively correlated with the spine-head volume measured by the fluorescence intensity of mEGFP (Fig. 3d). Since most AMPARs in pyramidal cells of the hippocampus or neocortex contain GluA233,34, our quantitative imaging is consistent with the functional mapping of AMPAR-mediated postsynaptic responses evoked by single spine stimulations35. We also successfully detected iHalo-GluA2 signals at dendritic spines (Fig. 3c), with its fluorescent peak located between those for sHalo-GluA2 and mEGFP (Supplementary Fig. 5j). The integrated signal intensity of iHalo-GluA2 was variable from spine to spine, and proportional to the spine-head volume (Fig. 3e). Moreover, a tight positive correlation between sHalo-GluA2 and iHalo-GluA2 was noted at individual spines (Fig. 3f). This expression pattern for GluA2 was the case for GluN1. We found accumulations of sHalo-GluN1 signals at the spine-head with the integrated intensity proportional to the spine-head volume (Fig. 3g, h), consistent with functional NMDAR-mediated currents triggered by single-spine stimulations36. We also detected weak iHalo-GluN1 signals at dendritic spines with a proportional relationship to the spine-head volume and sHalo-GluN1 signals (Fig. 3g, i, j). Thus, our in vivo dual labeling for HaloTag-fused endogenous transmembrane proteins with different HTLs is a powerful technique to differentially visualize surface and intracellular subpopulations at individual synapses in single neurons.

Simultaneous, synapse-level quantification of surface and intracellular AMPARs enables precise evaluation of AMPAR-mediated synaptic plasticity. When integrated with single-cell synaptome analysis, this approach offers a comprehensive assessment of synaptic plasticity in individual neurons. Using this methodology, we next performed single-cell synaptome mapping of surface and intracellular GluA2 in SynGAP1 knockout (KO) neurons, a well-established model of neurodevelopmental disorders37,[38](https://www.nature.com/articles/s41467-025-65813-w#ref-CR38 “Llamosas, N. et al. Syngap1 regulates experience-dependent cortical ensemble plasticity by promoting in vivo excitatory synapse strengthening. Proc. Natl Acad. Sci. USA 118 https://doi.org/10.1073/pnas.2100579118

(2021).“). While this model is known to impair AMPAR-mediated synaptic plasticity, it remains unclear whether the impairment is widespread across synapses or restricted to a specific subset within neurons. To address this, we used the CRISPR-Cas9 system to simultaneously introduce Halo-GluA2 knock-in and SynGAP1 knockout (Supplementary Fig. 6a, b). We then mapped both sHalo-GluA2 and iHalo-GluA2 at thousands of synapses in control and SynGAP1-KO pyramidal cells of the primary somatosensory cortex (Fig. 3k and Supplementary Fig. 6c). Compared with control neurons, SynGAP1-KO neurons exhibited a global decrease and increase in iHalo-GluA2 and sHalo-GluA2/iHalo-GluA2 (S/I) ratio, respectively, at the single-cell level, indicating a selective and robust reduction in the intracellular pool of GluA2-containing AMPARs at the majority of synapses in SynGAP1-KO neurons (Fig. 3l–n and Supplementary Fig. 6c). These findings suggest that AMPAR-mediated synaptic plasticity is broadly affected rather than confined to specific synaptic populations in SynGAP1-KO neurons. Overall, our single-cell synaptome mapping approach provides a quantitative representation of the synaptic landscape within individual neurons, offering valuable insights into synaptic phenotypes associated with neurodevelopmental disorders.

Single-cell multiplexed synaptome mapping of different endogenous synaptic proteins

Since HaloTag is compatible with other self-labeling protein tags, such as SNAP-tag or CLIP-tag without cross-labeling39, multiplexed chemical tag labeling in the same cell could expand the applicability of the protein labeling toolkit. We confirmed no difference in the signal pattern between HaloTag- and SNAP-tag-mediated protein labeling (Supplementary Fig. 1o–q). To test if SLENDR enabled multiplexed protein tagging with HaloTag and SNAP-tag in the same cells, we simultaneously targeted HaloTag and SNAP-tag into the N-terminals of GluA2 and CaMKIIα, respectively (Supplementary Fig. 7a). We found 3.3% of SNAP-CaMKIIα-positive neurons co-labeled with Halo-GluA2 (n = 61) in cortical slice cultures (Supplementary Fig. 7b), suggesting an efficient multiplexed chemical tag labeling of two different proteins in the same neurons.

The AMPA/NMDA ratio is known to be highly correlated with synaptic plasticity, but it is not feasible to estimate the AMPA/NMDA ratio in single excitatory synapses. To quantitatively image both AMPARs and NMDARs in the same dendritic spines, we next targeted HaloTag and SNAP-tag to GluN1 and GluA2 in pyramidal cells of the somatosensory cortex layer 2/3 (Fig. 4a). We found dually-labeled pyramidal cells for Halo-GluN1 and SNAP-GluA2 with different color ligands (100 nM JF646-HTL and 10 nM TMR-Star SNAP-tag ligand (TMR-STL)) in fixed brain slices at postnatal day 27 or 35 (P27 or P35) (Supplementary Fig. 7c). Both Halo-GluN1 and SNAP-GluA2 accumulated at the same dendritic spines labeled for mEGFP (Fig. 4b). Consistent with Fig. 3, Halo-GluN1 and SNAP-GluA2 integrated intensities were proportional to the spine-head volume (Supplementary Fig. 7d, e), and positively correlated with each other at individual spines (Fig. 4c). We noted a low correlation between the Halo-GluN1/SNAP-GluA2 ratio and spine-head volume (Fig. 4d). This low correlation may be due to differences in the distribution of this ratio between small and large dendritic spines. Indeed, the CV for the SNAP-GluA2/Halo-GluN1 ratio is higher at small dendritic spines with the bottom 20% of mEGFP intensity, compared to large dendritic spines with the top 20% of mEGFP intensity (CV = 0.68 vs 0.44, 40 spines from three mice for each), suggesting greater variability in the ratio among smaller spines. Intriguingly, some small dendritic spines accumulated Halo-GluN1 signals but not SNAP-GluA2 signals (Fig. 4b, arrows, d), which may represent “silent” synapses that contain NMDARs but no AMPARs40.

For more comprehensive mapping of functional AMPARs at excitatory synapses, we performed dual labeling of surface Halo-GluA2 (sHalo-GluA2) and intracellular PSD95-SNAP labeled with membrane-impermeable (100 nM FLAG2-JF646-HTL) and permeable (10 nM TMR-STL) ligands in single pyramidal cells of the somatosensory cortex layer 2/3 at P15 or P16 (Fig. 4e). We obtained z-stacks from four dually-labeled pyramidal cells derived from different animals (Supplementary Fig. 7f). PSD95-SNAP-labeled puncta were very often co-labeled for sHalo-GluA2 (Fig. 4f). Interestingly, the intensity of sHalo-GluA2 signals varied remarkably, with no detectable levels at a few PSD95-SNAP-labeled puncta (Fig. 4f, arrows). To quantify the amount of surface GluA2 at individual excitatory synapses, we semi-automatically registered 1004-1296 PSD95-SNAP-labled puncta as synaptic ROIs (Fig. 4g and Supplementary Fig. 7g). We found that the size of synaptic ROIs was tightly correlated with their integrated intensity for PSD95-SNAP over serial images (Supplementary Fig. 7h). Since the integrated intensity of fluorescently-labeled PSD95 is known to be proportional to the excitatory PSD size41, we concluded that the synaptic ROI size serves as a proxy of the excitatory PSD size. We next calculated and mapped the “signal density” of PSD95-SNAP and sHalo-GluA2 by normalizing their integrated intensity to the synaptic ROI size (Fig. 4h, i and Supplementary Fig. 7i, j). While the integrated intensity represents the total amount of PSD95-SNAP and sHalo-GluA2 at individual synapses, the signal density reflects the local concentration of these proteins. Notably, the signal density of sHalo-GluA2 was highly variable compared with that of PSD95-SNAP (CV: 0.53-0.63 for sHalo-GluA2 and 0.29–0.32 for PSD95-SNAP). When examining the relationship between the signal density and synaptic ROI size, we found a strong positive correlation for PSD95-SNAP, whereas sHalo-GluA2 density showed only a weak correlation (Fig. 4j, k and Supplementary Fig. 7k, l). Neither signal density measure correlated with the distance from the soma within the analyzed imaging volume (~ 60 µm from the soma) (Fig. 4l, m and Supplementary Fig. 7m, n), suggesting that both PSD95 and surface AMPARs preferentially accumulate at larger synapses regardless of their spatial location.

We then ranked synaptic ROIs based on their sHalo-GluA2 signal density, and grouped them into high (top 5%, presumably potentiated), medium (middle 90%), and low (bottom 5%, presumably silent) sHalo-GluA2 density groups40 (Fig. 4n). The average synaptic ROI size of the high sHalo-GluA2 density group was 1.3–2.4 (1.86 ± 0.23) times higher than the medium sHalo-GluA2 density group, while the low sHalo-GluA2 density group was 0.36–0.46 (0.43 ± 0.02) times lower. These findings suggest that potentiated synapses are relatively larger and silent synapses smaller in PSD size (Fig. 4o, p and Supplementary Fig. [7o](https://ww