Abstract
Acute myeloid leukemia (AML) is a genetically and phenotypically heterogeneous disease, with diverse therapeutic responses. Ex vivo drug sensitivity testing (DST) allows direct functional profiling of patient-derived leukemia cells and may complement genomic biomarkers to guide treatment. While previous DST studies have demonstrated clinical utility in Western populations, limited data exist in Asian cohorts. We conducted a DST-based feasibility study in a Korean AML population to (1) quantify patient-specific drug responses, (2) correlate functional profiles with clinical outcomes and genomic alterations, and (3) identify potential mutation-specific drug sensitivities.
Bone marrow aspirates from 58 AML patients were subject to DST using a viability-based assay across a panel of up to 21 anti-leukemic agents, including cytotoxic, targeted, and epigenetic drugs. Drug response metrics (IC₅₀, Emax, AUC) were derived for each sample–drug pair. Targeted sequencing identified mutations in TP53, FLT3-ITD, NPM1, RUNX1, ASXL1, DNMT3A, IDH1, and IDH2. Among the entire cohort, 31 patients received treatment with available clinical outcome data. Patients were categorized as newly diagnosed (ND, n=17) or relapsed/refractory (R/R, n=14) and stratified by treatment regimen: Cytarabine + anthracycline, hypomethylating agent (HMA) + venetoclax, or gilteritinib.
DST was technically feasible in 45 viable samples. Drug responses were highly variable across patients, with dasatinib, venetoclax, gilteritinib, and mitoxantrone showing the greatest inter-patient variability. Genomic stratification revealed established associations: TP53 mutations correlated with multi-agent resistance, while NPM1 and IDH1/2 mutations were associated with increased sensitivity to venetoclax-based combinations. Novel findings included strong ex vivo sensitivity to sorafenib in IDH2-mutated cases (P <0.05) and to midostaurin in RUNX1-mutated samples (P <0.05)–associations not previously reported.
When stratified by disease status, R/R patients showed inconsistent correlations between DST profiles and clinical outcomes, likely due to clonal evolution and residual disease complexity after prior treatment. In contrast, ND patients exhibited clearer associations. Among ND patients treated with venetoclax + hypomethylating agents (HMA), ex vivo sensitivity to venetoclax, azacitidine, and decitabine showed strong correlation with overall survival (OS), with ROC-AUCs of 0.87, 0.67, and 0.87, respectively. Kaplan-Meier survival analysis based on DST-derived risk groups revealed significant differences: Venetoclax (hazard ratio, HR of 9.1, P<0.05), azacitidine (HR >10.0, P<0.01), decitabine (HR of 7.9, P<0.05). In ND AML patients treated with cytarabine + anthracycline, ex vivo sensitivity to mitoxantrone predicted OS with perfect discrimination (ROC-AUC = 1.00; HR >10, P<0.05). These results suggest that in treatment-naïve patients, DST can identify clinically meaningful responders and high-risk individuals with strong prognostic implications.
This study demonstrates the feasibility and predictive potential of ex vivo DST in a Korean AML cohort. In newly diagnosed patients, DST-based functional profiling was strongly associated with survival outcomes and enabled risk stratification within clinically homogeneous treatment groups. Integration with genomic data confirmed known drug–mutation relationships and revealed novel sensitivity patterns, including IDH2–sorafenib and RUNX1–midostaurin. These findings support the incorporation of DST into precision oncology frameworks and highlight its utility in both therapeutic guidance and biomarker discovery. Prospective validation in larger cohorts is warranted.
