Prognostication in acute leukemia at follow up is important for risk stratification and disease monitoring. Morphologic evaluation was the gold standard for remission evaluation, but many patients relapsed despite lack of morphologic evidence of increased blasts. Initial studies demonstrating that detection of IgH/TCR sequences in patients acute lymphoblastic leukemia correlated with increased rate of relapse suggested that tracking low level evidence of disease can identify patients with higher risk disease despite morphologic remission.1 Subsequent studies demonstrated that very low level of measurable (formerly referred to as minimal) residual disease (MRD) is associated with worse outcomes in a wide variety of hematologic diseases. 2, 3, 4, 5

Currently, most assays used for MRD assessment rely on identifying and tracking populations with phenotypic or genomic aberrancies that correspond to leukemic blasts.6 Multiparametric flow cytometry assays (MFC-MRD) show several characteristics that was ideal for MRD detection including rapid turnaround time, direct quantitation of MRD, assessment of potential immunotherapy markers, and the technology and expertise are already established at most cancer centers. There are two main approaches for MRD assessment by flow cytometry. The first, the leukemia associated immunophenotypes (LAIP) approach, relies on the immunophenotype of blasts at disease diagnosis. Identification of the blast population by the LAIP approach most commonly tracks aberrations such as lineage infidelity, asynchronous antigen expression, antigen overexpression, and abnormal Forward Scatter/Side Scatter distribution.7 This immunophenotype is then tracked through subsequent specimens. This method requires a diagnostic sample which determines trackable flow markers and assumes that there are aberrancies that allow for identification of the leukemic blasts. The difference from normal (DfN) approach identifies immunophenotypic shifts and aberrancies compared to expected maturation patterns in the bone marrow.8 This method can be used in the absence of a diagnostic specimen and is especially useful in cases of clonal shift, however, it requires expert level knowledge of normal maturation patterns. Most practices favor the combined use of approaches.

Genomic-based methods rely on the identification of leukemia-specific genetic aberrations, such as chromosomal translocations, point mutations, and gene expression profiles, to detect and quantify the residual disease. While fluorescence in situ hybridization (FISH) and karyotyping are routinely used for the detection of genetic aberrations in the diagnosis and classification of new acute leukemias, these methods are not sensitive enough for MRD tracking. As such, the most genomic-based methods for MRD tracking in acute leukemia currently utilize quantitative PCR (qPCR) and next-generation sequencing (NGS) based methods. In particular, NGS-based MRD assays allow for tracking of multiple sequences at an unprecedented level of sensitivity and allow for retrospective analysis using FFPE or unprocessed aspirate smears. Limitation of genomic analysis methods is that it requires prior knowledge of the aberrancies from a positive specimen which may not be available. Summary of the different technologies is listed in Table 1.

Detection of MRD in patients with B-ALL following induction and consolidation therapy is strongly associated with morphologic relapse and poor outcomes. 9, 10, 11 MFC-MRD is the most widely used technique within the United States and North America for in B-ALL patients and are generally a derivative of standardized panels that are associated with trials sponsored by the Children's Oncology Group (COG) and The St. Jude Children's Research Hospital, and the EuroFlow Consortium.12, 13, 14 A common diagnostic challenge in assessing for B-ALL using MFC-MRD is difficulty in distinguishing neoplastic cells from their benign B cell progenitors (hematogones) that are present in ∼80% of bone marrow specimens and are often expanded post-chemotherapy or post-transplant.15 While there are numerous candidates that show promise in distinguishing leukemic blasts from hematogones and a subset of these are incorporated into some standardized panels such as the St. Jude (CD44, CD24, CD73 and CD123) and Euroflow panels (CD73 and CD123), many have not been incorporated into routine clinical use. 16, 17, 18

More recently, increasing use of targeted immunotherapies against B cell antigens has provided challenges in detection by flow cytometry, as there is considerable overlap between ideal target and detection antigens. CD19 has long been the favored antigen for identification of B cells and B-ALL blasts due to its expression early in B cell development and expression in essentially all cases of B-ALL.19 Chimeric antigen receptor T cell therapies directed against CD19 (CAR19) and the bispecific T-cell engager blinatumomab directed against CD19 are FDA approved as second line therapies in both pediatric and adult B-ALL and has shown an impressive response rate, but 30-60% of patients eventually relapse without further treatment. 19,20 CD19 downregulation or loss is seen in 9-25% and 28% of relapse cases after CAR19 and blinatumomab therapy, respectively, and sequential use of blinatumomab and CAR19 is associated with CD19-negative relapse.21,22 Loss of CD19 poses considerable challenges to detection of blasts by flow cytometry, hence alternative gating strategies must be used in patients after anti-CD19 immunotherapy. CD22 expression is detected in approximately 93% of B-ALL and its expression remains largely unchanged after anti-CD19 therapy.19,20 Gating with CD22 and CD24 has shown to be a reliable alternative for MRD monitoring following anti-CD19 immunotherapy, however, monitoring of KMT2A rearranged B-ALLs, which tend to express lower levels of CD22 and are often negative for CD10 and CD24, is very challenging. 23,24 Monitoring patients with KMT2A rearranged B-ALLs is further complicated by the fact that CAR19 and blinatumomab can induce upregulation of myeloid antigens or a complete lineage switch.25, 26, 27, 28 As such, MFC-based MRD assessment of KMT2A B-ALLs may require additional panels interrogating myeloid markers for aberrant expression of CD15, CD13, and CD33 as well as assessment for the presence of an abnormal myeloid or monocytic blast population. 29,30 Another diagnostic pitfall of CD22/CD24 gating is the need to recognize a normal CD19-negative B cell precursor population that can be present following anti-CD19 immunotherapy. These “stage 0 hematogones” express bright CD22, dim/negative CD10, CD38, and CD34 but lack CD24. Distinguishing this population from residual leukemia can be challenging but unlike most B-ALL blasts, stage 0 hematogones show bright CD22 as well as CD38 expression levels that is similar to normal B cell progenitors.23

The lack of sustainable response with CAR19 and blinatumomab also poses a problem where there is sequential targeting of multiple B cell antigens in patients with resistant disease. CD22 is the target of the antibody-drug conjugate inotuzumab and CAR22 is being developed as an alternative for patients with CD19-negative disease.29,30 Like CD19, selective pressure of the therapeutics also results in CD22 downregulation, hence tandem use of immunotherapy agents can make assessment by MFC-MRD increasingly challenging. 21,30,31 Several studies are underway to address this issue of antigen-negative relapse, but it is unclear whether alternative approaches such as dual antigen CARs will be successful.32

Genomic methods for MRD detection in B-ALL includes sequential tracking of T-cell and B-cell receptor gene rearrangements, PCR-based detection of recurrent gene fusions such as BCR::ABL1 or KMT2A. It is estimated that greater than 90% of B-ALLs demonstrate an immunoglobulin rearrangement, TCR rearrangement, and/or deletions in the T cell receptor, which makes the vast majority of B-ALLs trackable by this method.33 Molecular based methods are powerful in that they are sensitive; however, it does not necessarily discriminate between normal and abnormal clonal populations. This most commonly happens when a clonal T cell population is detected through TCR sequencing of B-ALL specimens. These unrelated populations can be identified since the frequency of the clone remains relatively stable and does not correlate with morphologic evaluation, hence correlation with other findings is critical in avoiding misinterpretation as persistent disease. Similarly, monitoring of MRD through detection of genetically defined rearrangements such as the BCR::ABL1 fusion transcript in Philadelphia+ B-ALLs is challenging since cell sorting studies show that this difference is at least in part attributed to BCR::ABL1 expression in non-ALL cells, such as myeloid cells.34,35 Indeed MRD levels detected by BCR::ABL1 shows some level of discordance when compared to Ig/TCR NGS-MRD and must be interpreted with caution.36

Identifying a clinically actionable threshold and preferred method of MRD testing is an ongoing challenge. There is conflicting data regarding whether MRD sensitivity thresholds beyond 10−4 confer actual advantages at the post-induction timepoint. 37, 38, 39, 40 In a pre-transplant setting, NGS-MRD negativity can identify a group of B-ALL patients with low risk of relapse, and is a superior method of prediction than MFC-MRD.41 In a post-transplant setting, one study found that MRD ≥ 10−6 is highly associated with relapse and poor survival.42 It is tempting to speculate that the prognostic discordance between the two techniques could be in part due to sensitivity differences, but further studies will be required to clarify these differences.

Unlike ALL, clinical significance of MRD assessment in AML is less defined. The negative predictive value of MRD in prognosis of relapse post-transplantation is becoming clear, although remains imperfect, wherein MFC-MRD negative status during the disease course and at the time of transplant is significantly associated with better outcomes and lower risk of relapse, compared to MFC-MRD positive patients and patients with morphologic evidence of disease.43, 44, 45 However, the clinical significance of MFC-MRD positivity is still debatable but currently appears to indicate a prognosis intermediate between MRD negative disease and morphologic evidence of disease. 46 MFC-MRD panels and methodology used across institutes are strikingly heterogeneous in the absence of national standardization. Remarkably, albeit not surprisingly, the prognostic value of MRD in AML is dependent on the experience level and approach of the laboratory performing the test. A PETHEMA study examining the prognostic impact from data generated from decentralized, local hospitals and laboratories performing MFC-MRD testing in AML found wide heterogeneity among hospitals in protocols, differing in number of markers, number of tube combinations, and even approach (i.e. LAIP vs DfN vs both). The result of such heterogeneity is overall reduced prognostic value when compared to prior large studies performed at single centers. 45

The dynamic nature of disease progression and post-treatment relapse is often reflected in flow immunophenotypic shifts. Some studies estimate that immunophenotypic changes occur in the majority of AMLs after relapse, and included both gains and losses of populations of AML cells, and gains and losses of antigens. 47, 48, 49, 50 As such, reporting out the LAIP as a composite immunophenotype of multiple subpopulation can be misleading when evaluating subsequent follow-up specimens (figure). Most guidelines, therefore, recommend the integration of LAIP and DfN. 47 Despite these efforts, MFC-MRD is challenging in AML. For instance, monitoring AML with monocytic differentiation is notoriously difficult as it lacks markers of immaturity such as CD34 and CD117. 51 Efforts to identify more reliable markers for monoblasts such as IRF8 are underway but the immunophenotypic overlap between reactive monocytes and monocytic leukemias remains a problem. 51,52 Overexpression of CD56 or loss of CD14 that is seen in monoblasts can also be observed in both reactive and neoplastic conditions and its expression patterns may change depending on the clone in use. 53 Moreover, transient reactive monocytosis, which occurs after stem cell transplant, GCSF administration, infection, chemotherapy, and many other settings, can also present these changes.54, 55, 56 Therefore, relying on these changes alone is not specific for disease recurrence. There are also challenges that stem from the technical complexity of flow cytometry assays. For instance shifts in the intensity of HLA-DR, CD64, or CD11b expression are commonly seen in monocytic leukemias, but these changes are often subtle and consistent evaluation requires strict instrumentation calibration and standardization.

In addition to LAIP and DfN methods, there is emerging evidence that detection of small, malignant populations of progenitor cells termed “leukemic stem cells” (LSCs) can serve as an additional valuable prognostic data point and a more reliable population for MRD tracking. LSCs are characterized by their self-renewal, proliferation, and differentiation capabilities and are crucial for sustaining leukemic blasts. 57 This population is thought to have immune invasive properties and increased resistance to chemotherapy, thus serving as a potential source of relapse.58,59 Enumeration of CD34+CD38- LSCs appear to have prognostic value and therapeutic targeting of LSCs is an area of intense research.60 Despite these findings LSC tracking has not been universally adopted in routine clinical MRD analysis due to its rarity and lack of a standardized definition of what constitutes an LSC population.60 While CD34+CD38- compartment is thought to be the main reservoir of LSCs, studies have demonstrated that CD34+/CD38+ and CD34-negative LSCs have been described.61,62 Additionally, discriminating between normal hematopoietic progenitors and LSC is difficult since LSCs are heterogeneous and show a range in immunophenotypic findings.60,63 Furthermore, abnormal CD34+CD38- LSCs are also detected MDS and CML hence it is unclear whether LSC enumeration will allow for discrimination between an underlying myeloid neoplasm in patients with AML with myelodysplasia related changes without incorporation of other markers.64,65 There are several studies that aim at measuring LSCs using a wide range of differentially expressed antigens such as CLL1, Tim3, CD45RA, CD123, CD11b, CD22, CD7, and CD56.63 Low numbers of LSCs also pose a challenge with studies proposing evaluation of 5-10 million cells in a single tube for optimal sensitivity.66

Approximately 90% of patients with AML have at least one mutation at diagnosis, however there is no universal molecular target in AML that is analogous to Ig/TCR gene rearrangements seen in ALL.59,67 Many of the challenges seen in MRD assessment in AML are the result of this heterogeneity. Even within AMLs with defining genetic abnormalities, the data for optimal timing and clinically significant threshold levels are largely limited to determining the prognosis with serial measurements, and stable, low level MRD positivity may not trigger a change in management.4,68 For instance, MRD negativity in AML with t(8;21) patients at the end of treatment and during follow up after completion of therapy is a valuable independent favorable prognostic factor for relapse risk and overall survival, but only 6 of 25 patients who converted from MRD-negative to MRD-positive clinically relapsed.69 The optimal threshold of MRD positivity appears to vary based on the timepoint that it is detected as well as the presence of other mutations.68 Patients with NPM1 mutated AML show excellent long term survival regardless of FLT3 status if they are MRD-negative pre-transplant while low NPM1 MRD positivity, relapse rates were low unless these patients also harbored a FLT3 ITD.70

In addition to the issues with thresholds and timing of MRD measurement, there are challenges that are becoming more evident as we learn more about the complex genomic landscape of AML and clonal evolution after treatment. For instance, approximately 10% of patients AML with mutated NPM1 demonstrate wildtype NPM1 at relapse.71, 72, 73 Careful consideration must be given to which mutations should be tracked, as not all mutations are suitable for this purpose. Persistent mutations in genes associated with clonal hematopoiesis, DNMT3A, TET2, and ASXL1 (known as DTA mutations) have not been shown to confer a risk of relapse in post induction chemotherapy or post-transplant patients but is associated with poorer prognosis.74 Overall, these findings suggest that molecular MRD analysis likely needs development of a more multifactorial approach that is currently used in MFC-MRD.