From theory to therapy: Predicting response to chemotherapy with multiomic longitudinal profiling of peripheral blood in AML Free

Acute myeloid leukemia (AML) is an aggressive hematological malignancy with 5-year survival rate of less than 32%, mainly due to treatment failure and disease relapse. Thus, a systematic framework that allows us to monitor treatment response is urgently needed. Previously, we applied a state-transition modeling framework to represent AML disease evolution as messenger RNA (mRNA) and microRNA (miRNA) transcriptome trajectories in their respective state-spaces, partitioning samples into health, transition, and leukemic states based on mathematically defined critical points. These state-spaces were constructed using time-series RNA-seq data collected from a murine model of AML which mimics the fusion geneCBFB::MYH11 (CM) created by chromosome 16 inversion.

To test the hypothesis that response to chemotherapy could be predicted by mRNA and miRNA transcriptome dynamics using a state-transition framework, we performed an experiment and mathematical analysis of longitudinal sample collections and multiomic profiling of peripheral blood from CM knock-in mouse model (Cbfb^56M/+/Mx1-Cre) before, during, and after chemotherapy, as follows. After detection of overt AML (circulating cKit+ > 20%), CM mice (n=9) were treated with a “5+3” combination of cytarabine (50mg/kg/day, 5 days) and daunorubicin (1.5mg/kg/day, 3 days), modeling the standard “7+3” chemotherapy. Peripheral blood samples were collected weekly and were subjected to bulk RNA-seq and miRNA-seq. We used PCA to project the samples into a reference state-space constructed using untreated CM mice and C57BL/6 WT controls. The mRNA and miRNA transcriptome trajectories were first analyzed independently and then together to form a 2D multiomic state-space.

After chemotherapy, we observed movement of both mRNA and miRNA transcriptomes from a leukemia state back towards a healthy state, before relapsing back to overt AML. Interestingly, we observed a desynchronization of mRNA and miRNA transcriptome dynamics post-chemotherapy. Specifically, the mRNA transcriptome immediately moved from a leukemia state to a health state, with an average peak response at week 4 (n=7; before week 4, n=2), however, a delayed response of 2 or more weeks was observed in the miRNA transcriptome trajectories, which continued to move towards a disease state post-treatment before moving towards a health state and reached a maximum response at later time-points (4 weeks, n=3; 5 weeks, n=2; 6 weeks and after, n=4). These desynchronized dynamics post-chemotherapy were enabled by the state-transition critical points and appeared more evident when visualized in a 2D multiomic state-space.

To identify genes, miRNAs, and pathways altered by chemotherapy, we performed differential gene expression and gene set enrichment analyses. These analyses revealed that metabolic pathways such as OXPHOS, fatty acid metabolism and glycolysis are downregulated after chemotherapy and upregulated during relapse (p<0.05).

We also performed a weighted gene co-expression network analysis on mRNAs and miRNAs to identify features with similar expression dynamics. We found a set of 34 miRNAs which were significantly upregulated and responsible for the delayed response of the miRNA transcriptome dynamics post-chemotherapy. Out of these 34 miRNAs, 23 are found in the DLK1-DIO3 imprinted region, located on chromosomes 14q32 in humans and 12qF1 in mice, which has been associated with stress response, pathogenesis of APL, solid tumors, and type 2 diabetes. These miRNAs have not previously been associated with effects of standard-of-care chemotherapy in AML. The co-localization of these miRNAs to a region of the genome suggests a potential explanation for genome instability and chemotherapy resistance.

In summary, this study provides a longitudinal multiomic characterization of genomic instability and desynchronization of patterns of mRNA and miRNA expression induced by standard-of-care “7+3” therapy in AML, guided by mathematical analysis and critical points.