Multiparameter analysis of stimulated human peripheral blood mononuclear cells: a comparison of mass and fluorescence cytometry

Abstract

Mass and fluorescence cytometry are quantitative single cell flow cytometry approaches that are powerful tools for characterizing diverse tissues and cellular systems. Here mass cytometry was directly compared with fluorescence cytometry by studying phenotypes of healthy human peripheral blood mononuclear cells (PBMC) in the context of superantigen stimulation. One mass cytometry panel and five fluorescence cytometry panels were used to measure 20 well-established lymphocyte markers of memory and activation. Comparable frequencies of both common and rare cell subpopulations were observed with fluorescence and mass cytometry using biaxial gating. The unsupervised high-dimensional analysis tool viSNE was then used to analyze data sets generated from both mass and fluorescence cytometry. viSNE analysis effectively characterized PBMC using eight features per cell and identified similar frequencies of activated CD4+ T cells with both technologies. These results suggest combinations of unsupervised analysis programs and extended multiparameter cytometry will be indispensable tools for detecting perturbations in protein expression in both health and disease.

Introduction

Fluorescence cytometry has driven forward our understanding of cell biology in human immune monitoring and disease studies for decades by quantitatively characterizing single cells based on cell surface and intracellular features (1–5). Mass cytometry is a new quantitative single cell flow cytometry approach that employs antibodies conjugated to stable isotopes of metals and time of flight mass spectrometry as a detection technology (6, 7). Due to the precision of mass resolution, hundreds of features can theoretically be measured on each cell simultaneously using a mass cytometer. Recently mass cytometry has emerged as a powerful tool for high dimensional single cell analysis that has been used to characterize diverse populations of immune cells (8–15).

While several studies have highlighted the potential of mass cytometry for describing cellular subsets in great detail (8–14), only a few of these studies have directly compared mass cytometry with traditional fluorescence cytometry for evaluating human cell populations (10, 15–17). Furthermore, despite its promise, mass cytometry is still a relatively new technology, and extensive optimization of panel design, protocols, and analysis workflows will be required to acquire and appropriately analyze the vast amount of data generated (18–20). Here a direct comparison of mass cytometry and traditional fluorescence cytometry is described in detail for human subjects. A panel of 20 well-established surface markers of lymphocytes was used to assess whether mass cytometry provided equivalent per-marker and per-subset information on a one-to-one basis with traditional fluorescence cytometry. Unstimulated and stimulated human PBMC from six donors were analyzed with five established fluorescence cytometry panels in our laboratory and one newly optimized mass cytometry panel. The results of the two platforms were highly concordant, suggesting that mass and fluorescence cytometry will be complementary technologies used for characterizing the complex, dynamic cellular phenotypes that exemplify immune responses.

Materials and Methods

Cell isolation and culture

Peripheral blood mononuclear cells (PBMC) from healthy donors (N = 6) were isolated using density gradient separation (Ficoll-Paque™ Plus, GE Healthcare, Piscataway, NJ, USA). PBMC were pelleted by low speed centrifugation (400 × g), resuspended in media composed of 90% fetal bovine serum (Atlanta Biologicals, Norcross, GA, USA) containing 10% DMSO (Sigma-Aldrich, St. Louis, MO, USA), frozen slowly in the vapor phase of liquid nitrogen in multiple cryotubes, and stored in liquid nitrogen, as previously described (21). The Vanderbilt University's Institutional Review Board approved this study, and all individuals provided written informed consent.

Individual PBMC cryotubes were thawed in 2 mL of warm phosphate buffered saline (PBS, Gibco, Life Technologies, Grand Island, NY, USA), pelleted by centrifugation (650 × g), divided for immediate ex vivo phenotyping or phenotyping following 16 hours of in vitro SEB (EMD Millipore, Billerica, MA) stimulation, and then pelleted again before resuspension in room temperature PBS (ex vivo) or R10 media (in vitro) at 10 × 10⁶ cells/mL. R10 media contained RPMI 1640 Medium (Gibco), 2 mM L-glutamine (Gibco), 50 µg/mL penicillin (Gibco), 50 µg/mL streptomycin (Gibco), 10% FBS, and 10 mM HEPES buffer (Thermo Fisher Scientific, Waltham, MA, USA). Cells for ex vivo staining were further divided among flow cytometry tubes (Falcon 2052, BD-Biosciences, San Jose, CA) for fluorescence or mass cytometry staining, described below. Cells for in vitro culture were stimulated by addition of SEB to achieve a final concentration of 1 µg/mL in 200uL of 10 × 10⁶ cells/mL in 48-well flat bottom culture plates (Costar, Corning Incorporated, Corning, NY, USA). After 16 hours of incubation at 37°C in a 5% CO₂ incubator, cells were removed from the plate, washed twice in PBS, and stained as described below.

Fluorescence cytometry

For each healthy donor, 2 × 10⁶ PBMC were stained in 200µL PBS. PBMC were incubated first with a viability dye for 10 minutes (LIVE/DEAD Aqua, Life Technologies), washed once in PBS, and then stained with combinations of fluorescently-tagged antibodies (Table 1). For ex vivo phenotyping, cells were stained with Panels 1–5 from Table 1 (for antibody information see Table S1). For phenotyping following in vitro stimulation, cells were stained with Panels 3–5 from Table 1 at 16 hours after addition of SEB. After staining, all cells were washed twice in PBS and fixed with 2% paraformaldehyde (PFA, Electron Microscopy Services, Fort Washington, PA, USA) and refrigerated up to 24 hours until analysis on the Special Order Research Product (SORP) BD LSRFortessa (BD Biosciences, San Jose, CA) at the Vanderbilt Flow Cytometry Shared Resource.

Table 1.

Fluorescence cytometry instrument and antibody panel information

Instrument characteristics			Reagent panels
Laser emission and output power	Bandwidth transmitted to PMT	Fluorochrome	1	2	3	4	5
639nm (40mW)	750 – 810	APC-Cy7, APC-A750	HLA-DR	CD8	HLA-DR	HLA-DR	CD8
	663 – 677	APC		CD3	CD3	CD20	CD3
488nm (50mW)	505 – 550	FITC	CD57	CD45RO	CD25	CD27
561nm (150mW)	685 – 735	PE-Cy5.5		CD127
	600 – 620	PETR	CD14				CD45RO
	750 – 810	PE-Cy7			CD38	CD38
	655 – 685	PE-Cy5			CD62L		CD4
	575 – 590	PE	CD16	CD27	CD69	CD86	CD40L
404nm (100mW)	430 – 470	PB, BV421, V450	CD3	CCR7	CD8		PD-1
	685 – 735	BV711	CD19			CD19
	595 – 620	BV605		CD4	CD4
	505 – 550	Aqua	Live/dead	Live/dead	Live/dead	Live/dead	Live/dead

Five fluorescence cytometry panels were designed to measure PBMC populations with 21 parameters comparable to those measured in mass cytometry: 20 fluorophore-labeled antibodies and 1 viability marker. Panel 1: PBMC subsets; Panel 2: T cell memory subsets; Panel 3: T cell activation; Panel 4: B cell activation; Panel 5: T cell activation and exhaustion.

Mass Cytometry

For each healthy donor, 2 × 10⁶ PBMC were stained in 50 µL PBS. PBMC were incubated first with a viability reagent (50 µM cisplatin, Enzo Life Sciences, Farmingdale, NY, USA) in 1 mL serum-free RPMI for 3 minutes. Cisplatin was quenched by washing once with RPMI containing 10% FBS followed by two washes in PBS. A master mix containing 21 antibody-metal conjugates (Table 2, Table S1) was added to each sample (50 µL total staining volume) and incubated at room temperature for 25 minutes. Cells were then washed twice with PBS, fixed for 10 minutes with 1.6% PFA at room temperature, washed once with PBS, and then permeabilized at −20°C in 1 mL 100% cold methanol for 20 minutes. Following permeabilization, cells were washed at 800 × g, stained with 250 nM Iridium intercalator (22) (Fluidigm/DVS Sciences, Sunnyvale, CA) for 16 hours at 4°C, washed twice in PBS, washed once with ddH₂O, and then re-suspended in 500 µL ddH₂O for mass cytometry analysis that day. Cells were filtered immediately before injection into the mass cytometry using a 35µm nylon mesh cell-strainer cap (BD Biosciences).

Table 2.

Mass cytometry panel to identify PBMC populations

	Target	Metal	Mass	Cell type
1	CD19*	Neodymium (Nd)	142	B cells
2	CD40L*	Neodymium (Nd)	143	Activated T cells
3	CD4	Neodymium (Nd)	145	T helper cells
4	CD8*	Neodymium (Nd)	146	Cytotoxic T cells
5	CD20*	Samarium (Sm)	147	B cells
6	CD38	Neodymium (Nd)	148	Activated lymphocytes
7	CD62L	Europium (Eu)	153	Activated lymphocytes
8	CD86*	Gadolinium (Gd)	156	Activated lymphocytes
9	CCR7*	Terbium (Tb)	159	T cell memory subsets
10	CD14*	Gadolinium (Gd)	160	Monocytes
11	CD69	Dysprosium (Dy)	162	Activated lymphocytes
12	HLA-DR	Dysprosium (Dy)	163	APCs; activated T cells
13	CD45RO	Dysprosium (Dy)	164	T cell memory subsets
14	CD16*	Holmium (Ho)	165	NK cells
15	CD27*	Erbium (Er)	167	T cell memory subsets
16	CD25	Thulium (Tm)	169	Activated lymphocytes
17	CD3	Erbium (Er)	170	T cells
18	CD57*	Ytterbium (Yb)	172	T cell memory subsets
19	PD-1	Ytterbium (Yb)	174	Exhausted T cells
20	CD127*	Ytterbium (Yb)	176	T cell memory subsets
21	CD45	Samarium (Sm)	154	White blood cells
22	Nuc acid --Ir	Iridium (Ir)	191/193	DNA+ Cells
23	Cisplatin	Platinum	195	Viable cells

One panel was used to measure 23 parameters using mass cytometry: 21 metal-conjugated antibodies, 1 DNA marker, and 1 viability marker (cisplatin). Bolded antibodies were custom conjugated using MaxPar Metal Conjugation Kits. “Cell type” indicates the type of cell that was identified for correlative purposes with fluorescence cytometry data.

Asterisks (*) denote when different clones were used in mass and fluorescence cytometry.

Bolded antibodies were custom conjugated to metals using purified antibodies from BioLegend and metal-labeling kits from Fluidigm.

Markers #21–23 were not used in direct comparison to similar fluorescence parameters. Conjugates were chosen to minimize crosstalk.

Samples were analyzed using a CyTOF 1.0 (Fluidigm Sciences, Sunnyvale, CA) and CyTOF software version 5.1.615 (Fluidigm) at the Vanderbilt Flow Cytometry Shared Resource. Dual count calibration (on the “data”) and noise reduction (cell length 10–75, lower convolution threshold 10) were applied during acquisition.

Data processing and statistical analysis

All fluorescence and mass cytometry FCS files were uploaded and evaluated using Cytobank software and established methods (20, 23). Data were transformed to arcsinh scales with varying cofactors: mass cytometry cofactors ranged from 15 to 50 while fluorescence cytometry cofactors ranged from 150 to 3,000. Software compensation was applied to all fluorescence cytometry FCS files. For viSNE analysis in Figure 3, 150,000 total cells were analyzed and equal cell numbers were sampled from each FCS file. 8-parameter viSNE maps were created using the 8 antibodies listed in Panel 3 of Table 1. GraphPad Prism software (GraphPad, La Jolla, CA, USA) was used to determine Spearman’s rank correlation coefficient rho (ρ) between fluorescence and mass cytometry values (Table 3).

Figure 3. viSNE identified activated T cells in 8-dimensional analysis of fluorescence and mass cytometry.

(A–B) viSNE plots show unstimulated and SEB stimulated PBMC compared according to 8 proteins (Table 2, Panel 3) detected by fluorescence cytometry (A) or mass (B) cytometry. For each cell, color indicates the intensity of the labeled protein on a rainbow heat scale (arcsinh scales). Activated CD4⁺ T cells (CD3^lo/−CD4^medCD8⁻CD69^hiCD25^hi) are outlined in black. Maps from one representative donor are shown. (C) Maps displaying density of cells from the same donor by fluorescence and mass cytometry highlight the absence of activated cells in the unstimulated condition. (D) SEB induced T cell activation in all 6 individuals, with similar percentages measured by fluorescence and mass cytometry (Wilcoxen matched pairs t-tests p=n.s).

Table 3.

Correlations of fluorescent and mass cytometry analysis of percent positive cells for proteins measured on healthy human PBMC

			Spearman rank		Pearson		Frequency detected (fluorescent)				Frequency detected (mass)
Antibody	Gated off of	n	ρ (rho)	p	r	p	Range	Mean (ex vivo)	Mean (+ SEB)	Average MFI	Range	Mean (ex vivo)	Mean (+ SEB)	Average MMI
CD57*	Live CD16+	6	1.00	0.003	0.99	<0.0001	16.15–58.52	33.28	n/a	10178	17.83–51.91	31.18	n/a	290
CD27	Live CD3+	6	1.00	0.003	1.00	<0.0001	53.14–89.51	75.05	n/a	13995	52.27–85.02	71.25	n/a	54
CD45RO	Live CD3+	6	0.94	0.017	0.94	0.005	31.98–59.18	46.98	n/a	5811	30.71–64.83	48.64	n/a	861
CD3	Singlets	12	0.92	<0.0001	0.92	<0.0001	24.36–74.87	62.24	39.28	5759	30.36–71.58	62.36	36.81	262
CD62L	Live CD3+	12	0.90	0.0002	0.93	<0.0001	32.79–79.58	48.60	55.24	18435	28.93–79.03	41.18	53.16	62
HLA-DR	Live CD20+	12	0.88	0.0003	0.87	0.0002	31.22–98.89	72.74	97.59	20843	56.50–97.07	74.18	91.46	1225
CD69	Live CD4+CD3+	12	0.86	0.0006	0.95	<0.0001	0.15–38.46	1.51	29.84	2650	0.26–57.98	0.87	36.27	178
PD-1	Singlets	12	0.85	0.008	0.72	0.007	6.1–19.16	9.39	13.72	1915	11.65–19.55	16.73	14.69	19
CCR7*	Live CD3+	6	0.83	0.058	0.96	0.002	35.47–88.22	59.32	n/a	1946	34.41–80.48	59.12	n/a	72
CD38	Live CD8+CD3+	12	0.83	0.001	0.96	<0.0001	1.62–86.03	37.64	39.72	8987	3.62–57.63	24.89	22.67	37
CD40L*	Singlets	12	0.81	0.002	0.83	0.0008	0.01–33.58	0.52	18.47	1286	1.47–22.37	4.84	18.13	14
CD25	Live CD4+CD3+	12	0.80	0.003	0.92	<0.0001	5.21–30.94	8.58	22.72	1402	0.33–62.31	3.22	33.93	180
CD20*	Singlets	12	0.78	0.004	0.80	0.002	4.15–13.21	7.33	7.82	10756	2.54–11.8	6.91	7.29	145
CD16*	Singlets	6	0.77	0.100	0.67	0.14	4.31–16.92	11.22	n/a	10133	6.55–16.47	9.86	n/a	283
CD86*	Live CD19+	12	0.76	0.006	0.97	<0.0001	1.87–88.30	3.20	77.48	21763	9.42–80.56	13.06	68.29	54
CD19*	Singlets	12	0.73	0.009	0.78	0.0030	4.07–12.24	6.66	7.46	7264	3.42–13.8	6.15	9.29	307
CD4	Live CD3+	12	0.72	0.01	0.74	0.006	51.40–90.09	75.31	73.92	3685	47.41–78.39	63.10	57.25	255
CD8*	Live CD3+	12	0.72	0.01	0.79	0.002	9.17–21.49	15.35	13.55	4467	4.28–16.23	16.73	12.85	485
CD27*	Live CD20+	12	0.69	0.016	0.50	0.09	7.23–38.34	21.67	11.73	1607	5.51–27.04	18.17	11.44	26
CD127*	Live CD3+	6	0.60	0.240	0.75	0.08	79.16–93.92	86.05	n/a	2130	33.90–61.20	50.81	n/a	8
CD14*	Singlets	6	0.43	0.420	0.23	0.66	2.23–17.69	10.51	n/a	16727	3.42–16.47	11.04	n/a	122

Antibodies are listed in order of decreasing Spearman's rank correlation coefficient rho (ρ).

Asterisks (*) denote when different clones were used in mass and fluorescent cytometry.

Bolded antibodies were custom conjugated to metals using purified antibodies from BioLegend and metal-labeling kits from Fluidigm. The starting population of cells was used to determine the percent positive of cells for each protein.

N indicates the number of values used in the Spearman rank and Pearson analysis: N was 6 when only ex vivo data was used (antibodies solely used in Panel 1 or 2, Table 1); N was 12 when an ex vivo and in vitro value were used from each subject.

Results

Fluorescence and mass cytometry panels to track T cell identity

Panel design

Five fluorescence cytometry panels currently in use in our laboratory were used to measure 20 well-established cell surface markers chosen to provide a systematic view of T cell activation after SEB stimulation (Table 1). Fluorochrome and antibody conjugates were chosen based upon current availability in the laboratory and their compatibility with the BD LSRFortessa at the Vanderbilt Flow Cytometry Shared Resource.

A single mass cytometry panel was developed to measure the same set of 20 surface markers captured by the five fluorescence cytometry panels (Table 2). While mass cytometry avoids the severity of channel overlap that affects fluorescence cytometry, ‘crosstalk’ between channels exists. Crosstalk leads to false signals and must be taken into consideration when designing a panel for mass cytometry and gating cellular populations. The three sources of crosstalk result from variations in abundance sensitivity, isotope purity, and oxide formation (Fluidigm.com “Maxpar Panel Designer User Guide”). These types of crosstalk can contribute to signal in the M±1 and M+16 masses from the dominant signal at mass M. To minimize crosstalk within this panel, four of the 20 antibodies were custom conjugated to metals (Table 2, bolded).

Antibody titrations

Single antibody titrations were performed for all fluorochrome-conjugated antibodies (FCAs) and metal-conjugated antibodies (MCAs) as needed. As an example, Figure S1 shows the titration of CD4-Nd¹⁴⁵ and CD4-PETR (PE-Texas Red). CD4-Nd¹⁴⁵ was titrated from 0uL to 0.5uL (recommended amount) with DNA intercalator to identify single cells for analysis (Fig. S1A). The mean mass intensity (MMI) of the CD4-Nd¹⁴⁵+ population shifted from 6.63 to 264.48 while the MMI of the CD4-Nd¹⁴⁵-population stayed between −0.40 and −0.19 (Fig. S1A and S1C). The standard deviation of the CD4-Nd¹⁴⁵- population was always below 1 (Fig. S1A and S1C). With increasing antibody concentrations the frequencies of the CD4-Nd¹⁴⁵+ populations increased from 0.18% to 60.73% and stain index values increased from 11.78 to 149.16 (Fig. S1A and S1C).

A single antibody titration was also performed with CD4-PETR from 0uL to 2uL and FSC and SSA properties were used to identify single cells for analysis (Fig. S1B). The mean fluorescence intensity (MFI) of the CD4-PETR+ population increased from 715.13 to 57736.25 while the CD4-PETR- population shifted from 113.71 to 898.64 (Fig. S1B and S1C). The standard deviation of the CD4-PETR- population ranged between 100–1100 (Fig. S1B and S1C). The frequencies of CD4-PETR detected ranged from 0.56% to 55.42% and the stain index of CD4-PETR ranged from 2.92% up to 31.07% (Fig. S1B and S1C). The highest stain index was achieved at the 0.5uL concentration (Fig. S1B and S1C).

The remaining MCAs and FCAs were conjugated and titrated in a similar manner (Tables 1–2 and Table S1). For fluorescence cytometry, single antibody titrations were performed for all FCAs and final volumes were chosen based on stain index. For markers not typically expressed on resting T cells (e.g. CD69), antibody titrations were performed on cells stimulated in vitro with SEB. To optimally titrate antibodies in the mass cytometry panel, final concentrations were chosen based on the frequency of detected populations (considering their fluorescence counterparts), stain index, and crosstalk of each MCA into their M+1, M−1, and M+16 channels. For mass cytometry, custom conjugates were titrated in groups as they were created. PBMC were then stained with the full mass cytometry panel (Table 2) at recommended volumes and adjustments were made as needed for each antibody. Further titrations were done in groups that never included masses within 1 or 16 masses of each other. As needed for antibodies with non-bimodal distributions additional antibodies were included to determine optimal staining volumes. For example, final adjustments of PD-1 staining volumes were made after plotting PD-1 versus CD45RO on CD4+ T cells (Fig. S2).

Values derived from fluorescence and mass cytometry were closely correlated

Considerations for setting gates in mass cytometry

Several factors were taken into consideration when setting gates on mass cytometry data to ensure that only true signal was being reported. In the absence of background signal, the gate for a particular metal could theoretically be set at 10⁰. However, sources of background including nonspecific binding of antibodies and crosstalk from other channels require gates to be set at least at 10¹ since the abundance sensitivity for our instrument is 1% and the MMIs of the antibodies ranged up to 1000 (data not shown, Table 3). To help determine where a gate should be set, mass minus one (MMO) controls can be used to ensure that only signal from a single antibody is being detected within a gate. High-density antigens with bimodal staining patterns (MMIs between 10² and 10³) did not require MMOs since the mass intensity of the signal was significantly beyond 10¹. MMOs were especially important, however, when the MCA had a non-bimodal staining pattern, was a dim antigen, and was at the M+1, M−1, or M+16 position of a MCA with an intense, abundant signal.

Figure 1 illustrates how gates were set for such an antibody, CD25-Tm¹⁶⁹, using a mass minus one (MMO) control. The frequency of Tm¹⁶⁹+ events when PBMC were only stained with DNA intercalator and CD4-Nd¹⁴⁵ was 0.02% when the gate was set at 10¹ (dashed line gate) and 0% when the gate was set at 22 (solid line gate) which represents the actual gate used to identify CD25+ cells (Fig. 1A). When PBMC were stained with the full mass cytometry panel (Table 2) except CD25-Tm¹⁶⁹ –an MMO control—the frequency of Tm¹⁶⁹+ events when the gate was set at 10¹ was 2.63% (Fig. 1B, dashed line gate). This signal results from crosstalk into Tm¹⁶⁹ by the rest of the panel: most likely M−1 crosstalk from CD3-Er¹⁷⁰ and M+16 crosstalk from CD62L-Eu¹⁵³. When the gate was set at 22, however, the frequency of non-specific signal was reduced to 0.19% (Fig. 1B, solid line gate). This gate, which avoided analyzing artifact signals, was then used in the full panel to detect true CD25+ events (Fig. 1C).

Figure 1. MMOs guided gating for CD25 expression on live single CD4+ T cells.

(A) PBMC were stained only with DNA intercalator and CD4-Nd¹⁴⁵. Single cells were analyzed for CD25 expression. The gate with the dashed red line is set at 10, the same as in (B). The gate with the solid red line is set at 22, the same place as in (B–C) and Figure 2. (B) PBMC were stained with all the MCAs (Table 2) except CD25-Tm¹⁶⁹ (termed a mass minus one (MMO) control). Gates are set the same as in (A), and frequencies inside each gate represent non-specific signal in the Tm¹⁶⁹ channel. (C) PBMC were stained with all the MCAs (Table 2) with or without SEB stimulation (the same plots from Fig. 2 are shown).

Staining patterns and gates for all antibodies are shown in Figure S3. Final decisions for gating took into consideration many factors including crosstalk, fluorescence/mass counterpart staining, published staining pattern data and frequencies, and staining patterns on multiple cell populations (e.g. evaluating markers present on particular cell types and absent on others). The staining pattern for CD27 illustrates how gates were set for antibodies that were not at the M±1 or M+16 position of another antibody (Fig. S3B). Alternatively, the staining pattern for CD45RO-Dy¹⁶⁴ illustrates gating when considering M+16 crosstalk from CD38-Nd¹⁴⁸ and M±1 crosstalk from HLA-DR-Dy¹⁶³ and CD16-Ho¹⁶⁵ (Fig. S3C). An MMO was not required here, however, since the antigen was highly-expressed and displayed a bimodal staining pattern. This example further demonstrates the need for careful panel design.

Identification of live single cells

To directly compare mass and fluorescence cytometry, an equivalent starting population of live single cells was identified in healthy human PBMC (Fig. 2). In fluorescence cytometry (Fig. 2A), forward and side light scatter signal properties were used to identify intact single cells and exclusion of the LIVE/DEAD Aqua dye identified live single cells. In mass cytometry (Fig. 2B), event length and intercalator uptake were used to identify intact single cells and exclusion of cisplatin (24) identified live single cells. Although PBMC were stained with CD45-Sm¹⁵⁴ it was not included in analysis since we did not have a comparable marker fluorescently and correlations improved between the two technologies when identical phenotypic gating strategies were used. After single cells were identified, live CD3+ cells and CD4+ and CD8+ T cells were gated by mass and fluorescence cytometry (Fig. 2A–B). Both mass and fluorescence cytometry measured comparable increases in CD25 expression following SEB stimulation (Fig. 2C–D).

Figure 2. Gating schemes for fluorescence and mass cytometry.

Plots show PBMC from a single healthy human donor. (A) Representative biaxial plots show the gating scheme for fluorescence flow cytometry. Intact single cells were gated using forward and side scatter area and height properties. Single cells were then assessed for viability and expression of CD3. This population was further gated as CD8⁺ and CD4⁺ T cells. CD25 expression on CD4⁺ T cells was compared in PBMC from an individual healthy donor with or without SEB stimulation. (B) Representative biaxial plots show the gating scheme for mass cytometry. Single cells were identified using event length and intercalator uptake and then gated and compared as for fluorescence cytometry.

Correlation between fluorescence and mass cytometry data

Statistical correlation between fluorescence and mass cytometry was determined using Spearman’s rank (Table 3) for all 20 measured proteins. Samples analyzed in parallel by mass and fluorescence cytometry included 12 populations of PBMC from 6 individual healthy donors under 2 conditions (unstimulated ex vivo and 16 hours after in vitro SEB stimulation). Frequencies of cellular populations identified by 20 MCAs and FCAs were directly compared using biaxial gating plots (as in Fig. 2). The frequency of each antibody was gated from the same starting population, which is indicated in the “gated on” column (Table 3). The range of frequencies detected, mean frequencies of unstimulated and stimulated populations, and average intensity of each marker by fluorescence and mass cytometry is indicated (Table 3).

Statistically significant correlations were observed for all 9 proteins detected using the same antibody clones (custom or commercially conjugated). Eleven of the metal conjugated antibodies in the mass cytometry panel did not match the FCAs in our existing panels, (Table 3, asterisks) but 9 out of 11 of these antibodies still identified similar frequencies of populations by both technologies.

CD14 and CD127 were the only two antibodies that did not provide consistent values between the two technologies. We compared the two CD14 clones (Tük4-PETR and M5E2-FITC) using fluorescence cytometry and found that despite showing bimodal staining patterns they did not detect similar frequencies of CD14+ singlets (N=6, p=0.36, r=0.49). When we compared the two CD127 clones using fluorescence cytometry (R34.34-PECy5.5 and A019D5-PECy7) we found they did in fact detect similar frequencies of CD127+ T cells. Staining improved dramatically with a new lot of CD127-Yb¹⁷⁶: the MMI increased from 8 to 65 (Fig. S3V). We then evaluated the frequency of CD127+ T cells in six new subjects and demonstrated a high concordance between the two technologies (N=6, p=0.002, r=0.96). These additional experiments suggest the two discrepancies we found in Table 3 were not due to differences in mass and fluorescence cytometry, but far more likely to be due to differences in the particular antibody clone, or the particular lot of antibody-conjugate used.

viSNE identified similar frequencies of activated CD4+ T cell populations using data from each technology

Large data sets resulting from mass cytometry have resulted in the development of new computational approaches to analyzing complex single cell data. Having found a high degree of correlation between the two technologies, we next compared the ability of an unsupervised high-dimensional analysis program, viSNE, to analyze data sets derived from both mass and fluorescence cytometry. viSNE was developed for mass cytometry data and approximates high dimensional relationships using a two dimensional scatter plot, or map, where each dot represents a single cell (25). To read a viSNE map, one can visually identify a cluster or ‘island’ of cells and then determine the cellular identity based on marker expression. To determine whether viSNE could identify similar populations using mass and fluorescence cytometry data sets we chose to study frequencies of activated CD4+ T cells. The viSNE map in Figure 3 was generated with an equal number of cells from the FCS files of each healthy donor before and after stimulation. The viSNE analysis was restricted to eight markers of T cell activation and analyzed FCS files stained either with Panel 3 (Table 1) or the identical eight markers in the mass cytometry panel (Table 2). While cells identified by mass cytometry were tagged with greater than eight antibodies, only the same eight antibodies from the fluorescence T cell activation panel were considered when creating the viSNE maps with mass cytometry data (Fig. 3B).

When applied to fluorescence and mass cytometry data, viSNE created similar maps that demonstrated patterns of T cell activation. One fluorescence cytometry viSNE map (Fig. 3A) and one mass cytometry viSNE map (Fig. 3B) is each shown ten times highlighting the intensity of CD3, CD4, CD8, CD69 and CD25 on SEB stimulated (top rows) or unstimulated (bottom rows) PBMC. A gate was drawn (black outline) to highlight the population of activated CD4+ T cells. Cells within this island displayed characteristics of activated CD4⁺ T cells that included little or no CD3, no CD8, moderate CD4, and high expression of CD69 and CD25 (Fig. 3A–B). The activated cells are not present in the unstimulated condition (Fig. 3A–B) and this is further highlighted both by the frequency and density of cells within the gate on the viSNE maps (Fig. 3C). viSNE identified similar frequencies of activated CD4+ T cells analyzed by mass and fluorescence cytometry (p=0.007, ρ=0.74) and all subjects had cells that fell within this gate after stimulation (Fig. 3D).

Discussion

The ability to combine high-dimensional single cell biology with unsupervised analysis approaches is powering a new era of systems immunology. Here, a high-dimensional mass cytometry panel was developed evaluate T cell memory and activation. The frequencies of markers detected with 20 antibodies within the mass cytometry panel were compared on a one-to-one basis with antibodies from five fluorescence cytometry panels. The resulting data indicate that mass and fluorescence cytometry data are highly comparable. We also show that unsupervised viSNE analysis provides valuable insight into single cell data, regardless of the instrumentation used to collect that data. The 20 antibody T cell panel developed and validated in this study is expected to be particularly useful for detailed characterization of human T cell populations in a variety of settings such as longitudinal immune monitoring of viral infections, immune disorders, and cancer.

Mass cytometry has the potential to greatly expand the number of observable features on small populations of cells (26). Recent studies achieved 38- and 44-parameter single cell analysis using mass cytometry (8, 9). Alternatively, the number of measurable parameters using polychromatic fluorescence cytometry has increased to 20 and is growing still with the advent of new instruments and fluorochromes (27). We previously used multiparameter fluorescence flow cytometry with many of the same markers used in this study to evaluate the activation status of subpopulations of virus-specific T cells in memory compartments of peripheral blood (28–31) and cerebrospinal fluid (32, 33). Here, we used previously established panels in our laboratory that focused on T cell memory and markers of immune activation to provide a detailed comparison of mass and fluorescence cytometry.

A high dimensional mass cytometry approach provided equivalent per-marker and per-subset information when compared directly with traditional fluorescence cytometry (Fig. 2, Table 3). For example, the average difference in CD3+ cells detected by both technologies was 4.8% even though healthy variation before and after stimulation spanned a range of 25–75%. The findings in this study align well with published comparisons of mass and fluorescence cytometry (10, 16, 17). Other than expected differences due to using different antibody clones, the minor discrepancies observed between the two technologies (Table 3) likely resulted from differences in gating for ‘live single cells’. Prior to viability gating, fluorescence cytometry employed forward and side scatter while mass cytometry employed DNA content and cell length (Fig. 2). Overall, these results provide further support for the concordance between the two technologies.

We demonstrate that the high-dimensional visualization tool, viSNE, was still effective even in a ‘low-dimensional’ 8-parameter analysis. Amir and colleagues demonstrated previously that viSNE successfully identifies blood cell populations even when using non-canonical markers (25). This ability to detect obscure or unexpected cells is one of the most powerful attributes of new unsupervised analysis programs (34). In this study, viSNE identified activated CD4+ T cells based on their multidimensional phenotypes without requiring cells to express CD3. Additionally, viSNE returned comparable results with mass and fluorescence cytometry data considering 8 parameters, further strengthening the correlations of antibody detection between the two platforms. Going forward, familiarity with these tools, and learning their strengths and weaknesses, is likely to become a core skill for immunologists, especially since they apply well to any type ‘event list’ format data, such as single cell cytometry data from flow and imaging instruments.

One disadvantage to mass cytometry is that the samples must be destroyed for analysis, so this technology is not suitable for cell sorting. Hence, these should be viewed as complementary technologies. After identification of cell populations of interest with an extensive mass cytometry panel and high dimensional analysis with unsupervised algorithms, more focused fluorescence cytometry panels can be designed to sort cells for further analysis (19, 29, 30, 35). For this approach to work, however, it will be necessary to have matched panels of antibodies that can reliably detect the same markers with each technology, and we demonstrate that this is feasible.

This rigorous comparison of mass and fluorescence cytometry suggests that the technologies are highly comparable. Traditional biaxial gating and an unsupervised high-dimensional analysis approach, viSNE, identified similar patterns of protein expression and frequencies of cellular populations in superantigen stimulated human blood. These results demonstrate that multidimensional analysis using either platform will be particularly useful for the comprehensive characterization of cells, including cells with dynamic or unexpected cell phenotypes in health and disease.