28-Color single tube for flow cytometric assessment of myeloid maturation, myeloid neoplasia, and acute myeloid leukemia minimal / measurable residual disease

Abstract

Introduction:

Flow cytometry (FC) is an indispensable tool for myeloid neoplasia (MN) diagnosis, and cell of origin has clinical diagnostic and prognostic significance. However, the complex maturational pathways within the blast compartment complicate the detection of the abnormal population at the minimal/measurable disease (MRD) level using the difference from normal approach. The analysis could be simplified by separating the blast compartment into maturational-defined stages with relatively uniform phenotypes as reference populations. This requires a relatively extensive panel of antibodies to define maturational stages and simultaneously detect the common deviations from the normal pattern. We validated a single tube 28-color clinical assay for MN assessment and acute myeloid leukemia (AML) MRD detection assisted by the precise maturation stage assignment.

Methods:

The new assay uses a previously described 21 antigen backbone, with the additions of CD10, CD36, CD42b, CD45RA, CD90, CD105, and CD371. Validation was performed using 37 normal samples and 151 MN follow-up samples in a split-sample fashion comparing the new assay to the legacy 3-tube, 21-antigen assay. Dilution studies were performed to establish assay sensitivity. A new analysis framework relying on comparison to well-defined maturational stages was established for MRD analysis.

Results:

The assays showed 100% qualitative concordance with excellent quantitative concordance. Dilution studies yielded a limit of detection of 0.01%. The addition of new antibodies allowed for consistent comparisons of candidate abnormal populations to well defined normal maturation stages through traditional FC plots and Uniform Matrix Approximation and Projection assessments.

Conclusions:

This new single-tube, 28-color clinical assay allows for efficient MN assessment in clinical settings. It reliably detects MRD levels of abnormal myeloid cells because the expanded panel allows for precise comparison to defined lineage commitment maturational stages. Lastly, it provides high information density while reducing equipment use, reagent use, and technical labor.

1. INTRODUCTION

Acute myeloid leukemia (AML) is a prototypical myeloid neoplasm (MN) derived from neoplastic hematopoietic precursor cells; it is a heterogeneous disease in terms of genetic drivers, morphology, flow cytometry (FC) immunophenotype, and response to therapy.^1–4 Even with intensive chemotherapy therapies resulting in remission below the thresholds of conventional morphologic detection, many patients relapse.^1–6 Relapse after remission is caused by persistent leukemic blasts below the limits of conventional morphologic detection, which is known as minimal or measurable residual disease (MRD) and consistently associated with inferior outcomes.^5–8 Therefore, accurate and efficient detection of MRD is paramount in AML management.

Normal hematopoiesis begins with bone marrow (BM) hematopoietic stem cells (HSCs) that give rise to multipotent progenitors (MPPs). This population then branches into others with more restricted maturation potential including lymphoid-primed multi-potential progenitors (LMPPs), multi-lymphoid progenitors (MLP), common-lymphoid progenitors (CLPs), granulocyte-monocyte progenitors (GMPs), erythromyeloid progenitors (EMPs), and megakaryocyte-erythroid progenitors (MEP). All these populations express CD34, highlighting the heterogeneity within the CD34+ cell compartment. However, they are reliably identified using markers such as CD10, CD36, CD38, CD42b, CD45RA, CD90, CD105, and CD371.^9–13 These ultimately give rise to the major cell populations found in the peripheral blood (PB), and cells show predictable range of immunophenotypic changes as they mature.^9,14,15

Conversely, neoplastic hematopoiesis is associated with dysfunctional maturation resulting in immunophenotypic aberrancies detectable by FC, in populations spanning the full range of maturation, from HSCs to more mature populations.^5,6,14,16,17 Leukemia-associated immunophenotype approaches are frequently useful, but are limited when no original phenotype is available or immunophenotypes change after therapy.^17–21 The approach best works when supplemented with a “different from normal” (DfN) strategy which requires a thorough understanding of developmental immunophenotypes.^17,22–29 A holistic approach incorporating both LAIP and DfN elements is likely the most practically useful method.^5,6,30

Numerous studies have demonstrated the value of MRD detection by FC, and many AML related clinical trials show that MRD status is a predictor of both relapse and outcomes.^5–8,31 In our clinical FC laboratory we have used an established, 3-tube, 10-color, 21-antigen FC assay for MRD assessment.^24,25 However, MRD assessment is a complicated problem because of the branching nature of maturational pathways. Thus, MRD assays continue to evolve as the immunophenotypic definitions of different populations at different levels of maturation are refined.^17,30 While the number of useful antigens grows, so does the need for more comprehensive assays.

As the legacy FC assay for AML MRD and MN assessment requires multiple sample aliquots, redundant antibodies, and significant FC technologist time and labor, we sought a more efficient testing approach by unifying 3 separate tests into a single assay. Furthermore, as there is prognostic value in the determination of cell of origin or differentiation state in MNs, we also aimed to design an FC assay that could also detect the full range of known hematopoietic precursor populations.^{9–13,32–36} To support these aims, we designed a single-tube, 28-antigen clinical assay for standardized hematopoietic precursor detection, maturation assessment, and detection of MNs and AML MRD.

2. METHODS

2.1. SAMPLES

Samples were obtained from the Memorial Sloan Kettering Cancer Center FC laboratory under institutional review board-approved protocol. Diagnosis of MN was performed by an MSKCC hematopathologist incorporating FC, histomorphology, cytomorphology, cytogenetic, and molecular analysis, as appropriate to the most current WHO classification systems.^1,4

2.2. ANTIBODIES AND REAGENTS

Table 1 summarizes antibodies used for both the new 28-color assay and the legacy 3-tube, 10-color, 21-antigen assay. In short, antigens assessed include previously used CD2, CD4, CD5, CD7, CD11b, CD13, CD14, CD15, CD16, CD19, CD25, CD33, CD34, CD38, CD45, CD56, CD64, CD71, CD117, CD123, and HLA-DR, with additional new antibodies against CD10, CD36, CD42b, CD45RA, CD90, CD105, and CD371 (Clec12A). The first 21 listed antigens were used in the previously used 3-tube method, and the new antibodies used (CD10, CD36, CD42b, CD45RA, CD90, CD105) have been previously described.^24,25 Cocktails were prepared for both new and legacy assays and the cocktail stability of the 28-color assay was validated following the method described previously).³⁷

Table 1.

Immunophenotyping panels for new 28-color myeloid assay and legacy 10-color myeloid assay.

New 28-color Myeloid Assay	Legacy 10-color Myeloid Assay
	Tube #1	Tube #2	Tube #3
CD45 BUV395	CD45 BUV805	CD45 BUV805	CD45 BUV805
CD34 APC	CD34 APC	CD34 APC	CD34 APC
CD38 APC-A750	CD38 APC-A750	CD38 APC-A750	CD38 APC-A750
HLA-DR AF700	HLA-DR BUV395	HLA-DR BUV395
CD13 PE-Cy7	CD13 PE-Cy7	CD13 PE-Cy7
CD33 BV421	CD33 PE		CD33 PC7
CD15 BUV805	CD15 FITC
CD19 BUV615	CD19 BV421
CD71 BV480	CD71 APC-A700
CD117 PE	CD117 PC5
CD11b BV605		CD11b BV605
CD14 PC5		CD14 PC5
CD16 BUV737		CD16 APC-A700
CD64 BUV661		CD64 FITC
CD123 BB700		CD123 PE
CD2 BV711			CD2 BV421
CD4 BV650			CD4 APC-A700
CD5 BB660			CD5 BB700
CD7 BB515			CD7 BB515
CD25 PE-DAZZLE594			CD25 PE-Dazzle 594
CD56 BV786			CD56 PE
CD10 BB790
CD36 BV750
CD45RA BUV496
CD105 BB630
CD371 (celc12a) BUV563
CD42b V547
CD90 BYG710

2.3. CELL PROCESSING / STAINING AND ACQUISITION

Flow cytometry staining for the initial method evaluation was performed using standard protocols with simultaneous ammonium chloride lysis and fixation.^24,25 In brief, up to 200 uL of blood or bone marrow containing approximately 2 million cells were stained with a cocktail of antibodies for 15 minutes at room temperature, followed by ammonium chloride lysis with 0.25% formaldehyde for 15 minutes. Cells were washed with phosphate-buffered saline with bovine serum albumin and sodium azide (PBA) and cell pellets were resuspended using 100 uL of PBA. Fluorescence minus one and single stain testing were performed using the same methodology. Sample and cocktail stability, and linearity assessment were also done according to previously described methods.³⁷ The legacy assay was acquired on BD LSR Fortessa X-20 flow cytometers (BD Biosciences, Franklin Lakes, New Jersey, USA) with up to 500,000 events per tube, while the new 28-color assay was acquired on BD FACSymphony A3 flow cytometers (BD Biosciences) equipped with 5 lasers capturing up to 1 million events or until sample exhaustion (See Supplemental Table S1 for detailed configuration).

2.4. INSTRUMENT QUALITY CONTROL

The quality control of the instrument was performed following manufacturer’s recommendation and the protocol established by the laboratory. Briefly, both CS&T (BD Biosciences) and Spherotech rainbow beads (Spherotech Inc, Lake Forest, Illinois, USA) were acquired, and the target mean values for each parameter were tracked. The PMT was adjusted for parameters that shifted out of the validated acceptable range to ensure the compensation stability. The compensation for the assay was verified using a clinical sample on daily basis.

2.5. FLOW CYTOMETRY ANALYSIS

Manual analysis was performed using Woodlist software (generous gift of Dr. Brent Wood, Children’s Hospital of Los Angeles). The initial gating strategies were applied to defined the different lineage of cell populations (see Supplemental Figure 1). Intensities and expression of markers for myeloid blasts, mature monocytes, mature granulocytes, eosinophils, basophils, plasmacytoid dendritic cells (pDCs) were reviewed on normal samples, recorded, and compared between the legacy and new methods. Erythroid precursors, immature B-cells, CD4+ T-cells, and NK cells were also compared. The normal immunophenotypes of maturational stages of various cell types were defined using these normal samples, and deviations from the normal immunophenotypes of these cell types were considered abnormal.^{5,9–13,24,25,30} For abnormal samples, the percentage of the abnormal blast population and their immunophenotype were recorded and compared between methods.

2.6. UNIFORM MATRIX APPROXIMATION AND PROJECTION ANALYSIS (UMAP)

To demonstrate relevant biological relationships with the CD45/side scatter defined “blast” compartment we performed additional analysis using UMAP to synthesize two-dimensional data from high dimensional flow cytometry data. Cells from a dim CD45 gate in the 28-color myeloid assay from 5 normal control BM cases were exported as compensated .fcs files, down-sampled for equal representation (3000 cells per file), concatenated, and normalized for all fluorescence parameters using GaussNorm and plotted using UMAP using OMIQ software pipeline (OMIQ, Boston, MA, USA).^38,39

3. RESULTS

3.1. IDENTIFICATION OF HEMATOPOIETIC CELL SUBSETS

The increased number of antigens allowed efficient detection of precursor cell populations according to previously published immunophenotypes, including using expression of combinations of CD10, CD13, CD34, CD38, CD42b, CD45RA, CD71, CD90, CD105, CD117, CD123, CD371, and HLA-DR (See Figure 1).^{9,11,13,24,25} Cells with CD34+, CD38−, CD45RAm, CD90+ phenotype were defined as HSCs, while the MPP compartment was defined by a CD34+, CD38−, CD45RA−, CD90− phenotype. More differentiated precursor subsets were detected with LMPP compartment defined by an immunophenotype of CD10−, CD34+, CD38−, CD45RA+, CLP with CD10+, CD19−, CD34+, CD38+, CD45RA+, and GMP with CD10−, CD34+, CD38+, CD45RA+.

Figure 1.

A. Diagram of CD34+ hematopoietic cell maturation based on immunophenotypes in new 28 color myeloid assay. Maturation begins with HSCs which progress to MPPs and can then be split by CD45RA expression or lack thereof. CD45RA+ LMPP cells can differentiate into MLP, CLP, and GMP populations. CD45RA− EMP cells can differentiate into EoBMP and MEP populations; MEPs can further differentiate into MKP and EryP. B. Maturation stages demonstrated on histogram plots generated from the new assay. CD34+ cells can be first split into CD38−/dim and CD38+ populations. Within the CD38−/dim compartment, HSCs can then be identified by expression of CD90 and absence of CD45RA. MPPs are identified by absence of CD45RA and CD90. LMPP/MLP populations are characterized by CD45RA expression without CD90. LMPP and MLP can be split by expression of CD10 and CD117. Within the CD38+ compartment, CLP can be identified by expression of CD10 and CD45RA. GMP are identified by CD45RA expression without CD10. EMP population is negative for both CD10 and CD45RA. Within the EMP population, MEP can be further refined using absence of CD13 and CD123. Finally, MEP can be refined into EryP by CD105 expression, while MKP can be refined by expression of CD42b. Abbreviations: HSC: hematopoietic stem cell; MPP: multipotent progenitor; LMPP: lymphoid-primed multi-potential progenitors; MLP: multi-lymphoid progenitor; CLP: common lymphoid progenitor; GMP: granulocyte-monocyte progenitor; EMP: erythromyeloid progenitor; MEP: megakaryocyte-erythroid progenitor; EryP: erythroid precursor; MKP: megakaryocyte precursor. Figure adapted from Roshal and Gao, Am J Clin Pathol, 2024, in press.

EMP were initially defined with CD10−, CD34+, CD38+, CD45RA−, and CD90−. These precursors reportedly give rise to erythroid and megakaryocyte lineages, but also basophils, eosinophils, and mast cells. The myeloid committed progenitors in this compartment, referred to as eosinophil/basophil/mast cell precursors (EoBMP) could be identified via expression of Clec12a, CD13, and CD123. MEP were refined by excluding cells with CD13 and CD371 expression. The MEP population could be further refined into committed erythroid precursors (EryP) that are CD71 bright and CD105+ and committed megakaryocytic precursors (MKP) could be refined with an immunophenotype of CD42b+ and CD105−.

Additional precursor populations were detectable using previously described methods.^24,25 We also generated UMAP assessment using cells from a dim CD45 gate in the FC assay from 5 normal control BM samples. Cells were concatenated and plotted in 2 dimensions and using heat-maps for individual antigens identified markers important to detection of various precursor populations. As expected, the important antigen patterns for the detection of various precursor populations matched the expected immunophenotypes (See Figure 2).^24,25 These subsets were used for comparison to suspected abnormal populations in the context of MRD for more precise DfN analysis

Figure 2.

UMAP heatmap plots showing cells with dim CD45 from 5 combined normal bone marrow samples. Cell populations are clustered based on antigen expression, red-arrows and labels point out specific populations. Abbreviations: UMAP: uniform matrix approximation and projection.

3.2. ASSAY SENSITIVITY AND CELL RECOVERY USING SPIKED SAMPLES

To establish analytical sensitivity of the new 28-color myeloid FC assay in AML MRD testing, dilution experiments were performed. Five normal bone marrow aspirate samples, without FC evidence of MN but with normal CD34 and CD117 positive blasts, were spiked with five samples containing abnormal myeloid blasts with variable immunophenotypes at targeted dilution points of 1%, 0.1% and 0.01%. As suggested, our proposed LLoQ (0.1%) was placed in the mid-range of the dilution points.^40,41 Blanks without the spike, original abnormal samples, and the admixed samples were stained using 28-color myeloid tube following the same established staining protocol. Triplicates were performed at the proposed LLoQ of 0.1%. See an example of blank and admixed sample analysis in Supplemental Figure 2.

Near the targeted 0.1% dilution point, the final spiked median targeted abnormal myeloid population was measured with a median of 0.07% (range 0.04 to 0.09%) of total WBC. The median abnormal cell population detected by the new assay was 0.05% (range 0.03–0.06%) of WBC, with a median coefficient of variation (CV) of 7.42% (range 2.84–13.43%). The median number of detectable cells was 203 (range 116–477) with a CV of 7.15% (range 1.09–17.26%). The median abnormal cell recovery in the spiked samples was 68% (range 63–86%), in a median of 415,757 (348,591 to 880,686) WBCs acquired. These CVs were under the generally acceptable CV of 20%, validating an LLOQ of 0.05%.⁴²

To determine LLoD, We first performed limit of blank (LOB) experiments according to previously described standards, with a calculated LOB of 8 events or 0.002% of WBC.⁴² The LLoD based on calculation of LOB + 1.645SDblank or Mean blank + 3SDblank) was 12 events or 0.003% of WBC. To verify the samples at 0.01% are still above LLoD, spiked samples with abnormal population 10-fold below the proposed LLoQ (0.01%) were examined. The median of targeted abnormal population pre-staining was 0.007% (range 0.004 to 0.009%) with a median of 27 cells (range 16–63 cells). The median of detected abnormal population post-staining was 0.007% (range 0.003–0.008%) with a median of 26 cells (range 13–59 cells). All 5 spiked samples had the events detected above the calculated LLoD at 0.003% or 12 events, validating the assay can successfully detect abnormal myeloid blasts at least 0.01% or 20 events (below the validated LLoQ at 0.05%). According to the dilution experiment, we verified that the assay’s LLoQ is 0.05%, with the conservatively set LLoD of 0.01% or 20 events (whichever comes first), in a minimum of 400,000 WBCs acquired. However, it is important to note that not all myeloid stem cell disorders show abnormal immunophenotype with this assay due to heterogeneity of the immunophenotypes of various MNs.

3.3. QUALITATIVE DETECTION OF ABNORMAL POPULATIONS IN CLINICAL SAMPLES INCLUDING AML MRD SAMPLES

Split-sample analysis was performed on 151 other samples including 36 PB, 89 BM, 11 tissue, 10 CSF, and 5 other fluid samples (See Supplemental Table 2) and analyzed blindly by two different analysis. Of these, 92 contained populations of abnormal myeloid cells, and included 40 cases with blast proportion >20% and 52 cases with blast proportion <20% by both cytomorphology and immunohistochemistry. Of those with <20% blasts, there were 37 cases with blasts <5%. There was 100% concordance between the two assays by qualitative detection of abnormal populations. There were 92 true positives detected, 59 true negatives, and no false negatives or false positives, resulting in 100% sensitivity and specificity.

Within this group of 151 samples tested, we also evaluated the concordance between legacy and new methods for AML MRD detection in 20 samples. We defined AML MRD as having a history of AML, being treated with any AML-directed therapy, and having <5% blasts by immunohistochemical and cytomorphologic assessment of the BM, with an abnormal myeloid blast population detected by the legacy FC method. Of these 20 patients, 9 patients had received chemotherapy, 8 had received only hypomethylating agents, and 5 had received a combination of therapies over long disease courses. 10 patients were status post allogeneic stem cell transplant. In these 20 true AML MRD cases, the two assays also showed 100% concordance, and this agreement was also seen in the 37 cases that included non-AML MNs where the total blast count was <5%.

3.4. NEW ASSAY QUANTITATIVE COMPARISON WITH LEGACY METHOD

The new assay’s quantitative performance was also compared to the legacy 3-tube assay using 37 normal samples.^24,25 The 37 normal samples had no evidence of MN involvement by morphologic, immunohistochemical, and FC evaluation, and consisted of 11 BM, 10 PB, 6 tissue, 6 CSF, and 5 other fluid samples. Correlation of percentages of the cell subsets in the 37 normal samples including myeloid blasts, mature monocytes, mature granulocytes, eosinophils, basophils, plasmacytoid dendritic cells (pDCs) were analyzed using Deming regression and Pearson’s correlation. Deming regression for CD34+ myeloid blasts showed a slope of 0.96 (95% confidence interval [CI] 0.84 to 1.07) and a Y-intercept of −0.003 (95% CI −0.04 to 0.3) (See Figure 3A). Other populations showed similar results with slopes close to 1 and intercepts close to 0 (see detailed results in Supplemental Figure 3).

Figure 3.

A. Deming regression of CD34+/CD19− blasts detected by the new assay compared to the legacy assay in a set of 37 normal samples, showing a slope of 0.96 (95% CI 0.84 to 1.07) and intercept of −0.003 (95% CI −0.04 to 0.03). Pearson’s correlation coefficient testing showed an r² of 0.98. B. Deming regression of all cases (n = 92) with abnormal blast proportions detected by the new assay compared to the legacy assay showing a slope of 1.03 (95% CI 1.01 to 1.05) and intercept of −0.35 (95% CI −0.61 to −0.10). Pearson’s correlation coefficient testing showed an r² of 1.00. This showed a slight fixed bias. C. However, this bias was not present in cases with <5% abnormal blasts (n = 37), which demonstrated a slope of 0.98 (95% CI 0.90 to 1.06) and an intercept of −0.02 (−0.07 to 0.03) with r² of 0.97, and therefore was considered clinically negligible.

Split-sample analysis was also performed on 151 other samples described in section 3.3 (See Supplemental Table 2). Correlation of percentages of abnormal blast populations by Deming regression and Pearson’s correlation showed a slope of 1.03 (95% CI 1.01 to 1.05), intercept of −0.35 (95% CI −0.61 to −0.1), and r² correlation coefficient of 1.00 in the abnormal samples (See Figure 3B). There was a very slight fixed bias seen in blast quantitation (95% CI 1.01 to 1.05) for all abnormal samples. Of the 151 cases, there were 37 with <5% abnormal myeloid blasts by FC. In this group, Deming regression and Pearson’s correlation showed a slope of 0.98 (95% CI 0.90 to 1.06) and intercept of −0.02 (95% CI −0.07 to 0.03%) and r² correlation coefficient of 0.97 (See Figure 3C). In cases with <5% blasts, there was no evidence of bias in the new method. Therefore, the very slight fixed bias noted in blast quantitation in the overall abnormal sample pool was considered clinically negligible. We performed the same correlation tests in these 151 samples with other populations including granulocytes, monocytes, eosinophils, basophils, and pDCs, with similar results (See Supplemental Figure 4).

3.5. INTER-OBSERVER VARIATION

To assess if trained observers could achieve concordant results in samples with MN, including AML MRD, 20 cases (10 positive with abnormal population ranging from 0.02% to 36.9% of WBC, and 10 negative cases) were selected for inter-observer comparison (See Table 3 and Supplemental Table 3). The cases included various immunophenotypes of the abnormal blasts to represent the heterogeneity of MN. Files were deidentified and provided to the observers to perform the analysis starting with raw data .fcs files. The analysis was conducted by five trained interpreters (QG, AB, OM, NN, and SJZ) blindly and independently using DfN approach, as no diagnostic phenotype was provided. QG is an attending flow cytometrist while additional interpreters are technologists proficient in myeloid analysis using the legacy method and received initial training in the new method, consisting of a minimum of 20 normal and 20 abnormal myeloid cases over a one-week period.

Table 3.

Interobserver agreement studies.

		Abnormal blast population proportion detected by observer					Qualitative result Positive/Negative		Case type
Case	Sample Type	QG	AB	OM	NN	SJZ	Positive	Negative
#1	PB	2.7%	2.8%	2.7%	2.7%	2.8%	5	0	CD34+CD117+ blasts
#2	BM	0%	0%	0%	0%	0%	0	5
#3	PB	0%	0%	0%	0%	0%	0	5
#4	BM	22.5%	19.8%	22.8%	19.3%	20.2%	5	0	CD34variable and CD117+ blasts
#5	BM	0%	0%	0%	0%	0%	0	5
#6	BM	0%	0%	0%	0%	0%	0	5
#7	BM	9.5%	10.2%	8.9%	8.8%	10.1%	5	0	CD34+CD117+ blasts
#8	BM	29.1%	29.3%	29.6%	29%	28.6%	5	0	CD34partialCD117+ blasts
#9	BM	0%	0%	0%	0%	0%	0	5
#10	BM	1.5%	1.4%	2%	2.4%	1.5%	5	0	sum of distinct abnormal CD117+ subset and CD34+ subset
#11	BM	0.2%	0.15%	0.25%	0.24%	0.2%	5	0	CD34−CD117+ blasts
#12	BM	0%	0%	0%	0%	1.0%**	1	4
#13	BM	0%	0%	0%	0%	0%	0	5
#14	BM	0.12%	0.095%	0.1%	0.12%	0.1%	5	0	Spiked sample; target population 0.1%
#15	BM	0.02%	0%*	0.04%	0.07%	0.02%	4	1	CD34+ with abnormal cell events >80 cells; molecular shows positive Runx1-RUNXT1 AML1-ETO fusion
#16	BM	0%	0%	0%	0%	0%	0	5
#17	BM	36.8%	19.6%	36.7%	26%	20.4%	5	0	AML with monocytic differentiation, blasts negative for CD34/CD117
#18	BM	0%	0%	0%	0%	0%	0	5
#19	BM	0.5%	0%*	0.47%	0.47%	0.7%	4	1	Spiked sample; target population 0.5%
#20	BM	0%	0%	0%	0%	0%	0	5

^*False negative result (2 total);

^**False positive result (1 total). Abbreviations: PB: peripheral blood; BM: bone marrow; AML: acute myeloid leukemia; MDS: myelodysplastic syndrome.

Fleiss kappa measurement was used to measure the reliability of agreement among the interpreters by categorizing each case as either positive or negative. The percentage of abnormal blasts recorded by each interpreter was used to calculate the intraclass correlation coefficient (ICC) to assess quantitative agreement. ICC estimates and their 95% confidence intervals were calculated based on a single-rating, absolute-agreement, 2-way-random-effects model.⁴³

Overall, 8/10 positive and 9/10 negative cases were consistently rated as positive or negative by the interpreters. Two cases with low levels of abnormal blasts (0.02% and 0.5%) were missed by one interpreter and one case that was negative for MN was misinterpreted as positive by another interpreter. The percent overall agreement was 94%, with the fixed-marginal kappa of 0.88 (95% CI 0.74 to 1.00), indicating a strong to almost perfect agreement. The ICC value obtained was 0.96 (95% CI 0.92 to 0.98) concluding excellent reliability. Two reviewers generally perform the analysis (technologist and pathologist). No cases had errors by two reviewers.

3.6. QUALITATIVE ASSESSMENT FOR MYELOID NEOPLASIA

Using the expanded panel in the new assay, we were able to clearly demonstrate the full range of bone marrow hematopoietic maturation stages, including the heterogeneous CD34 positive blast compartment (see Supplemental Figure 5). The added antigens allowed for estimation of abnormal cell populations into the previously defined immature precursor subsets. Some MN showed abnormal myeloid blasts with a phenotype suggesting HSC predominance (Figure 4A). Others demonstrated immunophenotypes in keeping with more differentiated precursors in the CD34 positive compartment such as GMP and MEP (Figure 4B–C). Many cases had abnormal blasts with immunophenotypes in keeping multiple different precursor compartments, and within the full cohort of 92 cases with abnormal myeloid blasts, they ranged from 0.01% to 99.6% of WBC. In the 37 cases with abnormal myeloid blasts and <5% blasts, the blasts in these 37 cases ranged from 0.02 to 4.7% of WBC.

Figure 4.

Examples of detectable abnormal myeloid blasts at levels <5% and identification of their likely cell of origin. A. This case of AML with low-levels of detectable MRD demonstrated an abnormal myeloid blast population (0.006% of WBCs) with an HSC-like immunophenotype (colored red), with abnormal expression of CD45RA, slightly bright expression of CD13 and CD33, and partial, aberrant co-expression of CD7 and CD25. This was corroborated by cytogenetic studies which detected a loss of chromosome 7 in few cells. B. An example of an AML with low-levels of MRD with a more GMP-like immunophenotype (CD34+/CD38+/CD10−/CD45RA+) measuring 0.08% of total WBC, showing abnormal CD13 expression without CD33 expression, bright CD117 expression, and CD45RA expression without concurrent CD371 expression. This finding was corroborated by molecular testing which detected persistent IDH2, STAT5B, and ARID2 mutations. C. An example of an MDS with low blasts and ring sideroblasts with an MEP-like immunophenotype (CD34+/CD38+/CD10−/CD90−/CD13−/CD123−/CD371−) measuring 2.2% of total WBC (colored red) and showing aberrant CD7 expression and uniformly bright HLA-DR. This case was corroborated by molecular testing which demonstrated mutations in ASXL1, IDH2, SETBP1, and SRSF2. Abbreviations: AML: acute myeloid leukemia; MRD: minimal/measurable residual disease; WBC: white blood cell; HSC: hematopoietic stem cell; GMP: granulocyte-monocyte precursor; MEP: megakaryocyte-erythroid progenitor.

There were also significant drifts between compartments in post-therapy AML samples positive for MRD. We analyzed 20 paired diagnostic vs. MRD-positive samples (See Table 2). In these 20 paired cases, 19 of 20 diagnostic samples contained blast populations with immunophenotypic features of multiple differentiation compartments. One case showed an exclusively GMP-like population which persisted at MRD. In the 19 mixed differentiation cases, MRD-positive samples had different distributions of abnormal blasts compared to the diagnosis. Notably LMPPs were relatively expanded at MRD stage (mean 24% vs 58%, p < 0.001, paired Student t-test), while GMP-like populations were relatively reduced (46% vs 26%, p = 0.01), while MEP and EMP populations were not detected at MRD stage (Figure 5A). An extreme example of compartment redistribution with lymphoid priming may be seen in case #8 where at diagnosis 53% of blast cells had LMPP phenotype, 46% had GMP phenotype, and 1% showed an MLP phenotype (Figure 5B), while exclusively MLP-like immunophenotype MRD was seen after multiple lines of therapy (blasts 0.02%) (Figure 5C). These findings confirm that blasts in AML are not homogeneous and that certain subpopulations may show a greater propensity to persist in MRD stage.

Table 2.

Precursor compartment categorization of abnormal leukemic cells on paired samples at time of AML diagnosis and post-therapy with MRD positivity.

	Diagnostic sample							MRD positive sample
Case	HSC/MPP	LMPP	MLP	GMP	EMP	MEP	Other	HSC/MPP	LMPP	MLP	GMP	Other
#1				100%							100%
#2		70%		30%					100%
#3				54%		34%	12%		85%	15%		Abnormal immature B-cells
#4		11%		89%					70%		30%
#5		60%		40%					60%		40%
#6	31%	4%		2%		63%		100%
#7		32%		54%		12%	2%		100%
#8		53%	1%	46%						100%
#9		50%		50%					100%
#10	93%	7%							83%		17%
#11		13%		87%					47%		53%
#12		8%		92%					37%		63%
#13				85%		15%					100%
#14	61%					39%		100%
#15	45%	15%		15%		25%			37%		63%
#16		15%		85%					73%		27%
#17		20%		80%					100%
#18	12%	1%		10%	62%		15%		73%		27%
#19		37%					67%		100%
#20		83%		10%		7%			100%

Most AMLs show multiple sub-populations in different differentiation compartments. After therapy, MRD positive samples often show shifts in compartment distributions. Abbreviations: MRD: minimal/measurable residual disease; HSC: hematopoietic stem cell; MPP: multipotent progenitor; LMPP: lymphoid-primed multi-potential progenitor; MLP: multi-lymphoid progenitor; CLP: common-lymphoid progenitor; GMP: granulocyte-monocyte progenitor; EMP: erythromyeloid progenitor; MEP: megakaryocyte-erythroid progenitor.

Figure 5.

Phenotypic drift after therapy. A. Comparison of majority of abnormal blast compartment distributions at AML diagnosis and after treatment at MRD stage. 20 cases were plotted with overall means represented in the pie chart. B. This case of AML (patient #8 from Table 2) demonstrated 35% abnormal blasts mostly with an overlapping LMPP-like (53%) and GMP-like (46%) immunophenotype, with variable CD38, abnormal expression of CD7, variable CD33, positive CD45RA, positive CD117, and absent CD10. A small minority (1% of blasts) showed expression of CD10 suggestive of a more MLP-like immunophenotype. C. After therapy, MRD was detected by flow cytometry in the same patient, with 0.02% abnormal myeloid blasts (red population, background normal blasts colored blue). At this time point, the abnormal blasts demonstrated a purely MLP-like (100%) immunophenotype, with dim CD38, abnormal expression of CD7, positive CD45RA, negative CD117, and now showing positive CD10. Both the original diagnostic AML and the MRD positive specimen demonstrated mutation in RUNX1. Interestingly, new cytogenetic abnormalities including gain of chromosome 8 in 1.7% of cells and loss of chromosome 20q in 4.7% of cells were detected by FISH in the MRD positive specimen but not the original diagnostic AML specimen.

Abbreviations: AML: acute myeloid leukemia; MRD: minimal/measurable residual disease; HSC: hematopoietic stem cell; MPP: multi-potent progenitor; LMPP: lymphoid-primed multi-potential progenitors; MLP: multi-lymphoid progenitor; GMP: granulocyte-monocyte progenitor; EMP: erythromyeloid progenitor; MEP: megakaryocyte-erythroid progenitor.

3.7. UTILIZATION IMPROVEMENTS VERSUS LEGACY METHOD

Using the previous 3-tube assay required up to 600 uL of sample per assay (up to 200 uL per tube), versus a maximum 200 uL per sample in a single tube in the new assay, a 3-fold reduction in sample utilization per test. Reagent savings were realized by the elimination of duplicative antibodies in the single vs. 3-tube approach. The assay also resulted in more efficient use of laboratory equipment and supplies. The assay also reduces FC technologist’s required labor time, though this is difficult to quantify precisely.

4. CONCLUSIONS/DISCUSSION

Flow cytometry based analysis is a rapid method for evaluating AML MRD. However, this evaluation often requires many markers in a panel, which is challenging to implement without splitting a sample into multiple tubes for testing on 8–10 color cytometers. A limitation of the multiple-tube approach is that only one or two specific lineage populations can be reviewed per tube. For instance, the evaluation of later-stage maturing population may be separated from the evaluation of earlier progenitors. Hematopoiesis, however, is a continuous and heterogenous process, and AML MRD analysis requires an evaluation within the context of overall maturation pattern of different myeloid lineage populations. Such a comprehensive analysis is difficult to achieve using multiple separate tubes.

Recently, higher parameter flow assays for AML MRD have proven to be effective, such as a 27-color assay using spectral FC that significantly improves the detection capability by increasing the information content, thereby enhancing confidence in MRD interpretation.⁴⁴ However, the panel primarily focuses on cross-lineage markers and myelomonocytic lineage markers for AML evaluation, which does not allow for a granular analysis of different maturational stages of hematopoietic progenitors. These progenitors often harbor aberrant residual leukemic cells that may show loss of cross-lineage markers after therapy. Other authors also validated a 12-color assay, but this still requires multiple tubes for analysis.²³

We refined the approach, performed more extensive validation, and clinically deployed a 28-color single tube assay on a conventional cytometer. The newly developed, 28-antigen FC panel for myeloid assessment successfully identifies the full range of known hematopoietic precursor populations and detects abnormal precursor populations in MNs agnostic to their stage of differentiation. This assay has LLoD of 0.01% of WBCs based on spiked sample experiments, allowing for use as a clinical MRD diagnostic tool, though a thorough assessment of the assay performance near the LoQ or LLoD requires further study. It is built using a similar backbone to a well-established myeloid FC assay and can be performed in virtually any sample type including BM, PB, lymph node, and other tissue samples, as well as fluid samples including fine needle aspirate specimens.^24,25,45

In the BM, there is significant heterogeneity within the immature cell compartment comprised of CD34+ hematopoietic precursor cells.^9,13 While our legacy testing method could detect some different populations within the CD34+ compartment, the addition of new antigens including CD10, CD36, CD42b, CD45RA, CD90, CD105, and CD371 in the new assay adds significant value to this assay in its ability to detect small early precursor populations with increased granularity. With these additions, the assay reliably separates precursor populations including HSC, MPP, LMPP, CLP, GMP, EMP, and MEP.^9–13 The addition of new markers to detect the known range of CD34+ precursor cells may also improve sensitivity and/or specificity in the diagnosis of MN and AML MRD. For example, CD36 may aid in the detection of erythroid dysplasia in MN.⁴⁶ In another example, CD45RA which is normally expressed in more differentiated blasts including GMP, LMPP, and CLP, may show dyssynchronous expression in earlier progenitors.⁴⁷

There is also interest in how differentiation of a leukemic cell of origin affects the disease.^32,35,36 For example, some models suggest that KMT2A (MLL) rearrangements occurring in HSCs are more likely to result in AML and less responsive to chemotherapy, compared to those occurring in GMP.^32,36 This assay allows future categorization of AMLs by corresponding precursor immunophenotypes. Many leukemic cell populations also show immunophenotypes that overlap multiple precursor compartments, which can shift after therapy. In the future, this assay may provide insights into which precursor phenotypes are most likely to survive therapy and contribute to clinical relapse.

In summary, this new assay shows excellent concordance with the previously used 3-tube assay, with excellent sensitivity and specificity for MRD detection. While this new assay provides increased information density and ability to detect more precursor cell types than the previously used assay, it simultaneous improves laboratory efficiency, saving valuable technologist labor, equipment costs, and flow cytometer machine time. The improvement in sample usage efficiency by using a single tube versus a 3-tube assay may further improve sensitivity for MRD detection, particularly in hypocellular samples which are especially relevant in an evolving therapeutic landscape in AML. A single-tube approach, grounded in a thorough understanding of the maturational stages of early progenitors, could help streamline the gating strategy, standardize the analytical skills among different analysts, and pave the way for automated MRD analysis in the future.