Abstract
In our high-sensitivity flow cytometric approach for systemic mastocytosis (SM), we identified mast cell event clustering as a new diagnostic criterion for the disease.
To objectively characterize mast cell gated event distributions, we performed cluster analysis using FLOCK, a computational approach to identify cell subsets in multidimensional flow cytometry data in an unbiased, automated fashion.
FLOCK identified discrete mast cell populations in most cases of SM (56/75 [75%]) but only a minority of non-SM cases (17/124 [14%]). FLOCK-identified mast cell populations accounted for 2.46% of total cells on average in SM cases and 0.09% of total cells on average in non-SM cases (P < .0001) and were predictive of SM, with a sensitivity of 75%, a specificity of 86%, a positive predictive value of 76%, and a negative predictive value of 85%.
FLOCK analysis provides useful diagnostic information for evaluating patients with suspected SM, and may be useful for the analysis of other hematopoietic neoplasms.
Systemic mastocytosis (SM) is a neoplasm of mast cells in which they proliferate in bone marrow and other organs. Based on 2008 World Health Organization (WHO) diagnostic criteria, the diagnosis of SM requires the presence of mast cell infiltrates in bone marrow or other extracutaneous organs (the major diagnostic criterion) and/or a number of minor diagnostic criteria, which include atypical mast cell morphology, the D816V KIT mutation, elevated serum tryptase, and aberrant CD2 and/or CD25 antigen expression.1 However, the major diagnostic criterion is not seen in 20% to 30% of indolent SM cases, and strict use of the current WHO criteria results in false-negative SM cases, especially in early disease stages with low disease burden.2,3
Recently, we described a high-sensitivity flow cytometric approach for evaluation of patients with SM that identified the presence of discrete mast cell event clustering as a new diagnostic criterion for the disease. All cases that exhibited event clustering within the mast cell gate were found to have fulfilled WHO diagnostic criteria for SM, whether or not the mast cells exhibited aberrant CD2 and/or CD25 expression.4 Utilization of this mast cell event clustering criterion, along with the presence of aberrant expression of CD2 and/or CD25, increased the diagnostic sensitivity of flow cytometric analysis for the diagnosis of SM to 95%, which exceeds the diagnostic sensitivity of bone marrow biopsy examination (85%) or the bone marrow aspirate examination (69%).4
Our subsequent statistical analysis found that cells within a broadly defined CD117/side-scatter characteristic (SSC) mast cell gate had significantly lower coefficients of variation (CVs) for CD117 mean fluorescent intensity and SSC in cases of SM compared with non-SM cases.5 Flow cytometrically identified mast cell event clustering was associated with significantly lower CVs for CD117 and SSC than cases without mast cell event clustering.5 A combined CV (CD117 + SSC) of 125 or less showed a sensitivity of 80% and a specificity of 80% for SM with a positive predictive value (PPV) of 67% and a negative predictive value (NPV) of 80%. Probability scores of having SM, based on CVs for SSC and CD117, were significantly higher in patients with SM than in those without SM (0.55 vs 0.17, respectively; P < .001).5
In our prior studies, flow cytometric and statistical analyses were based on manual gating of mast cell populations using a uniform, broadly defined mast cell gate, but not all SM cases showed mast cell event clustering or a consistent distribution of mast cells within the mast cell gate.4,5 To objectively characterize event distributions within the mast cell gate and identify discrete mast cell populations, we performed cluster analysis using FLOCK, a computational approach that uses algorithms and density-based clustering to identify cell subsets in multidimensional flow cytometry data in an unbiased, automated fashion, eliminating operator-dependent variability Figure 1.6 FLOCK has been employed to objectively identify B-cell and T-cell subsets, based on flow cytometric data, in patients following vaccination for tetanus, typhoid, and influenza vaccination; in toxic shock syndrome; and during acute and chronic virus infection.6–10 To our knowledge, FLOCK has not until now been used to objectively identify neoplastic cell populations from multiparametric flow cytometric analysis. In this study, we investigated whether FLOCK could identify discrete mast cell populations based on flow cytometric data and whether the presence of FLOCK-identified mast cell populations is associated with SM.
FLOCK analysis involves (A) partitioning of flow cytometry data based on multiple measured characteristics, such as fluorescence due to antibody binding or side scatter of light into equal-sized bins; (B) identification of data-dense regions that exceed a minimum density threshold (shown in red); (C) merging of neighboring data-dense regions into clusters, which are each assigned a unique color; and (D) final clustering of events by assignment to the nearest cluster. Each cluster has a centroid, which is the point whose coordinates are the averages of the corresponding coordinates of all the points within the cluster. FLOCK calculates the percentage of the total population present in each identified cluster. Reproduced with permission from Qian et al.6
Materials and Methods
Cases and Prior High-Sensitivity Flow Cytometric Analysis
The Brigham and Women’s Hospital hematology laboratory previously performed flow cytometric analysis on 199 consecutive bone marrow aspirate samples from patients being evaluated for SM.5 Seventy-five samples came from patients who fulfilled the diagnostic criteria for SM after complete evaluation, and 124 samples came from patients without SM who had various diagnoses, including nonclonal mast cell disorders, low-grade B-cell lymphoproliferative disorders, myeloid malignancies, and autoimmune disorders. Flow cytometric results from the bone marrow aspirates were previously correlated with the results of the concurrent bone marrow biopsy specimen, as well as molecular diagnostic, other laboratory, and clinical findings.5 The diagnosis of SM was based on the 2008 WHO Classification of Tumours of Haematopoietic and Lymphoid Tissues.1 Clinical information was collected through chart review. High-sensitivity flow cytometric analysis of mast cells was previously performed on a Becton Dickinson (Franklin Lakes, NJ) FACSCanto flow cytometer using the following antibodies: CD117-PE, CD2-APC, and CD25-FITC with respective fluorochrome isotype controls from BD Biosciences (San Jose, CA).5 The number of collected events per sample ranged from 20,000 to 3,518,900 (mean, 473,454 events; median, 200,000 events), and in most samples, the number of collected events was 200,000 or more. A broad, uniform mast cell gate was employed for all cases and included all CD117-bright events (x-axis between 104 and 105 on the logarithmic scale) and all events on the y-axis (side scatter characteristic [SSC], linear axis; Figure 1) regardless of the presence or absence of cluster formation. The number of events within the mast cell gate ranged from 13 to 87,615 and followed two distribution patterns: 50 cases showed mast cell event clustering, and 149 cases displayed scattered events within the mast cell gate.5 Studies were performed with the institutional review board’s permission.
FLOCK Analysis
FLOCK analysis was performed using flow cytometry standard format file data from the 199 specimens evaluated by flow cytometric analysis for SM (see above). FLOCK is available at the Immunology Database and Analysis Portal (www.immport.org). FLOCK analysis was employed to investigate CD117/SSC flow cytometry data, determine the number of FLOCK-identified cell populations present, determine if a discrete population was identified in the CD117/SSC mast cell gated area, and, if so, determine the size of the mast cell population identified by FLOCK. Cell populations that were uniformly distributed throughout the broadly defined CD117/SSC mast cell gate or that were included as part of other FLOCK-identified populations outside of the mast cell gated area were not regarded as discrete mast cell populations. In a small subset of SM cases, FLOCK identified a uniform, discrete population within the mast cell gated area and overpartitioned it into more than one population (data not shown). In such cases, mast cell population size comprises the sum of the sizes of the uniform, discrete but partitioned total mast cell population.
Statistical Analysis
Statistical analysis was performed using GraphPad Prism version 5.02 (GraphPad Software, La Jolla, CA).
Results
FLOCK Analysis
Seventy-five specimens from patients who fulfilled WHO diagnostic criteria for SM and 124 cases from patients negative for mastocytosis after diagnostic workup (non-SM) were evaluated by FLOCK analysis and compared with prior flow cytometric analysis. There was no significant difference in the number of cell populations identified by FLOCK from the CD117/SSC flow cytometry data in SM cases (18.6 ± 7.5 populations) and non-SM cases (19.0 ± 8.0 populations; P = .6835, Mann-Whitney). Representative FLOCK results for SM and non-SM cases are shown in Figure 2.
Representative flow cytometry findings and FLOCK analysis findings in systemic mastocytosis (SM) and non-SM cases. A, An SM case showing a significant population of cells with event clustering within the broadly defined CD117/side-scatter characteristic (SSC) mast cell gate by flow cytometric analysis (left panel) and with a discrete mast cell population identified by FLOCK analysis (right panel). B, An SM case showing a small population of cells with event clustering within the broadly defined CD117/SSC mast cell gate by flow cytometric analysis (left panel) and with a discrete mast cell population identified by FLOCK analysis (right panel). C, A non-SM case without event clustering within the broadly defined CD117/SSC mast cell gate by flow cytometric analysis (left panel) and without a discrete mast cell population identified by FLOCK analysis (right panel).
In 73 cases (56 SM and 17 non-SM cases), FLOCK identified discrete populations within the flow cytometric CD117/SSC mast cell gate, while 126 cases (19 SM and 107 non-SM cases) showed no discrete populations in the CD117/SSC mast cell gate. All 48 SM cases that showed flow cytometry–based event clusters exhibited discrete mast cell populations by FLOCK analysis (Figure 2). These discrete mast cell populations ranged in size from 0.03% to 28.56% of total cells (mean, 2.85%) and correlated with flow cytometry–based total CD117/SSC mast cell gated events (R2 = 0.9699) Figure 3. FLOCK identified discrete mast cell populations in eight (30%) of 27 SM cases without apparent flow cytometry–based event clusters Figure 4A, which ranged in size from 0.02% to 0.30% of total cells (mean, 0.11%). Combining all FLOCK-positive SM cases, FLOCK-identified mast cell populations ranged in size from 0.02% to 28.56% of total cells (mean, 2.46%).
Comparison of mast cell population size by FLOCK analysis and CD117/side-scatter characteristic (SSC) mast cell gated events by flow cytometric analysis, showing correlation between the number of events within the broadly defined CD117/SSC mast cell gate by flow cytometric analysis (y-axis) and the FLOCK analysis–identified mast cell population (x-axis; R2 = 0.9699).
Representative FLOCK analysis findings and flow cytometry findings in systemic mastocytosis (SM) and non-SM cases. A, An SM case showing no discernible event clustering within the broadly defined CD117/side-scatter characteristic (SSC) mast cell gate by flow cytometric analysis (left panel) but with a mast cell population identified by FLOCK analysis (right panel). B, A non-SM case showing a small population of cells with event clustering within the broadly defined CD117/SSC mast cell gate by flow cytometric analysis (left panel) and with a mast cell population identified by FLOCK analysis (right panel). C, A non-SM case showing no discernible event clustering within the broadly defined CD117/SSC mast cell gate by flow cytometric analysis (left panel) but with a mast cell population identified by FLOCK analysis (right panel).
FLOCK identified discrete mast cell populations in two of two non-SM cases with flow cytometry–based event clusters, which accounted for 0.20% and 0.01% of total cells Figure 4B. In addition, FLOCK identified discrete mast cell populations in 15 (12%) of 122 non-SM cases without apparent flow cytometry–based event clusters Figure 4C, ranging in size from 0.03% to 0.16% of total cells (mean, 0.09%). Combining all non-SM cases with FLOCK-identified discrete mast cell populations, the populations ranged in size from 0.01% to 0.20% of total cells (mean, 0.09%). There was a statistically significant difference in FLOCK-identified mast cell population sizes in SM cases compared with non-SM cases (P < .0001, Mann-Whitney) Figure 5.
Comparison of FLOCK analysis–based mast cell population sizes in systemic mastocytosis (SM) (mean population size, 2.51%) and non-SM cases (mean population size, 0.09%; P < .0001).
Receiver operating characteristic (ROC) curve analysis of FLOCK-based mast cell population size as a diagnostic criterion for SM was performed based on discrete mast cell population sizes determined by FLOCK analysis; the results are shown in Figure 6. The area under the ROC curve was 0.84 (95% confidence interval, 0.78–0.91; P < .0001). Based on the presence of any FLOCK-identified mast cell population, sensitivity was 75%, specificity was 86%, PPV was 76%, and NPV was 85% for SM. For FLOCK-identified mast cell populations with a size 0.07% or more of the total population, sensitivity was 67% and specificity was 90% for SM. For FLOCK-identified mast cell populations with a size 0.15% or more of the total population, sensitivity was 53% and specificity was 98% for SM.
Receiver operating characteristic (ROC) curve analysis of FLOCK-based mast cell population size as a diagnostic criterion for systemic mastocytosis (SM). The area under the ROC curve was 0.84 (95% confidence interval, 0.78–0.91; P < .0001). For SM, the sensitivity was 75%, specificity was 86%, positive predictive value was 76%, and negative predictive value was 85% for the presence of any FLOCK-identified discrete mast cell population. As the FLOCK-identified mast cell population size increases, sensitivity for the diagnosis of SM decreases and specificity increases (eg, for mast cell populations with a size 0.07% or more of the total population, sensitivity was 67% and specificity was 90% for SM, and for mast cell populations with a size 0.15% or more of the total population, sensitivity was 53% and specificity was 98% for SM).
Comparison With Histologic Findings
When FLOCK results were compared with bone marrow biopsy findings, all SM cases with apparent flow cytometry–based event clusters and FLOCK-identified discrete mast cell populations more than 0.15% of total cells and six of 10 cases with FLOCK-identified discrete mast cell populations 0.15% or less of total cells had mast cell aggregates present in the bone marrow biopsy comprising less than 5% to 80% of total cellularity (mean, 38% of total cellularity), with one exception. A case with a FLOCK-identified discrete mast cell population of 0.91% showed increased, scattered mast cells, including spindle-shaped cells, which accounted for 5% of total cells, but no mast cell aggregates. The remaining SM cases contained scattered mast cells and no discrete mast cell aggregates.
All SM cases with FLOCK-identified discrete mast cell populations without apparent flow cytometry–based event clusters exhibited mast cell aggregates in the bone marrow biopsy comprising 5% to 20% of total cellularity, with the exception of one case with loose fibrosis and a mast cell infiltrate that comprised 90% of total cellularity (FLOCK-identified discrete mast cell population of 0.3%). For SM cases without FLOCK-identified discrete mast cell populations and with available bone marrow biopsy findings, 13 of 19 exhibited mast cell aggregates in the bone marrow biopsy specimen, ranging in size from less than 5% to 20% of total cellularity (mean, 7.5% of total cellularity).
Discussion
In this study, we employed FLOCK, a computational approach that uses algorithms and density-based clustering to identify cell subsets in multidimensional flow cytometry data, to identify and quantitate mast cell populations in an unbiased, automated fashion. Previously, we found that patients who exhibited discrete mast cell event clustering by high-sensitivity flow cytometric analysis, in conjunction with aberrant CD2 and/or CD25 expression, all fulfilled WHO diagnostic criteria for SM following a complete evaluation. In fact, with the addition of event clustering as a diagnostic criterion, flow cytometric analysis was more sensitive and specific in detecting SM than bone marrow biopsy examination, the current gold standard for diagnosing SM.4 In our prior study, statistical analysis showed that the population of cells within a broadly defined mast cell gate had lower CVs for CD117 and SSC in SM cases compared with non-SM cases, as well as in cases with mast cell event clustering compared with cases without mast cell event clustering.5 Here we sought to determine whether an automated, objective, algorithmic approach for the identification of discrete mast cell populations could be used and whether the results of such an approach would correlate with the presence of SM.
In our analysis, FLOCK was able to identify discrete mast cell populations in most cases of SM (75%) but only a minority of non-SM cases (14%). FLOCK identified discrete mast cell populations in a number of cases in which event clustering was not detected by flow cytometric analysis alone. We also found that FLOCK-identified mast cell population size was significantly different in SM cases compared with non-SM cases (2.46% vs 0.09%, P < .0001). A shortcoming of FLOCK analysis is the absence of assessment of the relative density of mast cell distributions within the CD117/SSC flow cytometric space. Preliminary analysis of the density of FLOCK-identified mast cell populations, based on a calculation of the total number of cells within the area occupied by the mast cell population (from measurements of printed two-dimensional FLOCK data), finds a mean density of 1,688 cells/cm2 in SM cases and 157 cells/cm2 in non-SM cases (P = .0003, Mann-Whitney; unpublished data). Consideration of the density of FLOCK-identified mast cell populations may improve the diagnostic accuracy of FLOCK analysis in the assessment of possible SM cases.
FLOCK analysis was more accurate than statistical analysis of flow cytometric data at predicting the presence of SM. In our prior study, event clusters within the mast cell gate were associated with significantly lower CVs for SSC and CD117 in SM vs non-SM cases. A combined CV (CD117 + SSC) of 125 or less showed a sensitivity of 80% and specificity of 80% for SM, with a PPV of 67% and an NPV of 80%.5 In the current study, the presence of FLOCK-identified discrete mast cell populations was predictive of SM, with a sensitivity of 75%, specificity of 86%, PPV of 76%, and NPV of 85%. However, based on ROC analysis of FLOCK data, relatively large-sized FLOCK-identified mast cell populations are highly specific for SM: populations with a size 0.07% or more of the total population are 90% specific for SM, and populations with a size 0.15% or more of the total population are 98% specific for SM.
Comparison of FLOCK findings with bone marrow biopsy findings revealed that almost all SM cases with FLOCK-identified discrete mast cell populations more than 0.15% of total cells and 10 of 15 cases with FLOCK-identified discrete mast cell populations 0.15% or less of total cells had mast cell aggregates present in the bone marrow biopsy specimen. In general, there was histologic evidence of greater bone marrow involvement by SM in cases with FLOCK-identified discrete mast cell populations than in SM cases without FLOCK-identified discrete mast cell populations; however, the absence of FLOCK-identified discrete mast cell populations did not rule out bone marrow involvement by SM, which may be due in part to sampling artifact as well as marrow fibrosis, which may be seen in SM. Hence, assessment of bone marrow biopsy histologic and immunophenotypic findings remains important for the evaluation of patients with suspected SM.
Previously, FLOCK was created and used to identify T-cell and B-cell subsets in patients in various immunologically challenged states, such as postvaccination, and during bacterial or viral infections. FLOCK was able to identify and quantitate specific lymphocyte subsets and to allow comparison of these subpopulations between individuals. To our knowledge, FLOCK analysis of flow cytometric data has not been employed until now to identify neoplastic cell populations in hematopoietic neoplasms. Identification of neoplastic populations usually relies on the identification of a neoplastic cell immunophenotype based on expression of specific antigens, which often recapitulate the expression of hematopoietic lineage markers at specific states of differentiation (eg, the myeloblast state or the follicular center cell state). Similarly, high-sensitivity flow cytometric analysis for minimal residual disease assessment in acute lymphoblastic leukemia is based on the detection of aberrant antigen expression or a leukemia-associated immunophenotype.11,12 In the evaluation of patients with suspected SM, a high-sensitivity flow cytometric approach has been used to identify mast cells with an aberrant CD2 and/or CD25 immunophenotype, indicative of neoplastic mast cells.13–15 However, in our prior studies, we found that the identification of discrete mast cell event clusters by flow cytometry correlated with the presence of an aberrant mast cell phenotype and was also present in patients without an aberrant mast cell immunophenotype who otherwise fulfilled the WHO diagnostic criteria for SM.4,5 For example, in the current study, 10 cases of SM demonstrated event clustering by flow cytometric analysis and discrete FLOCK-identified mast cell populations, without aberrant antigen coexpression. This finding—event clustering and discrete FLOCK-identified populations within specific, flow cytometrically defined and gated populations—may itself be indicative of the presence of a neoplastic population and suggests a new approach for the flow cytometric assessment of specimens from patients with suspected hematopoietic neoplasms. Preliminary FLOCK analysis of bone marrow aspiration specimens from patients with possible involvement by a plasma cell neoplasm suggests that the number and size of discrete FLOCK-identified plasma cell populations are increased in positive specimens compared with negative specimens.
In conclusion, FLOCK analysis identifies discrete mast cell populations that correspond to flow cytometric mast cell event clustering in most SM cases, is predictive of SM and the presence of mast cell aggregates in the bone marrow biopsy specimen, and provides useful information for the diagnosis of patients with suspected SM, although a negative FLOCK result does not rule out bone marrow involvement. FLOCK analysis may also be useful for the objective identification of neoplastic cell populations in other hematopoietic neoplasms.
References
