Single-cell immune profiling of PBMCs by CyTOF

Frozen PBMCs and SFMCs were thawed and stained with a T cell focused panel of 37 heavy metal-conjugated antibodies (Supplementary file 2), as previously described (Chew et al., 2019 (link)), and analyzed by CyTOF-Helios (Fluidigm, San Francisco, CA). Briefly, PBMCs were stimulated with or without phorbol 12-myristate 13-acetate (150 ng/ml, Sigma-Aldrich) and ionomycin (750 ng/ml, Sigma-Aldrich) for 4 h, and blocked with secretory inhibitors, brefeldin A (1:1000, eBioscience) and monensin (1:1000, BioLegend) for the last 2 h. The cells were then washed and stained with cell viability dye cisplatin (200 μM, Sigma-Aldrich). Each individual sample was barcoded with a unique combination of anti-CD45 conjugated with either heavy metal 89, 115, 141, or 167, as previously described (Lai et al., 2015 (link)). Barcoded cells were washed and stained with the surface antibody cocktail for 30 min on ice, and subsequently washed and re-suspended in fixation/permeabilization buffer (permeabilization buffer, eBioscience) for 45 min on ice. Permeabilized cells were subsequently stained with an intra-cellular antibody cocktail for 45 min on ice, followed by staining with a DNA intercalator Ir-191/193 (1:2000 in 1.6% w/v paraformaldehyde, Fluidigm) overnight at 4°C or for 20 min on ice. Finally, the cells were washed and re-suspended with EQ Four Element Calibration beads (1:10, Fluidigm) at a concentration of 1×10⁶ cells/ml. The cell mixture was then loaded and acquired on a Helios mass cytometer (Fluidigm) calibrated with CyTOF Tunning solution (Fluidigm). The output FCS files were randomized and normalized with the EQ Four Element Calibration beads (Fluidigm) against the entire run, according to the manufacturer’s recommendations.
Normalized CyTOF output FCS files were de-barcoded manually into individual samples in FlowJo (v.10.2), and downsampled to equal cell events (5000 cells) for each sample. Batch run effects were assessed using an internal biological control (PBMC aliquots from the same healthy donor for every run). Normalized cells were then clustered with MarVis (Kaever et al., 2009 (link)), using Barnes Hut Stochastic Neighbor Embedding (SNE) nonlinear dimensionality reduction algorithm and k-means clustering algorithm, as previously described (Chew et al., 2019 (link)). The default clustering parameters were set at perplexity of 30, and p<1e−21. The cells were then mapped on a two-dimensional t-distributed SNE scale based on the similarity score of their respective combination of markers, and categorized into nodes (k-means). To ensure that the significant nodes obtained from clustering were relevant, we performed back-gating of the clustered CSV files and supervised gating of the original FCS files with FlowJo as validation. Visualizations (density maps, node frequency fingerprint, node phenotype, and radar plots) were performed through R scripts and/or Flow Jo (v.10.2). Correlation matrix and node heatmaps were generated using MarVis (Kaever et al., 2009 (link)) and PRISM (v.7.0).