Leukemia stemness and co-occurring mutations drive resistance to IDH inhibitors in acute myeloid leukemia

Abstract

Allosteric inhibitors of mutant IDH1 or IDH2 induce terminal differentiation of the mutant leukemic blasts and provide durable clinical responses in approximately 40% of acute myeloid leukemia (AML) patients with the mutations. However, primary resistance and acquired resistance to the drugs are major clinical issues. To understand the molecular underpinnings of clinical resistance to IDH inhibitors (IDHi), we perform multipronged genomic analyses (DNA sequencing, RNA sequencing and cytosine methylation profiling) in longitudinally collected specimens from 60 IDH1- or IDH2-mutant AML patients treated with the inhibitors. The analysis reveals that leukemia stemness is a major driver of primary resistance to IDHi, whereas selection of mutations in RUNX1/CEBPA or RAS-RTK pathway genes is the main driver of acquired resistance to IDHi, along with BCOR, homologous IDH gene, and TET2. These data suggest that targeting stemness and certain high-risk co-occurring mutations may overcome resistance to IDHi in AML.

Fig. 1. Mutational landscape of IDHi-treated AML patients and their association with clinical response.

A Landscape of high-confidence somatic mutations detected in baseline samples by sequencing with a 295-gene panel. Legend for the best response is located at the top left, while legend for the mutation classification is located at the top right. Baseline mutation data are available for 59 patients. B Forrest plot showing enrichment of the mutations at baseline against complete remission (CR) by logarithmic odds ratio. Two-sided Fisher’s exact test was performed. *P < 0.05 (P = 0.012 for RUNX1; P = 0.037 for TF). Circles (center of the error bars) represent odds ratios. The error bars represent 95% confidence interval of odds ratio. Baseline mutation data are available for 59 patients, out of which 11 achieved CR. Genes mutated in three or more patients are plotted. TS tumor suppressor, TF transcription factors. Source data are provided as Source data files.

Fig. 2. Analysis of DNA methylation at baseline samples reveals two distinct clusters associated with treatment response.

A Consensus k-mean clustering of promoter methylation data at baseline revealed two distinct clusters. Methylation data are based on methylation beta value. Promoter CpG probes from top 1% most variably methylated CpG probes were selected for the analysis. Responders were defined as patients achieved best response of CR, CRp, MLFS, PR, and HI. Nonresponders were defined as patients achieved best response of PD and SD. B Top: Box plot comparing mean methylation beta value of top 1% most variably methylated CpGs among baseline samples for cluster 1 (N = 36) and cluster 2 (N = 15). IDH1/2 wild-type AML samples (N = 8) are used as control. Box plot shows the minimum, first quartile (Q1), median, third quartile (Q3), and maximum. Two-sided Student’s t test was performed. Bottom: Density distribution of top 1% most variably methylated CpG probes with methylation beta values comparing baseline samples of cluster 1 and cluster 2. Two-sided Kolmogorov–Smirnov test was performed. IDH1/2 wild-type AML samples (N = 8) are used as control. C Forrest plot showing enrichment of the mutations at baseline against cluster 2 by logarithmic odds ratio. Circles (center of the error bars) represent odds ratios. The error bars represent 95% confidence interval of odds ratio. Baseline mutation data with clustering information are available for 51 patients, out of which 15 is in cluster 2. D Bar plot comparing the overall response (OR) and CR rate between cluster 1 (N = 36) and cluster 2 (N = 14) patients. Two-sided Fisher’s exact test was performed. E Metascape analysis of hypermethylated promoter DMPs. F Starburst plot showing integrated analysis of gene expression and promoter methylation changes between cluster 1 and cluster 2. Of 215,521 promoter CpGs, 704 showed statistically significant differential methylation and differential expression between cluster 1 and cluster 2. A total of 558 of 704 double significant CpGs showed promoter hypermethylation and downregulation of the gene expression in cluster 2. G Gene set enrichment analysis (GSEA) comparing gene expression profiles between the two clusters revealed upregulation of genes associated with leukemia stem cells (LSCs) in cluster 2. Source data are provided as Source data files.

Fig. 3. Leukemia stemness is associated with primary resistance to IDHi.

A List of top driver transcription factors (TF) and signaling genes (SIG) identified by NetBID2 analysis by comparing gene expression profiles between cluster 1 and cluster 2 (P < 0.01 was used as the significance cutoff). Drivers identified for cluster 1 and cluster 2 are colored with red and blue, respectively. B Heatmap of NetBID-based activity of top drivers in cluster 2. Samples in cluster 1 (CL1) are labeled as red, while cluster 2 (CL2) are labeled as blue. C LSC17 score was calculated for each baseline sample and compared between patients achieving CR (N = 9) vs. not (N = 30) and OR (N = 16) vs. not (N = 23). *P < 0.05 (P = 0.011 for CR vs. non-CR; P = 0.037 for OR vs. non-OR). Box plot shows the minimum, first quartile (Q1), median, third quartile (Q3), and maximum. D, E Receiver operating curve (ROC) for predicting CR or OR with LSC17 score, RUNX1 mutation status, RAS-RTK mutation status, and ELN cytogenetic risk classification. Source data are provided as Source data files.

Fig. 4. DNA methylation changes after IDHi.

A Longitudinal trend of methylation level at baseline (BL) and posttreatment (POST) for all, cluster 1 and cluster 2 patients. Responders and nonresponders are color coded. Two-sided Student’s t test was performed. B Box plot showing maximum reduction of plasma 2HG levels after IDHi treatment (%) in cluster 1 (N = 17) and cluster 2 (N = 10) patients, as well as responders (N = 13) and nonresponders (N = 14). Box plot shows the minimum, first quartile (Q1), median, third quartile (Q3), and maximum. Two-sided Student’s t test was performed. C Scatterplot showing correlation of the longitudinal methylation changes between cluster 1 and cluster 2 patients. Each dot represents a CpG probe and was colored based on its significance in the longitudinal differentially methylation test in either cluster 1 or cluster 2 patients. The X-axis represents the differential methylation level between BL and POST samples (i.e., beta value at BL minus beta value at POST) in cluster 1 patients and the Y-axis represents the differential methylation level between BL and POST samples in cluster 2 patients. D Scatterplot showing correlation of the intercluster methylation differences between BL and POST time points. Each dot represents a CpG probe and was colored based on its significance in the intercluster differentially methylation test at either BL or POST time points. The X-axis represents the differential methylation level between cluster 1 and cluster 2 in BL samples (i.e., beta value of cluster 1 minus cluster 2) and the Y-axis represents the differential methylation level between cluster 1 and cluster 2 in POST samples. E Venn diagram showing the overlapped DMPs between cluster 1 and cluster 2 at baseline (BL) and posttreatment (POST). Among 2621 DMPs hypermethylated in cluster 2, 2156 overlapped between BL and POST, suggesting that most of the hypermethylated DMPs in cluster 2 were the same before and after treatment. F Scatterplot showing correlation of the longitudinal methylation changes between responders and nonresponders. Each dot represents a CpG probe and was colored based on its significance in the longitudinal differentially methylation test in either responders or nonresponders. The X-axis represents the differential methylation level between baseline and response samples in responders (i.e., beta value at BL minus beta value at POST) and the Y-axis represents the differential methylation level between baseline and non-response samples in nonresponders. G GSEA analysis comparing gene expression of posttreatment samples between cluster 1 and cluster 2 showed that LSC genes are still upregulated in cluster 2 posttreatment. Source data are provided as Source data files.

Fig. 5. Selection of resistant mutations accompanies relapse after IDHi.

A Longitudinal mutation landscape plot showing mutation acquisitions in 16 out of the 18 tested relapsed cases. Each column represents an individual case with differentially shaped triangles representing mutation status in either baseline or relapse. B Bar plot showing percentage of tested relapsed cases with acquired mutations in various genes and pathways. C The longitudinal trajectory of mutation VAFs, bone marrow (BM) blast counts, absolute neutrophil count (ANC), hemoglobin (HGB) counts, and platelet (PLT) counts in UPN2394529 (C). Line plots show mutation VAFs and ANC/HGB/PLT counts. Blue shades represent BM blast counts. D Single-cell landscape of selected mutations in UPN2394529. Each column represents one individual cell. A total of 1000 cells scale bar is shown on the top left. E GSEA comparing gene expression data from RNA sequencing between baseline and relapse samples showing significant enrichment of E2F targets, TNF alpha signaling via NF-kappa B, and G2M checkpoint genes, in relapse samples. Source data are provided as Source data files.

Fig. 6. Heterogeneous patterns of genetic and epigenetic evolution in AML patients treated with IDHi.

A–E Multidimensional longitudinal plot of mutation VAFs, bone marrow (BM) blast counts, absolute neutrophil count (ANC), hemoglobin (HGB) counts, platelet (PLT) counts, 2HG level, and DNA methylation level in UPN1825001 (A), UPN2463247 (B), UPN2297625 (C), UPN2620771 (D), and UPN2370759 (E). Line plots show mutation VAFs, ANC/HGB/PLT counts, and 2HG level. Blue shades represent BM blast counts. Violin plots show the methylation distribution. Violin plot shows the minimum, first quartile (Q1), median, third quartile (Q3), and maximum. Source data are provided as Source data files.