International cooperative study identifies treatment strategy in childhood ambiguous lineage leukemia

Cases with acute leukemias (ALs) are typically diagnosed as belonging to either the lymphoid or myeloid lineage. However, 2% to 5% of AL cases fulfill the definition of ambiguous lineage leukemia (ALAL). The following 4 different types of lineage ambiguities have been described: (1) single-population mixed-phenotype leukemias (MPALs; also known as biphenotypic ALs), which share immunophenotypic features of lymphoblastic and myeloid AL (acute lymphoblastic leukemia [ALL] and acute myeloid leukemia [AML], respectively); (2) bilineal ALs, in which 2 separate clones of different lineages coexist; (3) undifferentiated ALs, in which the criteria of ALL and AML are not met due to a paucity of positive markers of any of the 2 diseases; and (4) ALs with an early switch to a different lineage, in which the phenotype of leukemic cells changes to a different lineage during treatment (usually during the early phases).^1-3  This type of ALAL was not included in this study, unless this switch was combined with 1 of the 3 aforementioned ALAL presentation features.

Most of these leukemias fulfill 1 of the 2 international definitions of MPALs, both of which principally rely on flow cytometry. An older definition by the European Group for Immunophenotyping of Leukemia (EGIL) uses a spectrum of antigens weighted by their presumed significance.^4,5  A newer definition by World Health Organization (WHO) utilizes fewer antigens^6,7  (supplemental Table 1, available on the Blood Web site). In general, the prognosis of MPALs is slightly worse than that of non-MPALs on the same type of treatment.^8-11 

The spectrum of underlying genetic aberrations is wide. Whereas fusion genes with KMT2A (MLL) contribute to more myeloid or lymphoid phenotypes partly depending on the fusion partner,¹²  other fusions are typically lymphoid with a few myeloid markers and only occasionally have a full ambiguous phenotype (ETV6/RUNX1) or analogically with myeloid lineage (RUNX1/RUNX1T1). These fusion genes, however, are present only in a minority of ALALs,⁹  and the majority of ALALs are genetically heterogeneous. Next-generation sequencing techniques are likely to discover not only further heterogeneity but also more recently identified recurrent genetic aberrations such as the ZNF384 gene involving fusions, which occur especially in B-myeloid ALAL.^13-15,36  Mutations in TP53, which are generally more frequent in adult leukemias, were expectedly found mostly in adult ALALs.¹⁶ 

The aim of the present study was to collect a detailed data on biological features, treatment, and outcome of ALALs in order to set up treatment recommendations that can be tested prospectively.

Eighteen centers participated in the iBFM-AMBI2012 study. These centers were located in Australia (single hospital), Austria, Brazil (single hospital), Czechia, Germany, Greece (single hospital), Israel (single hospital), Italy, Netherlands, the NOPHO group (representing Denmark, Estonia, Finland, Norway, Sweden, Iceland, and Lithuania), PINDA (Chile), Poland, SAHOP (Argentina), Slovakia, St. Jude Children’s Research Hospital and Baylor College of Medicine (United States, 2 institutions), Ukraine, and the United Kingdom. Each center supplied data on consecutive children to a Web-based database. The periods of referral differed between the centers and lasted 1 to 12 years, except for Greece, which reported 3 isolated cases (supplemental Table 2). Patients diagnosed before 2002 or after 1 June, 2015 were excluded from this study. The fulfillment of the inclusion criteria was then verified by 3 of the authors (V.d.H., J. Stancikova, and O.H.). Except when excluded patients are explicitly mentioned, all analyses address included patients only.

Patients with AL diagnosed or referenced in one of the participating centers within the study period before 18 years of age were considered for analysis. Definition of a single-population biphenotypic/mixed lineage leukemia or bilineal leukemia or undifferentiated leukemia was fulfilled. For single-population ALAL, the immunophenotypic criteria of EGIL and/or WHO were met. Cases with >1 significant population, each belonging to a different lineage, were considered bilineal. Blast populations that did not fulfill the criteria of B, T, or myeloid lineages were considered undifferentiated. Subsequently, we excluded patients who had a genetic disease that predisposes to leukemia. Lastly, we excluded patients who received only palliative treatment or other treatment not considered as ALL type, AML type, or combined type (supplemental Table 3).

For event-free survival (EFS) calculations, the uncensored event was death of any cause, relapse of leukemia, and secondary malignancy. The censoring times included time to a treatment type change unless we explicitly mention that treatment changes are disregarded. In other instances, censoring times reflected the end of follow up without an event. For univariate comparisons of EFS, a log-rank (Mantel-Cox) test was used. For the analysis of possible benefit of stem cell transplantation (SCT), a delayed-entry statistics was applied, in which all events (deaths, relapses, or secondary malignancy onsets) preceding the median time to SCT for the analyzed subcohort were considered censored times.

For multivariate Cox regression, the following variables were considered: referring center, age, sex, number of white blood cells (WBCs) per liter, central nervous system infiltration, fulfillment of WHO/EGIL criteria, categorization to single population/bilineal/undifferentiated, estimate of lineage by Association of Pediatric Hematology Oncology/Berlin-Frankfurt-Münster flow,¹⁷  positivity of KMT2A/AFF1, lineages involved using EGIL or WHO criteria, frequency of blasts, and positivity of CD1a, CD2, CD3, intracellular (i) CD3, CD5, CD7, CD8, CD10, CD117, CD11c, CD13, CD14, CD15, CD19, CD20, CD22, iCD22, CD24, CD33, CD64, CD65, iCD79a, intracellular immunoglobulin M, iLysozyme, intracellular myeloperoxidase (iMPO), T-cell receptor αβ, T-cell receptor γδ, and intracellular terminal deoxynucleotidyltransferase. Factors with no correlation with the outcome in univariate pre-assessment (P > .2) were considered irrelevant, leaving WBC count and positivity of KMT2A/AFF1, CD2, CD10, CD11c, CD22, iCD22, CD24, CD64, iMPO, and intracellular terminal deoxynucleotidyltransferase included in the analysis for ALL-treated patients. For AML-treated patients, the following factors were considered relevant and included in the analysis: age at diagnosis, WBC count, and positivity of KMT2A/AFF1, CD2, iCD3, CD7, CD14, CD19, CD22, iCD79a, and iMPO.

In total, data on 274 patients were available. A CONSORT diagram of the entire study is shown in Figure 1. Of these, 233 cases fulfilled the inclusion criteria. The exclusion of noneligible cases (n = 41) did not noticeably affect the 5-year EFS of the entire cohort (62% ± 3.6% and 65% ± 3.8% before and after exclusion, respectively). The predisposing genetic diseases that resulted in exclusion from the cohort (n = 8) comprised Fanconi anemia (n = 2), Down syndrome (n = 4), and GATA2 mutations (n = 2). Patients excluded for not having received or not having been assigned to any type of therapy (n = 4) or with a missing information regarding primary therapy (n = 3) partially overlapped with the ones excluded for predisposing genetic features. The rest of the article describes only the included patients. Demographic and diagnostic information is listed in Table 1, together with the primary chemotherapy. In total, 176 patients had single-population ALAL, 45 had bilineal ALAL, and 12 had undifferentiated leukemia. Of the single-population ALAL, 103 fulfilled WHO and EGIL definitions, 42 fulfilled EGIL only, and 31 fulfilled WHO only. Distribution of cases into genetic categories demonstrates the heterogeneity of ALAL and the correlation with the physicians’ choice of primary treatment type or lineages involved (supplemental Table 4).

Due to the specific nature of ALAL, various treatment protocols or experimental therapies were applied. Supplemental Table 5 lists the different choices for initial therapy.

The outcome of patients who started on ALL-type treatment was superior (5-year EFS, 80% ± 4.0%) to that of patients who started on AML-type or combined-type (ALL/AML) treatment (5-year EFS, 36% ± 7.2% and 50% ± 12%, respectively; Figure 2). No significant prognostic difference was observed between AML- and combined-type primary therapies. Treatment toxicity was in general relatively high, accounting for 18 out of 63 deaths, and was seemingly evenly distributed between ALL-, combined-, or AML-type primary therapy (7.4%, 8%, and 8.3% patients on the respective primary treatment). In total, 69 children underwent SCT, and 20 of them (29%) died (4 [5.8%] from toxicity and 16 [23%] from leukemia progression).

The genetic categories of the patients document a diversity of ALAL. The genetic findings of the patients in the study are shown in correlation to the primary type of therapy and type of ambiguity in supplemental Table 4.

Patients with bilineal ALAL (Figure 3) fared better on ALL therapy than on AML therapy (5-year EFS, 65% ± 12% and 14% ± 13%) and in general fared worse than single-population ALAL (5-year EFS, 84% ± 4.2% and 43% ± 8.9% for ALL and AML therapy, respectively). Supplemental Figure 1 demonstrates that the better prognosis on ALL-type primary therapy holds true for ALAL defined by the EGIL criteria only and for ALAL defined by WHO criteria or both.

As patients with molecular genetic findings that are considered to be unequivocally ALL (ETV6/RUNX1 or TCF3/PBX1) or AML (RUNX1/RUNX1T1, PML/RARA, or CBFB/MYH11)^12,18  should be guided as such in treatment recommendations, we have not included these patients in the subsequent analyses. In patients with BCR/ABL1 fusion, ALL-type treatment with tyrosine kinase inhibitors is indicated,¹⁹  and BCR/ABL1⁺ cases were also excluded from further analyses. Typically, TCF3/PBX1 leads to nonambiguous ALL and CBFB/MYH11 leads to nonambiguous AML. In line with this rule, no patient in our study was identified with TCF3/PBX1 or CBFB/MYH11. In total, 31 (13%) patients had one of the above-named gene fusions. For the remaining 202 patients (87%), we searched for the most significant parameters that would correlate with prognosis.

Among the patients without the gene fusions specified in the previous paragraph, 124 and 53 cases received ALL- and AML-type treatment, respectively. For each of these major types of treatment separately, a multivariate Cox regression analysis was performed. Of all factors considered (see Patients and methods for list of variables and their univariate preassessment), WBC count and iMPO were prognostically significant in ALL-type primary treatment and WBC count and CD19 were prognostically significant in AML-type primary treatment (Tables 2 and 3, respectively).

After identifying factors associated with differences in risks of events within one type of treatment, we searched for the ones that correlated with the most significant prognostic difference between ALL- and AML-type therapy (supplemental Table 6). Out of these parameters, the most striking difference between ALL- and AML-type primary therapy was found among CD19⁺ cases (Figure 4A,C). Among patients with positivity of CD19 (which included a positivity or a partial positivity in ≥1 blast population in cases with bilineal ALAL), the outcome strongly correlated with type of treatment. Whereas the 5-year EFS of children who received ALL type primary therapy was 83% ± 5.3%, the 5-year EFS of children who started with AML or combined type of treatments was inferior (0% ± 0% and 28% ± 14%, both P < .0001). After AML-type primary therapy, no patient was alive and without an event beyond 1.4 years after the diagnosis. We thus consider cases with CD19 indicated for a non-AML primary treatment. Details on CD19⁺ cases with AML-type primary treatment are described in supplemental Information 1. Among patients who started on a combined-type treatment, there was a high proportion of KMT2A/AFF1⁺ ALAL (supplemental Table 4), which largely drives the poorer prognosis of this subset. Among KMT2A/AFF1⁻ cases with CD19, the 5-year EFS did not differ between children with a combined-type treatment (71% ± 17%, n = 8) and ALL-type treatment (84% ± 5.2%, n = 84; P = .38).

If we consider patients with above-listed gene fusions, and/or those with CD19⁺ blasts assigned to recommended primary treatments, 33% remain to be addressed (76 out of 233 patients in our cohort). The primary therapy was of the ALL, combined, and AML type in 35 (46%), 9 (12%), and 32 (42%) cases, respectively. In total, 18 events occurred in this subset of patients with an ALL-, combined-, and AML-type primary therapy (4, 1, and 13 events, respectively).

The 2 main types of primary therapy (ALL and AML type) were compared for CD19⁻ cases (n = 67). As we sought to find a parameter that would positively define a biologic subset, we analyzed the impact of the remaining (after exclusion of CD19) antigens that define lymphoid involvement in both the EGIL and WHO classification. This means that cases were selected with positive CD7 and CD3 (intracellular or surface) or with at least 2 positive antigens of CD10, intracellular CD79a, and CD22 (intracellular or surface). This selected subcohort had a very good outcome on either ALL- or combined-type primary therapy (Figure 4B,D).

Among the remaining 10 patients, 5 children fulfilled criteria for undifferentiated leukemia. Although the number of patients was too low to make reliable conclusions, the ALL-type primary treatment appeared particularly inefficient; of the 3 patients on ALL-type primary treatment, the type of treatment was changed between days 15 and 19 of treatment in 2 of them, and both died within <3 months of the diagnosis. The third patient died 2 years after diagnosis. In contrast, 4 of 5 patients who received AML-type primary treatment were alive, although only 1 of them in the first complete remission (CR1). The treatment was changed from AML type in 2 patients (at days 15 and 34 of treatment). Finally, both patients on combined-type treatment remained on it and were in CR1.

In 29 patients, the treatment type (ALL/combined/AML) was changed after 8 to 126 days of therapy (median, 32 days; supplemental Figure 2). The reasons for these treatment-type changes mostly reflected response to the primary treatment. Although, as expected, the outcome of patients whose type of treatment changed was poorer, some of the patients whose treatment type changed were brought to durable remission in each of the types of primary treatment.

Another major treatment factor was SCT. In total, 69 patients underwent SCT, and 58 of them were transplanted in the CR1. Using a delayed-entry survival analysis, no significant prognostic benefit could be demonstrated in the ALL, AML, or combined type of primary treatment (Figure 5).

We then analyzed specifically patients with a high leukemia load at the end of induction (ALL-, AML-, or combined-type primary treatment). In total, 31 patients had 5% leukemic cells or greater. This relatively low number of patients did not allow separation by type of primary treatment. Of the 31 patients, 12 underwent SCT (irrespective of the remission status at the time of SCT). The 5-year EFS of these transplanted children was 83% ± 15%, compared with 29 ± 16% among nontransplanted patients (P > .05); in a delayed-entry analysis, 5-year EFS was 83% ± 15% and 36% ± 19%, respectively (P > .05).

Patients who started on ALL-type therapy would be considered for a treatment change if their estimated outcome was poorer on the continued ALL treatment than on the changed treatment type. Such a poor response predictor could be anticipated especially if the ALAL patients with a poorer response to initial treatment had a markedly poorer prognosis than similar patients with nonambiguous ALL. Therefore, we evaluated the proportion and EFS of ALAL patients with a poor treatment response. Among CD19⁺ and CD19⁻ ALAL cases, the proportion of patients with a poorer leukemia clearance was 25% and 50% at day 8 (poor prednisone responders [PPRs]; outcome is shown in Figure 6), 32% and 60% at the end of induction (minimal residual disease [MRD] ≥0.1%; outcome is shown in supplemental Figure 3A-B,E-F), and 13% and 10% at week 12 (MRD ≥0.01%; outcome is shown in supplemental Figure 3C-D,G-H), respectively.

By definition, a poorer blast clearance at day 8 is found in PPRs. The 5-year EFS of CD19⁺ PPRs with ALAL was 47% ± 14%, which appears inferior to that of nonambiguous precusor B (pB) ALL (68% ± 4.6%²⁰ ). As the principal question was whether the type of treatment should be changed, we analyzed the number and outcome of children whose type of treatment changed at any time after PPR status was set. Of the 23 CD19⁺ ALAL PPRs, the treatment type was changed in 11 and remained ALL type in 12, leading to continued CR1 in 4 and 6 cases, respectively. The 5-year EFS of CD19⁻ PPRs was (90% ± 9.5%), which is definitely not inferior to that reported in nonambiguous T ALL (63% ± 5.7%²⁰ ).

At the end of induction, the EFS of CD19⁺ and CD19⁻ cases with MRD ≥0.1% was 60% ± 13% and 79% ± 11% (supplemental Figure 3A-B), respectively, which appears to show an outcome noninferior to that reported for nonambiguous ALL cases (cumulative incidence of relapses, 36% ± 3.6% and 32% ± 3.6% for pB at 5 years²¹  and T ALL at 7 years,²²  respectively) and possibly reflects increased treatment intensities in the MRD higher-risk patients treated by contemporary protocols.

At week 12, the 5-year EFS of CD19⁺ cases with MRD ≥0.01% was 50% ± 19%, but both CD19⁻ patients with MRD ≥0.01% died (one from toxicity and one from leukemia) (supplemental Figure 3C-D). Whereas the data of CD19⁺ ALAL patients do not indicate a marked difference of nonambiguous pB ALL with high MRD,²³  the grim prognosis of the 2 CD19⁻ cases calls for more data.

We present the largest study to date on childhood ALAL involving children with 3 types of ambiguity. Keeping these patients together in one cohort allows for overlapping cases; eg, it depends on the investigator whether they consider cells with distinct phenotype as 2 separate populations (and thus a bilineal leukemia) or they consider one leukemic population that differentiates into 2 phenotypic states. We have shown previously that even quite phenotypically distinct populations might result from one clone.²⁴ 

Before the first attempts to standardize the ALAL definitions by immunophenotyping in the 1990s, the value of morphological examinations was also investigated.²⁵  While ALAL typically presents with ambiguous morphological features, there are cases with typical ALL or AML appearance. In an early study by Pui et al,²⁵  which used morphologic and cytochemical diagnostic criteria for treatment selection, 2 of the 4 nonresponders to AML-type treatment later responded to ALL-type therapy. However, morphology was not the focus of more recent studies, including this one.

So far, treatment recommendations are mostly based on scarce observations or on assumptions based on the biological similarities of leukemic cells to physiological hematopoiesis. In physiological granulocytic development, MPO is a prominent molecule and serves as a traditional marker of myeloid differentiation. However, its expression was documented in ALL cells on the messenger RNA and protein levels,²⁶  and although it was missing in more mature B or T cells, it was also found in messenger RNA of sorted physiological pre-B cells (recalculated data from Novershtern et al²⁷ ). Moreover, there is no evidence that MPO expression is directly connected to a better response to AML-type therapy. In fact, cases of MPO⁺ leukemia that respond well to ALL-type therapy are documented in the literature^26,28  and in this study. Given the variability of MPO expression, it is not surprising that there has been a debate over the cutoff for the diagnostic MPO positivity.^29-31 

This study documents the diagnostic heterogeneity of ALAL and demonstrates that the therapeutic approaches vary. These results, together with the retrospective nature of our study, led us to interpret our data with caution. We observed a superior outcome of children treated with ALL-type therapy, which is in line with previous studies.^8,9,32-35  This finding should not automatically imply that ALL-type therapy is superior for any subset of ALAL. However, the striking failure of AML-type therapy to cure any child with CD19⁺ leukemia in the absence of AML-specific gene fusions leads us to recommend other types of chemotherapy in this relatively large subset of ALAL. However, without a prospective study, we are unable to estimate the prognosis of these ALAL patients on ALL-type treatment. For the remaining 33% of patients, the support for our recommendation is less clear, and the ultimate answer could only come after being tested prospectively. We have established an international platform, which has allowed testing the prognosis of children with ALAL treated according to predefined rules.

Given that the phenotype of ALAL is in between classical ALL and AML, it has always been tempting to speculate that these leukemias would respond best to therapy that comprises elements of both ALL and AML types of treatment. This speculation assumes that what has been proven empirically for “pure” ALL or AML would analogically hold true for borderline cases. One problem is that this assumption might be wrong. Second, both ALL and AML types of treatment are close to maximum tolerable intensity. Therefore, by intending to combine these treatments, we either compromise their efficacy or increase the risk of toxicity beyond acceptable margins.

Collectively, the data lead to an algorithm, the details of which can be tested prospectively (Figure 7). After addressing ALL- and AML-specific fusions (13% of our cohort), positivity of CD19 should be considered. Due to the dismal outcome of CD19⁺ cases on AML-type primary treatment, this primary treatment type is not recommended. Although the EFS of CD19⁺ patients on combined-type primary therapy was significantly poorer than that of patients on ALL-type primary treatment, this difference was largely driven by a high proportion of KMT2A/AFF1⁺ cases; among KMT2A/AFF1⁻ cases, the outcome corresponding to the combined-type treatment is not different from the outcome of ALL-type primary treatment. Thus, combined-type primary treatment is a possible alternative for CD19⁺ ALAL. For the CD19⁻ patients, the bases for recommendations were less straightforward. In presence of lymphoid markers, the outcome on ALL-type primary therapy was excellent, which should be considered. However, a minority of children on AML-type primary therapy had an outcome similar to that of nonambiguous AML, leaving this option open of AML-type therapy for cases with additional myeloid features. Very rarely, ALAL presented without CD19 or the listed lymphoid features, and the outcome was poor, especially on ALL-type treatment, making the other 2 types of primary treatment preferred. Altogether, these treatment recommendations should be tested prospectively in the setting of multidisciplinary leukemia diagnostics.

In addition to the selection of the primary treatment type, another key question is whether and when to change the type of treatment. The outcome of ALAL patients with a poorer blast clearance at day 8 (CD19⁻ PPRs) or at the end of induction (both CD19⁺ and CD19⁻ cases) was not inferior to what is expected for nonambiguous ALL. Although CD19⁺ PPRs among ALAL do appear to have a worse prognosis than analogous nonambiguous ALL cases, in half of these cases, the type of treatment changed after day 8; while this change did not bring a clear benefit in a low number of patients, a similarly low number of patients stayed on ALL-type therapy. Notably, our data show that proportion of PPRs is higher (23 of 92 [25%] patients with CD19⁺ ALAL and 15 of 30 [50%] patients with CD19⁻ ALAL on ALL-type treatment with known prednisone response) than in nonambiguous ALL (6.4% and 34% in B-cell precursor ALL and T ALL, respectively).²⁰  Children who remain on ALL-type treatment should thus often be treated in high-risk strata.

For evaluation of SCT benefit specifically among patients with a high leukemia burden at the end of induction, the number of patients in our cohort was limited, and the possible benefit was not significant. With all these limitations, our observation points to the fact that patients with very poor response to induction treatment may be transplanted and rescued. Major treatment-type modifications are thus to be considered only in children who do not achieve CR1 by the end of induction. Such modifications comprise either SCT or, after re-evaluation of blast phenotype, change of treatment type. The possibility to transplant children with a poor treatment response is in line with a recent smaller study, which found a benefit of SCT in adult ALAL, although not using delayed-entry analysis.³² 

In summary, a multicenter international study was used to outline treatment recommendations for children with ALAL. These recommendations should be tested prospectively in a multinational study, and eventually, each patient with ALAL will be assigned to their optimal treatment.

The medical care provided by physicians, technicians, and nurses to the patients in the study in the participating hospitals is greatly appreciated.

This study was supported by the Czech Ministry of Health (15-28525A, conceptual development of research organization 00064203, University Hospital Motol, and RVO-64165), Ministry of Education, Youth and Sports NPU-I nr.LO1604, Charles University Center of Excellence 204012, GACR P302/12/G101, The Capital Region of Copenhagen (Global Excellence Funding), and the Israel Cancer Association.

Contribution: O.H. and J. Stary designed the study; O.H., V.d.H., E.M., J.T., M.Z., J.I., B.K., H.I., A.V.M., J.G.R., M.S.F., L.D.-P., M.D., B.B., G.B., M.C., M.E.C., N.M., S.E., S.I., D.L., T.F., A. Kolenova, P.S., O.K., K.R.R., S.P., A. Kattamis, E.d.C., R.R., D.R., J.K.C., M.S., and J. Stary performed primary or reference diagnostics and/or treatment of the patients; O.H., V.d.H., I.J., and A.L. performed data collection; V.d.H., I.J., K.B., A.M., B.K., H.I., K.S., S.S., M.S.F., J.M., M.D., B.B., M.C., M.E.C., S.E., A. Kolenova, O.K., S.P., and D.R. performed clinical data generation and processing; O.H., V.d.H., J. Stancikova, I.J., E.M., V.C., and B.V. performed data analysis (including statistics); O.H., V.d.H., J. Stancikova, B.V., I.J., M.Z., K.B., A.M., T.B.A., Z.Z., J.T., K.S., A.V.M., M.S., and J. Stary interpretation data; O.H., J. Stancikova, J. Stary, and B.V. wrote the manuscript; and all authors edited and revised the manuscript.

Conflict-of-interest disclosure: M.S. received grants from Sigma-Tau and Baxalta, and H.I. received a grant from Shire. The remaining authors declare no competing financial interests.

Correspondence: Ondrej Hrusak, Department of Pediatric Hematology/Oncology, Second Faculty of Medicine, Charles University, University Hospital Motol, V Uvalu 84, 150 06 Praha 5, Czech Republic; e-mail: ondrej.hrusak@lfmotol.cuni.cz.