Highly Sensitive Spatial Proteomics with Multicolor Cleavable Fluorescent Tyramide

Abstract

High-resolution single-cell spatial proteomics offers transformative insights into cellular diversity, architecture, interactions, and functions within complex biological systems. However, the existing multiplexed protein imaging platforms face challenges such as limited detection sensitivity, constrained target multiplexing capacity, or technically demanding. To address these issues, we report a highly sensitive spatial proteomics approach, using multicolor cleavable fluorescent tyramide and off-the-shelf antibodies. This method employs horseradish peroxidase (HRP) to enzymatically deposit distinct fluorophores to stain varied target proteins. Through reiterative cycles of target labeling, fluorescence imaging, and fluorophore cleavage, this approach allows numerous proteins profiled at the optical resolution in the same specimen. Utilizing this technique, we quantified 38 proteins within a human formalin-fixed paraffin-embedded (FFPE) tonsil tissue, which represents the highest target multiplexing capacity achieved to date using tyramide signal amplification (TSA) methods. Analysis of ∼500,000 individual cells in the same tissue revealed distinct cell clusters based on their protein expression profiles and spatial microenvironment. By mapping the cells back to their original tissue locations, we observed specific tissue subregions are composed of unique cell clusters. Furthermore, we also studied the cell–cell interactions and found the cells from the same cluster often showed strong association, while the cells in the varied clusters usually avoided contact.

Introduction

Complex biological systems, such as solid tumors, developing embryos and brain tissues, usually consist of diverse cell types with distinct molecular and functional profiles. , Spatial proteomics has emerged as a powerful approach to investigate the function, regulation, and interactions of these heterogeneous cell populations. While mass spectrometry and microarray enable extensive protein profiling, they require the extraction of proteins from their native environment. Consequently, the critical spatial information on the proteins is lost. Immunofluorescence, on the other hand, preserves the protein localization information by quantifying them at their original cellular contexts. However, traditional immunofluorescence is constrained by spectral overlap among the common fluorophores, which limits the simultaneous detection to only a few proteins per sample.

To advance single-cell spatial proteomic analysis, techniques such as cyclic immunofluorescence and mass spectrometry imaging have been explored. These methods − enable quantification of numerous proteins within their native cellular environments in single cells. However, such existing technologies still suffer from some limitations. In some approaches, antibodies are directly conjugated with fluorophores or metal isotopes, which often lack sufficient signal amplification. This can reduce sensitivity, making it challenging to detect low-abundance proteins or analyze samples with significant autofluorescence. Other strategies amplify signals using antibodies conjugated to oligonucleotides, haptens, or horseradish peroxidase (HRP). Despite improved sensitivity, the availability of chemically modified antibodies for many targets is limited, and developing and validating conjugated antibody panels is often a technically complex, time-consuming, and costly process.

To achieve sensitive spatial proteomics using off-the-shelf unmodified primary antibodies, we developed a multiplexed protein imaging technique with cleavable fluorescent tyramide (CFT). This method begins by labeling target proteins with unconjugated primary antibodies, followed by staining with secondary antibodies labeled with HRP and CFT. After imaging, the fluorophores are chemically removed, and antibodies are stripped from the sample, enabling repeated cycles of staining and imaging. In this approach, HRP catalyzes the conversion of CFT into short-lived reactive radicals that form covalent bonds with tyrosine residues in close proximity. This enzymatic reaction significantly amplifies the signal intensity. Nonetheless, with only one CFT successfully developed using Cy5 as the fluorophore, the cleavage reaction and the specimen imaging have to be carried out after every protein is stained. These accumulated steps after many cycles could result in potential tissue damage and limit the target multiplexing capacity. Additionally, the prolonged assay time will also lead to reduced sample throughput.

Here, we present the development of multicolor CFT for highly sensitive spatial proteomics. Instead of only imaging one protein per analysis cycle, multiple proteins can be stained using varied CFT with distinct fluorophores. Then, all the different proteins stained in one cycle can be imaged simultaneously. Afterward, the multicolor fluorophores in the whole specimen can be removed together by one cleavage reaction. In this way, the number of imaging and cleavage steps in a given experiment is dramatically reduced. As a result, the tissue integrity is better preserved, and more proteins can be profiled in the sample. Moreover, it also significantly shortens assay time and enhances sample throughput. Using this approach, we successfully profiled 38 proteins within single cells from human formalin-fixed paraffin-embedded (FFPE) tonsil tissue. Their unique protein expression patterns and cellular microenvironments allowed the classification of approximately 500,000 cells into distinct clusters. Moreover, we identified that specific cell clusters were enriched in particular subregions of the tissue, revealing the presence of spatially organized cellular neighborhoods within the tonsil architecture.

Experimental Section

General Information

Chemicals and solvents were obtained from Sigma-Aldrich (St. Louis, MO, USA) or TCI America (Portland, OR, USA) and used without further purification. Bioreagents were purchased from Abcam (Cambridge, UK), Invitrogen (Waltham, MA, USA), or Novus Biologicals (Littleton, CO, USA), unless otherwise specified.

Deparaffinization and Antigen Retrieval of FFPE Tonsil Tissues

After a 1 h baking at 60 °C, tonsil FFPE tissue slides (NBP2–30207, Novus Biologicals, Littleton, CO, USA) underwent deparaffinization through three sequential xylene treatments, each for 10 min. The slides were subsequently immersed in 50/50 xylene/ethanol, 100% ethanol, 95% ethanol, and 70% ethanol sequentially for 2 min each, and then rinsed with deionized water. Heat-induced antigen retrieval (HIAR) was then performed using a microwave. Slides were treated with antigen retrieval citrate buffer (Abcam ab64236) and heated at high power (700 W, level 10) for 2 min and 45 s, followed by low power (140 W, level 2) for 14 min, then allowed to cool to room temperature. Endogenous horseradish peroxidase (HRP) activity was blocked by incubating the slides in 3% H₂O₂ prepared in PBT (0.1% Triton X-100 in 1× PBS) for 10 min. Residual H₂O₂ was removed with two washes in PBT.

Reiterative Protein Staining in FFPE Tissues

To minimize nonspecific interactions, tissue slides were pretreated with antibody-blocking buffer (0.1% Triton X-100, 1% BSA, and 10% normal goat serum) for 30 min at room temperature. Slides were then washed three times in PBT.

The antibody labeling and staining procedure for FFPE tissue was initiated by mixing 1 μL of antibody (Supplementary Table S1) with the primary antibody labeling solution from Spatomics, following their instructions. The labeled antibody solution was diluted to the desired working concentration using antibody-blocking buffer and applied to the FFPE tissue. Incubation was performed for 45 min at room temperature. Then, the slides were washed with PBT for three times. Cleavable fluorescent tyramide (CFT) was prepared by mixing 1 μL of 0.5 mM CFT with 1 μL of 0.3% H₂O₂ in 200 μL of 0.1 M boric acid PBT solution. This solution was applied to the tissue, which was incubated for 10 min at room temperature and subsequently washed three times with PBT. To inactivate residual HRP activity, the tissue was treated with HRP blocking solution from Spatomics for 30 min at room temperature, followed by three washes with PBT.

The labeling procedure was repeated iteratively for additional target proteins using different antibodies and CFT fluorophores. Once all target proteins for a given staining cycle were labeled, the tissue was counterstained with DAPI (1 μg/mL in PBT) for 10 min, washed three times with PBT, and mounted with Prolong Diamond Antifade Reagent for fluorescence microscopy imaging of the stained proteins.

Signal removal was performed by immersing the slide in PBS to detach the cover glass. The tissue was then incubated with 0.1 M tris (2-carboxyethyl) phosphine (TCEP) solution in PBT (pH adjusted to 9.0 with 1 M sodium hydroxide) for 30 min at 40 °C, followed by three washes with PBT. A subsequent incubation with 0.1 M 1,3,5-Triaza-7-phosphaadamantane (PTA) solution in PBT was conducted under the same conditions to ensure complete fluorophore cleavage, followed by additional PBT washes. The slide was mounted with Prolong Diamond Antifade Reagent and imaged to verify signal removal.

Multiplexed in situ protein staining was achieved by repeating the iterative staining and signal removal procedures. After completing staining cycles that required antigen retrieval at pH 6.0, a final antigen retrieval step was carried out using Tris-EDTA buffer (pH 9.0) in a microwave, allowing for the staining of antibodies requiring higher pH retrieval conditions. This process facilitated the multiplexed detection of multiple target proteins within the same tissue sample.

Imaging and Data Analysis

A 4× objective-equipped Nikon epifluorescence microscope was used to image formalin-fixed, paraffin-embedded (FFPE) tissue samples. Images were captured using a CoolSNAP HQ2 camera and C-FLL large field of view filter sets: DAPI (Nikon 96369) (for DAPI and Tyramide-N₃-Coumarin), FITC (Nikon 96372) (for Tyramide-N₃-Fluorescein), Cy3 (Nikon 96374) (for Tyramide-N₃-Cy3), Cy5 (Nikon 96376) (for Tyramide-N₃-Cy5), and Cy7 (Nikon 96377) (for Tyramide-N₃-Cy7).

The acquired multiplexed fluorescence images were processed and aligned across different imaging cycles using the NIS-Elements Imaging Software. To ensure accurate spatial alignment, the DAPI-stained nuclear images from each cycle were used as a reference. Image registration was performed using an automated feature-based alignment algorithm to minimize spatial shifts between cycles. For large tissue sections, image stitching was conducted using NIS-Elements Imaging Software to maintain spatial continuity. Background subtraction was applied using a rolling ball filter (radius = 50 pixels) to remove uneven illumination.

Single-cell segmentation was carried out using a two-step approach involving nuclear segmentation followed by cytoplasmic and whole-cell boundary detection. Nuclei were identified using the DAPI channel, and segmentation was performed using CellProfiler. Overlapping nuclei were separated using a watershed algorithm to refine individual nuclear boundaries. Cytoplasmic segmentation was achieved by expanding the nuclear masks by 10 pixels to capture perinuclear protein expression.

Following segmentation, single-cell protein expression quantification was conducted using CellProfiler. The mean, median, and integrated fluorescence intensity of each protein marker were measured within segmented cellular regions. To account for variations in staining intensity and batch effects, per-channel normalization was applied using z-score normalization. The processed single-cell data were stored as a feature matrix (cells × markers) in comma-separated value (CSV) files (Supplementary Table S2). All processed data, including single-cell feature matrices, spatial maps, and raw images, were saved in CSV and OME-TIFF formats to ensure reproducibility and facilitate downstream computational analysis.

For clustering analysis, dimensionality reduction was performed using optimized t-distributed stochastic neighbor embedding (Opt-SNE), as described in the literature. This Opt-SNE analysis was performed using the following parameters: perplexity = 100, θ = 0.8, maximum iterations = 500, early exaggeration stopping at iteration 250, learning rate = 10,000, and early exaggeration factor = 100. To visualize tissue architecture, pseudocolored images were generated using ImageJ.

Spatial analysis was conducted to assess cell–cell interactions and microenvironmental organization. Cellular neighborhoods were identified by computing nearest-neighbor distances within a 20 μm radius of each cell. The number of cells belonging to different clusters within each cell’s neighborhood was used for further clustering, resulting in subclustered Opt-SNE plots.

Results and Discussion

Platform Design

This highly sensitive spatial proteomics technique involves six steps, as illustrated in Figure . First, off-the-shelf primary antibody mixed with the Spatomics HRP conjugation reagent is applied to bind to the target protein. Next, the protein target is stained with the first CFT. Subsequently, HRP is deactivated, without damaging the integrity of the epitopes and fluorophores. Then, steps one to three are repeated sequentially to detect all the protein targets using different colored CFT within the current analysis cycle. In the fifth step, images are captured using a fluorescence microscope, providing detailed single-cell protein expression data. To assist with image alignment, nuclei are costained with DAPI and imaged simultaneously with the protein targets. Finally, the fluorophores are chemically cleaved to prepare for the next cycle. By iteratively performing protein labeling, fluorescence imaging and signal removal, this method enables sensitive and multiplexed protein quantification at the single-cell level within intact tissues.

1.

Highly sensitive and multiplexed protein imaging with multicolor cleavable fluorescent tyramide (CFT). In each cycle, the first target is stained with HRP labeled antibodies and CFT. After HRP deactivation, another target is stained using CFT with a different fluorophore. These steps of protein labeling and HRP deactivation are repeated until each target in the first cycle is stained. Finally, images are captured, and all the fluorophores are chemically cleaved, to initiate the next analysis cycle.

To enable the multicolor protein imaging, we designed and synthesized five CFT with varied fluorophores (Figure ). These fluorophores are tethered to tyramide through an azide-based small molecule linker. , Such short linker ensures CFT can still be recognized by HRP as good substrates, and the generated CFT radical can efficiently couple to the tyrosine in close proximity. Additionally, this linker also allows the fluorophores to be effectively cleaved, without damage the integrity of epitopes. The detailed synthesis () and absorption/emission spectra () of the five CFT is described in the .

2.

Chemical structures of CFT with different fluorophores.

Efficient Staining with Multicolor CFT

To assess whether the synthesized CFT can be successfully applied for protein labeling, we stained protein NPM1, BRCA1, CD55, SQSTM1 and Histone H4 sequentially on a tonsil FFPE tissue (Figure ). These five targets were stained by Tyramide-N₃-Coumarin, Tyramide-N₃-Fluorescein, Tyramide-N₃-Cy3, Tyramide-N₃-Cy5 and Tyramide-N₃-Cy7, respectively. The obtained staining results are consistent with the ones generated by conventional immunohistochemistry, suggesting HRP can recognize the multicolor CFT as good substrates.

3.

Fluorescence microscopy images of 5 different proteins stained with 5 different CFTs on a single tonsil FFPE tissue. (a) NPM1 is stained with Tyramide-N₃-Coumarin. (b) BRCA1 is stained with Tyramide-N₃-Fluorescein. (c) CD55 is stained with Tyramide-N₃-Cy3. (d) SQSTM1 is stained with Tyramide-N₃-Cy5. (e) Histone H4 is stained with Tyramide-N₃-Cy7. (f) Overlapped image of (a)-(e). Scale bar, 100 μm.

Effective HRP Deactivation

After each protein is stained, HRP is deactivated using the Spatomics HRP quencher. The previous protein staining signals did not reappear in the other fluorescence imaging channels (Figure ), indicating HRP is efficiently deactivated. Additionally, the protein targets were successfully stained following the cyclic HRP deactivation step, and imaged together after all the five proteins are labeled. These results suggest the HRP quencher does not damage the integrity of the epitopes or the fluorophores.

Efficient Cleavage with Multi-Color CFT

To evaluate the fluorophore cleavage efficiency of the multicolor CFT, we incubated the stained tissue with the mild reducing reagents tris (2-carboxyethyl) phosphine (TCEP) and 1,3,5-Triaza-7-phosphaadamantane (PTA). Following cleavage, almost all the signals in the five fluorescence channels are simultaneously removed (Figure ). These results suggest the azide-based linker can be efficiently cleaved within single cells of FFPE tissues, regardless of the varied cellular locations or the different fluorophores. And we have demonstrated that the integrity of the protein epitopes is preserved after the TCEP and PTA treatment. , Thus, by cyclic protein staining and signal cleavage, many distinct proteins can be profiled in single cells in situ.

4.

(a) Fluorescence images of staining and cleavage for different protein targets stained with different CFT on a single tonsil FFPE tissue. Scale bar, 20 μm. (b) Fluorescence intensity profiles corresponding to the indicated arrow positions in (a).

Multiplexed Protein Imaging of Human FFPE Tissue

To validate this method for multiplexed protein imaging, we analyzed 38 distinct proteins in human FFPE tonsil tissue using the multicolor CFT, Tyramide-N₃-Fluorescein, Tyramide-N₃-Cy3 and Tyramide-N₃-Cy5. With off-the-shelf primary antibodies, we successfully detected all 38 proteins with subcellular resolution. In this experiment, only the well-validated primary antibodies (Supplementary Table S1) are selected, and they are used directly without further chemical modifications, ensuring the high staining specificity of our assay. Indeed, the observed staining patterns (Figure and Supplementary Figure S7) using our method closely matched those produced by traditional immunohistochemistry methods. Additionally, using the off-the-shelf primary antibodies directly also enables the rapid development of new antibody panels for various other studies. With the staining signals enhanced by HRP-mediated signal amplification, our approach maintains the exceptional sensitivity of TSA assays. This enables reliable protein detection even in highly autofluorescent FFPE tissues, requiring only millisecond exposure times with a 4× objective lens. Such efficiency allows imaging of 1–2 cm² tissue sections within 5 min. With multiple proteins profiled in each cycle, the number of imaging and signal cleavage steps are also significantly reduced. Such minimization in both the imaging time and the cycle number lead to shorter assay time and enhanced sample throughput. And by successfully profiling 38 proteins in a single tissue section, this method demonstrates the highest target multiplexing capacity reported among TSA-based assays.

5.

Zoomed-in view of 38 different proteins stained with CFT on the same FFPE tonsil tissue. Scale bar, 250 μm.

The target multiplexing capacity of this technique depends primarily on the number of imaging cycles and the quantity of proteins analyzed per cycle. Our previous studies confirmed that the signal removal and antibody stripping treatments preserve epitope integrity. Additionally, we recently demonstrated that at least 28 iterative cycles can be successfully carried out on a single tissue sample. In this study, five varied protein targets were visualized simultaneously in each cycle. With spectral unmixing, over 15-color imaging can be achieved. As all the organic fluorophores we tried have been successfully applied to develop CFT, we expect any spectrally distinguishable fluorophores can be used together to increase the number of proteins profiled in one cycle. These capabilities combined position our platform to analyze potentially hundreds of protein targets within a single specimen.

Our approach and single-cell proteomics by mass spectrometry (MS) offer complementary strengths in studying cellular protein expression. MS-based approaches provide unbiased, high-depth proteomic profiling, enabling the detection of thousands of proteins per cell without the need for preselected markers. However, MS lacks spatial resolution, requires extensive sample preparation (including protein isolation and purification), and suffers from limited sensitivity for low-abundance proteins. On the other hand, our method enables highly sensitive spatial proteomics, allowing researchers to visualize potentially hundreds of proteins in their native tissue context with subcellular resolution. Despite its advantages in spatial mapping, our approach is limited by antibody availability, requiring validation to ensure the staining specificity and reproducibility, and it cannot detect proteins outside the selected panel. Integrating these two approaches could provide a more comprehensive view of cellular proteomics. For example, MS-based discovery proteomics can identify novel biomarkers and protein networks, while our method enables spatial validation and functional mapping in the tissue architecture. Or specific regions of tissue samples can be first identified as regions of interest by multiplexed protein imaging. Then more comprehensive protein profiling can be performed in such regions by MS. These integrations would be particularly valuable in applications like tumor microenvironment studies, immune profiling, and biomarker validation, where both quantitative proteomic depth and spatial organization are critical for understanding the biological system.

Cell Type Classification and Their Spatial Distribution

With the obtained single cell in situ protein expression profiles, we explored the heterogeneity and spatial organization of cell types within the human tonsil tissue. With the unique expression signatures of the 38 varied proteins (Figure A), we classified the ∼500,000 cells into 12 distinct clusters (Figure B) through dimensionality reduction and clustering using the OptSNE algorithm. By mapping these clusters back to their tissue coordinates (Figure C and Supplementary Figure S8), we observed that specific subregions of the tonsil tissue were dominated by particular clusters. For instance, cluster 10 was concentrated in the epithelium; clusters 4 and 6 were prevalent in the submucosa and connective capsule; clusters 3 and 5 were enriched in lymphoid follicles; cluster 2 was predominant in germinal centers; and clusters 1, 7, 8, and 11 were restricted to connective tissues. These results demonstrate the feasibility of applying our method to classify diverse cell types and uncover their spatial distributions in formalin-fixed paraffin-embedded (FFPE) tissues.

6.

(a) Based on their single cell expression profiles of 38 proteins, (b) ∼500,000 cells in a human tonsil tissue are classified into 12 different clusters. (c) Anatomical locations of each cell in the 12 identified cell clusters.

Cell–Cell Interaction in Human Tonsil Tissue

With the cell types classified at their native tissue locations, our method also facilitates the analysis of cell–cell interactions between the varied cell clusters. To study these interactions, we defined a “cell neighborhood” as the area encompassing all cells located within 20 μm of a reference cell. For the approximately 500,000 reference cells within the human tonsil tissue, we analyzed the composition of their neighborhoods by counting the number of surrounding cells belonging to different clusters. We then calculated the correlation coefficients between these cell counts for every pair of clusters. When visualized as a heatmap (Figure ), this analysis highlighted selective associations and avoidance patterns between specific cell clusters, revealing nonrandom spatial relationships among the cell populations. The analysis revealed strong associations between specific clusters, such as clusters 1 and 7, 3 and 7, and 5 and 8. Notably, cells within the same cluster usually exhibited high levels of association (Figure , diagonal), while cells from different clusters tended to remain spatially segregated. These observations suggest that homotypic adhesion between cells of the same type likely plays a critical role in organizing the structural framework of human tonsil tissue.

7.

Heatmap of cell–cell interaction strength. The upper triangle displays the calculated correlation coefficients of the cell numbers in each cell neighborhood. The lower triangle shows the colors corresponding to the obtained correlation coefficients.

Additionally, cells within the same cluster can be further divided into subclusters based on the composition of their surrounding cells in their neighborhoods. Mapping these subclusters back to their tissue locations revealed that different subclusters of the same parent cluster occupy distinct regions within the tonsil tissue (Supplementary Figures S9–S19). For example, cluster 12 was divided into four subclusters: 12a, 12b, 12c and 12d (Figures A and B). Subcluster 12a and 12d were primarily localized in the submucosa and connective capsule; subcluster 12b was predominantly found in the epithelium connective tissues; subcluster 12c was concentrated in connective tissues (Figure C). These results demonstrate the ability of our approach to refine cell classification and analyze cell–cell interactions by considering their microenvironment.

8.

(a) Based on the identities and numbers of their neighboring cells, (b) cells in cluster 12 were further divided into different subclusters. (c) Anatomical locations of each cell from the varied subclusters.

Conclusions

This study introduces the development of multicolor cleavable fluorescent tyramide, enabling fast and sensitive spatial proteomics analysis. Our strategy addresses a critical limitation of the cyclic TSA assay: only one protein quantified per cycle. Using our method, we classified cells within human tonsil tissue into distinct clusters based on their unique protein expression profiles. Our analysis revealed that different tonsil subregions are composed of specific cell clusters. Additionally, we examined interactions between cell clusters, identifying patterns of association and avoidance. Importantly, cells within each cluster could be further subdivided into subclusters by analyzing the neighboring cell populations. These results demonstrate the potential of our approach for classifying cell types and subtypes by protein profiling and contextual cellular interactions. Such advancement will enhance our understanding of cell heterogeneity and its implications for disease diagnosis and patient stratification.

The versatility of the multicolor cleavable fluorescent tyramide extends beyond protein imaging, with potential applications in sensitive and multiplexed nucleic acid − and metabolic imaging. By integrating these technologies, our platform enables comprehensive in situ profiling of DNA, RNA, proteins, and metabolites at single-cell resolution in intact tissues. Furthermore, combing a programmable microfluidic system with a standard fluorescence microscope could lead to an automated imaging platform. Collectively, these advancements establish a highly multiplexed molecular imaging platform, which will bring new insights in biology and medicine.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/cbmi.4c00099.

• Experimental details, synthetic schemes, and tSNE maps (PDF)

• Single cell protein expression (ZIP)

The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

The authors declare the following competing financial interest(s): J.L. and J.G. are inventors on patent applications filed by Arizona State University that covers the method of using cleavable fluorescent tyramide for multiplexed biomolecule analysis. J.G. is a co-founder of Spatomics.