Development of an unbiased, semi-automated approach for classifying plasma cell immunophenotype following multicolor flow cytometry of bone marrow aspirates

Abstract

Background

Despite increased usage of multiparameter flow cytometry (MFC) to assess diagnosis, prognosis, and therapeutic efficacy (minimal residual disease - MRD) in plasma cell neoplasms (PCNs), standardization of methodology and data analysis is suboptimal. We investigated the utility of using the mean and median fluorescence intensities (FI) obtained from MFC to objectively describe parameters that distinguish plasma cell (PC) phenotypes.

Methods

In this retrospective study, flow cytometry results from bone marrow aspirate specimens from 570 patients referred to the Myeloma Institute at UAMS were evaluated. Mean and median FI data were obtained from 8-color MFC of non-neoplastic, malignant, and mixed PC populations using antibodies to CD38, CD138, CD19, CD20, CD27, CD45, CD56, and CD81.

Results

Of 570 cases, 252 cases showed only non-neoplastic PCs, 168 showed only malignant PCs and 150 showed mixed PC populations. Statistical analysis of median FI data for each CD marker showed no difference in expression intensity on non-neoplastic and malignant plasma cells, between pure and mixed PC populations. ROC analysis of the median FI of CD expression in non-neoplastic and malignant PCs was used to develop an algorithm to convert quantitative FI values to qualitative assessments including “negative”, “positive”, “dim”, and “heterogeneous” expression.

Conclusions

FI data derived from 8-color MFC can be used to define marker expression on PCs. Translation of FI data from Infinicyt software to an Excel worksheet streamlines workflow and eliminates transcriptional errors when generating flow reports.

Introduction

Detection of antigen expression on plasma cells by multiparameter flow cytometry (MFC) is an important adjunct in the diagnosis and classification of plasma cell neoplasms (PCNs) (1–4), prediction of disease progression (5–7), prognosis (3, 5, 8, 9), and therapeutic response via minimal residual disease (MRD) monitoring (2, 10, 11). Despite increasing use of MFC in the management of PCNs, utilization in clinical practice is limited by standardization of MFC methods (10), the need for extensive expertise to analyze flow cytometric data, and potential variability in the immunophenotypes of normal and malignant plasma cells (9, 12, 13).

Based on recent consensus reports (2, 14), a panel of eight antibodies is recommended to distinguish malignant plasma cells (mPCs) from normal counterparts and includes CD38, CD138, CD45, CD19, CD56, CD27, CD81, and CD117. mPCs are defined immunophenotypically by aberrant expression of at least two antigens, with diminished expression of antigens normally present, and/or increased expression of antigens not present on non-neoplastic PCs (nPCs). For example, mPCs show decreased expression or loss of CD45, CD19, CD27, and/or CD81 and may aberrantly express CD56, CD117, and/or CD20. However, there is overlap in the intensity and expression profile of mPCs with reactive PCs, plasma cells in monoclonal gammopathy of undetermined significance (MGUS), and B-cell lymphomas with plasmacytic differentiation (13, 15). Thus, heterogeneous expression of CD markers can result in immunophenotypic overlap between malignant and non-neoplastic PCs. Moreover, under conditions where the malignant clone is small, e.g. MRD, phenotypic overlap with nPCs may interfere with distinguishing the malignant population.

Overall, MFC is valuable approach for providing accurate diagnostic and prognostic assessment of PC neoplasia. However, analysis of the large and complex MFC data sets is associated with increased time, effort, and potential variability in interpretation. The objectives of this retrospective study were to develop an algorithm that delineates the optimal flow intensity descriptions of negative, positive, dim, and/or heterogeneous based on fluorescence intensity data generated by 8-color MFC and automates describing the mPC immunophenotype in bone marrow aspirates.

Materials and Methods

Case Selection

This study was approved by the Institutional Review Board at the University of Arkansas for Medical Sciences (UAMS). A total of 570 bone marrow aspirates from patients referred to the Myeloma Institute at UAMS for evaluation/follow-up for plasma cell neoplasm were submitted for MFC analysis from August 2015 to February 2016.

Flow Cytometry

Bone marrow aspirate cells collected in EDTA were processed for single tube, 8-color immunophenotyping using the following monoclonal antibodies: CD38 fluorescein isothiocyanate (FITC), CD20 phycoerythrin (PE), CD27 peridinin chlorophyll protein-cyanine 5.5 (PerCp-5.5), CD19 phycoerythrin -cyanine 7 (PECy-7), CD56 allophycocyanin (APC), CD81 allophycocyanin-H7 (APC-H7), CD45 Horizon V450 and CD138 Horizon V500 (BD Biosciences, San Jose, CA). Briefly, 50–200 microliter aliquots of unwashed bone marrow aspirate cells were incubated with the 8 antibody cocktail for 15 minutes. Red cells were lysed for 10 minutes (BD FACS Lyse reagent), the cell suspension centrifuged, washed with 2 mls of PBS then resupended in 300 ul of BD Stabilizing Fixative. Since mean FI can decrease over time, all flow analyses were performed within 4 hours of collection using a 3-laser FACSCanto II (BD Biosciences, San Jose, CA) and analyzed using DiVa 6.1.1 and Infinicyt flow cytometry software v1.8 (Cytognos, Salamanca, Spain). Instrument calibration was performed daily using Rainbow Beads (Spherotech, Lake Forest, IL).

Flow Analysis

Initially, 1.0 × 10⁵ total events were acquired and assessed for sufficient percentage of PCs for analysis. If PCs were less than 0.1% events, data acquisition continued until up to 2.5 × 10⁶ total events had been collected (at least 2 × 10⁶ viable events for MRD assessment). Forward and side scatter plots were analyzed to remove unlysed red cells, nonviable cells and debris (“doublets”) (16). Nonspecific binding of antigens to other hematopoietic cell populations were excluded from further analysis. In most cases, a large gate was drawn to include the total PC population based on light scatter properties and expression of CD38, CD138 and CD45.

Discrimination between immunophenotypically normal and aberrant PC populations was further defined using CD19, CD56, CD20, CD27 and CD81 expression as previously described (2, 16). For cases with two PC populations, automatic population separator (APS; Infinicyt software) multidimensional analysis was incorporated into the final analysis. The percent of malignant and non-neoplastic PCs was calculated by dividing the number of PCs in each group by the total number of viable nucleated bone marrow cells in each analyzed sample. In parallel, CD19 positive lymphocytes and hematogones were quantified as a quality control for the presence of bone marrow aspirate cells.

Data Analysis

MFC values (arithmetic mean and median FI) were recorded for each antibody. A Receiver-Operator Curve (ROC) was generated from the median FI of each antibody using Prism 6 software (GraphPad Inc, La Jolla, CA). Parameters derived from flow and ROC analyses were used to develop an algorithm for classifying expression of each marker as negative, dim, positive, and heterogeneous.

Results

Using eight antibodies (CD38, CD138, CD45, CD19, CD56, CD27, CD81, and CD20) (2, 14), 570 bone marrow aspirates were evaluated by MFC. nPCs and mPCs were identified based on consensus antigen expression profiles using the gating scheme illustrated in Figure 1. nPCs were defined as those with ≤1 antigen discrepancy from the most commonly described nPC immunophenotype: CD138, CD38, CD45, CD19, CD27 and CD81 positive, and CD20 and CD56 negative (9, 13, 16, 17). Including PC populations with dim or positive/heterogeneous expression of CD45 (6%) or CD19 (6%), CD27 (8%), or CD56 (3%), nPCs were identified in 402 bone marrow aspirates in proportions ranging from 0.016% to 3%. nPCs were present as pure populations in 252 cases. mPCs were defined by aberrant over- or under- expression of at least two antigens (16). mPCs were detected in 318 total bone marrow aspirates in percentages ranging from 0.005% to 70% of analyzed events. mPCs were present as pure population in 168 cases and mixed with nPCs in 150 cases (Figure 2).

Figure 1.

Gating strategy for discriminating bone marrow PC. A) Use of FSC-A vs SSC-A plots to remove debris and doublets. B) Selection of nPC by sequential gating of CD38 vs CD45, CD38 vs CD138, and CD38 vs SSC. C) Selection of mPC from nPC by sequential gating of CD19 vs CD38, CD45 vs CD38, and CD56 vs CD38. D) Illustration of the two distinct PC populations with different immunophenotypes.

Figure 2.

Example of PC profiles using multiparameter flow and Infinicyt software. PC populations from a representative bone marrow aspirate were identified using the gating strategy described in Fig 1. mPCs (red) comprised 1.61% and nPCs (blue) comprised 0.34% of total analyzed events. Green dots represent CD19 positive cells. Shown are dot plots for the different antigens detected in the mixed PC population.

Evaluation of the median FI values of individual CD markers in pure populations of nPCs (n = 252) and mPCs (n= 168) showed no statistical difference compared to bone marrow aspirates with mixed PC populations (n = 150). Therefore, median FI data with nPCs (n=402) and mPCs (n =318) were combined for each antigen and used for subsequent analyses. To optimize antigen expression determination in PC populations, the median FI for individual CD markers were evaluated using a ROC analysis. Optimal median FI thresholds to define populations of negative expressing cells, and differences between nPCs and mPCs were selected for each marker by maximizing sensitivity and specificity values for median FIs of nPCs and mPCs (CD19 and CD138), thresholding using as a negative control individual cases in which the mPCs were identified as negative for the expression of individual antigens (i.e., CDs 81, 27, 45, 56, 38); or thresholding using neoplastic plasma cells from known cases as positive control cells for CD20. Expression thresholds were not used for CD138 as this antigen is expressed on all PCs. Because CD19 is normally expressed on nPCs but not mPCs, the discriminating threshold is actually an expression threshold for which mPCs are the negative control. The optimal median FI thresholds and their associated sensitivity and specificity values as determined from the ROC analysis are depicted in Figure 3 and summarized in Table 1.

Figure 3.

Median FIs for CD marker expression. Shown are the median FIs for individual cases with mean and standard error of the mean for non-neoplastic (nPCs) and malignant plasma cells (mPCs). Also shown are the median FIs for negative control cells (−ctl: CD38, CD45, CD56, CD27 and CD81) or positive control cells (+ctl: CD20). Dashed lines represent the optimal thresholds determined from ROC analysis of each marker for negative expression (lower) and nPCs versus mPCs discrimination (upper).

Table 1.

Parameters from ROC Analysis

Marker	Expression Threshold log median FI1	Sensitivity (%) Specificity (%) AUC	Discriminating Threshold log median FI2	Sensitivity (%) Specificity (%) AUC	Aberrant Pattern
CD138	NA	NA	4.005	53.46	NA
				53.48
				0.547
CD38	2.972	99.69	3.986	80.82	Dim
		100		80.60
		1.0		0.883
CD45	3.01	25.79	3.592	88.99	Negative
		100		91.04
		0.621		0.940
CD19	3.171	98.74	NA	NA	Negative
		95.27
		0.993
CD56	2.424	66.98	3.872	56.29	Positive
		100		99.50
		0.867		0.802
CD20	2.019	84.85	2.831	13.52	Positive/Dim
		100		100
		0.993		0.831
CD27	2.754	45.60	3.424	88.68	Negative/Dim
		100		88.78
		0.779		0.944
CD81	2.683	54.09	3.221	86.48	Negative/Dim
		100		88.56
		0.850		0.922

¹Determined using negative control cells

²Determined from comparison between nPC and mPC

The derived optimal median FI thresholds were used together with the observed arithmetic mean FI and median FI for each CD marker to develop an algorithm that semi-automates the designation of marker expression (Figure 4). Using standard flow descriptors, this algorithm uses the optimal median FI thresholds to define negative, dim, or positive cell phenotypes as follows: when the median FI for a CD marker was less than the negative expression threshold, the phenotype was defined as “negative”; when the median FI for a CD marker was greater than the discriminator threshold, the phenotype was classified as “positive”; and for CD markers with an median FI between these threshold values, the phenotype was defined as “dim”. In addition, if the ratio of the arithmetic mean FI and the median FI was >0.65, the phenotype was characterized as heterogeneous. The frequency that these individual flow phenotypes applied to nPCs and mPCs when this algorithm was applied to all cases in our data set is summarized in Table 2. When applied to the populations of mPCs identified in the original 318 bone marrow aspirates, the algorithm indicated aberrant expression of at least 2 antigens in 100% of cases. When applied to the nPC populations of the original 402 flows, the algorithm indicated that 53 (13%) cases showed at least two antigens expressed that overlap with mPCs. Importantly, reactive PC populations express dim or positive/heterogeneous expression of CD45, CD19 and/or CD56 (12, 13). As illustrated in Figure 5, this algorithm can be incorporated into an Excel worksheet to automate description of the expression intensity of individual CD markers in mPCs.

Figure 4.

Algorithm based on optimized flow parameters for defining phenotypic marker expression.

Table 2.

Immunophenotype of non-neoplastic (nPCs) and malignant plasma cells (mPCs)

	Non-neoplastic PCs n=402 (%)					Malignant PCs n= 318 (%)
	P	P/H	D	D/H	N	P	P/H	D	D/H	N
CD138	402 (100)	0	0	0	0	179 (56.3)	0	138 (43.5)	1 (0.2)	0
CD38	333 (82.0)	0	69 (17.2)	0	0	70 (22.0)	1 (0.3)	246 (77.4)	0	1 (0.3)
CD45	379 (94.3)	14 (3.4)	3 (0.8)	6 (1.5)	0	43 (13.5)	5 (1.6)	24 (7.6)	57 (17.9)	189 (59.4)
CD19	379 (94.3)	19 (4.7)	1 (0.2)	3 (0.8)	0	5 (1.6)	1 (0.3)	247 (77.7)	57 (17.9)	8 (2.5)
CD56	2 (0.5)	2 (0.5)	8 (2.0)	261 (65.0)	129 (32.0)	180 (56.6)	7 (2.2)	16 (5.0)	51 (16.1)	64 (20.1)
CD20	1 (0.2)	3 (0.8)	80 (20)	29 (7.2)	289 (71.8)	29 (9.2)	31 (9.7)	153 (48.1)	22 (6.9)	83 (26.1)
CD27	371 (92.3)	3 (0.8)	18 (4.5)	9 (2.2)	1 (0.2)	45 (14.1)	10 (3.1)	79 (24.8)	52 (16.4)	132 (41.5)
CD81	374 (93.0)	6 (1.5)	20 (5.0)	2 (0.5)	0	53 (16.7)	15 (4.7)	100 (31.4)	66 (20.8)	84 (26.4)

P = Positive; P/H = Positive/heterogeneous; D = Dim; D/H = Dim/heterogeneous; N = Negative Aberrant patterns are indicated in bold

Figure 5.

Example worksheet utilizing algorithm for describing multiparameter flow results. PC populations from the bone marrow aspirate dot plots depicted in Figure 2 were distinguished using the gating strategy described in Figure 1. The algorithm (Figure 4) developed from optimized ROC analysis (Figure 3, Table 1) was applied to the mean and median FI for each CD marker (left) in mPCs and the resulting descriptions presented in a table (right).

Discussion

Due to its pivotal clinical utility in the diagnosis and management of PCNs, a standardized methodology and reporting of PC immunophenotypes is essential. Eight color-MFC immunophenotyping allows discrimination of mPCs from their normal counterparts in bone marrow aspirate cells despite overlap in PC phenotype with reactive PCs (12, 13) and changes in antigen expression with therapy (9, 12, 18). The conventional method of reporting flow cytometry data is observer-dependent and difficult to standardize. Based on the mean and median FI of consensus surface CD markers (14) on PCs, an algorithm was created to automate identification of PC phenotype.

We found that CD138 expression was positive in both nPCs and mPCs with no significant differences in expression intensity (Figure 3). This confirms previous reports (16, 17) that CD138 is a useful marker to identify PCs, but not distinguish between nPCs and mPCs. Similar to CD138, the majority of nPCs and mPCs express CD38 (Figure 3). Consistent with previous reports showing decreased CD38 expression in mPCs compared to nPCs (13, 16, 17), we detected dim CD38 expression more commonly in mPCs (Table 2). However, due to the overlap in CD38 expression intensity in nPC and mPCs, dim CD38 expression alone does not discriminate well (AUC = 0.883) between nPCs and mPCs.

The reported frequency of CD45 expression in nPCs and mPCs is variable (2, 12, 18). In our study, CD45 showed a broad range of expression intensity especially in mPCs (Figure 3). Based on the threshold MFIs used in our algorithm (Table 1), nPCs were positive (94%), positive/heterogeneous (3%), dim (~1%), or dim/heterogeneous (~2%) for CD45 expression (Table 2). In contrast to nPCs, the majority of mPCs were negative for CD45 (59%), consistent with previous results (19). Similar to previous reports (8, 20), CD45 expression was positive (14%), positive/heterogeneous (2%), dim (8 %) or dim/heterogeneous (18%) in mPCs. The minimal overlap in CD45 expression intensity between nPCs and mPCs makes the MFI for CD45 a powerful discriminator (AUC = 0.940) of mPCs.

Consistent with Rawstron (2) and Peceliunas (13), CD19 was either positive (94%) or positive/heterogeneous (5%) in the majority of nPCs. It was reported that CD19 expression is absent in 95%–97.5% of mPCs (2, 17, 19). Similarly, in this study the majority of mPCs were dim to negative for CD19 (98%). The overall frequency of CD19 expression on mPCs in this study (3%) is within the range (3%–11%) reported previously (8, 17, 18). CD19 expression is the strongest discriminator (AUC = 0.993) between nPCs and mPCs in our panel.

CD56 expression is negative in the majority of nPCs; whereas the majority of mPCs are positive for this marker (2, 8, 17, 20). Recent reports indicate that up to 15% of reactive/non-neoplastic PCs express CD56 (2, 13). In our study, 3% of nPCs showed positive, positive/heterogeneous or dim CD56 expression (Table 2). In one report, CD56 was reported to be positive in up to 37% of nPCs (12). CD56 expression was aberrantly expressed (positive) in 57% of mPCs analyzed in our study. Other studies have shown aberrant CD56 expression on mPCs in 72–78% of cases (2, 17, 19, 20). Although positive CD56 expression was often associated with mPCs, the heterogeneity in CD56 staining and overlap with nPC makes CD56 MFI one of the weaker discriminators (AUC = 0.802) in our MFC panel.

Aberrant CD20 expression (positive or dim) is reported in 12–30% of mPCs (2, 8, 13, 17, 19, 20). In our study, aberrant CD20 expression on mPCs was detected in approximately 67% of cases. In contrast, CD20 was negative in the majority of nPCs and positive or positive/heterogeneous in 4 cases of nPCs. As with CD56, positive CD20 expression was strongly associated with mPCs, however, heterogeneity in staining and overlap with nPCs results in CD20 MFI being a modest discriminator (AUC = 0.831) between mPCs and nPCs.

Expression of CD27 in nPCs is usually strongly positive (2). In our analysis, the majority of nPCs were positive for this marker (92%). In contrast to nPCs, aberrant (negative or dim) CD27 expression has been reported in 40–68% cases of mPCs (2, 17, 19). In our cohort, the majority of mPCs showed negative (42%), dim (25%) or dim/heterogeneous (16%) CD27 expression. CD27 expression in nPCs relative to mPCs, makes the MFI for CD27 a strong discriminator (AUC = 0.944) of mPCs and nPCs.

CD81 is expressed on nPCs, linked to CD19 expression and inversely correlated with prognosis in PCM (5, 21). In nPCs, CD81 expression was positive in 93% of cases. Similar to previous reports (5, 19), CD81 was positive 17% in mPC cases, and negative in approximately 26%. The majority of nPCs showed dim or dim/heterogeneous CD81 expression, whereas no nPCs were negative for CD81 (Table 2). Thus, similar to CD27, the expression of CD81 in nPCs relative to mPCs makes the MFI for CD81 is a good discriminator (AUC = 0.922) of mPCs and nPCs.

In summary, we analyzed 570 PC phenotypes in bone marrow aspirates using single-tube, 8-color MFC. ROC analysis of mean and median FI of CD expression in nPCs and mPCs was optimized to discriminate these PC populations, and an algorithm developed (Figure 4) to convert quantitative FI values to the qualitative assessments “negative”, “dim”, “positive” and “heterogeneous”. MFI data and plasma cell enumeration derived from Infinicyt software are incorporated to an Excel template which applies the algorithm (Figure 5). MFI data and CD expression (positive, dim, negative and/or heterogeneous) for each antigen on mPCs are provided to pathologists reviewing two-dimensional flow dot plots (Figure 2). The description of expression patterns for individual CD markers determined using this algorithm was comparable to other published reports. Given the limitation that specific flow MFIs will need to be empirically determined for each application, once flow procedures (e.g., conjugated fluorochromes, antibody incubations, time from collection to analysis, instrument settings, gating strategy, etc.) are rigorously standardized, the approach described here should be applicable to other clinical settings using MFC.