Persistent postremission clonal hematopoiesis shapes the relapse trajectories of acute myeloid leukemia

Abstract

Mutations found in acute myeloid leukemia (AML) such as DNMT3A, TET2, and ASXL1 can be found in the peripheral blood of healthy adults, a phenomenon termed clonal hematopoiesis (CH). These mutations are thought to represent the earliest genetic events in the evolution of AML. Genomic studies on samples acquired at diagnosis, remission, and at relapse have demonstrated significant stability of CH mutations after induction chemotherapy. Meanwhile, later mutations in genes such as NPM1 and FLT3 have been shown to contract at remission, and in the case of FLT3 often are absent at relapse. We sought to understand how early CH mutations influence subsequent evolutionary trajectories throughout remission and relapse in response to induction chemotherapy. We assembled a retrospective cohort of patients diagnosed with de novo AML at our institution that underwent genomic sequencing at diagnosis, remission, and/or relapse (total N = 182 patients). FLT3 and NPM1 mutations were generally eliminated at complete remission but subsequently reemerged upon relapse, whereas DNMT3A, TET2, and ASXL1 mutations often persisted through remission. CH-related mutations exhibited distinct constellations of co-occurring genetic alterations, with NPM1 and FLT3 mutations enriched in DNMT3Amut AML, whereas CBL and SRSF2 mutations were enriched in TET2mut and ASXL1mut AML, respectively. In the case of NPM1 and FLT3 mutations, these differences vanished at the time of complete remission yet readily reemerged upon relapse, indicating the reproducible nature of these genetic interactions. Thus, CH-associated mutations that likely precede malignant transformation subsequently shape the evolutionary trajectories of AML through diagnosis, therapy, and relapse.

Graphical abstract

Figure 1.

Charting the genomic evolution of de novo AML at diagnosis, remission, and REL. (A) Frequently mutated genes and karyotype aberrations at time of diagnosis in the Penn AML cohort (total n = 182 patients). Patients are annotated by the treatments received throughout the course of the disease and by ELN 2022 risk classifications. (B) Distribution of the number of serial genomic profiles obtained for each patient, expressed as a percentage of the total cohort. All patients included in the cohort underwent genomic profiling at least twice, with more than half having ≥3 matched genomic samples. The number of patients in each category is annotated above. (C) Distribution of the number of patients with a genomic profile at each stage of AML disease progression, expressed as a percentage of the total cohort. The number of patients represented in each category is annotated above. (D-G) Comparison of cohort-level mutation (mut) frequencies across different disease time points; (D) diagnosis (n = 182) vs CR1 (n = 119); (E) CR1 vs REL1 (n = 76); (F) diagnosis vs REL1; and (G) REF1 (n = 27) vs REL1. Mut frequencies are calculated from all samples at each time point, irrespective of patient-level sample matching. Point sizes are scaled by statistical significance (Fisher's 2-sided exact test) and colored based on mut frequency. Asterisks indicate P < .05. Dashed lines denote equality between disease stages. ATO, arsenic trioxide; ATRA, all-trans retinoic acid; chr, chromosome; ELN, European LeukemiaNet; HDACi, histone deacetylase inhibitor; HMA, hypomethylating agent; IDHi, IDH inhibitor; MEC, mitoxantrone etoposide and cytarabine.

Figure 2.

Muts associated with CH are persistent at remission. (A) Bar plot depicting the percentage of muts identified in diagnosis that were also identified at CR1. Numbers to the right of each bar indicate the proportion of variants initially found at diagnosis that were subsequently detected at CR1. Genes are grouped into their relevant biological categories as follows: DNA damage (TP53, ATM); DTAI (DNMT3A, TET2, ASXL1, IDH2, and IDH1); splicing (SRSF2, U2AF1, and ZRSR2); PRC/RUNX (BCOR, BCORL1, RUNX1, and EZH2); cohesin (SMC1A, RAD21, and STAG2); and signaling (CSF1R, FLT3, NF1, KRAS, NRAS, BRAF, KIT, PTPN11, JAK2, CSF3R, and CBL). (B) As in panel A, but individual genes are shown. (C) Violin plot of VAFs for persistent variants at CR1. (D-F) Scatterplot detailing patient-matched VAFs at diagnosis (x-axis) and CR1 (y-axis) for (D) IDH1 and IDH2; (E) DNMT3A, TET2, and ASXL1; and (F) FLT3, NPM1, and NRAS. Each point represents 1 variant in a specific patient, matched across time. DTAI, DNMT3A, TET2, ASXL1, IDH1/2; PRC, polycomb repressive complex.

Figure 3.

Signaling muts undergo dynamic losses and gains from diagnosis through REL. (A) Bar plot depicting the relative proportions of different mut trajectories between diagnosis and REL1 in individual patients (n = 76), filtered for genes with at least 4 variants identified across the cohort. Colors indicate whether the mut was stable (gray), lost (yellow), or gained (teal) from diagnosis through REL. The number to the right indicates the total number of variants identified among paired diagnosis and REL samples for the indicated gene; within each section of the bar plot, the numbers indicate the number of variants within each category. (B) Violin plot depicting the difference in VAFs (ΔVAF) between REL and diagnosis. Negative values indicate a lower VAF at REL, whereas positive values indicate a higher VAF. (C-E) Scatterplot indicating VAF at diagnosis (x-axis) and REL1 (y-axis) for (C) IDH1 and IDH2; (D) DNMT3A, TET2, and ASXL1; and (E) FLT3, NPM1, and NRAS. Each point represents 1 variant in a specific patient, matched across time. (F-I) Contingency tables evaluating the association between FLT3i treatment (F), STAG2 muts at diagnosis (G), PTPN11 muts at diagnosis (H), or NRAS muts at diagnosis (I) with FLT3 mut loss in panels F-G, FLT3 mut gain in panel H, or WT1 mut loss in panel I. In the event that a patient had multiple variants in the same gene, the variant with the largest change in VAF between diagnosis and REL was retained. Statistics were determined by Firth penalized logistic regression models; for panels G-H, FLT3i treatment status was included as a covariate in the model. OR, odds ratio.

Figure 4.

Early muts in DNMT3A, TET2, and ASXL1 differentially shape the subsequent evolution of AML from diagnosis (diag) through REL. (A) Upset plot indicating the number of patients at diag (n = 182) with muts in DNMT3A, TET2, and ASXL1. The number of patients per group is indicated above each bar. (B) Co-mut analysis of FLT3, NPM1, CBL, and SRSF2 in relation to DNMT3A, TET2, or ASXL1 across the entire cohort. Dots are color-coded by logORs and size-scaled by statistical significance (Fisher's 2-sided exact test). Asterisks denote P < .05. (C) Bar plot detailing the frequency of FLT3, NPM1, CBL, or SRSF2 muts in a cohort of patients with untreated MDS, stratified by DNMT3A, TET2, and ASXL1 mut status. Statistical significance was assessed by Fisher's 2-sided exact test. mut freq, mutation frequency; OR, odds ratio.

Figure 5.

Patterns of AML genomic evolution from diag to REL. (A-C) Fish plots detailing the expansion and contraction of specific variants within individual patients from diag to REL. (D) Classification of evolutionary patterns at the time of REL (left) or REF disease (right) across the entire cohort. (E) Bar plot detailing the type of REL patterns observed in patients jointly stratified by DNMT3A, TET2, and ASXL1 mut status.