# Pediatric acute myeloid leukemia tumor composition predicts patient outcomes at diagnosis and reveals mechanisms of resistance to chemotherapy

## TL;DR

This study shows that analyzing the composition of tumors in pediatric acute myeloid leukemia can predict patient outcomes and treatment response, offering new strategies for personalized therapy.

## Contribution

The study introduces a novel risk prediction strategy combining tumor subclone detection with existing biomarkers to improve outcomes in pediatric AML.

## Key findings

- Expanding and transforming subclones at diagnosis predict patient outcomes and chemotherapy response.
- Combining subclone detection with cytogenetic biomarkers improves risk prediction, even for patients in remission.
- Outcome-predictive subclones show elevated gene expression programs linked to chemoresistance.

## Abstract

Although most pediatric acute myeloid leukemia (pAML) patients achieve complete remission with standard-of-care chemotherapy, overall outcomes are poor, and 40% will eventually relapse. Improved methods for risk assessment at diagnosis and alternative therapies are needed to improve outcomes for these patients. Toward these objectives, we characterized the clonal composition of pAMLs, identifying subclones that expand or transform between diagnosis and relapse. We further showed that the abundance of these expanding and transforming subclones in diagnostic samples is predictive of patient outcomes and, similarly, predicts response to chemotherapy and targeted therapies in patient samples and patient-derived xenograft models. Moreover, gene expression programs previously associated with pAML chemoresistance are recurrently elevated in these predictive subclones. Consequently, we propose a novel strategy for improving pAML risk prediction at both diagnosis and during therapy that combines the detection of outcome-predictive tumor subclones in pAML blood or bone marrow with cytogenetic biomarkers and residual disease assessment. Critically, we showed that this combination dramatically improved risk prediction, including for patients who achieve complete remission after chemotherapy. Moreover, through our analyses of outcome-predictive pAML subclones, we identified potential personalized targeted therapies for pAML patients based on the composition of their tumors.