Validation Study of Analytical Methods for Multiparameter Flow Cytometry-Based Measurable Residual Disease Assessment in Acute Myeloid Leukemia

Abstract

The standardization of multiparameter flow cytometry-based measurable residual disease (MFC-MRD) assessment in acute myeloid leukemia (AML) lacks clear criteria to define leukemia-associated immunophenotypes (LAIPs). In addition, the most specific/sensitive aberrations used to define LAIPs are often partially expressed by the leukemic clone at diagnosis, raising questions about their reliability for accurate MRD quantification. To address this, we investigated whether the quantification of LAIP+ cells reflects residual disease in cases of partial LAIP expression. The following two MFC-MRD approaches were evaluated by comparing their results to RT-qPCR for NPM1 mutations: (1) the LAIP-method, wherein all cells within the patient-specific template created at diagnosis are counted without further gating; (2) the LAIP-based different-from-normal (DfN)-method, wherein cells+ for LAIP-specific aberrant markers are further selected. A total of 125 bone marrow samples from 25 NPM1-mutated AML patients were studied. Our data demonstrate that the LAIP-based DfN-method improves the MFC-MRD accuracy and comparability with molecular MRD. ROC analysis identified cut-offs of 0.034% and 0.095% to discriminate positive/negative results in patients receiving intensive chemotherapy and hypomethylating agents, respectively. We also found distinct accuracy degrees based on the LAIP-specific aberrant markers used for MRD assessment. These results refine the MFC-MRD method and highlight the importance of therapy-specific MRD cut-offs and LAIP classification based on specificity and sensitivity.

Keywords: acute myeloid leukemia; leukemia-associated immunophenotypes; measurable residual disease; multiparameter flow cytometry.

Figure 1

Flow chart illustrating the selection of AML patients, the application of two MFC-MRD approaches (LAIP- and LAIP-based DfN-methods) for MRD monitoring, and the comparison to RT-qPCR quantification of NPM1-mutated transcripts.

Figure 2

Gating strategy to create a specific analysis template for an AML patient at diagnosis. Representative example of the gating strategy used to select/identify and characterize the leukemic population (Pt#8, Table 1; Panel 3, Table S2), including doublets exclusion (P1), definition of white blood cells (WBCs) as CD45⁺ cells (P2), selection of leukemic blasts based on CD45 expression (P3) and FSC vs. SSC features (P4), and clustering of immunophenotypic characteristic of AML cells by sequential gates (from P5 to P10). Sequential gates must include the total leukemic population and absence or partial and total expression of markers.

Figure 3

Description of the two MFC-MRD analytical methods. (A): MRD quantification in the cases of total LAIP expression of AML cells at diagnosis (90–100% of AML cells, Pt#23, Table 1). After applying the patient-specific analysis template, the cell population included in the last gate of the hierarchy (in this case, P10) is directly quantified and identified as MRD cells (LAIP-method). (B): MRD quantification in the cases of partial LAIP expression of AML cells at diagnosis (20–90%, Pt#7, Table 1). Using the patient-specific analysis template, the cell population included in the last gate of the hierarchy (in this case, P10) is analyzed. The LAIP-method quantifies all cells included in gate P10 as MRD cells, while the LAIP-based DfN-method selects only cells positive for the LAIP-specific aberrant marker expressed at baseline (LAIP+ cells).

Figure 4

Comparison of MRD results obtained by RT-PCR for NPM1 mutations (NPM1-MRD) and the two MFC-MRD approaches. (A): Representative histogram of the percentage of positive/negative MRD samples (n = 125) obtained with NPM1-MRD and MFC-MRD, using LAIP- and LAIP-based DfN-methods (Chi-square test; *** p ≤ 0.001). (B): Concordance between the LAIP- and LAIP-based DfN-method results and the NPM1-MRD outcomes. The results are shown as stacked histogram and expressed as percentage of both positive/negative (purple/pink) and false-positive/negative (blue/green) MRD samples (Chi-square test; ** p ≤ 0.01). The percentages of concordance, false-positive/negative, sensitivity and specificity, and of the positive/negative predictive value (PPV, PNV) of the MFC analytical methods are reported.

Figure 5

Comparison of MRD results obtained by RT-PCR for NPM1 mutations (NPM1-MRD) and the two MFC-MRD approaches according to the percentage of LAIP expression (total or partial) of AML cells at diagnosis. (A): Concordance between the results of MFC-MRD (LAIP-method only) and NPM1-MRD in the evaluation of MRD samples from patients with total expression (90–100% of AML cells) of LAIP-specific aberrant lineage markers in AML cells at baseline (n = 4 patients; n = 11 MRD samples). (B): Concordance between LAIP- and LAIP-based DfN-methods and NPM1-MRD results in the evaluation of MRD samples from patients with partial expression (20–90% of AML cells) of LAIP-specific aberrant lineage markers in AML cells at baseline (n = 21 patients; n = 114 MRD samples). The results are shown as a stacked histogram and expressed as a percentage of both positive/negative (purple/pink) and false-positive/negative (blue/green) MRD samples (Chi-square test; ** p ≤ 0.01). The percentages of concordance, false-positive/negative, sensitivity and specificity, and of the positive/negative predictive value (PPV, PNV) of the MFC analytical methods are reported.

Figure 6

Accuracy of the two MFC-MRD analytical methods in evaluating MRD samples from patients with partial expression (20–90% of AML cells) of LAIP-specific aberrant markers in AML cells at diagnosis (n = 21 patients; n = 114 MRD samples). (A,B): ROC curves obtained by distributing the leukemic blast percentages estimated from the LAIP-based DfN- and LAIP-methods according to the NPM1-MRD outcome. Area under curve (AUC), Standard Error, 95% confidence interval, and p-value of each ROC curve were reported.

Figure 7

Accuracy of the two MFC-MRD analytical methods according to therapeutic setting. (A,B): Concordance between the results of MFC-MRD, using LAIP- and LAIP-based DfN-methods, and NPM1-MRD in evaluating MRD samples from patients receiving intensive chemotherapy regimens (post-CHT MRDs, n = 62) and Ven+HMA-based therapies (post-Ven+HMA MRDs, n = 58). The results are shown as a stacked histogram and expressed as a percentage of both positive/negative (purple/pink) and false-positive/negative (blue/green) MRD samples (Chi-square test; * p ≤ 0.05). The percentages of concordance, false-positive/negative, sensitivity and specificity, and of the positive/negative predictive value (PPV, PNV) of the MFC methods are reported. (C,D): ROC curves obtained by distributing the leukemic blast percentages estimated from the LAIP- and LAIP-based DfN-methods according to the NPM1-MRD outcome, analyzing post-chemotherapy (post-CHT MRDs, n = 62) and post-Ven+HMA-based therapy (post-Ven+HMA MRDs, n = 58) MRD samples. Area under curve (AUC), Standard Error, 95% confidence interval, and p-value of each ROC curve were reported.