Cumulative disruption of regulatory genes precedes whole-chromosome gains in acute myeloid leukemia

Whole-chromosome gains are common in myeloid malignancies, yet the genomic states that predispose cells to their emergence remain unclear. Genes whose perturbation induces phenotypically obvious numerical chromosomal instability in functional screens are rarely mutated in cancer patients [ 1 , 2 ], suggesting that subtler indirect drivers may be at play. To define somatic events associated with whole-chromosome gains ( trisomies ), we focused on myeloid malignancies where they occur as sole cytogenetic events in 8–9% of patients [ 3 , 4 ], permitting analysis of numerical chromosomal instability in relative isolation [ 5 ]. Towards this end, we defined mutations enriched in trisomy-positive myeloid malignancies by integrating somatic data from three independent cohorts comprising 1650 cytogenetically normal (CN) and 229 trisomy (TRI) patients. In each cohort, patient inclusion and classification were based on clinical karyotyping with trisomy defined by the presence of one (or at most two) whole-chromosome gains. To minimize confounding from other forms of chromosomal instability, exclusion criteria removed patients with monosomies or evidence of structural chromosomal instability. For in-house acute myeloid leukemia (AML) cohorts [ 6 , 7 ], genomic data from 304 donors profiled using the same 544 pan-cancer gene panel were reanalyzed (Fig. 1A, B , S1 ). Thirteen genes were more frequently mutated in trisomy ( ASXL1 , SRSF2 , MACF1 , NRAS , NAV3 , MED12 , TPR , ETV6 , MAP3K13 , NFE2L3 , WRN , NOTCH3 , SOS1 ; Fisher’s exact test p < 0.05, Table S1 ), with ASXL1 , SRSF2 and MACF1 remaining significant after multiple testing correction (FDR < 0.2). For robustness, we performed meta-analysis combining these results with analogous analysis of AML data from Bottomly et al. [ 8 ], and myelodysplastic syndrome (MDS) data from Bernard et al. [ 9 ] using Fisher’s method. This identified ASXL1 , SRSF2 , STAG2 and EZH2 as significantly enriched in trisomy-positive cases after multiple testing correction (Fig. 1C , Table S2 ). Additional regulatory genes including RUNX1 , ETV6 , U2AF1 , NRAS and IDH2 showed similar trends but did not remain significant. Together, these findings indicate recurrent mutations in regulatory genes are preferentially associated with trisomy in myeloid malignancies. Fig. 1: Regulatory gene mutations enriched in trisomy-positive myeloid malignancies. Full size image A , B Oncoplots summarizing the mutational landscape of cytogenetically normal (CN) and trisomy (TRI) patients in the in-house AML cohort, with genes ordered by mutation frequency across both groups. C Genes recurrently enriched in TRI were identified across three independent cohorts (in-house, Bottomly et al., Bernard et al.) using Fisher’s exact test; p-values were combined by Fisher’s method and adjusted using the Benjamini–Hochberg procedure. Dot size indicates log₁₀(p-value), color reflects enrichment score (ES), and bordered dots indicate p < 0.05. Individual cohort statistics are provided in Table S1 and combined statistics in Table S2 . D Volcano plot of RNA sequencing data comparing ASXL1 mt and ASXL1 wt chronic myelomonocytic leukemia (CMML) samples, showing global bias toward upregulation in ASXL1 mt . Genes annotated to chromosome segregation (GO:0007059) and upregulated in ASXL1 mt are highlighted, with the most significant genes labeled. E Treemap of Gene Ontology (GO) Biological Process terms enriched among ASXL1 mt upregulated genes, highlighting convergence on chromosome segregation. F Segregation gene score derived from genes upregulated in ASXL1 mt and annotated to chromosome segregation (GO:0007059) is elevated in ASXL1 mt SRSF2 mt double-mutant samples. Methodological details and gene list are provided in Supplementary text and Table S4 . Since truncating variants near C-terminus of ASXL1 were the most significant finding in the meta-analysis, we examined their potential contribution to chromosomal segregation–related processes. Previous studies show that ASXL1 interacts with the cohesin complex through its ASXM1/2 domain to regulate sister chromatid separation [ 10 ]. However, most mutations in myeloid malignancies occur downstream of this region, and in knock-in models produce clonal haematopoiesis–like phenotypes without overt numerical chromosomal instability [ 11 ]. To avoid over-interpreting these observations, we examined processes altered in ASXL1 mt patients while remaining agnostic to causality. Reanalysis of multi-omics data from ASXL1 mt versus ASXL1 wt chronic myelomonocytic leukemia (CMML) patients [ 12 ], revealed upregulation of genes involved in chromosome segregation and nuclear division in ASXL1 mt (Fig. 1D, E , Table S3 , S4 ). Epigenetic data showed broader dysregulation involving metabolic processes, chromosome segregation, chromosome organization and RNA processing, suggesting widespread regulatory perturbation rather than a single segregation-specific mechanism (detailed in Supplementary text , Fig. S2 , Tables S5 , S6 ). The prolonged latency before overt malignancy in mouse models, their phenotypic diversity [ 11 ], and early occurrence of ASXL1 mutations in human clonal hematopoiesis together suggest ASXL1 mutations alone may be insufficient to induce overt chromosomal segregation errors, and that additional genetic or epigenetic alterations are likely required. RNA-seq-based variant calling identified frequent mutations in TET2 and SRSF2 (Table S7 ). Segregation gene scores, derived from genes upregulated in ASXL1 mt samples and annotated to chromosome segregation (GO:0007059, gene list included in Table S4 ) were significantly higher in ASXL1 - SRSF2 double-mutant samples (Tukey-adjusted p < 0.05), indicating that these samples disproportionately contribute to the observed differences (Fig. 1F ). However, to our knowledge, combinations of ASXL1 mutations with additional genes have not been reported to induce overt numerical chromosomal instability. These observations therefore suggest that segregation-related transcriptional differences occur within a broader dysregulated state shaped by additional genetic and epigenetic alterations. To resolve temporal ordering of somatic mutations relative to trisomy acquisition, we profiled 51,092 cells across 16 AML patients using the Mission Bio Tapestri platform for targeted single-cell DNA sequencing (scDNA-seq). A custom amplicon panel captured recurrent mutation hotspots in myeloid malignancies and enabled single-cell ploidy inference through broad genomic representation of chromosomes 4, 8, 11, 13, and 21, which are most frequently gained in trisomy AML (Table S8 , Fig. S3 ). Clinical characteristics are summarized in Table S9 . Because scDNA-seq is noise-prone [ 13 ], we benchmarked single-cell data against matched bulk data to validate variant calls ( Supplementary text ). Variant allele frequencies were concordant between assays supporting reliable mutation detection ( R = 0.77, p < 1 × 10 −4 , Fig. S4 , Table S10 ). Single-cell ploidy inference posed an analytical challenge as existing copy-number-variation callers were affected by non-uniform chromosomal representation and variability in amplicon performance, yielding results inconsistent with clinical karyotyping. To address this, we developed a customized approach that uses sequencing coverage from amplicons in non-trisomy-targeted chromosomes to normalize coverage at the single-cell level. This strategy corrects for technical noise while preserving biological signal. Normalized coverage values were incorporated into logistic regression models leveraging cohort-wide karyotype information to train chromosome-specific trisomy predictors (Fig. 2A ). Within each sample, Gower distances were calculated from mutation profiles and trisomy scores, followed by K-medoid clustering to group cells with similar genomic features (Figs. S5 – S20 ). For each patient, inferred clonal relationships were consolidated into phylogenetic trees constructed under the infinite sites model to resolve the temporal hierarchy of genomic events (Fig. 2B ). Validation using the tumor phylogeny inference tool SCITE confirmed consistency of the inferred phylogenetic structures (Fig. S21 ). Fig. 2: scDNA-seq analysis indicates cumulative regulatory gene mutations preceding trisomy in myeloid malignancies. Full size image A Overview of the single-cell trisomy inference framework. For each chromosome θ, samples were stratified by clinical karyotyping into those with or without trisomy-θ. Cells from trisomy-negative samples were used to train chromosome-specific trisomy-score models, which were then applied to classify cells in trisomy-positive samples. B Phylogenetic trees for 16 AML samples, with each node representing a clone. Arrows indicate lineage relationships from parental to descendant clones, illustrating the sequence of mutational events. The number of cells in each clone is shown in adjacent bubbles, with colors denoting trisomy status: blue, absence; orange, emergence; red, presence. The star next to SRSF2 indicates mutations imputed following phylogeny inference. In AML-11, trisomy 8 was reported at diagnosis; stars next to trisomies 13 and 21 indicate additional chromosome gains likely arising during treatment prior to scDNA-seq sampling. C Treemap of Gene Ontology (GO) Biological Process terms enriched in trisomy-preceding genes, highlighting convergence on regulation of transcription by RNA polymerase II. D Gene set enrichment analysis (GSEA) of cellular phenotypes from CRISPR-Cas9 screen data showing shared enrichment of trisomy-preceding genes for increased cell and nucleus size, reduced nuclear DAPI mean, greater cell DAPI mass displacement, and reduced nucleus DAPI-γH2AX correlation during interphase. E Relative likelihood estimates from evolutionary simulations using CINner across combinations of baseline missegregation rates (y-axis) and mutation-associated gain in missegregation rates (x-axis). Likelihoods are calculated relative to the most likely parameter combination, with higher values indicating better agreement with the observed karyotype distribution and mutation timing relative to trisomy across cohorts. Trisomy-positive cells were identified in 14 of 16 patient samples, with twenty-one unique genes mutated prior to chromosomal gains. Among these, ASXL1 , SRSF2 , and RUNX1 were the most frequent recurrent events. Trisomy-preceding genes broadly grouped into transcription regulators ( ETV6 , MED12 , RUNX1 , SF3B1 , SRSF2 , TPR ), epigenetic regulators ( ASXL1 , BCORL1 , CREBBP , DNMT3A , EZH2 , IDH1/2 , TET2 ), and other genes ( ATM , CALR , FAT1 , FLT3 , NRAS , RHBDF2 , SRGAP1 ). Enrichment analysis showed that trisomy-preceding mutations were over-represented in genes regulating RNA polymerase II-mediated transcription and chromatin organization (Fig. 2C , Table S11 ). To characterize cellular phenotypes shared among trisomy-preceding genes, we analysed data from high-throughput CRISPR-Cas9 screen targeting 5,072 genes and quantifying effects across thirty-four microscopy-derived cellular/DNA organization and morphology features [ 2 ]. Ten trisomy-preceding genes were represented in this dataset ( ETV6 , MED12 , RUNX1 , SF3B1 , SRSF2, TPR, CREBBP , EZH2 , FLT3 , NRAS ). Across the phenotypic space examined, these genes were preferentially enriched for greater nuclear and cellular area accompanied by lower mean interphase nuclear DAPI intensity, consistent with enlargement without a commensurate increase in DNA content (Fig. 2D , Table S12 ). Additionally, greater interphase DAPI mass displacement and lower DAPI–γH2AX correlation was observed, indicating poor chromatin condensation with domain-specific DNA damage that is uncoupled from bulk DNA density. These features are consistent with subtle reorganization of nuclear architecture and replication/transcription-associated stress, with phenotypes such as anaphase bridges, lagging chromosomes or micronuclei formation consistent with defective chromosome segregation. Together, these results indicate that trisomy-preceding genes are enriched for a subset of phenotypes among many assayed. These phenotypes are consistent with subtle perturbations of chromosomal segregation rather than overt instability. Inspection of clonal hierarchies revealed that all fourteen trisomy-positive cases harbored at least one mutation in the identified regulators occurring before or within the same clonal branch as chromosome gain, and 10 of 14 (71%) carried more than one such mutation. These observations indicate that combinations of regulatory mutations, rather than individual lesions, are consistent with increased susceptibility to chromosomal segregation errors. Co-mutation enrichment analysis of MDS cohort, which provided sufficient sample size, showed that combinations of RNA and epigenetic regulators were enriched in trisomy cases (Table S13 ). To examine whether accumulation of mutations influences trisomy risk, we used cancer evolutionary simulation framework CINner [ 14 ], with model parameters reflecting genetic heterogeneity in myeloid malignancies (details in Supplementary text ). Although these simulations provide a simplified representation and are constrained by the modest size of the scDNA-seq cohort, the results were most consistent with a model in which accumulation of mutations progressively increases chromosomal missegregation risk. The most likely parameter combinations were 10 142 -fold more likely to generate the observed clonal structures than the null model in which these mutations do not influence missegregation probability (Fig. 2E , S22 – S24 , Table S14 ). Taken together, our study indicates that whole-chromosome gains, in the absence of structural chromosomal instability, are preferentially associated with mutations in regulatory genes. Across myeloid malignancies, these mutations were significantly enriched in trisomy-positive cases, implicating disruption of transcriptional and chromatin organization as a predisposing state. By resolving clonal hierarchies using single-cell DNA sequencing, we inferred event order and showed that these mutations typically arise before chromosome gain. Moreover, most trisomy-positive samples harbored multiple such alterations, indicating that cumulative perturbation rather than any single lesion may bias cells toward chromosome gain. Although our analyses focus on whole-chromosome gains, this framework suggests a broader susceptibility to chromosome missegregation, consistent with the roles of RNA processing and chromatin organization in maintaining transcriptional restraint during cell division [ 15 ]. Perturbation of these processes, rather than overt failure of mitotic machinery, appears sufficient to tip the balance toward numerical instability. Together, these findings support a model in which whole-chromosome gain arises as a cumulative consequence of transcriptional and chromatin regulatory imbalance rather than abrupt mitotic failure.