Phenotypic Complexity of the Human Regulatory T Cell Compartment Revealed by Mass Cytometry

Regulatory T cells (Treg) are an essential component of the healthy immune system, limiting host mediated immunopathology caused by the host in response to a diverse range of immunological and inflammatory challenges. Tregs constitute 5–10% of the healthy human peripheral blood CD4⁺ T cell compartment (1) and are routinely identified by the use of a combination of phenotypic markers: CD4, CD25, and the canonical transcription factor FOXP3 (2, 3). Tregs maintain peripheral tolerance by a number of direct and indirect suppressive mechanisms, targeting a wide variety of immune effector cell types (4–6). A reduction in the number or functional capacity of Tregs can disrupt the delicate immunological balance between immune response and regulation and result in immune-mediated pathology and autoimmunity (7, 8). Intensive research has helped inform a consensus of the critical role Tregs play in both health and disease. The clinical manipulation of Tregs is currently under intense investigation in autoimmunity, transplantation, graft-versus-host disease, and anticancer therapy (9–13). However, it is not possible to isolate viable Tregs based on FOXP3 expression (given its intranuclear location). Current protocols rely on isolation of highly pure Tregs defined as CD4⁺CD25^highCD127^low cells for in vitro studies focused on Treg biology and functionality (10) or as the starting material from which to expand Tregs for adoptive cell therapy, which involves the isolation and expansion of Tregs from an individual followed by reinfusion (9, 11, 13). However, it is now clear that in humans, Tregs are in fact a heterogeneous mixture of cellular subphenotypes reflecting different states of maturation, differentiation, and activation (reviewed in Refs. 14–16). Several studies have identified cellular markers differentially expressed within the FOXP3⁺ T cell (and CD25^hiCD127^lo) population that delineate cells with distinct developmental states, capacities, and methods of suppression, homing properties as well as targets of suppression. For example: expression of CD45RA and CD25 delineates naive and Ag-experienced Tregs (17); ICOS delineates Tregs that produce IL-10 or TGF-β (18); and CD39/CD73 identifies Tregs capable of cAMP- or adenosine-mediated suppression (19). Furthermore, it has recently been suggested that different populations of Tregs that mirror Th cell subsets by specific expression of chemokine receptors may be involved in selectively suppressing their effector T counterpart (20). It is therefore possible that different Treg subpopulations may be best suited for treating different conditions, and it is therefore essential to gain a full understanding of the constituent subcellular phenotypes present in the Treg population.

To date, such studies have been hampered by a number of technical limitations. Standard flow cytometric methods using fluorophore-conjugated Abs can measure multiple parameters at the single-cell level; however, the number of parameters that can be used simultaneously is limited due to spectral overlap. Conversely, transcriptional profiling increases the number of parameters measured, but lacks single-cell resolution (21). The recent advent of mass cytometry offers single-cell resolution using metal-conjugated Abs bound to cells that are then atomized, ionized, and finally detected by time-of-flight mass spectrometry (22). Mass cytometry avoids limitations associated with spectral overlap and can theoretically detect in excess of 40 parameters per cell, offering increased resolution at the single-cell level.

In this study, we have used a panel of metal-conjugated Abs specific for surface markers associated with Treg function and phenotype, in conjunction with mass cytometry to interrogate the Treg population at high resolution. The results have been analyzed using both traditional biaxial plot-based gating and unbiased bioinformatic approaches to interrogate the multidimensional characteristics of the human Treg compartment. Our data reveal a high degree of phenotypic complexity within this defined regulatory population. Additionally, we show that whereas classical biaxial gating of such multidimensional data restricts interpretation of such datasets, a combination of viSNE (23) projection and FLOwClustering without K (FLOCK) (24) clustering can be used for assessing novel cell populations by mass cytometry from a phenotypically complex population of cells. Using these analyses, we show the presence of 22 distinct subpopulations of Tregs, which were reproducibly identified in multiple individuals. This study demonstrates for the first time, to our knowledge, the multidimensional phenotypic complexity of the human Treg compartment at high resolution.

PBMCs were prepared from leukophoresis cones from four anonymous donors by Ficoll-Paque density-gradient centrifugation (Lymphoprep; Axis-Shield, Oslo, Norway). PBMCs were stained with anti-CD4 (L200) PerCP, anti-CD25 (2A3) PE, and anti-CD127 (HIL-7R-M21) Alexa Fluor 647 and sorted on a BD FACSAria II flow cytometer and FACSDiva software (BD Biosciences) to obtain pure populations of CD4⁺CD25^highCD127^low (Treg) from each sample.

PBMC stained with anti-CD4 (L200) PerCP, anti-CD25 (2A3) PE, anti-CD127 (HIL-7R-M21) Alexa Fluor 647, and anti-CD45RA (L48) FITC and were then treated with the FOXP3 staining buffer kit (eBioscience) according to the manufacturer’s instructions. FOXP3 was stained using anti-FOXP3 (236a/E7) v450 (BD Biosciences) and anti-FOXP3 (PCH101) e450 (eBioscience). Samples were then acquired on a BD FACSAria II flow cytometer using FACSDiva software (BD Biosciences) and files analyzed using FlowJo (Tree Star).

Metal-conjugated Abs were obtained directly from DVS Sciences. Abs not available prelabeled were made in house using purified Abs obtained from BioLegend and labeled in house using MaxPar labeling kits (DVS Sciences), according to the manufacturer’s instructions. Alternatively, when both purified and metal-conjugated Abs were unavailable, PE- or allophycocyanin-conjugated primary Abs (BioLegend) were used in conjunction with anti-PE or anti-allophycocyanin metal-conjugated Abs (DVS Sciences), respectively. The full staining panel is shown in Supplemental Table I.

Sorted cells were transferred to Eppendorf tubes, centrifuged at 500 × g for 5 min, washed in 1 ml staining buffer (500 ml low-barium PBS and 2.5 g protease-free BSA; Sigma-Aldrich), and then the cell pellet was incubated in 500 μl 1× intercalator (DVS Sciences) in low-barium PBS, for 15 min at room temperature, to allow for dead cell exclusion. Postincubation, cells were washed twice in cell 1 ml staining buffer, prior to resuspending the cell pellet in 10 μl Kiovig Fc block (5 mg/ml; Baxter) for 10 min at room temperature. Postincubation, cells were again washed in 1 ml staining buffer and stained initially with titrated fluorescently conjugated Abs for 30 min at 4°C in the dark. Postincubation, cells were washed and stained with the panel of metal-conjugated Abs using previously titrated values, in a total volume of 100 μl, and incubated at room temperature for 30 min. Cells were then washed twice in staining buffer and the resuspended in 500 μl 1.6% methanol-free paraformaldehyde (Electron Microscopy Sciences) in PBS overnight at 4°C. Postincubation, cells were centrifuged 600 × g for 7 min. The supernatant was then aspirated, and the cell pellet was resuspended in 500 μl 1× Ir intercalator (DVS Sciences) in 0.3% saponin (Sigma-Aldrich) in 500 ml staining buffer and incubated for 20 min at room temperature. Postincubation, cells were washed once in 500 μl PBS and two subsequent washes in 1 ml Milli-Q water. Cells were then adjusted to 0.5 × 10⁶ cells/ml in Milli-Q water and passed through a 35–70 μmol mesh, prior to the addition of EQ beads (DVS Sciences) according to the manufacturer’s instructions. Samples were then acquired and analyzed immediately using a CyTOF mass cytometer as described previously and files exported in flow cytometry file (FCS) format for analysis of data.

Data were normalized using Rachel Finck's MATLAB normalizer (25). Normalized data were then gated in Cytobank according to the strategy in Supplemental Fig. 1D).

All viSNE analyses were performed using CYT (23). viSNE analysis used the Barnes-Hut implementation of the t-distributed stochastic neighbor embedding (tSNE) algorithm to reduce high-dimensional data into two dimensions, maintaining complex parameter relationships and single-cell resolution. The viSNE analysis and associated maps shown in Fig. 2 were generated using all events from all samples and 26 parameters. Unsupervised clustering was performed on the same dataset, using FLOCK, an unbiased model-independent, grid-based partitioning, and density distribution analysis that algorithmically identifies cellular populations within the high-dimension cytometric data (24). The analysis was performed using all events for each sample and the same 26 parameters used for the viSNE analysis with the addition of both tSNE dimensions generated by viSNE. Details of each individual viSNE and FLOCK analysis, including parameter inclusion and cell number, are shown in Supplemental Table II. Frequency tables and heat map plots were generated in R using custom scripts.

Biaxial gating of previously described Treg populations (Fig. 1) was compared with the FLOCK-generated clusters (Fig. 3). Comparative tables were generated in R to assess overlap between the biaxial gating of the five previously described Treg populations and the 22 populations identified by unsupervised clustering. Cell count values were normalized to the total number of cells within each manually gated population. Cell tables were then colored according to the enrichment p value computed using the Fisher exact test. A low p value indicated the corresponding FLOCK cluster is enriched for cells belonging to the corresponding biaxially derived subpopulation. The computed p values were adjusted for multiple testing using Benjamini-Hochberg correction. Adjusted p values <0.05 were deemed significant.

PBMCs prepared from leukapheresis cones were used to isolate Tregs based on the expression of CD4⁺CD25^highCD127^low by FACS according to the gating strategy in Supplemental Fig. 1A. Treg defined in this manner comprised 5.6–8.0% of the peripheral blood CD4⁺ T cell compartment (mean 7.0%; n = 4) (Supplemental Fig. 1B). We next determined the expression of the canonical Treg transcription factor FOXP3 within this population by flow cytometry (Supplemental Fig. 1C). The range of FOXP3 expression within the CD4⁺CD25^highCD127^low population was 87.4–92.0% (mean 89.4%; n = 4).

We next used a panel of heavy metal-conjugated Abs specific to a range of markers associated with human Tregs (Supplemental Table I) and mass cytometry to define the human Treg compartment in higher resolution than previously possible. Data sets were normalized using the inclusion of a bead control within each acquisition shown in Supplemental Fig. 1D and gated according to the strategy presented in Supplemental Fig. 1E for all further analyses.

A number of phenotypic markers have been previously used to define and characterize subsets within the human Treg compartment, including CD45RA (17), CCR4 (12), CD39 (26), HLA-DR (27), and CD161 (28, 29). We initially sought to confirm that these previously defined populations identified by standard flow cytometry were also observable using CyTof technology. All five previously described Treg subsets were reproducibly detected in each sample examined (mean percentage of Treg compartment: CD45RA, 15.75; CCR4, 38.51; CD39, 28.92; HLA-DR, 7.27; and CD161, 4.45% (n = 4) (Fig. 1A, 1B). Furthermore, detailed characterization of these 5 established Treg subsets using all 25 analysis parameters revealed characteristics consistent with published reports and known features of these subsets. Such analysis also provided enhanced and novel phenotypic detail regarding the composition of these defined Treg subpopulations in terms of the 25 analysis parameters shown in Fig. 1C. For example, reduced CD45RA expression was noted in the CCR4, CD39, HLA-DR, and CD161 Treg subpopulations, consistent with the notion that these are Ag-experienced populations. Conversely, CD45RA⁺ Tregs showed reduced expression of CD147 (30) and CCR4, as expected. These observations suggest that this approach identifies cell populations determined by standard flow cytometric methodologies and thereby validates its application in this context.

The opportunity to interrogate the human Treg compartment with an extended panel of markers on a single-cell basis requires specific tools to maximize the interpretation of the n-dimensional nature of the data set. viSNE has been recently shown to allow simultaneous characterization of high-dimensional data while maintaining single-cell resolution, projecting cells on a biaxial plot in terms of how similar they are based on all parameters simultaneously (23). Consequently, we used viSNE to analyze the composition and expression profile of the human Treg compartment. Initially, viSNE analysis was performed on all events from all four samples pooled and analyzed together to produce a consensus map of the Treg compartment (Fig. 2A). The overall profile of individual maps for each sample was remarkably similar to that of the initial consensus plot (Fig. 2B). However, although similar in terms of the general shape and arrangement of cells, the contours within the viSNE map from each individual sample revealed different densities of cells within different portions of the plots, suggesting variation in the relative proportions of Treg subpopulations between individuals.

We next examined the intensity of expression of each individual parameters within the consensus viSNE map using color as a third dimension (Fig. 2C). Expression patterns of these Treg-associated markers showed a dynamic range of expression profiles that we could describe as discrete, clustered, or diffuse. Expression of some markers were restricted to cells localized within a discrete single region of the viSNE map (e.g., CD73 and CD161); other markers displayed a clustered appearance, with high expression localized in several distinct areas of the map (e.g., latency-associated peptide [LAP], CD103, CD31, CCR4, CXCR3, and CD39), whereas other markers showed a more diffuse pattern of expression (e.g., CD27, CD62L, and CD38). Interestingly, the expression of LAP and CD31 were associated mainly within the CD45RA⁺ naive subset. In contrast, CD103 expression was present in both the CD45RA-negative and -positive populations. A clear gradient of CD147 expression was observed, inverse to the direction of CD45RA expression. Additionally, we generated viSNE plots to examine parameter expression patterns for each individual sample (Supplemental Fig. 2A–D).

Clearly, our data show the human peripheral blood Treg compartment is phenotypically complex and heterogeneous. Next, we asked whether the Treg compartment could be rationally divided into distinct populations of Tregs based on expression of all analysis parameters. To avoid bias due to assumptions based on previous knowledge, cluster analysis was performed in an unsupervised manner using FLOCK (24). This analysis used the same 26 analysis parameters used to produce the viSNE plots in Fig. 3, in addition to the two dimensions (tSNE) generated by the dimensionality reduction software. Clusters identified by FLOCK were then projected back into viSNE and subpopulation analysis performed (Fig. 3). FLOCK analysis revealed a total of 22 Treg subpopulations within the data set (Fig. 3A). Each subpopulation was present in each of the samples analyzed (Fig. 3B). Interestingly, the majority of these subpopulations were observed as a similar proportion of the total Treg population in all four samples; however, others showed a greater degree of interindividual variation (e.g., cluster 5 and cluster 20).

We next assessed the median expression of all 25 analysis parameters for each of the 22 FLOCK-generated clusters to visualize phenotypic characteristics of each cluster and investigated the relationship between clusters by unbiased hierarchical analysis (Fig. 3C). Notably, in this hierarchical analysis, the first separation is driven by differential expression of CD45RA and discriminate previously described naive (CD45RA⁺) from Ag-experienced (CD45RA⁻) Treg populations. However, each of these two large clades is also divided into multiple subclades and leaves based upon differential expression of multiple markers. To enable future functional investigation of Treg populations in this study by FACS, we sought to elucidate Boolean gating strategies capable of defining individual clusters we identified in this study by FLOCK (Supplemental Fig. 2E, 2F). For 11 Treg clusters, gating strategies were determined down to individual cluster level. For a further nine clusters, we identified gating strategies capable of defining three highly related leaves of the dendrogram, each composed of three individual clusters. No simple gating strategy could be determined for the remaining two clusters (clusters 1 and 18). An example of the gating strategy from one representative donor (donor 1) is also shown (Supplemental Fig. 2F).

Traditional gating analysis of high-resolution multiparameter data using biaxial plots is limited, as it only examines two parameters at a time. In contrast, viSNE examines all parameters simultaneously and, in conjunction with FLOCK, can be used to project the results of the composition of a cell population based on the expression of all parameters examined simultaneously, gauging a better description of the cellular composition of a sample.

Consequently, we next examined the relationship between classical biaxial gating of the five previously described Treg populations (shown in Fig. 1) to the 22 Treg populations identified by viSNE-FLOCK clustering analysis (shown in Fig. 3). The overlap between both analyses is shown as the enrichment p value computed using the Fisher exact test, in which p values were adjusted for multiple testing using Benjamini-Hochberg correction (adjusted p values <0.05 were deemed significant). A perfect correlation between the analyses would appear as an overlap box representing 100% of identified events. Treg populations identified by biaxial analysis are generally more heterogeneously composed and do not map to a single clustered population generated by viSNE-FLOCK (Fig. 4). However, the overlap between the biaxial gating and the clustered analysis does show a high degree of significance for some Treg populations. For example, the CD161⁺ Treg population, is almost entirely represented by three clusters; (7, 10, and 17; totaling 93%), with the vast majority of CD161 expression represented by cluster 7 (86%). Similarly, CD45RA⁺ (naive) Tregs were largely represented by 6 clusters (totaling 95%), comprising of 3 major clusters 19 (29), 20 (22), and 21 (38%), and 3 minor clusters 1 (3.8), 4 (1.6), and 22 (0.9%). Interestingly, four of these populations clearly cluster together at the bottom of the dendrogram (clusters 19, 20, 21, and 22), whereas the other two populations form a distinct bifolious clade (clusters 1 and 4), well separated from the other naive Tregs (Fig. 4B). Similarly, the HLA-DR subpopulations of Tregs were mainly represented in 6 clusters (totaling 84%), including 4 major, clusters 6 (14.67), 11 (35.27), 12 (25.13), and 13 (5.89%), and 2 minor, clusters 15 (2.48) and 22 (0.41%). In contrast, Treg populations defined by gating on CCR4- and CD39-positive cells were represented in multiple clusters (CCR4, 12 clusters; CD39, 11 clusters), suggesting these are highly heterogeneous populations.

Adoptive Treg therapy is an attractive emerging treatment for controlling aberrant immune responses in autoimmunity and transplantation. In this procedure, Tregs are isolated from a patient, expanded in vitro, and then reinfused (9, 11). Because its intranuclear location precludes the use of FOXP3 as a marker, isolation of Tregs for clinical use is achieved based on a surface phenotype of CD4⁺CD25^highCD127^low (10). In the current study, we have isolated Tregs in this manner and examined the expression of an extensive panel of lineage-specific surface markers by mass cytometry to determine the high-dimensional phenotypic composition of the Treg compartment at a single-cell level, revealing unprecedented phenotypic complexity and diversity and describing distinct subpopulations, providing initial insight into Treg subset composition and heterogeneity.

The use of biaxial analysis in multidimensional datasets uses only two parameters at a time and as such may miss important characteristics and underestimate the complexity of high-dimension datasets produced by mass cytometric methods. Analysis by viSNE allows projection of multiparameter data on a two dimensional plot, while maintaining single-cell resolution and both the multidimensionality and complex parameter relationships of the data (23). The analysis of multidimensional data by viSNE has been previously applied to human bone marrow, where the authors showed excellent separation and distinction of the constituent cell populations. Although purified Tregs may be thought of as a highly related group of cells, the application of visNE to our dataset showed good separation of distinct subpopulations of Tregs. Additionally, we used a combination of FLOCK (24) and viSNE to perform cluster analysis, revealing 22 individual clusters within our dataset. Similar to Amir et al. (23), we found that FLOCK was better able to identify populations more reproducibly and objectively than manual identification of Treg clusters, which was more cumbersome, speculative, and time consuming.

Our data show nonuniform expression of the panel of Treg markers we examined. Instead, we demonstrate a complex and diverse expression profile for the majority of parameters analyzed. In some cases, high levels of expression of a particular marker delineated a single population of cells. For example, CD73 and CD161 identified distinct Treg subpopulations, clustering far apart in the viSNE analysis (Fig. 2). In other instances, high levels of expression of a particular marker were present in multiple subpopulations of cells identified in the viSNE analysis (e.g., CD39, CD103, and CXCR3). Other markers showed a more graded expression profile across the viSNE plot. For example, we observed a gradient of CD147, CCR4, and ICOS expression antiparallel to that of CD45RA. We also observed low-level expression of markers previously associated with Tregs such as CTLA4 (21, 31), in which our data show only a small number of cells constitutively expressed CTLA4 at the cell surface. This could be due to using cells directly ex vivo in our study, with no stimulation performed and restricting our phenotypic analysis to surface-bound expression, which would not detect preformed CTLA4 present in intracellular vesicles or those trafficked to the cell surface after T cell activation, which has been shown in human Tregs previously (32, 33).

A number of reports have shown that Tregs are composed of subpopulations based on a more limited range of phenotypic markers (12, 17, 20). Using CD45RA and FOXP3 expression, Miyara et al. (17) demonstrated a population of naive Tregs and two populations of memory Tregs that exhibited differential methylation of the TSDR region, a measure of FOXP3 constitutive expression and, as such, Treg stability. As expected, our data showed that the naive Treg compartment (CD45RA⁺) did not express markers associated with Ag-experienced Tregs [CCR4 (12), CD39 (19, 26, 34), and HLA-DR (27, 35)] and displayed reduced expression of a number of other markers (ICOS and CD147). However, the inclusion of multiple markers revealed that the CD45RA⁺ population of Tregs is composed of a number of subpopulations itself. These CD45RA⁺ subpopulations were mainly differentiated based on differential expression of CD31, CD103, CCR10, and LAP. CD31 expression in Tregs has been shown on cells phenotypically and molecularly defined as recent thymic emigrants (36, 37). Accordingly, hierarchical clustering performed with our data showed CD31 expression was entirely restricted to cluster 22, and represented the initial branching of the CD45RA⁺ subpopulations (clusters 19–22) to reveal recent thymic emigrant Tregs. In contrast, both CD103 and LAP expression in Tregs have been shown in activated cells, denoting Ag-experienced Tregs (38–40). The presence of these subpopulations within the CD45RA⁺ fraction may reflect truly naive Tregs preprogrammed toward different regulatory lineages. Alternatively, the continuum of CD45RA expression observed in our analyses, in conjunction with CD103 and LAP expression, may represent transient expression of phenotypic markers as Tregs differentiate.

The expression of CCR4 within Tregs has been previously designated as effector Tregs that express high levels of FOXP3 and low levels of CD45RA (12). Similarly, in our study we observed CCR4 expression on Tregs mainly restricted to CD45RA⁻ cells. However, we also identified multiple subpopulations of CCR4⁺ cells that can be distinguished based on expression of a range of markers including HLA-DR, CD39, and CD73, suggesting specialization of Tregs suited to function in response to a range of immune insults, immunological environments, or scenarios, as has been postulated by others (20).

Interestingly, phenotypically distinct populations of Treg have been described that reflect their CD4⁺ effector T cell counterparts (20). These functionally suppressive bone fide Treg, coexpressed cytokines, transcription factors, and chemokine receptors classically associated with Th effector lineages, including Th1-like Treg cells that expressed CXCR3. Similarly, our data also show the presence of CXCR3-expressing Treg in two distinct clusters 5 and 13, differing in the expression of CD38 and CD39, potentially reflecting further specialization within the Th1-like Treg compartment. During preliminary experiments, we observed low-intensity CCR6 staining (data not shown), below the resolution required for further analysis and was insufficient for inclusion of CCR6 in the study we present in this study. Consequently, this prevented analysis of Th17-like Treg (CCR6⁺CCR4⁺), Th22-like Treg (CCR6⁺CCR4⁺CCR10⁺), and Th2-like Treg (CCR6⁻CXCR3⁻CCR4⁺) also identified by Duhen et al. (20).

Tregs capable of producing proinflammatory cytokines such as IL-17 have been described in both health and disease and, importantly, throughout immunological development and associated with the expression of CD161 (29). Furthermore, we have shown that the IL-17–producing, CD161⁺ subpopulation of Tregs remain stable in terms of maintaining regulatory function in a proinflammatory environment (28). Our analysis in this study revealed discrete clustering of CD161 expressing Tregs into one major (cluster 7) and one minor (cluster 21) subpopulation. Cells within cluster 21, but not cluster 7, also included some CD45RA expression, which recapitulates what we have shown previously regarding CD161 expression within both CD45RA⁺ and CD45RA⁻ population of Tregs (28). These observations raise the possibility that CD161-expressing Tregs may be preprogrammed prior to contact with their Ag for the first time.

We also observed the expression of CD39 by a number of Treg subpopulations. CD39 is an ectoenzyme (nucleoside triphosphate diphosphohydrolase-1) that degrades ATP to AMP and is expressed on a subset of highly suppressive memory-like Tregs (19, 26). In murine Tregs, CD39 is coexpressed with CD73, an ectonucleotidase, which results in the collaborative extracellular degradation of ATP to adenosine (34, 41). In contrast, human Tregs can independently express CD39 and CD73 (26). Our data show CD73-expressing Tregs that do (cluster 16) or do not (cluster 4) also coexpress CD39.

Tregs are known to mediate suppression via a wide array of contact-dependent and -independent mechanisms (4, 42, 43). Due to the nature of mass cytometry (vaporization of the cell sample), functional characterization of the Treg subpopulations we identified was not possible in the current study. It is possible that particular subpopulations of Tregs identified may use distinct mechanisms of suppression or preferentially mediate suppression of different effector populations. Similarly, subpopulations identified in this study may represent various Treg developmental states or stages of commitment to the regulatory lineage. These represent pivotal questions in understanding the immunobiology of the Treg compartment that will require both application of conventional flow cytometry and the use functional assays of regulatory function and stability.

Treg frequency in the peripheral blood is known to increase (44, 45) or decrease (7, 8) in the context of a range of clinical conditions or immunomodulatory therapies (46–48). The application of the methods and analysis presented in this current study in the context of autoimmunity, infectious diseases, transplantation, and cancer could potentially accelerate our understanding of the roles of Treg population composition during particular disease states or in response to therapeutic interventions. Future application of mass cytometry to the human Treg compartment in both health and disease will require greater subject numbers than presented in this preliminary study, which represents the first deep immunophenotypic examination of human Treg. The inclusion of increased numbers of subjects in future studies will support observation and establishment of potential disease-specific differences in the composition of the human Treg compartment. Similarly, the clinical application of in vitro–expanded Tregs is under intensive investigation in the context of a wide spectrum of human diseases including autoimmunity, transplantation, and graft-versus-host disease (9–11). The investigatory approaches we present in this study could be pivotal to completely profiling the Treg compartment and identifying the optimal population or combination of subpopulations that may be more efficacious within the context of a particular immune insult, provide a stratified approach with regards to Treg cell therapy, and, importantly, further our understanding of Tregs during both health and disease.

We thank Tom Hayday, Helen Graves, and P.J. Chana for technical assistance in cell sorting.

The authors have no financial conflicts of interest.