Project EVOLVE: an international analysis of postimmunotherapy lineage switch, an emergent form of relapse in leukemia Open Access

With increasing use of antigen-targeted immunotherapies in B-cell acute lymphoblastic leukemia (B-ALL), such as the CD19-directed bispecific T-cell engager blinatumomab and chimeric antigen receptor (CAR) T cells, emergence of lineage switch (LS) as a mechanism of immune escape has been uniquely perplexing.^1,2 Defined as the transformation of an acute leukemia morphologically and immunophenotypically from 1 lineage to another, while maintaining original cytogenetic and/or molecular markers of the original diagnosis, postimmunotherapy LS has most commonly been reported as B-ALL transforming to acute myeloid leukemia (AML).^1,3 Importantly, LS is distinguished from treatment-related, secondary or second neoplasms, which are clonally distinct from patients’ original disease and typically arise from a novel oncogenic event, therapy induced or otherwise.⁴ 

LS of acute leukemia was first described after standard chemotherapy, however, the incidence remains unknown.^5-9 Most commonly, LS has been associated with KMT2A-rearranged (KMT2Ar) leukemias, a cytogenetic alteration found in both AML and ALL, including in up to 70% of infant B-ALL in which lineage plasticity is well established.^5,10,11 Indeed, KMT2Ar B-ALL has a more stem cell–like phenotype and emergence of AML after B-ALL–directed chemotherapy alone has been described.^8,10 Although KMT2Ar is relatively rare beyond infant ALL,¹² with increasing use of antigen-targeted immunotherapies, incidence of both KMT2Ar-associated^1,13-15 and non–KMT2Ar-associated LS appear to be rising.^13,16-20 Moreover, enhanced disease monitoring with more sensitive multiparametric flow cytometry (MFC) and next-generation sequencing (NGS) may also facilitate the increased recognition of LS events. Mechanisms underscoring LS are likely heterogenous; proposed pathogenesis includes underlying plasticity of leukemic progenitor cells, cellular reprogramming due to a variety of intrinsic (genetic/epigenetic changes)²¹ and extrinsic (cytokine stimulation) factors, as well as emergence of previously undetected myeloid clones.^3,22 

To date, analysis of LS specifically after antigen-targeted immunotherapy, has been limited to case reports and case series,²³ despite contributing to as many as 7.2% of relapses after CD19 CAR T cells.¹ Given increasing reports of postimmunotherapy LS, its diagnostic difficulty, and poor outcomes,^1,23 we established an international collaborative effort to understand how LS is identified and diagnosed, describe presentation and treatment approaches, and analyze overall outcomes for this emergent form of postimmunotherapy failure.

Project EVOLVE (Evaluation of Lineage Switch, an International Initiative; https://ccr.cancer.gov/pediatric-oncology-branch/carnation-consortium/project-evolve) was an institutional review board–exempt retrospective study developed to capture cases of LS after antigen-targeted immunotherapy. After institutional approval at the National Cancer Institute, National Institutes of Health (the organizing study site), the steering committee performed a global call for cases. Case collection occurred from 18 January 2023, through 31 October 2023, but allowed for submission of any case that occurred before this timeframe. Inclusion criteria were broad, allowing for patients with acute leukemia of all ages who developed LS after any form of antigen-targeted immunotherapy (inclusive of antibody-drug conjugates). Participating sites were provided an electronic data collection form for transmission of deidentified data related to demographics, prior immunotherapy, presentation and diagnosis of LS, treatment approach, and overall outcomes (supplemental Methods, available on the Blood website). Additionally, individual authors of previously published cases of LS (supplemental Table 1) were contacted for inclusion of their cases with incorporation of additional data as needed. Cases of therapy-related AML (t-AML) or secondary neoplasms were excluded, as were cases or published reports missing substantial data that precluded LS determination.

Given heterogeneity in LS presentation, we, in collaboration with expert hematopathologists, established definitions to facilitate case categorization and discriminate across the spectrum of immunophenotypically altered leukemias submitted for this project (Table 1). Each LS case was individually reviewed and categorized based on initial pathologic assessment, clinical team designation, and MFC-based immunophenotyping when provided. Local immunophenotypic data, when available, was systematically analyzed by centralized review (supplemental Methods).

LS was defined as an acute leukemia with immunophenotypic changes that (1) warranted reclassification to a different lineage (eg, B-ALL to AML) and (2) maintained clonal relatedness through retention of cytogenetic and/or molecular signatures from initial diagnosis, with or without acquisition of additional mutations. Assays used to confirm clonality varied based on institutional practices and included, but were not limited to, standard cytogenetics, polymerase chain reaction, and NGS (eg, clonoSEQ, targeted or whole-exome genetic analyses). LS cases were further stratified as “confirmed” or “probable,” with the latter reserved solely when clonal relatedness could not be confirmed but a treatment-related secondary event was unlikely based on presentation (eg, presentation of central nervous system [CNS] leukemia). Lineage-defining markers and designation of cases as acute leukemias of ambiguous lineage (ALAL) or mixed phenotype acute leukemia (MPAL) were based on the European Group for the Immunological Classification of Leukemias, the International Consensus Classification of ALAL,²⁴ and/or World Health Organization guidelines (supplemental Methods).^24,29 

The term “lineage drift” was applied to cases in which a substantial change from prior phenotype with evolution in cell-surface antigen expression was insufficient to warrant disease reclassification (eg, loss of CD19 with retention of CD22 and gain of CD33 without presentation of other myeloid markers). Loss of CD19 alone did not constitute lineage drift. Lastly, standard definitions were used to identify secondary neoplasms, such as t-AML or therapy-related myelodysplastic syndrome.²⁷ 

Analyses were primarily descriptive, using Excel and Prism Graphpad 10.0. Nonparametric analysis was performed for all comparisons using P <.05 to indicate statistical significance; the Fisher exact and the Mann-Whitney-Wilcoxon tests were used for categorical and continuous variables, respectively. Overall survival (OS) was calculated from time of LS diagnosis to death or last follow-up as of data cutoff using the Kaplan-Meier method.

A total of 92 cases were submitted from 43 institutions/groups, across 8 countries (Figure 1A-B). These included 55 new cases submitted for Project EVOLVE and 37 previously published cases^{13-17,19,26,30-48} for which additional data was provided by the original authors (supplemental Table 1). Three cases were subsequently excluded, 1 because of insufficient data, 1 occurring without receiving immune-targeted therapy, and 1 duplicate submission of a patient treated at 2 sites; thus 89 cases were analyzed. Across 89 cases, 75 (84.3%) met the definition of LS, and 10 (11.2%) were classified as lineage drift. Four cases determined to be t-AML/myelodysplastic syndrome were excluded (supplemental Results).

Across 70 cases of LS, 53 (75.7%) were B-ALL to AML and 17 (24.3%) were B-ALL to B/Myeloid MPAL/undifferentiated leukemia. The median age at initial B-ALL diagnosis was 11.9 years (range, 12 days to 76.5 years), and 37 (52.9%) were male (Table 2). Most patients were White (n = 48, 68.6%) and non-Hispanic (n = 43, 61.4%).

B-ALL cytogenetics (at diagnosis or preimmunotherapy relapse) included KMT2Ar in 45 (64.3%) cases (Figure 1C; supplemental Figure 1). Other nonmutually exclusive cytogenetic aberrations included BCR::ABL1 translocation in 4 (5.7%) patients, a TCF3::ZNF384 fusion or other ZNF384 rearrangement in 3 (4.3%), a CDKN2A mutation in 5 (7.1%), a TP53 mutation in 6 (8.6%), a CRLF2 rearrangement in 3 (4.3%), a PTPN11 mutation in 2 (2.9%), a DUX4 rearrangement in 1 (1.4%), and a IKZF1 deletion in 1 (1.4%) patient(s) (supplemental Table 2).

Myeloid antigen coexpression was seen frequently at baseline, defined by either the B-ALL diagnostic immunophenotype or at relapse. Across 62 patients with available flow cytometric and/or immunohistochemistry available for review, 39 (62.9%) expressed at least 1 myeloid and/or monocytic defining or associated antigen (per World Health Organization criteria; supplemental Methods), across whom 29 (74.4%) harbored KMT2Ar. CD15 was expressed by 22 patients (35.5%), however, 19 (86.4%) of these had KMT2Ar in which CD15 may be aberrantly expressed.⁴⁹ No patients met criteria for MPAL at baseline.

The immunotherapy used most proximal to diagnosis of LS included CAR T cells (n = 34, 48.6%), blinatumomab (n = 31, 44.3%), inotuzumab (CD22-targeted antibody drug conjugate, n = 4, 5.7%), and an investigational CD19-targeted antibody drug conjugate (n = 1, 1.4%; Figure 1D). Cumulative exposure to pre-LS immunotherapy included blinatumomab in 45 (64.3%) patients, inotuzumab in 17 (24.3%), and CAR T cells in 35 (50%; Figure 1E-F). Eleven (15.7%) patients received both blinatumomab and CAR T cells; and 4 (5.7%) received blinatumomab, CAR T cells, and inotuzumab. In addition, 5 (7.1%) received >1 CAR T-cell infusion, and 22 (31.4%) received at least 1 prior hematopoietic stem cell transplant (HSCT). A total of 69 (98.6%) patients received prior CD19 targeting and 20 (28.6%) received prior CD22 targeting (of whom all but 1 had received prior CD19 targeting).

Across patients with LS, 7 (10%) received an interval HSCT between the most proximal immunotherapy and development of LS. Among the 34 patients receiving CAR T cells, 29 (85.3%) received single-antigen CD19-targeting constructs, 3 received combination CD19/CD22-targeting, and 2 received CD22-targeted constructs. In patients receiving CAR T cells as the most proximal therapy, cytokine release syndrome of any grade⁵⁰ occurred in 30 of 34 (88.2%) and was grade ≥3 in 4 (11.8%) patients.

Disease burden before the most proximal immunotherapy was variable: 10 (14.3%) had a measurable residual disease (MRD)-negative complete remission (CR; 2 with isolated CNS disease), 19 (27.1%) had an M1 marrow (<5%), 11 (15.7%) had an M2 marrow (5%-25%), and 30 (42.9%) had M3 disease (>25%). Thirty-eight (54.3%) patients achieved an MRD-negative CR of their B-ALL after the most proximal pre-LS immunotherapy, with higher rates of response after CAR T cells than after blinatumomab (70.6% vs 35.5%, P = .0063).

LS emerged at a median of 1.5 months (range, 0-36.5) from the most proximal immunotherapy (Figures 1F and 2A) and a median of 14.0 months (range, 2.7-274.2) from initial diagnosis. Most LS, n = 57 (81.4%), emerged within 6 months of the most proximal immunotherapy, inclusive of patients who developed LS during active B-ALL targeting. When stratified by disease burden before most proximal immunotherapy, the time to LS was shorter in those with high disease burden (M2/M3 marrow) than those with low or no disease burden (M1/MRD negative), median 1.1 vs 4 months (P = .016).

Stratified by prior therapy (Figure 2A), the median time to LS was 1 month after blinatumomab (range, 0-32; n = 31) and 2.8 months after CAR T cells (range, 0.3-36.5; n = 34). Data on CAR T-cell persistence at the time of LS were not available. At least 7 cases of LS emerged during blinatumomab infusion and at least 6 cases emerged before or at the 1-month after CAR T-cell time point. Across the 13 cases in which LS presented ≥6 months after prior immunotherapy, the median time from proximal immunotherapy to LS was 11 months (range, 6.9-36.5).

At LS diagnosis, the median age was 14.5 years (range, 0.4-77.3). A substantial proportion (n = 20, 28.6%) of LS occurred in those who were infants (aged <1 year) at initial diagnosis. At the other end of the age spectrum, 15 (21.4%) cases occurred in those who were older adults (aged >39 years) at initial diagnosis (Figure 2B).

LS was confirmed through clonal relatedness in 64 (91.4%) cases and probable in 6 (8.6%; supplemental Table 3). B-cell receptor (BCR) gene rearrangements (tracked by polymerase chain reaction or clonoSEQ) were available in 25 and confirmed to be retained in 20 cases. Alongside original findings, new mutations first described at LS diagnosis included new TP53 mutations in 4 (5.7%) patients, new PTPN11 mutations in 3 (4.3%), and a new TCF3::ZNF384 fusion in 1 (1.4%) patient, the latter identified on RNA sequencing of LS and likely present but not evaluated at diagnosis. At least 7 patients (10%) demonstrated acquisition of complex cytogenetics (Figure 2C). Comprehensive information regarding cytogenetic and molecular alterations are in supplemental Table 2.

LS was most commonly identified in the bone marrow (n = 46, 65.7%). However, 3 cases (4.3%) of LS presented with isolated CNS extramedullary disease (EMD), 4 cases (5.7%) were isolated non-CNS EMD, 16 cases (22.9%) had combined marrow and EMD involvement, and 1 (1.4%) presented with peripheral blasts (Figure 2D).

Across 69 patients with evaluable pathology reports sufficient for central review of immunophenotype, all (with exception of 1 case of undifferentiated leukemia) expressed ≥1 myeloid and/or monocytic-defining or -associated antigens at LS. Most (59, 85.5%) expressed CD33, and over half of KMT2Ar cases (26, 57.8%) expressed CD15. B-lineage antigen expression of either CD19 or CD22 was seen in 16 cases. A summary of key cytogenetics/molecular and immunophenotypic features at baseline and at LS are summarized in Figure 3.

Treatment of LS was highly variable (Figure 4A; Table 3). When LS treatment was initiated (n = 65), an intensive AML-induction–type chemotherapy-based regimen (for example, cytarabine, daunorubicin, and etoposide; or fludarabine and cytarabine) was most common (n = 27, 41.5%). Regimens incorporating anti-CD33–targeted gemtuzumab ozogamicin were used in 14 (21.5%) patients, whereas 6 (9.2%) received less-intensive therapy with venetoclax and azacitidine. Low-dose chemotherapy (eg, low-dose cytarabine, or oral palliative agents) was used in 12 (18.5%) patients. Four (6.2%) received “other” treatments including splenectomy for non-CNS EMD, daratumumab to target CD38 expression, B-ALL maintenance chemotherapy, and CNS-directed therapy including radiation. Two patients (3.1%) received palliative care (specific interventions not provided).

In total, 20 (30.8%) patients receiving LS-directed therapy achieved a CR to first-line treatment (Figure 4B), inclusive of 2 patients treated with palliative intent. Collectively, 23 (35.4%) achieved a CR at some point (Figure 4C; supplemental Table 4), not all of whom achieved MRD negativity (insufficient data). The median number of treatment lines for LS, excluding HSCT, was 1 (range, 0-6). A gemtuzumab ozogamicin–containing regimen led to CR in 7 (30.4%) patients and AML induction chemotherapy led to CR in another 7 (30.4%). Interestingly, in 1 case of LS with CD22 retention, inotuzumab was used as second-line after failure to AML induction chemotherapy and led to a transient MRD⁺ CR. Menin inhibitors, given as second or later treatment in 6 patients with KMT2Ar LS (single-agent use in 5 patients), did not result in any CRs. Fourteen (21.5%) proceeded to HSCT for treatment of LS, with 5 alive in a CR. During LS-directed treatment, 7 (10.8%) patients reverted to B-ALL.

As of the data cutoff, 8 of 69 (11.6%) remain alive (1 patient lost to follow-up; Figure 4D). Causes of death (n = 61) included progressive/refractory LS (n = 52 of 61, 85.2%), treatment-related mortality (TRM) in CR (n = 5, 8.2%), TRM with disease (n = 3, 4.9%), and relapsed B-ALL (n = 1, 1.6%). The median OS after LS diagnosis was 4.8 months (Figure 4E). At a median follow-up of 47.4 months (range, 2.4-115.8), 6 patients are in a CR (including 2 with reversion and treatment of B-ALL with CD19 targeting) and 2 patients remain with active disease (1 with LS, and 1 with B-ALL; Figure 4F).

Given the substantial fraction of LS occurring in those with infant ALL (aged <1 year at initial diagnosis), we further characterized this population (n = 20). The median age at B-ALL and LS diagnosis was 0.4 years (range, 12 days to 0.9 years) and 1.6 years (range, 0.4-9.6) respectively, and all harbored a KMT2Ar. The median time from most proximal immunotherapy to LS was 1.4 months (range, 0-5.4), with 4 cases occurring during blinatumomab. A CR was obtained in 9 (45%) cases and 3 (15%) remain alive in CR at last follow-up.

The median age at initial diagnosis in those with KMT2Ar (n = 45) and without (n = 25) was 5.9 years (range, 12 days to 76.5 years) and 13.5 years (range, 1.1-68.4; P = .05) respectively. LS occurred at a median of 1.1 vs 3 months in those with KMT2Ar vs without (P = .16). Among those treated for LS, CR rates were relatively higher in those with KMT2Ar (17 of 40 [42.5%]) vs those without KMT2Ar (6 of 25 [24%] with CR, P = .18,). OS was uniformly poor, with or without KMT2Ar (7 [15.6%] and 1 [4%] survivor, respectively).

Rare immunophenotypes of LS included 3 cases of T-ALL to AML conversion, 2 after anti-CD7 CAR T-cell therapy and 1 after the anti-CD38 monoclonal antibody, daratumumab. The other 2 cases began as B-ALL, 1 of which developed a T-cell/myeloid MPAL⁴¹ and 1 with evolution to plasmablastic lymphoma, first in multiple EMD sites and eventually in the bone marrow⁴⁰ (Table 4). The median age at initial diagnosis and LS for these patients was 18.2 years (range, 14-37.7) and 22.1 years (range, 16.2-40), respectively. Three died from LS, 1 died from TRM, and 1 was alive in CR at time of data cutoff.

In cases of lineage drift, in which immunophenotypic changes were transient or incomplete (eg, partial loss of B-lineage markers or gain of myeloid phenotype), the median age at initial diagnosis and lineage drift presentation was 3.4 years (range, 1 day-17.1) and 4.2 years (range, 0.4-18), respectively (Table 4). Seven (70%) had a KMT2Ar at initial B-ALL diagnosis and all but 1 had confirmed retention of clonality. Half reverted to B-ALL at some point. At data cutoff, only 1 remained alive, and with active disease, suggesting a similarly poor outcome even without full LS.

With emergence of post immunotherapy LS as a variant of immune escape, this comprehensive analysis, representing, to our knowledge, the largest collection of LS cases to date, illustrates the diversity and difficulty in diagnosis and lays a foundation for future study of this complicated mechanism of relapse. Although the identification of LS relies on immunophenotyping by MFC, there is no standardized approach to establishing cytogenetic/molecular retention or clonal harbingers of evolving LS, especially because most cases of LS present unexpectedly. As such, establishing clonal relatedness was challenging and a multitude of assays were used, including cytogenetics and/or other forms of molecular testing. However, these were inconsistently sent at diagnosis and/or time of LS and varied by institution, diagnostic availability, and sample accessibility. Importantly, although the use of the clonoSEQ NGS platform by Adaptive Biotechnologies to track clonal BCR gene rearrangements is becoming more commonplace in the United States, including its use in upfront trials through the Children’s Oncology Group (ClinicalTrials.gov identifier: NCT03914625), in the peri-HSCT setting⁵¹ and after CAR T-cell therapy,⁵² it is not yet routinely used or available. Furthermore, because LS can emerge from different stages of lymphoid leukemogenesis, a B-cell receptor identified in the diagnostic specimen may not be present in the LS population if it originates from an earlier stage of leukemic development.²¹ In such cases, the absence of retention of a BCR gene rearrangement (eg, clonoSEQ) does not eliminate the possibility of clonal relatedness. Additionally, as clonoSEQ does not assess cell surface markers, it cannot be solely used in establishing immunophenotype essential to LS diagnosis. Notably, in cases in which LS emerged in the CNS, the ability to confirm clonal relatedness was limited.

Moreover, because the existing hematopathology classification systems of lineage-defining markers for LS or even ALAL did not always accommodate the spectrum of immunophenotypic changes that arose in these postimmunotherapy relapses,²⁴ we needed to establish a set of definitions (Table 1) to guide the categorization of LS. This additionally served to help characterize a subset of probable LS in which clonal retention could not be confirmed but t-AML was unlikely, and put forth a new category of “lineage drift” to describe cases in which immunophenotypic changes were transient or incomplete, a phenomenon which has previously been described.¹⁴ Although these definitions are not meant to be conclusive, they serve as a foundation to build upon.

Collectively, through this global initiative, several key observations emerged. First, nearly two-thirds of the cohort experiencing B-ALL to AML LS harbored a KMT2A aberration. Given the known association of KMT2Ar with infant leukemia as a driver mutation leading to LS, the overrepresentation of infant ALL in this data set is unsurprising.^5,10,23 The incidence of post immunotherapy LS in patients with KMT2Ar or infant ALL is not fully elucidated, and has ranged from 0% to 28.6% after CD19-directed CAR T cells across retrospective patient cohorts.^1,53,54 This association supports close monitoring for post immunotherapy LS and consideration of combined and/or multi-antigen approaches but does not preclude use of antigen-targeted therapies in infant ALL, particularly given the very good outcomes after use of CD19 targeting in this high-risk population.^53-55 

Beyond KMT2Ar, alternative cytogenetic and molecular aberrations were also found in association with postimmunotherapy LS. With an expanding list of B-ALL subtypes with so-called “lineage plasticity” that have a propensity to express myeloid markers either at diagnosis or during initial therapy, monitoring for LS in these subsets may be warranted.²³ For instance, the finding of BCR::ABL1 translocation in 4 patients, supports the hypothesis of LS occurring in genetic subtypes in which driving alterations in progenitor subsets enable plasticity^18,56 even when the lymphoid compartment is eradicated by B-cell targeting. Accordingly, whether certain patients harbored an undiagnosed chronic myelogenous leukemia remains a possibility, but our data set did not have consistent information regarding BCR::ABL1 breakpoints to help facilitate this determination. In B-ALL harboring the TCF3::ZNF384 fusion or ZNF384 rearrangements, which also has an established association with MPAL²⁵ and predisposition to LS⁵⁷ and frequent coexpression of myeloid markers (CD13 and CD33),⁵⁸ such patients may be more at risk of developing postimmunotherapy LS, as evident by 4 patients in this analysis.^13,17 Similarly, DUX4r (seen in 1 patient) and PAX5 P80R-mutated B-ALL have been associated with an early switch to a monocytic phenotype during induction chemotherapy.⁵⁹ Three cases of CRLF2r B-ALL LS were also identified, suggesting this may be a previously unrecognized B-ALL subtype that associates with LS.⁴⁷ Lastly, the identification of TP53 mutations, which confer high-risk disease and poor prognosis in both ALL and AML, necessitates further study to understand its role in the development of LS.²³ A recent multicenter analysis of 33 patients with LS (in whom a subset had post immunotherapy LS) demonstrated that beyond KMT2Ar, additional molecular and cytogenetic complexities were also seen, highlighting the evolving genetic landscape of LS.²³ 

Along these lines, given frequent myeloid coexpression,^10,60 baseline immunophenotype alone should not be considered a harbinger of postimmunotherapy LS but may warrant closer evaluation of an underlying genetic predisposition to LS.^21,61 For instance, CD371, a transmembrane protein typically expressed on normal monocytes, AML blasts, and leukemic stem cells, has also been found to be expressed by a subset of B-ALL (often with a DUX4r) and can be associated with both a transient early myelomonocytic switch in the favorable prognosis subtype DUX4r as well as an inferior response to therapy in other subtypes,⁶² similar to the experience with aberrant CD2-expressing B-ALL.⁶³ Future mechanistic studies to explore the association of LS with these rare cytogenetic and molecular aberrations beyond KMT2Ar are warranted,^31,61,64 as such, a biobank to capture LS samples to study phylogenetic pathways of leukemic evolution mediated by immune-targeted pressure is planned.

With a primary focus on post immunotherapy LS in B-ALL, the overrepresentation of CD19 targeting as the most proximal therapy reflects a selection bias toward targeted therapies most available and used. Given that a proportion of patients developed LS either during active blinatumomab infusion or in very close proximity to the first post–CAR T-cell restaging evaluation, whether LS is emerging from immune-pressure by direct antigen targeting, or whether effective clearance of B-ALL is facilitating the emergence of aberrant subclinical myeloid phenotypes, is unknown. This is especially complicated in patients receiving multiple antecedent immunotherapies. Although assessment of CD19 genomic alterations was not feasible, further evaluation may provide additional insight into immunophenotypic alteration after immune pressure, particularly with lineage drift. Additionally, although the inflammatory milieu associated with cytokine release syndrome has been hypothesized to facilitate LS transformation through interleukin-6 and other promyeloid cytokines,⁶⁵ further study is needed to elucidate the interplay of these various factors, especially in cases that revert to the original phenotype. The finding that LS occurs across the full age spectrum is particularly relevant given the genetic variability in underlying disease, especially considering that cases are likely underreported. In contrast, the limited racial and ethnic diversity across these cases more likely reflects disparities in trial enrollment or CAR T-cell access^66-68 and we anticipate that any patient receiving antigen-targeted immunotherapy is at risk of LS, especially with certain genomic subtypes.

Beyond the more common presentation of B-ALL to a myeloid phenotype, we also report on 5 rare immunophenotypes of LS (3 T-ALL to AML, 1 B-ALL to T/myeloid MPAL, and 1 B-ALL to plasmablastic lymphoma) and 10 cases of lineage drift, to designate an incomplete or potentially more evanescent change in immunophenotype. Raising awareness of novel associations of LS (eg, AML after CD7-directed CAR T cells or daratumumab for T-ALL), immunophenotypic evolution to circumvent antigen-targeted therapies will be problematic beyond B-cell malignancies. Similar to the subset of infant B-ALL, those with early T-cell precursor ALL also have a predisposition to LS, warranting close monitoring when treating early T-cell precursor ALL with antigen targeting.⁶⁹ Thus, ongoing vigilance to monitor for LS is essential as novel targets are pursued and immunotherapeutic options expand.

Lastly, outcomes were dismal, further emphasizing the urgent need to study LS. In those with LS of B-ALL to AML or ALAL/MPAL (n = 70), the median OS was only 4.8 months after LS diagnosis, which occurred rapidly at a median of 1.5 months from the most proximal immunotherapy. Only 6 patients are alive in a remission, 5 of whom received a consolidative HSCT after remission induction, suggesting that HSCT may be important to achieve cure. Notably, 7 of 8 patients who remain alive had a KMT2Ar leukemia, suggesting that salvage therapy may still be possible, even in this high-risk subtype. Outcomes for those with rare cases of LS and lineage drift were similarly poor. Treatment approaches were varied, without a unified approach, and responses were limited, transient, or incomplete, likely because of the intrinsic or acquired chemotherapeutic resistance from continued therapy. In general, use of a combined regimen of chemotherapy and targeted agents such as gemtuzumab ozogamicin and/or venetoclax appeared to be most effective. However, these combinations were diverse, making it difficult to provide a first-line recommendation. Given the potential of menin inhibition in KMT2Ar leukemia, despite the poor outcomes with the limited use as salvage therapy in this cohort, with recent US Food and Drug Administration approval of revumenib based on impressive outcomes in the AUGMENT-101 trial,⁷⁰ future study of these agents to treat LS are warranted.^71, A recent publication highlighting approaches to treatment of LS provides additional guidance on factors to consider when selecting therapy for LS.⁷² 

Although the incidence of LS cannot be determined from this analysis, the critical need for systematic capture and study of this rare relapse phenomenon cannot be overemphasized. Additionally, despite comprehensive assessment, in cases with missing information in which clonal relatedness could not be confirmed, a treatment-related event, especially in the setting of germ line mutations, remains as a remote possibility, further emphasizing the need to establish a diagnostic framework. Future efforts will focus on understanding the frequency of this form of immune escape, particularly as antigen-targeted therapies move into frontline⁷² and we identify additional leukemia subtypes with a myeloid predisposition. As we prospectively collect samples of LS and establish harmonized analysis, mechanistic evaluations of the biology of LS will be further explored. Given the heterogeneity of our findings (genomic characteristics, prior therapy, and interval to LS) it is likely that multiple mechanisms are responsible.

In conclusion, although antigen-targeted immunotherapies have revolutionized our ability to effectively treat acute leukemias, LS has emerged as a unique mechanism of relapse. As we collectively strive to characterize this phenomenon and improve upon the currently dismal outcomes, these nascent efforts are a first step toward achieving these goals. Ultimately, gaining a better understanding of the potential drivers of LS are critical to finding targetable pathways that can both prevent and treat this phenomenon by incorporating therapeutic approaches based on individual cytogenetic, molecular, and immunophenotypic signatures.

The authors gratefully acknowledge the study participants and their families, referring medical care teams, the faculty and staff of the National Institutes of Health (NIH) Clinical Center who provided their expertise in the management of the study participants; and the data managers, research nurses, and patient care coordinators involved with this work. The authors thank the following consortia and organizations for promoting case contributions to Project EVOLVE: CARnation, Pediatric Real-World CAR T-cell Consortium, Pediatric Transplantation and Cellular Therapy Consortium, Therapeutic Advances in Childhood Leukemia, Children’s Oncology Group, and the Center for International Blood and Marrow Transplant Research. Figures are courtesy of BioRender.com and GraphPad Prism 10. Additionally, the authors acknowledge the following individuals based on their contributions to this effort: Caleb and Lesly Alfaro, Brent L. Wood, Children’s Hospital Los Angeles, and Keck School of Medicine, University of Southern California, Los Angeles, CA; Jennifer J. Clark, HCA HealthONE, Rocky Mountain Hospital for Children; Jennifer Sheppard, Dehoorne Annelore, and Steendam Liesbeth, Ted and Eileen Pasquarello Tissue Bank in Hematologic Malignancies; and Amy Li, funding from Instituto de Salud Carlos III:ICI14/00224 ICI21/00033 competitive grants and from Proyecto ARI. More information on Project EVOLVE is available at https://ccr.cancer.gov/pediatric-oncology-branch/carnation-consortium/project-evolve.

This work was supported, in part, by the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research, and the Warren Grant Magnuson Clinical Center (ZIA BC 011823; N.N.S.).

The content of this publication does not necessarily reflect the views of policies of the Department of Health and Human Services, nor does mention of trade names, commercial products, or organizations imply endorsement by the US government.

Contribution: S.K.S., A.W.R., C.N.H., and N.N.S. led the data collection and data analysis, and contributed to the first version of this manuscript; S.K.S., A.W.R., C.N.H., K.L.D., S.G., E.J., A.J.L., and N.N.S., who are also the members of the Project EVOLVE steering committee, helped to plan this research effort and met regularly to review the analysis and provided critical input on the data collection and subsequent analysis; H.-W.W., A.E.K., C.M.Y., D.D.C., A.B., I.B., K. Murphy, and L.R. performed central review of all available pathology data available on immunophenotype at baseline and lineage switch; all listed authors provided critical insights for Project EVOLVE, either through case submissions and/or insights into the diagnostic approach, case-review, provider outreach, and data analysis; and all authors have reviewed and agreed to the content of the manuscript and contributed meaningfully to this effort.

Conflict-of-interest disclosure: H.A.-A. has served on advisory boards for Adaptive, Vertex, and Johnson & Johnson; and received study support from Adaptive. B.H.C. has received research funding from Deliver Therapeutics. D.S.D. has sponsored research with Syndax; is a consultant for Tempus, Amgen, and Y-mAbs Therapeutics; and serves on the advisory board of Day One Bio. F.E.C. is a consultant with SPD Oncology, Amgen, CTI BioPharma, AbbVie, MorphoSys, Association of Community Cancer Centers, PharmaEssentia, Bristol Myers Squibb, Geron, Sobi, and DAVA Oncology; received clinical trial grant support (principal investigator) to the University of Virginia from Amgen, Celgene, SPD Oncology, Sanofi, Bristol Myers Squibb, FibroGen, PharmaEssentia, BioSight, MEI Pharma, Novartis, and Arog Pharmaceuticals; and received travel grant support from DAVA Oncology. S.A.G. receives clinical research funding from Novartis, Cellectis, Kite, Vertex, and Servier; consults for Novartis, Eureka, and Adaptive; and has advised Novartis, Adaptimmune, Vertex, Allogene, Jazz Pharmaceuticals, and Cabaletta. E.M.H. is a consultant for Novartis. A.A.M. reports research funding from Bristol Myers Squibb, Stemline, Gilead, Incyte, and Novartis, paid to institution. K.R. served as consultant (from April 2023 to April 2024) for Sumitomo Pharma Inc for presenting on patient experience and training clinical research coordinators. A. Stevens reports research funding from AbbVie Pharmaceuticals and Gilead Pharmaceuticals. S.R. reports honoraria and/or travel support from Novartis, Servier, Celgene/Bristol Myers Squibb, Kite/Gilead, Pfizer, Clinigen, and Amgen; and reports being part of data and safety monitoring board in a clinical trial sponsored by Novartis, and of a data monitoring committee in a clinical trial sponsored by Autolus. I.A. served on an advisory board for Kite, Jazz, Syndax, Takeda, Wugen, Pfizer, and Adaptive; and reports research support from MacroGenic, AbbVie, and Jazz. N.B. reports honoraria from Amgen, Pfizer, Novartis, and Gilead. S.R.R. served as consultant on the data and safety monitoring committee for Pfizer, and steering committee for AbbVie. S.G. reports honoraria/speaker fees from Novartis and Autolus; reports patents with University College London Business and Autolus Ltd; and serves on a trial steering committee for Autolus Ltd. N.N.S. receives research funding from Lentigen, Vor Bio, and CARGO therapeutics; and has participated on advisory boards (no honoraria) for Sobi, Allogene, invoX, ImmunoACT, and Vor Bio. The remaining authors declare no competing financial interests.

Correspondence: Sara K. Silbert, Pediatric Oncology Branch, Center for Cancer Research, National Cancer Institute, National Institutes of Health, 9000 Rockville Pike, Building 10-CRC, 1W-5750, Bethesda, MD 20892; email: sara.silbert@nih.gov; and Nirali N. Shah, Pediatric Oncology Branch, Center for Cancer Research, National Cancer Institute, National Institutes of Health, 9000 Rockville Pike, Building 10-CRC, 1W-5750, Bethesda, MD 20892; email: nirali.shah@nih.gov.