Mapping the nanoscale organization of the human cell surface proteome reveals new functional associations and surface antigen clusters

Abstract

The cell surface is a dynamic interface that controls cell-cell communication and signal transduction relevant to organ development, homeostasis and repair, immune reactivity, and pathologies driven by aberrant cell surface phenotypes. The spatial organization of cell surface proteins is central to these processes. High-resolution fluorescence microscopy and proximity labeling have advanced studies of surface protein associations, but the spatial organization of the complete surface proteome remains uncharted. In this study, we systematically mapped the surface proteome of human T-lymphocytes and B-lymphoblasts using proximity labeling of 85 antigens, identified from over 100 antibodies tested for binding to surface-exposed proteins. These experiments were coupled with an optimized data-independent acquisition mass spectrometry workflow to generate a robust dataset. Unsupervised clustering of the resulting interactome revealed functional modules, including well-characterized complexes such as the T-cell receptor and HLA class I/II, alongside novel clusters. Notably, we identified mitochondrial proteins localized to the surface, including the transcription factor TFAM, suggesting previously unappreciated roles for mitochondrial proteins at the plasma membrane. A high-accuracy machine learning classifier predicted over 6,000 surface protein associations, highlighting functional associations such as IL10RB's role as a negative regulator of type I interferon signaling. Spatial modeling of the surface proteome provided insights into protein dispersion patterns, distinguishing widely distributed proteins, such as CD45, from localized antigens, such as CD226 pointing to active mechanisms of regulating surface organization. This work provides a comprehensive map of the human surfaceome and a resource for exploring the spatial and functional dynamics of the cell membrane proteome.

Figure 1.

Systematic proximity labeling of the Jurkat cell surface proteome reveals functional clusters on the cell surface. (A) Overview of the approach to spatially map the Jurkat cell surface proteome. DIA-NN: DIA by neural networks; GPF-DIA: Gas phase fractionation data independent acquisition. (B) Scatterplot of log fold change enrichment values and correlation coefficients for the known protein interactors (B) ITGB1/ITGA4 and (C) CD3 complex members CD3D and CD3G. (D) Distribution of pairwise Pearson correlation coefficient across all Jurkat surface protein pairs. Additional known protein interactions are highlighted. (E) Scatterplot of T-cell receptor subunit pairwise distances compared to their estimated Pearson correlation values from Jurkat surface proximity labeling experiments. (F) Circle plot of the clustered Jurkat cell surface proteome labeled with enriched functional annotations for each cluster and colors indicating cell surface characterization of individual surface proteins.

Figure 2.

Cell surface protein interactions are conserved across different cell types. (A) Numerical summary of cell surface proximity labeling done on Daudi B-lymphoblast cells. (B) Scatterplot of log fold change enrichment values and correlation coefficients for the MHC class 2 subunits HLA-DRA and HLA-DRB1 (C) and B-cell receptor SRK subunits CD79A and CD79B. (D) Venn diagram showing overlap of cell surface proteins identified on Jurkat and Daudi cells. (E) Scatterplot of pairwise Pearson correlation coefficients in Daudi and Jurkat cell lines. The dashed line represents a line of perfect correspondence where y = x. (F) Functional annotation counts of conserved protein-protein interactions in Daudi and Jurkat cell lines. Conserved protein-protein interactions are defined as protein pairs with Pearson coefficients greater than 0.7 and difference between the two cell lines of less than 0.1. (G) Circle plot of the clustered Daudi cell surface proteome labeled with enriched functional annotations for each cluster and colors indicating cell surface characterization of individual surface proteins.

Figure 3.

Detection and validation of maverick mitochondrial proteins on the cell surface of Jurkat cells. (A) Breakdown of annotated subcellular localizations for the Jurkat cell surface proteome. (B) Flow cytometry cell surface staining with an isotype control (grey) and anti-TFAM (black) antibody on Jurkat WT and TFAM KO cell lines. P-values were estimated using a Student’s T-test. (C) Cartoon of the cell surface proximity labeling experiment using TFAM as a bait. Volcano plot showing the results of cell surface proximity labeling using an anti-TFAM antibody. TFAM is highlighted in dark blue and other enriched proteins are shown in light blue. (D) Anti-TFAM flow cytometry cell surface staining results across 13 different cell lines. Fold change was estimated by comparing to isotype control binding levels on each cell line. (E) Anti-TFAM cell surface staining results for different cell types from 6 different donors. Fold change was estimated by comparing to isotype control binding levels. (F) Distribution of TFAM pairwise Pearson values with other detected surface mitochondrial proteins (grey) and all non-mitochondrial surface proteins (black). (H) Sub-mitochondrial counts for mitochondrial proteins detected on the cell surface. (H) Enriched GO biological process terms for cell surface mitochondrial proteins. (I) Description of the analysis done to identify native N-termini peptides from the mitochondrial transit sequence for mitochondrial proteins detected on the cell surface and bar plot showing the count of maverick mitochondrial proteins with peptides detected in the TPPRED3 predicted mitochondrial transit sequence (MTS).

Figure 4.

Design of a machine learning classifier for predicting cell surface protein associations. (A) Workflow for training a random forest classifier to identify cell surface associations. (B) Receiver operator characteristic (ROC) curve for 303 withheld gold standard protein pairs. Area under the ROC curve (ROC AUC) is 0.93. (C) Precision-recall curve for 303 withheld protein pairs. Area under the precision recall curve (PR AUC) is 0.62. (D) Distribution of Pearson correlation coefficients for protein pairs with high random forest classifier association predictions and low random forest classifier predictions. High association scores are defined as any protein pair with a score higher than 0.6. (E) Comparison of experimentally measured proximity labeling fold change (FC) enrichments for 49 bait experiments to the classifier predicted association scores of bait-prey interactions. Welch’s T-test statistic 16.69, and p-value 8.045e-49. (F) Kawada-Kawai network of classifier predictions for T-cell receptor subunits. Edge weight corresponds to association scores. (G) Kawada-Kawai network of classifier predictions for integrin alpha and beta proteins detected on the Jurkat cell surface. Edge weight corresponds to association scores.

Figure 5.

Prediction and characterization of IL10RB as a Type 1 Interferon regulator. (A) Distribution of classifier predicted association scores for all protein pairs with the prediction for IFNAR2 and IL10RB highlighted. (B) Cartoon of the Type 1 Interferon signaling transduction pathway. (C) Western blot of phosphorylated STAT1 (pSTAT1) abundance in IFNα2 stimulated Jurkat WT and IL10RB knockout lines at different interferon concentrations. (D) Intracellular flow cytometry measurement of cytosolic IRF9 abundance change in IFNα2 stimulated Jurkat wild type (black) and IL10RB knockout lines (grey) at various IFNα2 concentrations. Student’s t-test p-values for at the different concentrations are 0.027, 0.0036, and 0.035 for 0.1, 0.5, and 10 ng/mL respectively. (E) Cartoon workflow for monitoring interferon stimulated gene protein stimulation using quantitative proteomics when Jurkat WT and IL10RB KO cells were treated with 10 ng/mL of IFNα2 for 0, 2, 4, 8, 12, and 24 h. (F-H) Abundance change plots of selected ISGs (F) IFIT3, (G) IFI44, and (H) XAF1 in the Jurkat WT (black) and IL10RB KO (grey) cell lines. Relative fold change was estimated in comparison to the 0 h time point. (I) Proposed model for IL10RBs regulatory mechanism for Type I Interferon signaling by competing with IFNAR1 for binding of the Jak kinase family member Tyk2.

Figure 6.

Three-dimensional network modeling to predict the surface distribution of the cell surface proteome. (A) Design walkthrough of 3D network heatmaps where (i) a 2D Kamada-Kawaii network of high-quality proximity labeling bait proteins are (ii) transformed to lie on a 3D sphere where (iii) the normalized fold change of a given surface protein of interest is projected for each bait experiment (iv) and graph connectivity can be estimated for the entire cell surface proteome. (B) 3D network heatmap models of CD45, (D) CD81, and (E) CD226. (F-H) Representative confocal microscopy images of immunostained cell surface antigens CD45, CD81, and CD226. Antigens were visualized in the GFP channel using the dye Alexa Fluor 488. (H) Scatterplot comparing the protein abundance of Jurkat cell surface proteins measured using WGA-HRP for surface protein enrichment compared to their estimated graph connectivity. (I) Distribution of estimated graph connectivity for cell surface proteins based on their functional annotation.