Circulating Lymphocyte Subsets in Normal Adults are Variable and Can Be Clustered into Subgroups

Abstract

Background:

Flow cytometry is used to monitor lymphocyte subsets in both the clinical and research settings. An understanding of the degree of inter- and intrasubject variability of these populations is critical for data interpretation.

Methods:

Peripheral blood lymphocytes of 18 healthy adults were analyzed on two separate occasions using a multicolor flow cytometric panel with B, T, and NK cell markers. Variability was calculated using the coefficient of variation and compared between and within individuals using agglomerative clustering.

Results:

Each subject appears to have B and T cell subset profiles that are stable over the two time points, but differ from the profiles of other subjects. Thus, the range of measurements for a particular B or T cell subset is larger between subjects and narrower for an individual. In addition, the level of variability correlates inversely with the size of the lymphocyte subset. When lymphocyte profiles are analyzed by agglomerative clustering, replicate samples from the same individual tend to cluster. When single samples from different individuals are analyzed, individuals appear to cluster into different subgroups.

Conclusions:

Variability of lymphocyte subsets is usually greater between individuals than within a single individual and each person appears to have a characteristic profile of lymphocyte subsets. These results underscore the importance of obtaining a baseline value for each subject when investigating the impact of a treatment on lymphocyte subsets over time. These results also highlight the potential utility of cluster analysis as a tool for immune subset profiling and biomarker discovery.

Flow cytometry is often used to monitor lymphocyte subsets in subjects receiving immunomodulatory therapies. In this regard, the average proportion or absolute frequency of a lymphocyte subset is usually compared in groups of individuals before, during, and after treatment. Such comparisons assume that the differences observed between groups are greater than the differences between individual subjects within a single group. Yet, as we will demonstrate here, the level of interindividual variability for lymphocyte subsets is considerable and increases even further in smaller lymphocyte populations and with the use of absolute cell counts instead of percentages. In contrast, the level of intraindividual variability is lower. These findings indicate that intergroup comparisons potentially have limited sensitivity for the detection of biologically meaningful effects. These data support alternative study designs, such as longitudinal intraindividual analysis (i.e., fold change relative to a baseline measurement), and cluster analysis for the analysis of immunophenotyping data.

METHODS

Human Subjects

Peripheral blood lymphocytes were obtained from 18 apparently healthy controls. This group of subjects was used as a control group in a separate unrelated study evaluating B cell subsets and antibody repertoire in patients with systemic lupus erythematosus (SLE) (1). Subject demographics are summarized in Table S1 of the electronic supplement. The same immunophenotyping panel was performed on the healthy study subjects analyzed in this report at two separate time points ranging from 2 to 7 months apart. The first time point was used for comparison to subjects with lupus (described in Ref. no. 1) and the second time point was used for comparison to the first time point in this study. Subject inclusion criteria include good health and demographic characteristics (age, sex, and race) that matched fairly closely with our cohort of SLE patients. Exclusion criteria included malignancy, HIV disease or other acute or chronic viral infection, therapy with immunosuppressive agents including steroids and a history of autoimmune disease. Peripheral blood was obtained by routine venipuncture into sodium EDTA tubes. In parallel, blood samples were drawn and processed for absolute lymphocyte counts by Coulter counting (complete blood count with electronic differential) at the Hospital of the University of Pennsylvania. Informed consent was obtained from all patients. These studies were performed in accordance with a protocol that was approved by the University of Pennsylvania Institutional Review Board.

Flow Cytometry Methods

One hundred fifty microliter aliquots of the whole blood (<24 h old, stored at room temperature) were stained and processed for flow cytometry as described previously (2). Cells were stained with the following antibodies: anti-CD3-APC-AF750 (S4.1), anti-CD4-PETexas Red (S3.5), anti-CD5-PE-Cy5.5 (CD5–5D7), anti-CD19-Pacific Blue (SJ25C1), anti-CD24-PE-AF610 (SN3), anti-CD38-APC-Cy5.5 (HIT-2) [Invitrogen]; anti-CD3-Pacific Blue (UCHT1), anti-CD8-PE (HIT-8), anti-CD8-Pacific Blue (RPAT8), anti-CD10-PE (HI10A), anti-CD14-Pacific Blue (N5E), anti-CD16-PE-Cy7 (3G8), anti-CD19-PerCP-Cy5.5 (SJ25C1), anti-CD19-PE-Cy7 (SJ25C1), anti-CD20-PerCP-Cy5.5 (L27), anti-CD27-APC (MT271), anti-CD38-PE (HIT-2), anti-CD45-PerCP (2D1), anti-CD56-APC-AF700 (B159), anti-CD62L-PE-Cy5 (Dreg56), anti-IgD-FITC (IA62), anti-Igκ-PE (TB28–1), anti-IgM-FITC (G20–127), anti-IgM-APC (G20–127) [BD Bioscience]; anti-NKG2D-FITC (1D11), anti-CD23-APC (EBVCS2), anti-CD27-APC-AF750 (0323) [eBioscience]; and anti-NKp44-APC (Miltenyi Biotec; 2.29); 9G4-PerCP-Cy5.5 (kindly provided by M. Sanders, Dynavax Technologies Corporation, Berkeley, CA). The 9G4 antibody recognizes B cells that express V_H4–34 antibody heavy chains and was included as part of our earlier study on SLE patients (Ref. no. 1) because of its association with autoreactive B cells in SLE. Dead cell exclusion was performed using the Live/Dead Fixable Violet Dead Cell Stain Kit (Invitrogen). All 11-color flow cytometry (both samples from each of the 18 subjects) was performed on the same instrument (Becton Dickinson LSRII Benchtop Flow Cytometer) to control for machine variability. Three-color flow cytometry was performed in pilot experiments to compare 11-color to three-color data and evaluate gate placements (Fig. S1 in the electronic supplement). The three-color experiment shown in Figure S1 was performed on the Becton Dickinson FACS Calibur. The sample preparation and instrument operation were performed by the same single operator at all time points to minimize interoperator variability. Cytometer set-up and tracking beads (BD CST) were used daily in combination with BD FACSDiva software (v6.0 or later) by flow cytometry core facility staff to monitor the mean fluorescent intensities and CVs in each of the detectors. PMT voltages were adjusted accordingly and baselines were re-established with new bead lots. Compensation was performed digitally using cells stained with single color reagents. Typically, ~10,000 B cells were acquired per tube. Minimum, median, and mean event count data for the lymphocyte subsets analyzed in this study are presented in Table S2 of the electronic supplement.

Flow Cytometry Data Analysis

Flow cytometry data were analyzed using FlowJo (v. 8.2, Treestar Inc., Ashland, OR). Data analysis was performed by one person and checked by two others. For T cell subsets and NK cell enumeration, gate statistics and the same combination of antibodies from a single tube were used for all of the samples. After excluding dead cells and gating on lymphocytes utilizing forward scatter and side scatter, T cells were identified based on CD3 expression and T cell subsets were defined based on CD4 and CD8 expression into CD4+, CD8+, double positive (CD4+, CD8+), and double negative (CD4−, CD8−). NK cells were identified as CD3 negative, CD56, and/or CD16 positive lymphocytes. B cells were identified based on CD19 expression and B cell subsets were defined as transitional, naïve mature, mature activated, resting memory, and plasmablast populations, using CD27 and CD38 expression as described in the legend to Figure 2. Fluorescence minus one controls and marker expression on other cell types (such as CD27 staining on T cells) were used to define positive versus negative staining in pilot experiments. Three-color and 11-color flow cytometry was performed on healthy subjects and analyzed to establish the reliability of our B cell gating scheme, as shown in Figure S1 of the electronic supplement. The percentages of naïve mature, mature activated, and plasmablast subsets were comparable for the three-color and 11-color data, whereas the percentages of transitional cells tended to be slightly higher and percentages of resting memory cells slightly lower in the 11-color analysis. To minimize variation due to potential differences between the three-color and 11-color panels, all of the subsequent lymphocyte immunophenotyping studies were performed using the 11-color panel. For the analysis of the 18 subjects, B cell subsets were analyzed in two different tubes. The first B cell tube contained antibodies against IgD-FITC, IgM-APC, CD38-APCCy5.5, CD27-AF750, Igκ-PE, CD5-PECy5.5, 9G4-PerCP, CD19-PECy7, a dump channel, and a viability marker. The second B cell tube contained antibodies against IgM-FITC, CD20-PerCP, CD23-APC, CD38-APCCy5.5, CD27-AF750, Igκ-PE, CD24-PE-AF610, CD5-PECy5.5, CD19-PECy7, a dump channel, and a viability marker. Only tube 1 was run in the first time-point (except for subjects 17 and 18), and both tubes were run in parallel for all subjects at the second time point. Both tubes use the same antibody-fluorophore combinations for CD19, CD27, CD38, CD5, anti-kappa, viability (Invitrogen aqua), and the dump gate (CD3, CD8, and CD14 (all Pacific Blue)). Tube 1 was used to define the B cell subsets in all of the subjects for the first time point except subjects 17 and 18. For the second time point, tube 2 was used in subjects 2, 3, 4, 5, 6, 8, 9, 10, 14, 15, 16, 17, and 18 and tube 1 was used for the remaining subjects. When both tubes were available, the tube that provided the clearest discrimination between the B cell subsets was chosen for analysis. Gating was performed in an unbiased fashion for each of the samples. Gates were not copied within the same subject or between subjects. The placement of the gates was confirmed by analysis of other markers including CD24, CD5, and IgM. For example, in the case of IgM, naive mature cells express intermediate levels, whereas mature activated cells contain a mixture of IgM⁺ and isotype-switched (IgM⁻) cells (Fig. S1C). In a similar fashion, fluorescence intensity levels of CD24 and CD5 were used to distinguish transitional from naïve mature B cells. Transitional cells have higher levels of both of these markers than naive mature B cells. Absolute T, B, NK cell and B and T cell subset counts were obtained by multiplying the absolute lymphocyte count (obtained from a complete blood count) by the percentage of lymphocytes having a particular phenotype (obtained by flow cytometry).

Fig. 2.

B cell subsets in replicate samples from normal subjects. A: B cell subsets are plotted on the y-axis as percentages of CD19+ lymphocytes. B cell subsets are defined on the basis of CD27 and CD38 staining as follows: Transitional CD27−CD38++, Naïve Mature CD27−CD38+, Mature Activated CD27+CD38+, Resting Memory CD27+CD38−, and Plasmablasts CD27++CD38++. Subject numbers are given on the x-axis. Replicate samples are shown for each subject. B: B cell subsets (as defined in Fig. 2A) are plotted on the y-axis as a percentage of CD19+ lymphocytes. Dots indicate individual measurements. Vertical lines link the replicate measurements from each subject. Horizontal lines intersecting the vertical lines indicate the mean for each subject. The range of B cell subset percentages for all of the subjects is represented by the shaded box extending from the smallest individual mean to the greatest individual mean measurement.

Variability Analysis

Interindividual variability of a particular subset over time was defined using the coefficient of variation (CV), calculated as (standard deviation/mean of a particular subset) × 100. An unequal variance Student’s t test was used to compare the significance of the difference of within versus between individual pair differences. Variability was also evaluated using agglomerative clustering. Cluster analysis was performed in the R statistical programming environment (3) using the algorithm agnes (4). Cluster analysis on paired samples was performed on three different representations of the same data set:standardized percentages, percentages (defined above), and absolute cell counts (defined above). Standardization for each subset percentage consists of subtracting the arithmetic mean of the subset and dividing by the mean absolute deviation of the subset. The mean absolute deviation is the mean value of the absolute differences of the values from their mean. Cluster analysis on single samples was performed on lymphocyte subset percentages and not subjected to standardization.

Assessment of the Likelihood of Replicate Pairing

A Euclidean distance-based phylogenetic tree was built using the UPGMA algorithm with mean distances. Within the tree, we counted how many of the first neighbor couples corresponded to replicate samples from the same subject (hereafter referred to as a “pair”). The significance of this number was computed using a label mixing technique. The labels of all samples were scrambled one million times and the number of first neighbors that corresponded to a pair was computed in each tree. The maximal number of pairs was 5 in one million scrambles (for all trees built in the current analysis). Thus, the probability of observing more than 5 pairs is lower than about 1 × 10⁻⁶.

RESULTS

The Average Intrasubject Variability Is Significantly Less Than Intersubject Variability

To evaluate how lymphocyte subsets vary within and between individuals, we performed immunophenotyping on peripheral blood of 18 healthy adults at two time points that were at least 2 months apart. The absolute and relative counts of the major lymphocyte subsets are shown in Figures 1A and 1B, respectively. Most subjects have fairly similar profiles at each of the two time points. The similarity in lymphocyte profiles within individuals is more pronounced when the relative frequencies of B and T cell subsets are evaluated (Figs. 2A and 3A, respectively). The range of subset frequency measurements tends to be narrow in replicate samples from the same individual, whereas it is more variable in different individuals. This is illustrated graphically in Figure 2B for B cell subsets and in Figure 3B for T cell subsets. Replicate measurements, indicated by black dots connected by lines, are fairly close together for most subjects, whereas the ranges of values for all of the individuals are further apart. However, because the number of observations is larger for the interindividual analysis (n = 36) than the intraindividual analysis (n = 2 per subject), these data will tend to show a narrower range of measurements within individuals than between individuals. We therefore wished to compare intra- and interindividual variability using a complementary method (agglomerative clustering) that was not confounded by this issue. Clustering determines the relative similarities of samples. By evaluating the frequency with which samples from the same individual cluster or pair together versus the frequency with which a sample clusters with a sample from a different individual, it is possible to evaluate the likelihood that paired samples tend to be more similar to each other than they are to the general population.

Fig. 1.

B, T, and NK cell numbers and percentages in replicate samples from normal subjects. A: Absolute cell counts. Stacked bar graphs of T cells (CD3+ lymphocytes), B cells (CD19+ lymphocytes), and NK cells (CD3−, CD56+, and/or CD16+ lymphocytes). Cell counts (in thousands per microliter) are plotted on the y-axis versus the subject number on the x-axis. Each vertical bar represents a single sample. Samples from two different time points are shown for each subject. B: Relative cell counts (Percentages). Stacked bar graphs of T cells, B cells, and NK cells expressed as a percentage of lymphocytes on the y-axis versus the subject number on the x-axis.

Fig. 3.

T cell subsets in replicate samples from normal subjects. Repeat measurements of basic T cell subsets expressed as a percentage of CD3+ lymphocytes are presented for all subjects. A: Each grouping represents one subject and each bar represents a single sample. Replicate samples are shown for each subject. B: CD4+, CD8+, Double Positive (CD4+, CD8+), and Double Negative (CD4−, CD8−) T cell subsets are plotted on the y-axis as a percentage of CD3+ lymphocytes. Dots indicate individual measurements. Vertical lines link the replicate measurements from each subject. Horizontal lines intersecting the vertical lines indicate the mean for each subject. The range of T cell subset percentages for all of the subjects is represented by the shaded box extending from the smallest individual mean to the greatest individual mean measurement.

Agglomerative clustering starts by considering each sample as its own cluster. In a pair-wise fashion, the most similar instances are joined together into a larger cluster recursively until a single cluster containing all samples is achieved. The result is displayed in the form of a dendrogram. Figure 4 shows the resulting dendrogram when standardized lymphocyte subset data from both time points are clustered for all of the subjects. Standardization controls for the sizes and variances of populations (see Materials and Methods). Under the dendrogram is a map showing the lymphocyte subsets color-coded based upon their values expressed in standard deviations relative to the mean. This figure shows that the replicate samples from 10 of the 18 individuals pair together (occupy neighboring branches in the dendrogram). The chance of obtaining 5 pairs out of 18 is less than one in a million if the observations are unrelated (see Methods). Therefore, these data suggest that observations from individual subjects are correlated. The map below the dendrogram shows that there are two ways in which samples from the same individual can cluster together. One is for the individual to have subsets that differ from the rest of the group. For example, subject 17 has multiple subsets that are more than two standard deviations from the mean of the group. Other individuals who have more dissimilar lymphocyte subsets include 16, 10, and 2. These subjects are all found on one end of the dendrogram, where there are long branches, illustrating their dissimilarity to the other subjects. The other, nonmutually exclusive way in which samples from the same individual can cluster together is if the lymphocyte subsets are very similar in both samples. A low level of intraindividual variability is the most likely explanation for clustering of the samples from subjects 9, 8, 12, 4, and 6.

Fig. 4.

Cluster analysis with standardized percentage data. Agglomerative cluster analysis was performed on the percentages of T, B, and NK cells and the T and B cell subset percentage data (as defined in the legends to Figs. 2 and 3) but the data were first subjected to standardization. Standardization for each subset percentage consists of subtracting the arithmetic mean of the subset and dividing by the mean absolute deviation of the subset as defined in Methods. The data are displayed graphically below the dendrogram using colored squares that correspond to the lymphocyte subsets, color-coded based upon their values expressed in standard deviations relative to the mean. Samples from each subject are represented by a vertical bar with different colored squares.

Next, we performed agglomerative clustering on the basis of relative versus absolute lymphocyte subset frequencies. Figure 5 shows a map of the subsets, color-coded based upon their relative frequencies, and the clustering dendrogram, which reveals that replicate samples from 12 out of the 18 individuals cluster together. For absolute cell counts, 7 out of 18 replicate samples from individual subjects cluster together (Fig. S2). Qualitatively, the dendrograms for standardized, absolute, and relative lymphocyte subset data appear similar to one another (some of the differences in branch order can be resolved by pivoting branches of the dendrogram). Nine out of the 10 individuals whose samples clustered together based upon the standardized data also clustered together with the unstandardized percentage data. Six of the seven subjects whose samples clustered together based upon the absolute count data also clustered together when relative lymphocyte subset percentages were used.

Fig. 5.

Lymphocyte subset profiles from individual subjects cluster together. Lymphocyte profiles were subjected to agglomerative cluster analysis as described in Methods. A: A dendrogram displaying the resulting clusters is shown. Individual subject numbers are shown at the base of the dendrogram and link the dendrogram to the colored graphical display below. The percentages of the lymphocyte subsets are displayed graphically using colored squares that correspond to different percentage ranges. Samples from each subject are represented by a vertical bar with different colored squares. B: Subsets of the lymphocyte data were used for agglomerative clustering. General subsets are defined in the legend for Figure 1. B cell subsets are defined in the legend for Figure 2A. T cell subsets are defined in the legend for Figure 3A.

The highest level of intraindividual clustering (12 pairs) was achieved with the unstandardized lymphocyte percentage data. To gain further insight into this result, we repeated the cluster analysis with subsets of the data: the major lymphocyte subset fractions, the B cell subsets and the T cell subsets (Fig. 5B). Subjects 3, 5, and 14 clustered on the basis of the general subsets. Subjects 14, 17, and 16 clustered on the basis of their B cell subsets and subjects 3, 12, 10, 11, and 17 clustered on the basis of their T cell subsets. When all of these data were used collectively (Fig. 5A), subjects 14, 15, 5, 2, 4, 6, 8, 11, 10, 12, 16, and 17 clustered together. Thus, while some subjects can be clustered on the basis of distinctive lymphocyte subsets, for others such as 15, 2, 4, and 6, clustering was only revealed in the full data set. On the other hand, subject 3 clustered in the general and T cell subsets, but not when all of the subsets were incorporated into the clustering algorithm. In this subject, the B cell subsets are highly variable and may have had the effect of “canceling out” the correlation seen amongst the general and T cell subsets. To determine the level of variability within each lymphocyte subset, we analyzed the coefficient of variation for the percentage and absolute lymphocyte counts.

Variability Is Greater in Small Lymphocyte Subsets and for Absolute Cell Counts

For the basic lymphocyte subsets expressed as percentages, the coefficient of variation for T cells is 9%, B cells is 49%, and NK cells is 44% (please see Table S3 in the electronic supplement). Amongst the lymphocytes, the major subsets, CD4+ and CD8+ T cells, are less variable than the minor subsets, such as CD4+CD8+ T cells, CD4−CD8− T cells, B cells, and NK cells. The level of subset variability is inversely correlated with the size of the population (Fig. S3A). When absolute frequencies are analyzed, the coefficients of variation are higher than the corresponding relative subset coefficients of variation (Table S3). As with relative frequency variation, absolute subset frequency variation is inversely correlated with the size of the lymphocyte subpopulation (Fig. S3B). The finding that absolute count data yielded fewer intraindividual pairs was consistent with a greater level of variability seen in absolute lymphocyte counts. Furthermore, clustering is skewed toward the larger cell populations, which exhibit less variability. For example, CD4+ T cells are given more weight in the unstandardized clustering than plasmablasts. Thus in this analysis, the use of unstandardized subset percentage data appears to result in the most robust clustering, in which replicate samples from the same individuals tend to pair the most often.

Individuals Fall into Different Clusters on the Basis of Their Lymphocyte Subset Profiles

We next used nonstandardized subset percentage data to cluster single samples from individuals. By clustering single samples from individuals, we hoped to determine if individuals fell into one or more clusters on the basis of internal similarities in their lymphocyte subsets (Fig. 6). This analysis revealed three to four clusters amongst different individuals.

Fig. 6.

Cluster analysis with single samples from different individuals. Agglomerative cluster analysis of unstandardized lymphocyte subset percentage data is shown for the first sample of each of the 18 subjects.

DISCUSSION

There are many well-understood reasons for lymphocyte subset measurements to vary. These reasons include preanalytical variables such as the source, condition, and manner of preservation of the sample, the time of day that the blood was drawn, subject age, disease state, and medications (5–8). Analytical variables include instrumentation, instrument settings, the types and quantities of antibody-fluorophore conjugates, and the manner in which lymphocyte subsets are defined and electronically gated (5,9,10). We attempted to limit some of these sources of variation by using fresh whole blood, drawing all samples in the morning, using an identical panel of core antibodies and having a single operator perform all of the staining and run all of the samples on the same instrument. Furthermore, data analysis was performed using very similar gate settings. However, we did not run bead calibrators with each sample to standardize instrument settings between different days, so it is likely that some of the measurements could have been less variable. Another caveat to this study is that a two-platform technique was employed for absolute quantification: the absolute lymphocyte count was obtained from a CBC that was performed on the same blood sample, in parallel with the flow cytometry. Dual platform methods have greater variation than single platform methods for determining absolute cell counts and the use of the absolute lymphocyte count in particular is less reliable than the absolute white blood cell count (11). Thus, our finding that absolute cell counts have greater variability than lymphocyte subset percentages may be attributable entirely or in part to technical variations introduced by using a dual platform method. Nevertheless, the remarkable finding is not that lymphocyte subsets are variable but that the level of variability within subjects is much lower than between subjects.

This finding negates the standard concept of a reference interval for lymphocyte subsets; the reference interval for a population is quite broad compared with the baseline measurements for an individual. Therefore, in some instances, the use of reference ranges or average values will lead to erroneous conclusions, unless baseline measurements are obtained. For example, without a baseline measurement, one might incorrectly conclude that a high or low proportion of a particular lymphocyte subset is due to therapy, rather than being due to where that subject’s lymphocyte set-point tends to reside. Conversely, a small change in the proportion of a particular lymphocyte subset compared with baseline might be quite meaningful in an individual, even if the changed value still falls within the “normal” range. Thus, the use of a baseline measurement is preferable to an average or reference value when evaluating the effects of therapy on lymphocyte subsets in an individual over time. We acknowledge that patients who have had treatments or longstanding disease before enrollment into a study may have “baseline” samples that do not accurately reflect their true baseline in health. A very important extension of this work will be to determine how stable lymphocyte subset composition is in individuals with active disease, and ultimately in individuals at different stages of disease.

Another finding of this study is that not all lymphocyte subset measurements have a similar level of variability. Absolute cell counts and smaller lymphocyte subsets tend to have more inter- and intraindividual variability. This is illustrated by less robust clustering of paired measurements from individuals when absolute lymphocyte counts or standardized lymphocyte fractions are used. In the case of small populations, which are given similar weight in standardized lymphocyte fractions, small variations in staining intensity can lead to large relative changes. Others have also found that variability was proportionally greater for measurements of numerically smaller subsets of lymphocytes (5,6,10,12). In the case of absolute cell counts, additional sources of variation including analytical (e.g., the complete blood count (13), and the use of a dual platform method, as discussed above) and biological [e.g., diurnal shifts (14)], contribute to increased inter- and intraindividual variability. It is also possible that shifts in clonal populations and responses to immunological stressors are more apparent in smaller lymphocyte subsets and/or with absolute cell counts.

While the proportions of different lymphocyte subsets are stable in most subjects over the two time points, the degree of intrasubject variability does vary. Some of the subjects exhibit larger shifts in their subsets. It is unclear if this is due to unique unidentified circumstances in these subjects or if some individuals have inherently more variable lymphocyte subset set points. When the change in lymphocyte subset percentages was analyzed as a function of the length of time between time points, there was no positive or negative correlation (data not shown). Lymphocyte subsets and their level of variability also differ in subjects with diseases such as SLE (6,15–17). Therapy with corticosteroids and other immunomodulatory agents can have profound effects on lymphocyte subset distributions as well (2,18–21), further complicating longitudinal data interpretation in the setting of disease.

One of the most intriguing aspects of this analysis is that healthy individuals can be segregated into different groups on the basis of their lymphocyte subsets. Due to the small number of subjects studied, we were not able to correlate the clusters with demographic characteristics of the subjects such as age or race. The subjects analyzed in this study are mostly females of childbearing age (as described in Methods, these subjects were used as controls for an analysis of B cell subsets in patients with SLE and replicate samples were analyzed specifically for this study). Thus, the group of subjects is more uniform than a randomly recruited group of healthy control subjects with respect to age and sex but is not well controlled for race. The ability to cluster individuals suggests that there may be similarities between individuals who tend to fall into a particular cluster (such as race, HLA type, or other genetic factors, although race did not correlate in this small sample). Furthermore, there may be under-appreciated relationships between different lymphocyte subsets. Additional correlations may become apparent when larger numbers of individuals are analyzed.

Overall, these data suggest that each individual has a characteristic lymphocyte profile that defines his or her immunologic set point. Lymphocyte subset frequencies make up a multidimensional profile that can be studied longitudinally to determine the effects of disease or therapy on the immune system. The use of colored graphical displays or maps of lymphocyte percentages may be particularly useful in visualizing longitudinal trends. In a similar fashion, Petrausch and colleagues have used event acquisition histograms (FCOM^™ analysis, available on Winlist software) and applied hierarchical clustering to multiparameter flow cytometry data to demonstrate that this data analysis method reveals distinctive lymphocyte surface marker profiles in lymph node cells stimulated with different cytokine cocktails (22). More provocatively, it is possible that subjects with specific lymphocyte profiles are predisposed to develop certain diseases or immune responses.