The role of flow cytometry in the classification of myeloid disorders

Unclear cytopenia is the most common indication for BM examination and can have numerous causes, both secondary (reactive) and neoplastic. Chronic (autoimmune) disease, infection, nutritional deficiencies, and treatment-related cytopenia are common non-neoplastic causes. Among hematopoietic neoplasms, the most common causes of new onset cytopenia are MDS and acute myeloid leukemias (AML) in adults and acute lymphoblastic leukemia (ALL) in children. Mature B cell neoplasms can also lead to pancytopenia, but such presentations, apart from hairy cell leukemia, are rare unless in cases with significant BM infiltration. Anemia is a presenting hallmark of plasma cell myeloma, and concomitant thrombocytopenia and granulocytopenia can be seen in up to 10% of patients at the time of diagnosis. In these situations, FCIP is extremely helpful to either confirm or to rule out a hematological neoplasm, particularly in the acute clinical setting of unclear (pan)cytopenia and in patients presenting with cytopenia and secondary dysplasia, mimicking MDS. This basic diagnostic approach is illustrated in Fig. 1 and by the following three clinical cases:

The myelodysplastic neoplasms are a group of clonal hematopoietic stem cell diseases characterized by cytopenia, morphologic dysplasia, ineffective hematopoiesis, recurrent cytogenetic abnormalities in ≈50%, and somatic mutations in ≈90% of patients [1, 14]. The current gold standard and first step in the initial diagnostic work-up of unclear cytopenia/suspicious MDS, after the exclusion of other causes of cytopenia, is the cytomorphological examination of the peripheral blood (PB) and bone marrow (BM). Importantly, dysplasia is not specific for MDS and may be present along with other disease in the BM, drug-related, or secondary to other non-neoplastic conditions. The diagnosis of lower-risk MDS subtypes (without excess blasts) can be particularly challenging on morphologic grounds. Importantly, FCM can detect even subtle immunophenotypic aberrancies in the precursor and maturing erythroid, myeloid, and monocytic populations in BM aspirates.

Numerous publications have addressed immunophenotypic abnormalities in MDS patients over the past two to three decades. Most of the earlier studies were qualitative and correlative in nature, identifying FCM aberrancies that correspond to the diagnosis of MDS. Since about the 2010s, there has been a stronger advocacy within the FCM community for a quantitative approach and standardization among different laboratories. A series of consensus guidelines and multicenter studies have been published by the European Leukemia Net (ELN) MDS working group on the use of FCM in the diagnostic work-up of MDS, which were recently updated in a special issue of Clinical Cytometry, including pre-analytical, analytical, and technical considerations as well as research results from the members of the group [12, 21, 28, 29]. A multicenter prospective evaluation of FCM aberrancies by the ELNiMDS Flow working group led to the recognition of 17 immunophenotypic markers that were independently related to MDS. FCM showed 80% concordance with cytomorphology when applying a lower cut-off of three aberrant markers [12]. Importantly, this was independent of the number of affected cell lineages, in contrast to previously established guidelines and recommendations. However, there is no single specific FCM marker for MDS and the megakaryocytic lineage is best assessed by cytomorphology in combination with BM histology.

FCIP can help with blast enumeration and identify phenotypically aberrant blasts and be used for treatment follow-up by monitoring previously detected immunophenotypic abnormalities. The percentage of myeloid progenitor cells is informative, but should be correlated to the blast count by routine cytomorphology and BM histology including CD34 immunohistochemistry [25]. Aberrant immunophenotypes of CD34⁺ cells may indicate dysplasia per se and the finding of > 2% CD34⁺ myeloid progenitors was highly associated with MDS [12].

Several diagnostic MDS FCM scores have been published (recently reviewed in [19]), including the FCM scoring system, the Ogata score, the RED score and ELN-NEC, and the integrated flow score (iFS), the latter encompassing the analysis of myeloid progenitors, granulo- and monopoiesis, and nucleated erythroid cells. The most commonly used MDS FCM score is the Ogata score, based on four parameters [20]. Most MDS-FCM scores have been validated in clinical studies, comparing MDS patients to patients with secondary (reactive) cytopenias and normal controls. The iFS was shown to have the highest accuracy with respect to MDS diagnosis. [19].

The role of FCIP in the diagnosis and classification of myeloproliferative neoplasms (MPN) is not well defined. The integration of molecular findings with BM morphology and PB counts remains the cornerstone of diagnosis, including the BCR-ABL1 translocation in chronic myeloid leukemia (CML), the MPN-associated mutations JAK2 V617F, JAK2 exon 12, MPL 515L/K, and calreticulin (CALR) for the classical BCR::ABL1-negative MPN subtypes, and the presence of driver mutations in the colony-stimulating factor 3 receptor (CSF3R) in chronic neutrophilic leukemia. However, MPNs exhibit frequent immunophenotypic alterations in both myeloid progenitors and in the maturing myelomonocytic compartment (summarized in Table 1, including references from a literature review). Published data demonstrate that these changes are part of constellational findings in MPN and correlated with adverse clinical and morphological features, such as, for example, an increased blast percentage and an abnormal karyotype. Integration of the Ogata score using FCM analysis of the BM aspirate in patients with PMF resulted in improved prognostic stratification. The enumeration and immunophenotypic characterization of blasts by FCM remains important for the distinction between reactive versus neoplastic conditions and between an accelerated phase of MPN (MPN-AP) and de novo acute leukemia (AL). Importantly, patients with lower blast percentages (5–9%) may have a similar clinical course and prognosis to patients with MPN-AP, and FCIP can help in the early detection of progressive disease. Herborg et al. found that enumeration of circulating immature cells by FCM, including aberrant surface expression, was a promising discriminative tool in MPN diagnostics and a means of monitoring patients longitudinally [8].

In the ICC, immunophenotyping is now included as a criterium for the detection of lymphoblastic crisis in chronic myeloid leukemia (CML), using a threshold of 5% lymphoblasts in the PB or BM [1]. However, this cut-off is arbitrary, and some data suggest that the finding of any bona fide lymphoblasts should raise concern that a lymphoblastic crisis may be imminent. For this reason, a specific cut-off is not included in the WHO-HAEM5, while the presence of “increased” lymphoblasts still remains one of the criteria for blast phase in CML. Since FCM can identify aberrant lymphoid blast populations at much lower levels (< 0.1%), integration in the routine follow-up of CML patients using similar panels as for ALL MRD measurement should be considered.

FCIP can also assist in the diagnostic work-up of other subtypes within the MPN category, particularly in cases with a CMML-like clinical presentation. For example, the clinical and morphologic findings in CEL, NOS, may overlap those of other MPNs, MDS, and MDS/MPN, as illustrated by a previous BM workshop report [11]. In these situations, FCIP can provide hints for correct classification through the identification of MDS-related abnormalities and/or the analysis of monocytic subpopulations, as discussed in the following section. Recent studies recognize the potential value of proliferative and apoptotic indices as diagnostic and prognostic markers in MPN [18].

This category of myeloid neoplasms is defined by overlapping pathologic and molecular features of MDS and MPN, clinically manifesting with various combinations of cytopenia and cytoses. The prototype and most common MDS/MPN is chronic myelomonocytic leukemia. Abnormal partitioning of PB monocyte subsets has been introduced as a new supporting diagnostic criterion, based on the observation that an increase of “classical” monocytes (CD14⁺⁺, CD16⁻) was highly sensitive and specific for CMML [26]. Subsequent studies have validated these findings and confirmed the ability to distinguish CMML from both reactive monocytosis and other myeloid neoplasms presenting with monocytosis (Table 1, including references from a literature review). FCIP may also assist to separate the two CMML subtypes, myelodysplastic vs. myeloproliferative, at an immunophenotypic level by using MDS-adapted FCM panels according to ELN recommendations.

The ICC recognizes clonal monocytosis of undetermined significance (CMUS) as a CMML precursor condition, based on persistent monocytosis in the presence of a myeloid neoplasm-associated mutation(s), but without BM morphologic findings of CMML. While FCIP of the PB is not listed as a co-criterion, it could potentially be used as a screening test and predictive marker for the presence of a somatic mutation and for developing CMML.

Other flow cytometric alterations in CMML (see literature review in Table 1 for additional references) are well described and like those found in MDS. For example, expression of CD56 is highly sensitive and specific for a diagnosis of CMML, but only when combined with other immunophenotypic features, including reduced expression of myeloid antigens and ≥ 20% immature monocytes, since CD56 overexpression can also be seen in non-neoplastic conditions.

Outside of CMML, few larger studies have evaluated FCM findings in MDS/MPN overlap. Li et al. found that all studied patients with MDS/MPN demonstrated at least one abnormality by FCM; myeloblast abnormalities were the most common phenotypic aberrancy detected [16]. Although the findings were not entity specific, their presence was helpful in substantiating a diagnosis of a myeloid neoplasm and helping to exclude a reactive process. Finally, FCM can also be used as a predictive marker for treatment response in this group of patients.

The myeloid/lymphoid neoplasms with eosinophilia and TK gene fusions (M/LN-eo-TK) represent a rare and challenging group of hematological neoplasms with highly variable clinical and morphologic presentation and course of disease [1, 14]. M/LN-eo-TK frequently manifest as a chronic myeloid neoplasm with or without eosinophilia resembling CEL, NOS, other MPN, MDS/MPN, or MDS. Other presentations include T or B acute lymphoblastic leukemia/lymphoma (ALL), AML, the blast phase of MPN, or mixed-phenotype acute leukemia (MPAL). Importantly, cases may present with disparate stages of disease (chronic vs. acute or transformed), disparate cell lineages, as primary extramedullary (nodal/extranodal) disease, or manifest first during follow-up and/or after treatment of the original diagnosis. Abnormal mast cell proliferations, detected by FCM or immunohistochemistry (IHC), are frequent findings in M/LN-eo with any of the recurrent TK fusion genes. Taken together, FCIP plays an important and central role in the diagnostic work-up of M/LN-eo for the detection of aberrant (sub)populations or multilineage involvement, as illustrated by numerous single case reports, including children, multicenter studies, and recent BM workshop reports [23, 24].

Mastocytosis is characterized by neoplastic proliferation of abnormal mast cells in at least one organ system, including the skin and BM, and can be divided into cutaneous mastocytosis (CM), systemic mastocytosis (SM), and mast cell sarcoma (MCS) [1, 14]. Advanced SM compromises aggressive SM, SM with an associated myeloid neoplasm (SM-AMN), and mast cell leukemia (MCL). In both classifications (ICC/WHO-HAEM5), the major diagnostic criteria for SM remain largely based on morphology, supported by immunophenotypic analysis. FCIP can be used to identify mast cell populations, which are recognized by their unique phenotype (CD117⁺⁺, CD45⁺, CD38⁻, CD33⁺, CD13^±, CD11c⁺, CD11b^±, CD71⁺). Neoplastic mast cells usually demonstrate aberrant expression of CD25 and CD2, and may show an altered intensity of antigens which are normally expressed on mast cells. The recognition of an abnormal mast cell immunophenotype is one of the minor diagnostic criteria for SM in both classifications. CD30 expression, either by FCM or IHC, can be detected in up to 80–90% in SM and is now accepted as an additional marker in defining immunophenotypic aberrancy of MCs.

Normally, mast cells account for < 0.1% of total BM cells, and are usually < 2% of cells even in SM. Therefore, FCM analysis for mast cells require strategies similar to those used for the detection of measurable residual disease (MRD). FCIP provides a more sensitive method as compared to IHC for detecting abnormal mast cells and can be useful for the identification of circulating mast cells, present in nearly half of the patients with indolent SM and almost all patients with advanced SM [7]. Finally, FCM can also be helpful for the detection of an associated myeloid neoplasm (SM-AMN) or a concurrent lymphoid or plasma cell neoplasm.

Acute myeloid leukemia (AML) classification relies mainly on cytogenetic and molecular data [1, 14]. However, the overall AML classification structure continues to emphasize the integration of clinical, morphologic, immunophenotypic, and genomic findings. In AML, FCM remains an essential tool for rapid diagnosis and lineage assignment. Within the categories of AML not otherwise specified, NOS, (ICC)/AML with differentiation, FCM is necessary to confirm AML with minimal differentiation, to detect monocytic and megakaryoblastic differentiation and to diagnose mixed-phenotype acute leukemias (MPAL). Characteristic immunophenotypic findings in AML with main specific genetic abnormalities are well known and have been described and reviewed in previous classifications [27] and recent publications [17]. Importantly, FCM analysis, together with cytomorphology, is essential for prompt diagnosis and treatment of acute promyelocytic leukemia (APL). The combination of cytomorphology and FCIP permits the diagnosis with a high degree of certainty; however, distinction from APL mimics (e.g., NPM1-mutated, other AML with monocytic differentiation, KMT2A-rearranged AML) can be challenging. Several gating strategies have been explored to distinguish APL from AML subtypes with APL-like immunophenotypes, including the use of radar plots [6]. More recently, Fang et al. (2022) compared patients with APL to NPM1-mutated AML and suggested that CD2 and/or CD34 expression, along with uniform CD13 and CD64 positivity, is more consistent with microgranular APL [5].

AML with RUNX1 mutation has been associated with mixed-phenotype acute leukemia (MPAL) and with expansion of the plasmacytoid dendritic cell compartment. AML with myelodysplasia-related genetic abnormalities often carry immunophenotypic features like those described in MDS. However, the simultaneous occurrence of mutations and cytogenetic alterations makes it difficult to identify specific patterns associated with certain mutations. FCIP is also needed for the diagnosis of MPAL; however, criteria for lineage assignment are still a matter of debate and many cases previously assigned to this category with predominant myeloid blast populations will now be diagnosed as AML based on cytogenetic or mutation findings [3].

Finally, FCM plays an important role in MRD testing, which is a strong predictor of relapse and shorter survival in AML patients [9]. Nevertheless, MRD measurements are not yet used routinely to guide therapeutic decisions in AML, partly due to the lack of uniformity in methodologies used for MRD detection and interpretation. The ELN MRD working group has published guidelines for MRD analysis, including the harmonized use of an integrated diagnostic “leukemia-associated immunophenotype” (LAIP) and “different from normal” (DfN) approach with a set of MRD core markers [9]. Since relapses also occur in MRD-negative patients, further research is focused on measuring the frequency of residual leukemic stem cells, which may be resistant to therapy [13].

Apart from the role of FCM as a diagnostic and prognostic tool, it is also of value in patients being considered for targeted therapy and treatment with chimeric antigen receptor T cells (CAR-T) by demonstrating the expression of cell surface antigens (e.g., CD33, CD123) on leukemic blasts [10].

The application of AI strategies and ML for the analysis of complex FCM data, recently reviewed in [2, 15, 30], has opened a new, rapidly expanding era. Efforts to develop, validate, and disseminate automated computational methods for FCM data analysis can help overcome the limitations of manual analysis and provide efficient and data-driven diagnostic applications. An increasing number of mostly recent studies underline the applicability of AI tools for the analysis of FCM data in hematological malignancies (Table 2). Similarly, modern cyto-and histomorphology is evolving towards AI-assisted “digital microscopy,” allowing automated pattern recognition and classification of acquired cell images. A more detailed review of these studies is beyond the scope of this article. Briefly, ML models have demonstrated human-level performance using FCM data for the classification of chronic lymphoproliferative diseases and for the diagnosis, classification, and MRD testing of acute leukemias. AI applications in MDS include the use of convolutional neural networks to assess dysplasia, the development of an AI-assisted prediction score for MDS diagnosis [4], and models that provide a link between morphology, mutational status, and prognosis. Finally, markers for biological behavior, such as proliferative and apoptotic parameters, have been proposed as interesting candidates for incorporation into such automated approaches [18]. Data from the literature provide evidence that AI-driven FCM diagnostics allow reliable and comprehensive (multidimensional) automated analysis of large immunophenotypic datasets, and clinical applications appear time efficient and potentially harmonizable across laboratories.

Flow cytometry is an indispensable tool for quick diagnosis, classification, and follow-up of patients with myeloid neoplasms, particularly AML and MDS. Immunophenotypic alterations are already used as diagnostic criteria in several disease entities in the WHO/ICC classifications. FCIP should therefore be included into routine diagnostic algorithms in clinical hematology, supported by artificial intelligence approaches. Clinical validation of these computational approaches is ongoing and essential to realize their true potential for use in the clinical diagnostic laboratory.