Heterogeneity in refractory acute myeloid leukemia

Abstract

Successful clinical remission to therapy for acute myeloid leukemia (AML) is required for long-term survival to be achieved. Despite trends in improved survival due to better supportive care, up to 40% of patients will have refractory disease, which has a poorly understood biology and carries a dismal prognosis. The development of effective treatment strategies has been hindered by a general lack of knowledge about mechanisms of chemotherapy resistance. Here, through transcriptomic analysis of 154 cases of treatment-naive AML, three chemorefractory patient groups with distinct expression profiles are identified. A classifier, four key refractory gene signatures (RG4), trained based on the expression profile of the highest risk refractory patients, validated in an independent cohort (n = 131), was prognostic for overall survival (OS) and refined an established 17-gene stemness score. Refractory subpopulations have differential expression in pathways involved in cell cycle, transcription, translation, metabolism, and/or stem cell properties. Ex vivo drug sensitivity to 122 small-molecule inhibitors revealed effective group-specific targeting of pathways among these three refractory groups. Gene expression profiling by RNA sequencing had a suboptimal ability to correctly predict those individuals resistant to conventional cytotoxic induction therapy, but could risk-stratify for OS and identify subjects most likely to have superior responses to a specific alternative therapy. Such personalized therapy may be studied prospectively in clinical trials.

Keywords: acute myeloid leukemia; cancer heterogeneity; drug resistance.

Fig. 1.

OS and somatic mutation summary for the 154 cases of newly diagnosed AML patients. (A) Two groups of patients that had a CR (n = 111) or refractory (Ref, n = 43) response upon induction (cytarabine- and anthracycline-based) chemotherapy. The pie chart shows the proportion of patients in each group, with corresponding Kaplan–Meier (K-M) survival curves with log-rank test P values. (B) The 2017 ELN risk stratification summary. The pie chart and the stacked bar chart show the overall percentage and the percentage within the CR or Ref group, followed by K-M curves of three risk groups. (C) Heat map of all somatic mutations with cytogenetic information. Somatic mutations are grouped by gene functions/class (68). NPM1 and FLT3-ITD results were based on results of clinical testing, with other mutations based on exome and/or targeted DNA-sequencing panel results (20).

Fig. 2.

Genome-wide gene expression profile of pretreatment samples from patients. (A) Schematic illustration of study design. (B) Top 100 differentially expressed genes based on adjusted P value (FDR). Sample level expression values are mean-centered log2 counts per million with trimmed mean of M values (TMM) normalization.

Fig. 3.

LSC17 score and related survival. Patients are separated by the median score and labeled as LSC^hi (patients with scores greater than median) and LSC^lo (patients with scores lower than median). (A) LSC17 scores of AML patients in CR and refractory (Ref) groups, with t test P values. (B) Bar chart of LSC^hi and LSC^lo patient number in CR and Ref groups. (C) Kaplan–Meier (K-M) curves of LSC^hi and LSC^lo groups for CR patients. (D) K-M curves of LSC^hi and LSC^lo groups for Ref patients.

Fig. 4.

Expression profile of pretreatment samples from newly diagnosed AML patients reveals distinct subpopulations of refractory AML. (A) Hierarchical clustering of refractory groups reveals three distinct subpopulations (Ref1, Ref2, and Ref3). (B) LSC17 scores of complete responders (CR), Ref1, Ref2, and Ref3. (C) OS of CR, Ref1, Ref2, and Ref3 (P = 0.0018). (D) Top 100 differentially expressed genes in Ref1, Ref2, and Ref3 based on adjusted P value (FDR). Sample level expression values are mean-centered log2 counts per million with trimmed mean of M values (TMM) normalization. (E) OS of patients. The Kaplan–Meier (K-M) curves are plotted for all patients separated by LSC17 scores, by expression profile similar to Ref3 (patients similar to Ref3 are labeled as RG4^pos, with the others labeled as RG4^neg), and combined. (F) OS of TCGA cohort. The K-M curves are plotted for all patients separated by LSC17 scores, by expression profile similar to Ref3, and combined.

Fig. 5.

Characterization of enriched cp in refractory subgroups. (A) GSEA (MSigDB, C2 cp) of refractory subgroups (Ref1, Ref2, and Ref3) compared with the complete responder group. Top 20 gene sets with an FDR < 0.05 in at least one comparison are plotted in the heat map (***P < 0.01, **P < 0.05, *P < 0.1). Red and blue correspond to up-regulated and down-regulated pathways, respectively. GSEA of the top two pathways up-regulated (red) and down-regulated (blue) in Ref1 (B), Ref2 (C), and Ref3 (D) is illustrated. The normalized enrichment score (NES) and FDR are indicated. (E) Characteristic altered cellular processes in refractory AML. Red, blue, and green cells indicate Ref1, Ref2, and Ref3, respectively. (F) List of significantly up-regulated ABC transporters in Ref3. The FC (compared with the complete responders) is shown. Adj., adjusted.

Fig. 6.

Drug sensitivity landscape of refractory AML. (A) Schematic illustration of study design. A total of 103 pretreatment samples from newly diagnosed cases of AML that had either a CR (n = 74) or refractory (Ref) response (n = 29) upon induction chemotherapy were used in this study. The freshly isolated mononuclear cells of AML patients were exposed to seven-point drug dilution treatment against 122 small-molecule inhibitors. The drug sensitivity of these patient-isolated cells was examined. (B) Box plot illustrates treatment response of samples from FLT3-ITD and FLT3 wild-type (WT) AML patients against quizartinib (AC220), a drug known to improve the OS of FLT3-ITD–mutated AML patients. The receiver operating characteristic (ROC) curve indicates the significance of the quizartinib effect on FLT3-ITD patient samples. (C) Box plot illustrates treatment response between CR, Ref1, Ref2, and Ref3 patient samples against GW-2580 (P value: ANOVA between four groups) and venetoclax (P value: ANOVA between refractory subgroups). (D) Box plot illustrates treatment response between CR and refractory (Ref) or CR, Ref1, Ref2, and Ref3 patient samples against flavopiridol. ROC curves indicate the significance of the flavopiridol effect on patient samples. (E) Box plot illustrates treatment response of patient samples with an adverse or favorable ELN classification against flavopiridol. The ROC curve indicates the significance of the flavopiridol effect on samples from patients with an adverse or favorable ELN classification. (F) Top 100 differentially expressed genes of AML patient samples based on their sensitivity to flavopiridol (all with an adjusted P < 0.02). Sample level expression values are mean-centered log2 counts per million with trimmed mean of M values (TMM) normalization. Sensitivity to flavopiridol was defined based on the top 25% AUC and bottom 25% AUC. The resistant group has an AUC ≥ 164.6 (26 patients composed of 25 CR and 1 Ref patients) with a significant binomial test (P = 0.0015). The sensitive group has an AUC ≤ 100.88 (26 patients composed of 16 CR and 10 Ref patients) with no significant binomial test (P = 0.615).