Pharmacogenomic Profiling of Pediatric Acute Myeloid Leukemia to Identify Therapeutic Vulnerabilities and Inform Functional Precision Medicine

Abstract

Despite the expanding portfolio of targeted therapies for adults with acute myeloid leukemia (AML), direct implementation in children is challenging due to inherent differences in underlying genetics. Here we established the pharmacologic profile of pediatric AML by screening myeloblast sensitivity to approved and investigational agents, revealing candidates of immediate clinical relevance. Drug responses ex vivo correlated with patient characteristics, exhibited age-specific alterations, and concorded with activities in xenograft models. Integration with genomic data uncovered new gene-drug associations, suggesting actionable therapeutic vulnerabilities. Transcriptome profiling further identified gene-expression signatures associated with on- and off-target drug responses. We also demonstrated the feasibility of drug screening-guided treatment for children with high-risk AML, with two evaluable cases achieving remission. Collectively, this study offers a high-dimensional gene-drug clinical data set that could be leveraged to research the unique biology of pediatric AML and sets the stage for realizing functional precision medicine for the clinical management of the disease.

Significance: We conducted integrated drug and genomic profiling of patient biopsies to build the functional genomic landscape of pediatric AML. Age-specific differences in drug response and new gene-drug interactions were identified. The feasibility of functional precision medicine-guided management of children with high-risk AML was successfully demonstrated in two evaluable clinical cases. This article is highlighted in the In This Issue feature, p. 476.

Figure 1.

Drug response profile of pediatric AML. A, Heat map indicating the responses of 42 pediatric AML samples to 45 compounds represented by Z-score–transformed AUCs. Samples (rows) and drugs (columns) are ordered by unsupervised hierarchical clustering. Cluster A: highly active drugs with median IC₅₀ values <70 nmol/L. Cluster B: generally active drugs with median IC₅₀ values <700 nmol/L. Clusters C and D: drugs with bimodal activities. Cluster E: generally inactive drugs with sporadic exceptions. Cluster F: completely inactive drugs. The IC₅₀ distribution range for each compound is shown on the boxplot under the drug clusters, with median IC₅₀ values (white dots) and Cmax (red dots) depicted. B, Correlation of pretreatment variables with drug sensitivity for the entire patient cohort (n = 52). Drugs are clustered on drug family. Significant associations (P < 0.05) are connected by arrows (sensitive) or edges (resistant). Drug sensitivity was assigned based on median AUCs. C, Forest plot showing the performance of cytogenetic risk group or ex vivo resistance to chemotherapeutics (defined by median AUCs) for prediction of clinical outcome. Patients (n = 46) with >12 months follow-up were included for analyses. Odds ratio of event (relapse or death), 95% confidence interval, and P values are shown. Statistics: B, two-tailed, unpaired Student t test; C, logistic regression.

Figure 2.

Drug-sensitivity profiles of pediatric and adult AML. A, PCA of the response parameter (i.e., AUC values of 45 drugs) in pediatric (n = 42) and adult AML (n = 26) samples. B, Boxplots showing the AUC distributions of individual drugs, ordered by drug classes. Statistics: two-tailed, unpaired Student t test. P values comparing both AUCs and IC₅₀ values are indicated. Statistically significant values are shown in red. Asterisks mark statistically significant differences after correction for multiple comparison by the Benjamini–Hochberg procedure, with P values after FDR correction shown in blue. C, Fraction of pediatric (n = 52) and adult (n = 26) samples with major genetic alterations detected by targeted sequencing.

Figure 3.

Ex vivo drug activities capture in vivo responses. A, Dose–response curves of venetoclax on primary myeloblasts from 54 pediatric AML samples (diagnosis, n = 42; relapse, n = 12). Samples exhibiting sensitivity (red curves) or resistance (blue curves) were classified with respect to the median IC₅₀ value (6,674 nmol/L) across the patient cohort. Arrows mark the samples selected for animal modeling. B, NSG mice were transplanted with myeloblasts derived from a venetoclax-resistant (LEU280) or a venetoclax-sensitive (LEU350) sample to generate PDXs. Animals (4–5 mice/group) were randomized to receive vehicle or oral venetoclax (100 mg/kg, once daily, 5 days/week for 2 weeks). Treatments commenced on day 3 after transplantation. Circulating leukemic blasts were monitored serially by flow-cytometric detection of human CD45⁺CD33⁺CD19⁻ cells. Statistics: two-tailed, unpaired Student t test. P values are indicated. C, BM samples were collected on day 45 (LEU280) and day 89 (LEU350) posttransplantation for the enumeration of medullary leukemia. D, Fish plots showing the evolution of subclonal architecture for samples LEU280 and LEU350 before (P0) and after (P1) leukemia engraftment in the BM of control NSG mice at the overt disease stage (i.e., day 45 for LEU280 and day 89 for LEU350).

Figure 4.

Integration of genomic alterations with drug sensitivity. A, Mosaic plot showing the mutational landscape of 52 pediatric patients with AML. The earliest specimen was included for analyses for individuals with consecutive sampling. Recurrent mutations occurring in ≥2 patients are shown and ranked by their frequencies in the cohort illustrated by the bar chart. Patients are annotated with clinical features and mutations with types. Asterisks mark pathogenic or likely pathogenic variants. B, Gene–drug associations represented by a volcano plot showing their significance and effects between wild-type and mutant samples. Pathogenic or likely pathogenic variants with VAF >10% were included for the analysis. A P value cutoff of 0.05 (horizontal line) was applied to detect statistically significant associations. A Z score cutoff of 0 (vertical line) defines relative drug sensitivity (red circles) and resistance (blue circles). The size of circles represents the number of patients harboring the mutations. Circles annotated with green and black texts denote known and novel associations, respectively. C, Box plots showing the activities of BCL-2 inhibitors in pediatric AML stratified by KMT2C mutation status: wild-type (n = 29–35), benign or likely benign mutants (n = 8), pathogenic or likely pathogenic mutants (n = 3–4). Median IC₅₀ values are shown. D, Kaplan–Meier estimates of 3-year event-free survival (period from diagnosis to first relapse or death from any cause) in patients with wild-type (n = 40), benign (n = 8), and pathogenic KMT2C (n = 4). Statistics: B, one-way ANOVA; C, two-tailed, unpaired Student t test with correction for multiple comparison by the Benjamini–Hochberg method; D, log-rank test. *, P < 0.05; **, P < 0.01.

Figure 5.

Predictors of drug response. A, AML specimens were stratified by drug activities: (i) sensitive samples with AUC <25th percentile; (ii) intermediate samples with AUC between the 25th and 75th percentiles; and (iii) resistant samples with AUC >75th percentile. B, RNA-seq was performed on 48 specimens. DEGs in sensitive over resistant samples were identified for 36 drugs using FDR <0.05 and log₂ fold change >2 as the cutoffs. Drugs are ranked according to the total number of DEGs. Purple and green bars represent upregulated and downregulated DEGs, and the color intensity denotes fold changes. C, Volcano plots showing DEGs of venetoclax (n = 610) and YM155 (n = 245). Suppressed DEGs in sensitive samples are indicated by blue circles and augmented DEGs by red circles. The top 10 DEGs with the highest correlation with AUC values are marked with yellow frames with gene symbols listed. D, Correlation analyses between DEG (CPM) and drug sensitivity (AUC). DEGs of venetoclax (n = 98) and YM155 (n = 91) with correlation coefficients of < −0.5 or >0.5 are shown and ranked according to the magnitude of correlation. Positively correlated DEGs are indicated by pink bars, and negatively correlated DEGs by green bars. Arrow indicates the venetoclax target BCL-2. E, Intracellular BCL-2 and survivin expression in AML cell lines (n = 10) and samples (n = 8–9) was measured by flow cytometry and correlated with venetoclax and YM155 sensitivity, respectively. Statistics: B, C, DESeq2; D, E, Pearson correlation. *, P < 0.05; **, P < 0.01; ***, P < 0.001; ****, P < 0.0001.

Figure 6.

Precision medicine for high-risk pediatric AML. A, Flowchart showing the workflow of this pilot clinical trial, including patient numbers at each stage and reasons for dropping out of the study. B, Clinical progression of a child with relapsed AML receiving drug profiling-guided treatment. The disease course, treatment landscape, and laboratory investigations are depicted. C, Drug profiling results at first AML relapse. The bar chart shows the activities of 45 screened drugs ranked in descending IC₅₀ values. The horizontal line at 15 nmol/L was arbitrarily set to indicate highly active drugs. The table shows the top-hit drugs annotated with respective IC₅₀ values, AUC, C_max, and FDA-approval status. The drug recommended by the tumor board is indicated by an arrow in the bar chart and highlighted in yellow in the table. D and E, Clinical progression, drug testing results, and therapy recommendation for another case with relapsed MPAL. Abbreviations: BMT, bone marrow transplantation; DLI, donor leukocyte infusion; FLA, fludarabine and cytarabine; FLAD, fludarabine, cytarabine, and daunorubicin; HIC, high-intensity consolidation.