Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2021 Feb 1.
Published in final edited form as: J Immunol Methods. 2019 Nov 11;477:112667. doi: 10.1016/j.jim.2019.112667

High Throughput pSTAT Signaling Profiling by Fluorescent Cell Barcoding and Computational Analysis

Wanxia Li Tsai 1,*, Laura Vian 1,*, Valentina Giudice 2,3,*, Jacqueline Kieltyka 1, Christine Liu 1, Victoria Fonseca 1, Nathalia Gazaniga 1, Shouguo Gao 2, Sachiko Kajigaya 2, Neal S Young 2, Angélique Biancotto 4,5, Massimo Gadina 1
PMCID: PMC6981073  NIHMSID: NIHMS1545413  PMID: 31726053

Abstract

Fluorescent cell barcoding (FCB) is a multiplexing technique for high-throughput flow cytometry (FCM). Although powerful in minimizing staining variability, it remains a subjective FCM technique because of inter-operator variability and differences in data analysis. FCB was implemented by combining two-dye barcoding (DyLight 350 plus Pacific Orange) with five-color surface marker antibody and intracellular staining for phosphoprotein signaling analysis. We proposed a robust method to measure intra- and inter-assay variability of FCB in T/B cells and monocytes by combining range and ratio of variability to standard statistical analyses. Data analysis was carried out by conventional and semi-automated workflows and built with R software. Results obtained from both analyses were compared to assess feasibility and reproducibility of FCB data analysis by machine-learning methods. Our results showed efficient FCB using DyLight 350 and Pacific Orange at concentrations of 0, 15 or 30, and 250 μg/mL, and a high reproducibility of FCB in combination with surface marker and intracellular antibodies. Inter-operator variability was minimized by adding an internal control bridged across matrices used as rejection criterion if significant differences were present between runs. Computational workflows showed comparable results to conventional gating strategies. FCB can be used to study phosphoprotein signaling in T/B cells and monocytes with high reproducibility across operators, and the addition of bridge internal controls can further minimize inter-operator variability. This FCB protocol, which has high throughput analysis and low intra- and inter-assay variability, can be a powerful tool for clinical trial studies. Moreover, FCB data can be reliably analyzed using computational software.

Keywords: fluorescent cell barcoding, variability, phosphoproteins, computational analysis, phenotyping

1. Introduction

Fluorescent cell barcoding (FCB), a high-throughput flow cytometry (FCM) technique for multiplexed assays, was first developed for single-cell phospho-specific FCM (phosphoFlow) [1–6], and subsequently optimized for detection of intracellular cytokines [7–8] and immunophenotyping [1–10]. FCB minimizes staining variability and decreases both antibody consumption and required sample volumes [1,11]. The advantage in performing large-scale studies by the FCB technique is to generate more reliable FCM data by minimizing technical variability, as all samples and controls are processed and acquired collectively [1]. For these reasons, FCB is particularly applicable to translational studies, and new barcoding methods are being developed using amine-reactive or fixable viability dyes [9–10,12].

The analysis of cell-to-cell variation in gene and protein expression obtained from high content single cell techniques, such as mass cytometry and FCB, is challenging in terms of time, reliability, and reproducibility [13–15]. Manual analysis is subjective and difficult to reproduce due to inter-operator variability, subjectivity in choosing the sequence of marker combinations to explore, and the shape and boundaries of gates [16–17]. By contrast computational FCM techniques are more objective, allowing users to reproducibly identify cell populations and measure protein expression levels [18]. Several software tools have been developed for computational data preprocessing, transformation, analysis, and visualization of FCM data including standalone programs or platforms such as Cytobank and packages for Matlab and RStudio [19–27].

Standardization of protocols and antibody panels in FCM for phenotyping is important for more reliable, comparable, and reproducible data, and, especially critical for samples collected during clinical trials [28–31]. The use of FCB minimizes inter-assay variations, as all samples are acquired together, but a rigorous standardization and optimization of the protocol is still required before implementation of the technique both in clinical and basic research settings [1–10]. However, analysis of FCB data is usually conducted by conventional manual gating strategies using available commercial software products, which are designed to analyze individual samples by a gating sequence limited to one or two parameters at a time.

Here, we provide a method to assess intra- and inter-assay variability of FCB for immunophenotyping and phosphoFlow analysis, and for quantification and minimization of inter-operator variability. For FCB data analysis, we also compare percent of positive cells or median fluorescence intensity values obtained by conventional manual gating strategies to those calculated with machine learning methods in order to further minimize operator- and software-dependent variations, as well as significantly decrease analysis time.

2. Materials and methods

2.1. Reagents

The following FCB dyes were used: DyLight 350 NHS ester and Pacific Orange NHS ester (Thermo Fisher Scientific, Waltham, MA, USA). Antibodies used for surface staining were: mouse anti-human CD3-PerCP-Cy5.5 (clone SK7), mouse anti-human CD4-PE-Cy7 (clone SK3), mouse anti-human CD8-FITC (clone RPA-T8), and mouse anti-human CD20-APC-H7 (clone H1) (BD Biosciences, San Jose, CA, USA); and CD14-PE (clone M5E2) from BioLegend (San Diego, CA). Antibodies used for phosphoproteins were: pSTAT1(pY701)-Alexa Fluor 647 (clone 4a), pSTAT3(pY705)- Alexa Fluor 647 (clone 4/P-STAT3), and pSTAT5(pY694)- Alexa Fluor 647 (clone 47/Stat5 pY694) (BD Biosciences). Phosflow Lyse/Fix Buffer 5X, Phosflow Perm Buffer III, and Phosflow Barcoding Wash Buffer 4X buffers (BD Biosciences) were prepared and used according to manufacturer’s instructions. Phosflow Perm Buffer II (BD Biosciences) was diluted 1:1 with cold PBS and kept on ice before use. For PBMC stimulation, the following cytokines were used: recombinant human IL-10 (PeproTech, Rocky Hill, NJ); human Interferon-α (Cell Signaling Technology, Boston, MA); recombinant human IL-2 (Hoffmann-La Roche Inc, Nutley, NJ). The CTL Anti-Aggregate (CTL) wash buffer was from Cellular Technology Limited (Shaker Heights, OH).

2.2. PBMC isolation

Heparinized fresh whole blood was obtained from healthy donors of the NIH Clinical Center Department of Transfusion Medicine (n= 32) after informed consent was obtained in accordance with the Declaration of Helsinki [32] (NIH, Bethesda, MD, USA). Peripheral blood mononuclear cells (PBMCs) were isolated by Ficoll-Paque gradient centrifugation (GE Healthcare, Chicago, IL) with Leucosep centrifuge tubes (VWR, Radnor, PA), according to manufacturer’s instructions. Cells were frozen in medium containing 90% FCS and 10% dymethyl-sulfoxide (DMSO, Sigma-Aldrich, St. Louis, MO, USA) and stored in liquid nitrogen until use.

2.3. pSTAT stimulation

PBMCs were stimulated with cytokines for phosphorylated Signal Transducer and Activator of Transcription (pSTAT) activity [33–35]. Cryopreserved human PBMCs were thawed in a water bath at 37°C, washed twice with 10 mL of CTL buffer, according to manufacturer’s instructions, and counted with AO/PI (acridine orange/propidium iodide) staining solution. Viability and number of PBMCs were determined using a cellometer (Nexcelom Bioscience, Lawrence, MA). A total of 1.5 × 106 PBMCs were seeded at 15 × 106 /mL in a deep 96 well round-bottom plate. Cells were stimulated with human Interferon-α (Cell Signaling Technology) at 200 ng/mL for pSTAT1 stimulation, recombinant human IL-2 (Hoffmann-La Roche Inc) at 200 U/mL for pSTAT5, or recombinant IL-10 (PeproTech) 200 ng/mL for pSTAT3 and incubated for 20 min at 37°C. Untreated PBMCs were processed following the same protocol and used as negative control.

2.4. FCB and phosphoflow staining

Each dye was dissolved in DMSO at a final concentration of 500 μg/mL and stored at −80°C. Using the stock solution, FCB dyes were diluted with DMSO to have the following concentrations: 0, 30, and 250 μg/mL for DyLight 350 and 0, 15, and 250 μg/mL for Pacific Orange. A final volume of 40 μL/well was used for barcoding by combining 30 μL of sample and 5 μL of each dye or DMSO in order to have the final concentrations of 0, 6.25 (for DyLight 350) or 3.125 (for Pacific Orange), and 31.25 μg/mL. At the end of stimulation, untreated and treated PBMCs were fixed with 550 μL of 1X Lyse/Fix Buffer for 10 min at 37°C, centrifuged at 500g for 5 min, and then permeabilized with 600 μL of Phosflow Perm Buffer III on ice for 30 min. Subsequently, cells were washed twice with 600 μL of Phosflow Perm Buffer II (1:1 with PBS), resuspended in 35 μL of cold Phosflow Perm Buffer II and 30 μL were added to appropriate wells (1.29 × 106 cells/ 30 μL/well) with various concentrations of FCB dyes previously prepared in a U-bottom 96 well plate. After incubation on ice for 20 min in the dark, samples within each matrix were combined and washed twice with BD Phosflow Barcoding Wash Buffer (3 mL) by centrifugation at 500g for 5 min, followed by resuspension with BD Phosflow Barcoding Wash Buffer (50 μL) for antibody staining. After antibody titration (Supplementary Figure 1), cells were stained with 20 μL of CD14-PE (20 μg/mL), 40 μL of CD8-FITC (1.25 μg/mL), 40 μL of CD3-PerCP-Cy5.5 (0.6 μg/mL), 40 μL of CD20-APC-H7 (2.4 μg/mL), 10 μL of CD4-PE-Cy7 (0.3 μg/mL), and 40 μL of appropriate pSTAT-Alexa Fluor 647 (0.3 μg/mL for pSTAT1 and pSTAT3; 0.6 μg/mL for pSTAT5). Cells were incubated for 1 hr at room temperature in the dark with gentle shaking, washed with BD Phosflow Barcoding Wash Buffer (3 mL), and resuspended in 300 μL of the same buffer for acquisition. As non-barcoded controls, 1.5 × 106 cells were directly stained without fixation, permeabilization and barcoding. For each treated and untreated donor, 1.5 × 106 fixed and permeabilized cells were stained without barcoding. After staining, all samples were washed and acquired using BD Phosflow Barcoding Wash Buffer.

2.5. Data acquisition and analysis

LSR Fortessa cytometer (BD Biosciences) equipped with 6 lasers (ultraviolet, 355 nm; violet, 405 nm; blue, 488 nm; green, 552 nm; orange, 592 nm; and red laser, 628 nm) and BD FACSDiva software (v.8.0.1, BD Biosciences) were used for sample acquisition. Compensation was calculated using bead standards for each fluorochrome (anti-Mouse Ig, κ/Negative Control Compensation Particles Set, BD Biosciences) and barcoded cells with the highest concentration of each dye. An unstained sample was used as negative control for setting PMT voltages, and all samples were run using the same PMT voltages. A minimum of 300,000 lymphocytes were recorded. FlowJo software (v.10.0.7b, Treestar, Ashland, OR, USA) was employed for post-acquisition compensation and conventional flow cytometric analysis. Compensation matrix was exported from FlowJo as a CSV file (Supporting Information S1). FCM data are shown using biexponential transformation.

2.6. Conventional gating strategy

After post-acquisition compensation using FlowJo, lymphocytes and monocytes were identified using linear parameters (FSC-A vs SSC-A) and double cells were excluded (FSC-A vs FSC-H) (Supplementary Figure 1). Single cells were deconvoluted by plotting the FCB dye channels (Pacific Orange vs DyLight 350), identifying nine populations. On each barcoded lymphocyte sample, CD3 and CD20 expression were first investigated, and CD4 and CD8 expression was further studied on CD3+ cells. CD14+ cells were identified (CD14 vs FSC-A) on barcoded monocyte populations. pSTAT1, pSTAT3, or pSTAT5 expression was calculated on CD3+, CD4+, CD8+, CD20+, and CD14+ cells as median fluorescence intensity (MFI) values.

2.7. FCS file datasets and R packages

Computational workflows were built using datasets available on FlowRepository (FR-FCM-Z2Z2). The dataset included six FCS files from the current study: three unstimulated and three stimulated samples for each pSTAT. For computational analysis, RStudio (v. 1.1.456) and R software (v.3.5.1) were employed (RStudio, Inc., Boston, MA, USA) (28) and BiocInstaller (v. 1.32.1, Bioconductor) was used to update Bioconductor, CRAN, and GitHub packages (28–29). The following packages were used: flowType (v.2.20.0), ggcyto (v.1.10.0), ggplot2 (v.3.1.0), flowWorkspace (v.3.30.1), flowViz (v.1.46.0), flowClust (v.3.20.0), and flowCore (v.1.48.0). R codes were deposited in GitHub (https://github.com/ValeGiu076/FCB_computational_analysis.git).

2.8. Computational gating strategy

R packages were loaded into R session and FCS files read using the read.FCS function from the flowCore package. We began by removing debris using rectangular gates (rectangleGate function in flowCore) and subset function, and then cell clusters were identified using t-mixture modeling (flowclust function in flowClust R package; Supplementary Figure 2). Lymphocyte and monocyte clusters were identified based on SSC.A values, since monocytes are bigger and more complex compared to lymphocytes. In details, clusters were first split in two flowFrame objects, ordered by SSC.A values and then identified as monocyte cluster if SSC.A values were higher than those present in the other cluster. After biexponential transformation, unsupervised deconvolution was performed using FCB dye channels (Pacific.Orange.A and BUV395.A) and set the number of clusters (K) at 9, as implemented in the flowclust function of flowClust package. Barcoded populations were split in separate flowFrame objects and included in a list. On each barcoded lymphocyte population, unsupervised clustering was carried out to identify CD3+ and CD20+ cells using flowclust function setting the fluorescence channel parameters (APC.H7.A and PerCP.Cy5.5.A) and the number of expected clusters (K=3) for modeling. The three clusters were then ordered by APC.H7.A and PerCP.Cy5.5.A values, and the population with the highest maximum value of PerCP-Cy5.5.A was assigned as CD3+ cells, and CD4+ and CD8+ cells were further gated using the same strategy. While, the cluster with the highest value of APC.H7.A was assigned as CD20+ cells. Similarly, after deconvoluting monocytes, we identified CD14+ cells using linear parameters and CD14 expression by defining rectangular gates at the 1st quartile of the PE.A parameter.

2.9. Statistical analysis

Data were analyzed using Prism (v.8.0.1; GraphPad software, Inc., La Jolla, CA). Fluorescence values were reported as median fluorescence intensity (MFI), CV (CV = SD/mean of population), and MFI fold change. For each antibody tested, percentage of positive cells and MFI values were recorded for every barcoded population and matched non-barcoded controls. Variability between barcoded samples and matched controls, and sample-to-sample variability were defined and reported as previously described [9]. Percentages of positive cells or MFIs were correlated between barcoded and matched non-barcoded samples, or between conventional and computational methods by Pearson test and linear regression. Position variability was assessed using 1-way ANOVA with multiple comparisons corrected for FDR (False Discovery Rate). For inter-assay and inter-operator variability assessment, 1-way ANOVA and Pearson analysis were employed. A P-value < 0.05 was considered statistically significant.

3. Results

3.1. FCB efficiency

In this study, FCB was optimized using Pacific Orange and DyLight 350 dyes at concentrations of 0, 15 or 30, and 250 μg/mL to barcode nine human PBMC samples, and FCB efficiency was assessed by MFI fold increase calculation [9]. Lymphocytes were identified based on linear parameters (FSC-A vs SSC-A) and double cells were excluded (FSC-A vs FSC-H) (Supplementary Figure 3A). On single cells, FCB dyes were visualized separately using normalized cell number count histograms, and three peaks were shown according to FCB dye concentrations (Supplementary Figure 3B). For each peak of 12 separate matrices, MFIs and CVs were calculated in FlowJo software. Values were used to assess MFI fold increase as follows: MFI fold increase = [MFIpeak2 – CVpeak2]/[MFIpeak1 + CVpeak1][9]. MFI fold increase ranged from 4.1 to 11.3 achieving a complete peak separation without any overlap. CVs were higher in 0 μg/mL barcoded populations (21% and 58% for Pacific Orange and DyLight 350, respectively) compared to 15 or 30 and 250 μg/mL barcoded samples (15±1%) (Supplementary Figure 3C).

3.2. Surface marker intra-assay variability

After optimization of dye concentrations, intra-assay variability was assessed for combination staining with FCB dyes and surface and intracellular markers. Lymphocytes and monocytes were characterized using surface markers (CD3, CD4, CD8, CD20, and CD14), and phosphorylation levels of STAT1, STAT3, and STAT5 were measured in each T/B cell and monocyte population. One donor was used for barcoding all nine populations in a 3×3 matrix (p1 to p9; Figure 1A and Supplementary Figure 4) for each pSTAT (total of six matrixes: three untreated and three stimulated for each pSTAT). This experiment was repeated with three healthy donors. For each condition, a fresh sample without fixation, permeabilization and barcoding was stained only with surface marker antibodies and used as a control.

Figure 1. Surface marker intra-assay variability assessment for percentage of positive cells.

Figure 1.

Open in a new tab

One donor was employed for barcoding all nine populations in each of six matrixes (three with unstimulated samples and three stimulated for pSTAT1, pSTAT3, or pSTAT5). (A) Mean of percent of positive cells and SD were calculated for each position in a matrix (pi; p1 to p9) among all matrixes with unstimulated samples (total of nine values for each parameter; N) or stimulated samples. The range of variability was calculated as average of % of positive cells of all barcoded samples ± 2SD. The ratio of variability was defined as the ratio between the average of barcoded samples in every position in a matrix and the average of not-barcoded samples; values were considered as acceptable if between 0.8 and 1.2. (B) Percent of positive barcoded cells were compared to those from matched not-barcoded (w/o FCB) samples by Pearson analysis. (C) Ranges and ratios of variability were calculated for each population, and values outside the ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in black.

In order to exclude that methanol permeabilization could affect reliability and reproducibility of FCB data, we first compared percentage of positive barcoded samples to matched non-barcoded controls by Pearson correlation. For each donor, the average percentage of positive cells for all stimulated or unstimulated barcoded samples was compared to matched non-barcoded stimulated or unstimulated controls. The two staining methods were highly correlated (r = 0.993; P < 0.0001) (Figure 1B), confirming that methanol did not affect the staining process by increasing non-specific binding.

Range and ratio of positive stained cells were calculated to determine variations within a matrix compared to the arithmetic mean of barcoded samples or to non-barcoded controls, respectively. The range of variability was calculated as average of percent of positive cells of all barcoded samples ± 2SD. Average of percent of positive cells and SD were calculated for each position in a matrix (p1 to p9) among all matrices with unstimulated (total of nine values for each parameter) or stimulated samples. The ratio of variability was defined as the ratio between the average of barcoded samples in every position in a matrix and the average of non-barcoded samples (Figure 1A). Ratios within 0.80 and 1.20 were acceptable because a coefficient of variation <20% is related to physiological and technical variations [36–40]. A minor non-statistically significant variability was described for CD3+, CD8+ and CD20+ populations, and no ratios were <0.80 or >1.20. No intra-assay variability was documented for CD4+ and CD14+ cells (Figure 1C and Supplementary Figure 5). We also compared values for each position to those in position 1 (p1, barcoded only with DMSO) to determine if variability in percent of positive cells was due to specific dye concentration. No variations were described (all P > 0.1).

We further assessed the effects of FCB dyes on fluorescence intensity and intra-assay variability of fluorochromes used. MFIs from barcoded samples were compared to those from matched non-barcoded samples in order to confirm that FCB did not significantly affect fluorescence intensity of the fluorochromes used for staining. As expected, the data from the two techniques were highly correlated (r = 0.864; P < 0.0001) (Figure 2B). Subsequently, ranges and ratios of variability were calculated for MFIs of PE-, PE-Cy7-, FITC-, APC-H7-, and PerCP-Cy5.5-conjugated antibodies as described above, and populations barcoded with 0 μg/mL of dyes were used as controls (Figure 2A and Supplementary Figure 6). No MFIs were outside the ranges (Figure 2C). However, when ratios were defined, a statistically significant variability was described for FITC- and PE-conjugated antibodies. Position variability was also assessed by 1-way ANOVA corrected for multiple comparisons, and significant variations were confirmed in some positions for MFIs of FITC- and PE-conjugated antibodies. However, our FCB protocol produced a reproducible combination staining with Pacific Orange plus DyLight 350 with surface markers for T/B cell and monocyte identification.

Figure 2. Surface marker intra-assay variability assessment for median fluorescence intensity (MFI) values.

Figure 2.

Open in a new tab

One donor was employed for barcoding all nine populations in each of six matrixes (three with unstimulated samples and three stimulated for pSTAT1, pSTAT3, or pSTAT5). (A) Mean of MFIs and SD were calculated for each position in a matrix (pi; p1 to p9) among all matrixes with unstimulated samples (total of nine values for each parameter; N) or stimulated samples. The range of variability was calculated as average of MFIs of all barcoded samples ± 2SD. The ratio of variability was defined as the ratio between the average of barcoded samples in every position in a matrix and the average of barcoded samples in p1 (only DMSO). (B) MFIs for each fluorochrome were compared to those obtained from matched not-barcoded (w/o FCB) samples by Pearson analysis. (C) Ranges and ratios of variability were calculated for each population, and values outside the ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in black.

3.3. pSTAT intra-assay variability

We assessed intra-assay variability of pSTAT staining in combination with FCB and surface marker antibodies. Samples were treated with appropriate stimuli as described above, and MFIs of pSTAT1, pSTAT3, and pSTAT5 in CD4+, CD8+, CD20+, and CD14+ cells were obtained from barcoded populations of each stimulated and matched unstimulated sample (Supplementary Figure 7). Fold change (FC) was calculated as MFI of stimulated sample in pi / MFI of unstimulated sample in pi (matrix position, p1 to p9). Then, average and SD of all FC values in CD4+, CD8+, CD20+, and CD14+ cells were used to calculate the range of variability (average FC + 2SD). The ratio of variability was also defined as average FC in pi / average FC (Supplementary Figure 8). For pSTAT1, no intra-assay variability was described for CD4+ and CD14+ cells (Figure 3); while, minor variability was documented for CD8+ and CD20+ cells. For pSTAT5, there was no intra-assay variability for CD4+, CD8+, and CD14+ cells; a minor variability was documented for CD20+ populations. For pSTAT3, some variability was described for CD4+, CD8+, CD20+ and CD14+ cells. These variations were caused by an increased variability of MFI values in unstimulated samples (Supplementary Figure 6). Despite these variations in MFIs in specific populations, our FCB protocol remained highly reproducible for pSTAT signaling analysis in T/B cells and monocytes.

Figure 3. pSTAT intra-assay variability assessment.

Figure 3.

Open in a new tab

MFIs of pSTAT1, pSTAT3, and pSTAT5 in CD4+, CD8+, CD20+, and CD14+ cells were obtained from barcoded populations of each stimulated and matched unstimulated sample and fold change (FC) calculated. Mean and SD of all FC values in each population were calculated in order to identify the range of variability: Average FC ± 2SD. The ratio of variability was also defined as Average FC in pi (from p1 to p9) / Average of all FC. Values outside of these ranges were highlighted in black.

3.4. Inter-assay and inter-operator variability

After establishing minimal intra-assay variations of our FCB protocol, we investigated FCB reproducibility across different operators and days. PBMCs from eight different healthy donors (D1 to D8 in the same position across all matrixes; Figure 4A) were employed for FCB of a nine-sample matrix, and samples were appropriately stimulated for pSTAT1, pSTAT3, and pSTAT5. In position 9, one healthy donor used to measure intra-assay variability was placed in all matrices and considered an internal bridge control (Figure 4A). In addition, each of three operators performed FCB of one pSTAT by doing appropriate stimulation, fixation, permeabilization, FCB, and staining independently. This experiment was repeated four times on four separate days with a total of 32 healthy donors used. The same bridge control was used across the four days to assess inter-day variability.

Figure 4. Inter-operator variability assessment.

Figure 4.

Open in a new tab

(A) Each of three operators performed independently stimulation, fixation, permeabilization, FCB and staining for one pSTAT. Eight healthy donors (D1 to D8) were used for FCB in combination with surface marker and intracellular antibodies. An internal control sample (Int Ctrl) was added in position 9 of all matrixes and used as bridge sample to assess inter-assay and inter-operator variability. (B) Percent of positive cells from internal controls were compared between operators by 1-way ANOVA. Data are shown as mean±SD. (C) Percent of positive cells were also obtained from stimulated and unstimulated samples of each donor and compared by 1-way ANOVA in order to exclude variations between operators (Op1 to Op3). Pearson correlation analysis was carried out between percent of positive cells from stimulated samples (Stim) and those from unstimulated specimens (Unstim) in order to investigate inter-assay variability. A representative example of 1-way ANOVA (D) and Pearson analysis (E) for CD4+ cells is shown. (F) Fold change (FC) of each pSTAT in CD4+, CD8+, CD20+ or CD14+ cells were calculated for all 32 donors, and ranges of variability calculated as described in Figure 3. Values outside these ranges were highlighted in black.

The percentages of positive cells were first obtained from individual internal controls and compared by unpaired t-tests across operators. No significant differences were observed (Figure 4B). Similarly, percent of positive cells obtained from the other 32 donors were compared across operators by 1-way ANOVA, and no significant variations were described (Figure 4C). Pearson correlation analysis was performed between each stimulated and unstimulated sample, and high correlations were identified (Figure 4C–E and Supplementary Figure 9). FC and Log10FC of pSTAT1, pSTAT3, and pSTAT5 for all samples were calculated using MFIs of appropriate stimulated and unstimulated samples. For each population, average and SD of Log10FC were calculated and used to build heatmaps (Figure 4F). Total variability for pSTATs was 2.5%. Using an internal control across matrices, we were able to exclude technical variations between operators and, therefore, considered this variability as biological and not technical. In addition, stability and reproducibility of FCB was assessed by Pearson correlation of pSTAT values at different timepoints or assays in CD4+, CD8+, CD14+ or CD20+ cell populations (Supplementary Figure 10). Overall, the FCB protocol showed minimal variation across different operators and multiple assays, allowing greater sample collection and minimizing data acquisition variability for large data studies.

3.5. Intra-assay variability by computational analysis

To address the demand for rapid large-scale profiling, intra-assay variability was also assessed using semi-automated workflows. Results were compared to those obtained by conventional manual gating strategies in order to assess the feasibility of computational workflows for FCB data analysis. The percentages of positive cells determined by our computational analysis were highly correlated to those obtained from matched samples analyzed by conventional manual gating (r = 0.968; P < 0.0001; Figure 5A). The percentage of positive cells from computational gating was next compared to those from manual gating by Wilcoxon matched-pairs signed rank test and unpaired t-test (Figure 5B). No significant variations were observed for CD3+, CD4+, and CD8+ T cells; while, a significant difference was described for CD20+ and CD14+ cells. Intra-assay variability was assessed as explained above by first defining ranges and ratios of variability, and by using matched non-barcoded samples as controls (Figure 5C and Supplementary Figure 11). Since no statistically significant variabilities were detected, semi-automated workflows could be used for FCB data analysis with similar results of conventional analysis. The comparability of the computational gating to manual methods demonstrate the possibility of rapid and reproducible FCB data analysis by machine learning workflows. Therefore, our FCB protocol showed low intra- and inter-staining variability within an assay and across operators, and high reproducibility by computational data analysis which reduced workload and increased efficiency.

Figure 5. FCB computational data analysis.

Figure 5.

Open in a new tab

(A) Using semi-automated workflows, percent of positive cells were obtained and compared to those from matched barcoded samples analyzed by conventional manual gating strategy. (B) Percent of positive cells obtained using computational analysis (R) were also compared to those calculated by manual gating using FlowJo software (FJ) by unpaired t-test. (C) Intra-assay variability was assessed as described in Figure 1 by defining range and ratio of variability for each population and by using as controls matched not-barcoded samples. Values outside these ranges were highlighted in black.

4. Discussion

FCB is a multiplexed high-throughput flow cytometry technique which has been used for phosphoFlow analysis [1–6], detection of intracellular cytokines [7], and immunophenotyping of human PBMCs and mesenchymal stem cells [9–10]. In recent years, mass cytometry has gained popularity for phosphoprotein profiling, and has been frequently used for studies aimed at quantifying relative changes. However, CyTOF’s slow acquisition speed has created the need to control for time tracking and normalization compared to standard FCM rates. In addition, the costs associated with mass cytometry, including the costs of reagents, which are often not standardized if metal conjugation is done in-house, needs to be taken into account. Our efforts focused on precision, accuracy, and reproducibility of multiplexing of pSTAT profiling because we believe that the ability to combine standard FCM with multiplexed methods and cost efficiency makes FCB a feasible alternative. Therefore, we proposed and validated the use of a bridge control across FCB assays and showed a robust method to assess intra- and inter-operator variability of our FCB technique. Finally, we built a simple computational method to analyze FCB data in a faster and operator-independent way, which was also used to assess and verify the variability.

In previous work, we have optimized FCB for nine-sample barcoding using two sets of concentrations of three FCB dye combinations (0, 13 and 250 μg/mL or 1.56, 50 and 500 μg/mL) which allowed a good deconvolution with MFI fold increase >2 [9]. In this study, we implemented FCB for phosphoprotein analysis using DyLight 350 and Pacific Orange for barcoding nine samples. We also developed a standardized method for a robust assessment of intra- and inter-assay and inter-operator variability that could allow FCB to be applied to samples collected during clinical trials. Our FCB was optimized using Pacific Orange and DyLight 350 for surface and intracellular staining of human PBMCs. Dye concentrations were slightly changed to previous work (0, 15 or 30, and 250 μg/mL) in order to have MFI fold increase >3 (range, 4.1–11.3) for a complete and sharp separation of both lymphocytes and monocytes. As a result, FCB with DyLight 350 and Pacific Orange at 0, 15 or 30, and 250 μg/mL achieved complete deconvolution of both lymphocyte and monocyte populations.

FCM is a powerful tool for rapid phenotyping in clinical research; however, subjectivity and variability of FCM are well-known pitfalls, and many international organizations are working to improve comparability and reliability of FCM data across laboratories [16–17, 28–31]. Considering the need to standardize FCM techniques and minimize staining variability, here we provide a robust method for assessment of intra- and inter-assay variability of a nine-sample FCB using DyLight 350 and Pacific Orange co-stained with surface (CD3, CD4, CD8, CD20, and CD14) and intracellular (pSTAT1, pSTAT3, or pSTAT5) markers. We also assess different permeabilization methods which could allow sufficient permeabilization for pSTAT staining. Since antibody cocktails were added after permeabilization and barcoding, our method minimizes non-specific binding of surface marker antibodies to intracellular reservoirs. Our results show a high correlation between percent positive cells obtained from barcoded samples and those from matched non-barcoded samples stained only with surface marker antibodies used as controls. These results confirmed that FCB did not affect the staining process and different concentrations of methanol for permeabilization did not increase non-specific binding of intracellular reservoirs of surface markers [37]. For intra-assay variability assessment, one donor was barcoded for all nine populations in a matrix in order to exclude any variation in the percent of positive cells or MFIs due to biological differences between samples and not due to technical variations [13]. For each parameter, a range of variability was calculated, and values were considered as acceptable if within this range. Values obtained from barcoded samples were compared to those from matched non-barcoded specimens by calculating the ratio of variability which we utilized to directly compare to matched controls allowing only 20% of biological variation [36–40]. The use of two simple ranges for detecting outliers combined with standard statistical analysis could systematically and reliably identify intra- and inter-assay variability of FCB. Indeed, the employment of a single mathematical method could not find out variabilities within positions in a matrix or between barcoded and not-barcoded samples. In our case, some minor non-statistically significant variabilities were documented for CD3+, CD8+, and CD20+ cells for unstimulated specimens; no intra-assay variability was documented for MFIs of any fluorochrome studied. Significant variabilities in MFIs of FITC- and PE-conjugated antibodies were documented in positions 3 and 6, and also in position 9 for CD14-PE antibody. As CD14-PE was used to identify monocytes, variations in MFIs may be caused by an increased ability of this population to bind larger amount of FCB dyes and/or because of a higher cytoplasmic complexity [6,11]. These variations were described using other antibody clones and fluorochromes in both fresh and frozen PBMCs (e.g. BV650-conjugated CD14; data not shown). Due to this dye concentration-related variability of some antibodies, we suggest avoiding quantitative analysis of MFIs of antibodies that bind monocyte markers. However, we did not observe simultaneous variations in both range and ratio of variability for other antibodies. Intra-assay variability was also assessed for pSTATs. We documented low variability for pSTAT1 and pSTAT5; while, pSTAT3 was the most variable mainly caused by an increased variability of MFIs of unstimulated samples, especially in CD20+ populations. Nonetheless, our data suggest that the current FCB protocol for phosphoFlow analysis in T/B cell subpopulations and monocytes showed high robustness, reproducibility and reliability.

During post-acquisition analysis, accurate identification of target cells and gate boundaries are the major cause of inter-operator and inter-laboratory variability of FCM data analysis [16–17]. In order to measure and minimize inter-assay and inter-operator variability, nine-sample FCB for phosphoFlow analysis was performed by three operators independently from sample stimulation to antibody staining. Moreover, one of the healthy donors used for intra-assay variability assessment was employed as an internal “bridge” control and added to every matrix in position 9. The use of “bridge samples” for data normalization has already been proposed by several authors and for different assays [38]. Here, we employed bridge samples for the determination of accuracy, reproducibility, and reliability of FCB data from different operators for the development of protocols for clinical tests. The presence of significant variations in percent of positive cells or MFIs in the bridge control was considered as a rejection criterion for a particular run and the experiment was repeated. However, the robustness of our protocol was confirmed by the absence of significant differences between percent of positive cells or FC of pSTATs across operators for both internal controls and barcoded samples in all experiments.

The large datasets generated by FCB studies makes conventional FCM analysis challenging and poorly reproducible, mainly because of inter-operator variability [10, 24–25]. Computational FCM is becoming increasingly utilized, given that machine learning analysis is more reproducible and efficient due to simultaneous exploration of more than two parameters [24]. In this study, we investigated the applicability of a semi-automated machine learning workflow for analysis of combination staining with FCB dyes and surface and intracellular antibodies. We provided a simple R code for semi-automated FCB data analysis using available packages, such as flowCore and flowClust, and without post-acquisition compensation, which is a source of inter-assay variability [16]. The only guided step was the identification of lymphocyte and monocyte clusters which was followed by unsupervised analysis for deconvolution and identification of CD3, CD4, CD8, and CD20 cells by flowClust. For CD14+ monocytes, we employed the rectangleGate function from flowCore package, and we set the 1st quartile of chosen parameters (e.g. CD14-PE) as non-arbitrary threshold for cluster identification. Using this approach, data obtained from semi-automated workflows and conventional manual gating strategy were highly comparable. Some variations were observed for CD20+ cells possibly because of the very small SD, and for CD14+ monocytes because the threshold for gating was higher in the computational analysis compared to manual gating. Intra-assay variability of the data obtained from computational workflows was assessed, and minor variability was described in position 9 for CD3+ and CD8+ populations that could likely be adjusted by applying a post-acquisition compensation. However, this approach could be further implemented in order to develop a dedicated fully automated R package for FCB data analysis with minimal variations, such as by building an algorithm for auto-compensation of single barcoded populations for autofluorescence adjustment that could likely decrease intra-assay variability.

5. Conclusions

Analysis of single-cell high-throughput data is applied in many fields, including FCM. Similarly, standardization of pre- and post-analytical procedures and minimization of intra- and inter-assay variability are required in order to have reproducible and reliable data. FCB is a multiplexed high-throughput FCM technique which allow large-scale studies with minimal intraand inter- assay variability and may also reduce inter-operator variability by including a bridge sample across matrices used as a rejection criterion for a particular run. Our results demonstrated that FCB is a powerful tool especially for samples obtained during clinical trials, translational research and longitudinal studies. In addition, computational workflows may improve FCB data analysis by reducing time of analysis and inter-operator variability for gating boundaries.

Supplementary Material

1
2
3
4
5
6
7
8
9

Supplementary Figure 1. Conventional manual gating strategy. Lymphocytes and monocytes were identified using linear parameters (FSC-A vs SSC-A), double cells excluded (FSC-A vs FSC-H), and deconvolution carried out on each population using FCB dye channels (DyLight 350 vs Pacific Orange). On barcoded lymphocytes, CD3+ and CD20+ cells were first gated and CD4 and CD8 expression was further investigated on CD3+ cells. On barcoded monocytes, CD14+ cells were identified using linear parameter and corresponding fluorochrome channel (CD14-PE vs FSC-A). Then, on CD4+, CD8+, CD20+ and CD14+ populations, pSTAT expression was studied by calculating MFI values from stimulated and unstimulated samples. Data are shown using biexponential transformation except for linear parameters. pSTAT expression is displayed by normalized cell count histograms using unstimulated samples (light grey) and matched stimulated specimens (light red).

Supplementary Figure 2. Computational automated gating strategy. (A) Debris were first removed using rectangleGate function and then lymphocytes and monocytes were identified using linear parameters (FSC.A vs SSC.A) using flowClust function. (B) Deconvolution was carried out using FCB dye channels (BUV395.A vs Pacific.Orange.A). On each barcoded lymphocyte population, unsupervised clustering was carried out to identify CD3+ and CD20+ cells using flowclust function (APC.H7.A vs PerCP.Cy5.5.A), and CD4+ and CD8+ cells were further gated (PE.Cy7.A vs FITC.A). (C) Similarly, after deconvoluting monocytes, CD14+ cells were identified using linear parameters and CD14 expression by defining rectangular gates at the 1st quartile of the PE.A parameter.

Supplementary Figure 3. FCB efficiency. FCB efficiency was assessed by calculating MFI fold increase on lymphocyte population. First, lymphocytes were identified using linear parameters (FSC-A vs SSC-A), double cells excluded (FSC-A vs FSC-H), and deconvolution carried out using FCB dye parameters (DyLight 350 vs Pacific Orange). On single cells, Pacific Orange and DyLight 350 expression was displayed using normalized cell count histograms, and three sharper peaks were identified, according to each FCB dye concentration. On each peak, MFI and CV were obtained and used to calculate MFI fold increase as previously described [9]. FCB efficiency was assessed on 12 nine-sample matrixes barcoded with nine different healthy donors.

Supplementary Figure 4. Flowchart for intra-assay variability assessment. (A) Range of variability was determined for assessment of intra-assay variability. One donor was used to barcode all samples in six matrixes (three unstimulated and three stimulated for each pSTAT) for a total of three different healthy donors used (total of 81 values for each parameter for stimulated or unstimulated barcoded populations). First, average and SD were calculated for all stimulated or unstimulated samples for all donors, and the range of variability was defined as the average of all stimulated/unstimulated barcoded samples ±2SD. (B) Subsequently, we calculated averages of percent of positive cells or MFIs for each position (p1 to p9), and values were compared to corresponding ranges of variability. If values were within ranges, they were considered as acceptable. Next, the ratio of variability was calculated as the ratio between averages of percent of positive cells or MFIs for each position and the average of corresponding parameter from matched not-barcoded samples. If values were within 0.80 and 1.20, they were considered as acceptable. However, only if values were not-acceptable for both range and ratio of variability, high variability was defined for that position.

Supplementary Figure 5. Surface marker intra-assay variability for percent of positive cells. Corresponding raw data of Figure 1 for assessment of intra-assay variability for percent of positive cells. Mean of percent of positive cells and SD were calculated for each position in a matrix (pi; p1 to p9) among all matrixes with unstimulated samples (total of nine values for each parameter; N) or stimulated samples. Ranges and ratios of variability were calculated as described in Figure 1. Values outside these ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in light red.

Supplementary Figure 6. Surface marker intra-assay variability for MFIs. Corresponding raw data of Figure 2 for assessment of intra-assay variability for median fluorescence intensity (MFI) values. Mean of MFIs for each fluorochrome and SD were calculated for each position in a matrix (pi; p1 to p9) among all matrixes with unstimulated samples (total of nine values for each parameter; N) or stimulated samples. Ranges and ratios of variability were calculated as described in Figure 2. Values outside these ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in light red.

Supplementary Figure 7. pSTAT intra-assay variability for MFIs. Median fluorescence intensity (MFI) values of pSTAT1, pSTAT3 (A), and pSTAT5 (B) on CD4+, CD8+, CD20+, and CD14+ cells of stimulated and unstimulated samples for assessment of intra-assay variability. Mean of MFIs and SD were calculated for each position in a matrix (pi; p1 to p9) among all matrixes with unstimulated samples (total of nine values for each parameter; N) or stimulated samples. Ranges and ratios of variability were calculated as described in Figure 2. Values outside these ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in light red.

Supplementary Figure 8. pSTAT intra-assay variability for MFI Fold Change (FC). Median fluorescence intensity (MFI) values of pSTAT1, pSTAT3, and pSTAT5 from Supplementary Figure 5 were used to calculate mean MFI FC in each position (p1 to p9). Mean and SD were also calculated for all stimulated or unstimulated specimens in order to calculated ranges and ratios of variability. Values outside these ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in light red.

Supplementary Figure 9. 1-way ANOVA and Pearson analysis of percent of positive cells across operators. Percent of positive CD3+, CD8+, CD20+, and CD14+ cells were obtained from matrixes processed by three different operators and compared by 1-way ANOVA. Values from stimulated samples were also correlated to those from matched unstimulated specimens by Pearson analysis in order to assess inter-assay variability.

Supplementary Figure 10. Computational gating strategy. (A) Debris were removed using rectangleGate function of flowCore. After clustering using flowClust package, lymphocytes and monocytes were identified using the SSC-A values as cut-off, and subsequently, deconvolution carried out using flowClust (B). On each barcoded population, unsupervised clustering was employed for the identification of CD3+ and CD20+ cells, and then CD4+ and CD8+ populations on CD3+ cells. (C) On monocytes, deconvolution was first performed, and next CD14+ monocytes gated using rectangleGate function of flowCore. Cut-offs for CD14-PE and SSC-A were set at the 1st quartile for each parameter.

Supplementary Figure 11. Surface marker intra-assay variability for percent of positive cells by computational analysis. Corresponding raw data of Figure 5 for assessment of intra-assay variability for percent of positive cells by computational analysis. Mean of percent of positive cells and SD were calculated for each position in a matrix (pi; p1 to p9) among all matrixes with unstimulated samples (total of nine values for each parameter; N) or stimulated samples. Ranges and ratios of variability were calculated as described in Figure 1. Values outside these ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in light red.

Acknowledgements

The authors would like to thank Drs. Viviana Izzo, J. Philip McCoy, Giuseppe Sciumè, Francoise Meylan and Fred Davis for their careful reading of an earlier version of the manuscript and their valuable suggestions. This research was supported by the Intramural Research Program of the National Institute of Arthritis Musculoskeletal and Skin diseases (1ZICAR041181-10) and National Heart, Lung, and Blood Institute.

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Conflicts of Interest: The authors declare no conflicts of interest.

References

  • 1.Krutzik PO, Nolan GP. Fluorescent cell barcoding in flow cytometry allows high-throughput drug screening and signaling profiling. Nat Methods. 2006;3: 361–368. [DOI] [PubMed] [Google Scholar]
  • 2.Krutzik PO, Nolan GP. Intracellular phospho-protein staining techniques for flow cytometry: monitoring single cell signaling events. Cytometry A. 2003;55: 61–70. [DOI] [PubMed] [Google Scholar]
  • 3.Krutzik PO, Crane JM, Clutter MR, Nolan GP. High-content single-cell drug screening with phosphospecific flow cytometry. Nat Chem Biol. 2008;4: 132–142. [DOI] [PubMed] [Google Scholar]
  • 4.Kalland ME, Oberprieler NG, Vang T, Taskén K, Torgersen KM. T cell-signaling network analysis reveals distinct differences between CD28 and CD2 costimulation responses in various subsets and in the MAPK pathway between resting and activated regulatory T cells. J Immunol. 2011;187: 5233–5245. [DOI] [PubMed] [Google Scholar]
  • 5.Spurgeon BE, Aburima A, Oberprieler NG, Taskén K, Naseem KM. Multiplexed phosphospecific flow cytometry enables large-scale signaling profiling and drug screening in blood platelets. J Thromb Haemost. 2014;12: 1733–1743. [DOI] [PubMed] [Google Scholar]
  • 6.Davies R, Vogelsang P, Jonsson R, Appel S. An optimized multiplex flow cytometry protocol for the analysis of intracellular signaling in peripheral blood mononuclear cells. J Immunol Methods. 2016;436: 58–63. [DOI] [PubMed] [Google Scholar]
  • 7.Stam J, Abdulahad W, Huitema MG, Roozendaal C, Limburg PC, van Stuijvenberg M, Schölvinck EH. Fluorescent cell barcoding as a tool to assess the age-related development of intracellular cytokine production in small amounts of blood from infants. PLoS One. 2011;6(10):e25690. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Prussin C, Metcalfe DD. Detection of intracytoplasmic cytokine using flow cytometry and directly conjugated anti-cytokine antibodies. J Immunol Methods. 1995;188: 117–128. [DOI] [PubMed] [Google Scholar]
  • 9.Giudice V, Feng X, Kajigaya S, Young NS, Biancotto A. Optimization and standardization of fluorescent cell barcoding for multiplexed flow cytometric phenotyping. Cytometry Part A 2017;91(7):694–703. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10.Lekishvili T, Campbell JJ. Rapid comparative immunophenotyping of human mesenchymal stromal cells by a modified fluorescent cell barcoding flow cytometric assay. Cytometry A. 2018;93(9):905–915. [DOI] [PubMed] [Google Scholar]
  • 11.Krutzik PO, Clutter MR, Trejo A, Nolan GP. Fluorescent cell barcoding for multiplex flow cytometry. Curr Protoc Cytom. 2011;Chapter 6: Unit 6.31. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Akkaya B, Miozzo P, Holstein AH, Shevach EM, Pierce SK, Akkaya M. A Simple, Versatile Antibody-Based Barcoding Method for Flow Cytometry. J Immunol. 2016;197(5):2027–2038. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Burel JG, Qian Y, Lindestam Arlehamn C, Weiskopf D, Zapardiel-Gonzalo J, Taplitz R, Gilman RH, Saito M, de Silva AD, Vijayanand P, Scheuermann RH, Sette A, Peters B. An Integrated Workflow To Assess Technical and Biological Variability of Cell Population Frequencies in Human Peripheral Blood by Flow Cytometry. J Immunol. 2017;198(4):1748–1758. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Lu Y, Biancotto A, Cheung F, Remmers E, Shah N, McCoy JP, Tsang JS. Systematic Analysis of Cell-to-Cell Expression Variation of T Lymphocytes in a Human Cohort Identifies Aging and Genetic Associations. Immunity. 2016;45(5):1162–1175. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Niepel M, Spencer SL, Sorger PK. Non-genetic cell-to-cell variability and the consequences for pharmacology. Curr Opin Chem Biol. 2009;13(5–6): 556–561. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Pachón G, Caragol I, Petriz J. Subjectivity and flow cytometric variability. Nat Rev Immunol. 2012;12(5):396; author reply 396. [DOI] [PubMed] [Google Scholar]
  • 17.Maecker HT, McCoy JP, Nussenblatt R. Standardizing immunophenotyping for the Human Immunology Project. Nat Rev Immunol. 2012;12(3):191–200. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18.Saeys Y, Gassen SV, Lambrecht BN. Computational flow cytometry: helping to make sense of high-dimensional immunology data. Nat Rev Immunol. 2016;16(7): 449–462. [DOI] [PubMed] [Google Scholar]
  • 19.Aghaeepour N, Brinkman R. Computational analysis of high-dimensional flow cytometric data for diagnosis and discovery. Curr Top Microbiol Immunol. 2014;377: 159–175. [DOI] [PubMed] [Google Scholar]
  • 20.Aghaeepour N, Finak G; FlowCAP Consortium; DREAM Consortium, Hoos H, Mosmann TR, Brinkman R, Gottardo R, Scheuermann RH. Critical assessment of automated flow cytometry data analysis techniques. Nat Methods. 2013;10(3):228–238. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21.Candia J, Maunu R, Driscoll M, Biancotto A, Dagur P, McCoy JP Jr, Sen HN, Wei L, Maritan A, Cao K, Nussenblatt RB, Banavar JR, Losert W. From Cellular Characteristics to Disease Diagnosis: Uncovering Phenotypes with Supercells. PLoS Computational Biology 2013;9:e1003215. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22.Lo K, Brinkman RR, Gottardo R. Automated gating of flow cytometry data via robust model-based clustering. Cytometry A. 2008;73: 321–332. [DOI] [PubMed] [Google Scholar]
  • 23.Lo K, Hahne F, Brinkman RR, Gottardo R. flowClust: a Bioconductor package for automated gating of flow cytometry data. BMC Bioinformatics. 2009;10:145. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Diggins KE, Ferrell PB Jr, Irish JM. Methods for discovery and characterization of cell subsets in high dimensional mass cytometry data. Methods. 2015;82: 55–63. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Qiu P, Simonds EF, Bendall SC, Gibbs KD Jr, Bruggner RV, Linderman MD, Sachs K, Nolan GP, Plevritis SK. Extracting a cellular hierarchy from high-dimensional cytometry data with SPADE. Nat Biotechnol. 2011;29(10):886–891. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Ellis B, Haaland P, Hahne F, Le Meur N, Gopalakrishnan N, Spidlen J, Jiang M. flowCore: Basic structures for flow cytometry data. R package version 1.44.2. 2018. [Google Scholar]
  • 27.Finak G, Bashashati A, Brinkman R, Gottardo R. Merging mixture components for cell population identification in flow cytometry. Adv Bioinformatics. 2009:247646. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28.Spidlen J, Gentleman RC, Haaland PD, Langille M, Le Meur N, Ochs MF, Schmitt C, Smith CA, Treister AS, Brinkman RR. Data standards for flow cytometry. OMICS. 2006;10(2):209–214. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Kalina T, Flores-Montero J, van der Velden VH, Martin-Ayuso M, Böttcher S, Ritgen M, Almeida J, Lhermitte L, Asnafi V, Mendonça A, de Tute R, Cullen M, Sedek L, Vidriales MB, Pérez JJ, te Marvelde JG, Mejstrikova E, Hrusak O, Szczepański T, van Dongen JJ, Orfao A; EuroFlow Consortium (EU-FP6, LSHB-CT-2006–018708). EuroFlow standardization of flow cytometer instrument settings and immunophenotyping protocols. Leukemia. 2012;26(9):1986–2010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30.Kalina T, Flores-Montero J, Lecrevisse Q, Pedreira CE, van der Velden VH, Novakova M, Mejstrikova E, Hrusak O, Böttcher S, Karsch D, Sędek Ł, Trinquand A, Boeckx N, Caetano J, Asnafi V, Lucio P, Lima M, Helena Santos A, Bonaccorso P, van der Sluijs-Gelling AJ, Langerak AW, Martin-Ayuso M, Szczepański T, van Dongen JJ, Orfao A. Quality assessment program for EuroFlow protocols: summary results of four-year (2010–2013) quality assurance rounds. Cytometry A. 2015;87(2):145–156. [DOI] [PubMed] [Google Scholar]
  • 31.Valle A, Maugeri N, Manfredi AA, Battaglia M. Standardization in flow cytometry: correct sample handling as a priority. Nat Rev Immunol. 2012;12(12):864. [DOI] [PubMed] [Google Scholar]
  • 32.World Medical Association. World Medical Association Declaration of Helsinki: ethical principles for medical research involving human subjects. JAMA. 2013;310(20): 2191–2194. [DOI] [PubMed] [Google Scholar]
  • 33.Simons DL1, Lee G, Kirkwood JM, Lee PP. Interferon signaling patterns in peripheral blood lymphocytes may predict clinical outcome after high-dose interferon therapy in melanoma patients. J Transl Med. 2011;9:52. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Mahmud SA, Manlove LS, Farrar MA. Interleukin-2 and STAT5 in regulatory T cell development and function. JAKSTAT. 2013;2(1):e23154. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Niemand C, Nimmesgern A, Haan S, Fischer P, Schaper F, Rossaint R, Heinrich PC, Müller-Newen G. Activation of STAT3 by IL-6 and IL-10 in primary human macrophages is differentially modulated by suppressor of cytokine signaling 3. J Immunol. 2003;170(6):3263–72. [DOI] [PubMed] [Google Scholar]
  • 36.Begum H, Li B, Shui G, Cazenave-Gassiot A, Soong R, Ong RT, Little P, Teo YY, Wenk MR. Discovering and validating between-subject variations in plasma lipids in healthy subjects. Sci Rep. 2016;6:19139. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37.Shurubor YI1, Matson WR, Willett WC, Hankinson SE, Kristal BS. Biological variability dominates and influences analytical variance in HPLC-ECD studies of the human plasma metabolome. BMC Clin Pathol. 2007;7:9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Hadlow N, Brown SJ, Habib A, Wardrop R, Joseph J, Gillett M, Maguire R, Conradie J. Quantifying the intraindividual variation of antimüllerian hormone in the ovarian cycle. Fertil Steril. 2016;106(5):1230–1237. [DOI] [PubMed] [Google Scholar]
  • 39.Reed GF, Lynn F, Meade BD. Use of coefficient of variation in assessing variability of quantitative assays. Clin Diagn Lab Immunol. 2002;9(6):1235–1239. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40.Wu AH, Lu QA, Todd J, Moecks J, Wians F. Short- and long-term biological variation in cardiac troponin I measured with a high-sensitivity assay: implications for clinical practice. Clin Chem. 2009;55(1):52–8. [DOI] [PubMed] [Google Scholar]
  • 41.Behbehani GK, Thom C, Zunder ER, Finck R, Gaudilliere B, Fragiadakis GK, Fantl WJ, Nolan GP. Transient partial permeabilization with saponin enables cellular barcoding prior to surface marker staining. Cytometry A. 2014;85(12):1011–1019. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Biancotto A, Wank A, Perl S, Cook W, Olnes MJ, Dagur PK, Fuchs JC, Langweiler M, Wang E, McCoy JP. Baseline levels and temporal stability of 27 multiplexed serum cytokine concentrations in healthy subjects. PLoS One. 2013;8(12):e76091. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Hu Z, Jujjavarapu C, Hughey JJ, Andorf S, Lee HC, Gherardini PF, Spitzer MH, Thomas CG, Campbell J, Dunn P, Wiser J, Kidd BA, Dudley JT, Nolan GP, Bhattacharya S, Butte AJ. MetaCyto: A Tool for Automated Meta-analysis of Mass and Flow Cytometry Data. Cell Rep. 2018;24(5):1377–1388. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS1545413-supplement-1.TIF (10MB, TIF)
2
NIHMS1545413-supplement-2.TIF (10MB, TIF)
3
NIHMS1545413-supplement-3.TIF (7.5MB, TIF)
4
NIHMS1545413-supplement-4.TIF (6.9MB, TIF)
5
NIHMS1545413-supplement-5.TIF (6.9MB, TIF)
6
NIHMS1545413-supplement-6.TIF (4.1MB, TIF)
7
NIHMS1545413-supplement-7.TIF (7.3MB, TIF)
8
NIHMS1545413-supplement-8.tif (7.3MB, tif)
9

Supplementary Figure 1. Conventional manual gating strategy. Lymphocytes and monocytes were identified using linear parameters (FSC-A vs SSC-A), double cells excluded (FSC-A vs FSC-H), and deconvolution carried out on each population using FCB dye channels (DyLight 350 vs Pacific Orange). On barcoded lymphocytes, CD3+ and CD20+ cells were first gated and CD4 and CD8 expression was further investigated on CD3+ cells. On barcoded monocytes, CD14+ cells were identified using linear parameter and corresponding fluorochrome channel (CD14-PE vs FSC-A). Then, on CD4+, CD8+, CD20+ and CD14+ populations, pSTAT expression was studied by calculating MFI values from stimulated and unstimulated samples. Data are shown using biexponential transformation except for linear parameters. pSTAT expression is displayed by normalized cell count histograms using unstimulated samples (light grey) and matched stimulated specimens (light red).

Supplementary Figure 2. Computational automated gating strategy. (A) Debris were first removed using rectangleGate function and then lymphocytes and monocytes were identified using linear parameters (FSC.A vs SSC.A) using flowClust function. (B) Deconvolution was carried out using FCB dye channels (BUV395.A vs Pacific.Orange.A). On each barcoded lymphocyte population, unsupervised clustering was carried out to identify CD3+ and CD20+ cells using flowclust function (APC.H7.A vs PerCP.Cy5.5.A), and CD4+ and CD8+ cells were further gated (PE.Cy7.A vs FITC.A). (C) Similarly, after deconvoluting monocytes, CD14+ cells were identified using linear parameters and CD14 expression by defining rectangular gates at the 1st quartile of the PE.A parameter.

Supplementary Figure 3. FCB efficiency. FCB efficiency was assessed by calculating MFI fold increase on lymphocyte population. First, lymphocytes were identified using linear parameters (FSC-A vs SSC-A), double cells excluded (FSC-A vs FSC-H), and deconvolution carried out using FCB dye parameters (DyLight 350 vs Pacific Orange). On single cells, Pacific Orange and DyLight 350 expression was displayed using normalized cell count histograms, and three sharper peaks were identified, according to each FCB dye concentration. On each peak, MFI and CV were obtained and used to calculate MFI fold increase as previously described [9]. FCB efficiency was assessed on 12 nine-sample matrixes barcoded with nine different healthy donors.

Supplementary Figure 4. Flowchart for intra-assay variability assessment. (A) Range of variability was determined for assessment of intra-assay variability. One donor was used to barcode all samples in six matrixes (three unstimulated and three stimulated for each pSTAT) for a total of three different healthy donors used (total of 81 values for each parameter for stimulated or unstimulated barcoded populations). First, average and SD were calculated for all stimulated or unstimulated samples for all donors, and the range of variability was defined as the average of all stimulated/unstimulated barcoded samples ±2SD. (B) Subsequently, we calculated averages of percent of positive cells or MFIs for each position (p1 to p9), and values were compared to corresponding ranges of variability. If values were within ranges, they were considered as acceptable. Next, the ratio of variability was calculated as the ratio between averages of percent of positive cells or MFIs for each position and the average of corresponding parameter from matched not-barcoded samples. If values were within 0.80 and 1.20, they were considered as acceptable. However, only if values were not-acceptable for both range and ratio of variability, high variability was defined for that position.

Supplementary Figure 5. Surface marker intra-assay variability for percent of positive cells. Corresponding raw data of Figure 1 for assessment of intra-assay variability for percent of positive cells. Mean of percent of positive cells and SD were calculated for each position in a matrix (pi; p1 to p9) among all matrixes with unstimulated samples (total of nine values for each parameter; N) or stimulated samples. Ranges and ratios of variability were calculated as described in Figure 1. Values outside these ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in light red.

Supplementary Figure 6. Surface marker intra-assay variability for MFIs. Corresponding raw data of Figure 2 for assessment of intra-assay variability for median fluorescence intensity (MFI) values. Mean of MFIs for each fluorochrome and SD were calculated for each position in a matrix (pi; p1 to p9) among all matrixes with unstimulated samples (total of nine values for each parameter; N) or stimulated samples. Ranges and ratios of variability were calculated as described in Figure 2. Values outside these ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in light red.

Supplementary Figure 7. pSTAT intra-assay variability for MFIs. Median fluorescence intensity (MFI) values of pSTAT1, pSTAT3 (A), and pSTAT5 (B) on CD4+, CD8+, CD20+, and CD14+ cells of stimulated and unstimulated samples for assessment of intra-assay variability. Mean of MFIs and SD were calculated for each position in a matrix (pi; p1 to p9) among all matrixes with unstimulated samples (total of nine values for each parameter; N) or stimulated samples. Ranges and ratios of variability were calculated as described in Figure 2. Values outside these ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in light red.

Supplementary Figure 8. pSTAT intra-assay variability for MFI Fold Change (FC). Median fluorescence intensity (MFI) values of pSTAT1, pSTAT3, and pSTAT5 from Supplementary Figure 5 were used to calculate mean MFI FC in each position (p1 to p9). Mean and SD were also calculated for all stimulated or unstimulated specimens in order to calculated ranges and ratios of variability. Values outside these ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in light red.

Supplementary Figure 9. 1-way ANOVA and Pearson analysis of percent of positive cells across operators. Percent of positive CD3+, CD8+, CD20+, and CD14+ cells were obtained from matrixes processed by three different operators and compared by 1-way ANOVA. Values from stimulated samples were also correlated to those from matched unstimulated specimens by Pearson analysis in order to assess inter-assay variability.

Supplementary Figure 10. Computational gating strategy. (A) Debris were removed using rectangleGate function of flowCore. After clustering using flowClust package, lymphocytes and monocytes were identified using the SSC-A values as cut-off, and subsequently, deconvolution carried out using flowClust (B). On each barcoded population, unsupervised clustering was employed for the identification of CD3+ and CD20+ cells, and then CD4+ and CD8+ populations on CD3+ cells. (C) On monocytes, deconvolution was first performed, and next CD14+ monocytes gated using rectangleGate function of flowCore. Cut-offs for CD14-PE and SSC-A were set at the 1st quartile for each parameter.

Supplementary Figure 11. Surface marker intra-assay variability for percent of positive cells by computational analysis. Corresponding raw data of Figure 5 for assessment of intra-assay variability for percent of positive cells by computational analysis. Mean of percent of positive cells and SD were calculated for each position in a matrix (pi; p1 to p9) among all matrixes with unstimulated samples (total of nine values for each parameter; N) or stimulated samples. Ranges and ratios of variability were calculated as described in Figure 1. Values outside these ranges (< −2SD or > +2SD; or < 0.8 or > 1.2) were highlighted in light red.

NIHMS1545413-supplement-9.pdf (39.9MB, pdf)

RESOURCES