Single cells tell multiple tales in CML

In this issue of Blood, Krishnan et al describe a single-cell transcriptomic atlas of chronic myeloid leukemia (CML) at diagnosis.¹ By comparing groups with varying degrees of response or treatment failure on imatinib, they identified cellular features at diagnosis that were predictive of response to tyrosine kinase inhibitor (TKI) therapy (see figure).

CML has long been viewed as the paradigm of a monogenic cancer, defined by the BCR::ABL1 fusion gene. The opportunity to target the founding lesion of CML with BCR::ABL1 TKIs has led to near-normal life expectancy for most people with CML, yet this headline result conceals striking heterogeneity of outcomes ranging from sustained treatment-free remission (TFR) in 20% to 30% of individuals to blast crisis in around 5% of individuals. Advances in sequencing methods have made it possible to interrogate the biological heterogeneity of CML with ever-increasing resolution so that novel factors that influence treatment response can be identified. Krishnan et al used de novo clustering to identify 32 cell populations from the transcriptomic data, comprising both CD34⁺ cells and CD34^– mononuclear cells. Using differential expression, they were able to distinguish BCR::ABL1-positive primitive cells from the remaining wild-type counterparts. A machine learning approach was used to identify gene expression signatures in individual cell types that were associated with outcome.¹ 

In the leukemic stem and progenitor cell (LSC) compartment, an optimal response to TKI therapy was associated with erythroid differentiation bias.¹ In an important additional step, a flow cytometric approach was used to provide independent technical validation of the erythroid expression signature in LSCs. Conversely, an inflammatory and MYC-driven expression signature was associated with development of blast crisis,¹ in agreement with a prior study of single-cell transcriptomics in sorted LSCs from CML.² 

In the immune cell compartment, favorable outcomes were associated with expansion of adaptive-like, hyperfunctional natural killer (NK) cells, whereas inhibitory NK cells were increased in patients who developed blast crisis.¹ Increased numbers of immune effectors and decreased immune suppressors measured in peripheral blood after an optimal response to TKI treatment are associated with a higher probability of TFR.^3,4 The work of Krishnan et al provides evidence that this immune balance may, at least in part, be predetermined, rather than being an outcome of TKI treatment, and a net inhibitory immune profile may also be associated with the emergence of blast crisis.

Risk-adapted therapy in CML is substantially driven by molecular response milestones during treatment. Slower decline in BCR::ABL1 and failure to achieve early molecular response (EMR; BCR::ABL1 ≤10% by 3 months) are associated with higher rates of treatment failure and reduced chance of TFR.⁵ Although the first-line use of a more potent TKI reduces the incidence of EMR failure, the adverse impact of EMR failure on imatinib is only partially rescued by subsequent intensification of TKI treatment.⁶ A small proportion of patients develop early blast crisis even when treated first line with a more potent TKI, and for those individuals, novel therapies that might carry an increased risk of toxicity could be justified. In a machine learning logistic regression model, the authors developed a classifier for blast crisis with no false positives, and a sensitivity of 50% to 67%.¹ This cohort included only 6 patients, and the cases were diverse with respect to latency of blast crisis (76-2936 days) as well as lineage and genetics, so validation of these findings in a larger cohort is required. If patients destined for transformation or treatment failure could be identified at diagnosis, there would be an opportunity to design clinical trials of intensified first-line treatment to reduce this risk.

Recent studies have begun to dissect the relevance of genomic lesions in addition to BCR::ABL1, showing that some variants associated with blast crisis are already present at diagnosis.⁷ Variants in ASXL1 were found in 9% of 222 patients with chronic phase CML in a clinical trial using the more potent TKI, nilotinib, and were associated with a reduced incidence of major molecular response (BCR::ABL1 ≤0.1%).⁸ Variants in ASXL1 were detected in 23.8% of patients by Krishnan et al (whose series was enriched for higher-risk individuals), but were identified across all response cohorts, including those with optimal response.¹ One of the limitations of bulk sequencing is that it cannot distinguish between the 3 cellular contexts in which a mutation might occur. Additional variants could represent clonal evolution (BCR::ABL1 first), a preleukemic clone (BCR::ABL1 second), or independent clonal hematopoiesis (BCR::ABL1 wild type). In Ph-negative myeloproliferative neoplasms, it has been shown that the order of acquisition of mutations can influence disease biology.⁹ Larger single-cell sequencing studies may help to determine whether the prognostic effect of an additional variant in CML is influenced by its clonal relationship to BCR::ABL1.

Broader application of single-cell sequencing for clinical purposes is currently limited by cost and analytical complexity. Recognizing the need for simpler, more widely applicable methods to validate and apply these findings, Krishnan et al showed that flow cytometry or mass cytometry could be used to detect the predictive signatures from their transcriptomic analysis. Although larger single-cell sequencing studies may be undertaken in the future, the data contained in this report present multiple hypotheses that can be tested now. Some of the biological pathways that influence optimal response and TFR overlap with those that lead to treatment failure and blast crisis. Understanding and targeting these pathways has the potential to lead to improved outcomes at both ends of the response spectrum in CML.

Conflict-of-interest disclosure: D.M.R. has received research funding from Novartis and BMS and honoraria from Novartis, BMS, Takeda, Keros, and Menarini.