PDX models of relapsed pediatric AML preserve global gene expression patterns and reveal therapeutic targets

Patient-derived xenograft (PDX) models are commonly used for preclinical evaluation of targeted therapies. It is important to consider the fidelity with which these systems recapitulate the patient disease state. Currently, there is little information regarding how well pediatric AML PDXs mimic the global gene expression found in patients. Here we analyzed RNAseq data from a diverse series of high-risk pediatric AML PDXs, separately and compared to primary patient data. Unsupervised clustering of PDX data resulted in segregation according to KMT2A (MLL) status. Combined analysis showed PDX samples aligned with patient samples harboring similar genetics. Among KMT2A rearranged samples, we observed strong correlation of expression levels of nearly all expressed transcripts. Furthermore, matched patient/PDX pairs showed strong concordance, suggesting retention of sample-specific gene expression. Interestingly, our analysis uncovered previously unidentified cryptic CBFA2T3-GLIS2 rearrangement in two PDX models. Based on high BCL2 mRNA in these models, we tested the efficacy of venetoclax in combination with chemotherapy. While standard daunorubicin/ara-C treatment failed to produce benefits, CPX-351 decreased disease burden and prolonged survival, particularly when combined with venetoclax. These results validate PDX modeling of high-risk pediatric AML and highlight this system's utility for pre-clinical therapeutic discovery, especially for rare subtypes of disease.