RUNX1A isoform is overexpressed in acute myeloid leukemia and is associated with FLT3 internal tandem duplications

Abstract

RUNX1A is the shortest and least expressed of the RUNX1 three main isoforms (A, B, C); despite this, the leukemogenic role of its overexpression has been clearly described. Several studies have shown RUNX1A involvement in different blood cancers and pilot observations in acute leukemia have been reported. In this context, we evaluated RUNX1 isoforms expression in a cohort of acute myeloid leukemia (AML) patients, finding overexpression of RUNX1A and RUNX1B, with higher median levels in thrombocytopenic cases. No difference was observed for RUNX1C. RUNX1A overexpression is higher in more immature AML phenotypes. According to the mutational profile, FLT3 internal tandem duplication (ITD) positive cases have the highest RUNX1A levels and the presence of FLT3-ITD was the only molecular variable able to influence RUNX1A expression. RUNX1A overexpression is disease-related, associated with a specific transcriptional profile, and reappears at relapse, with no clear kinetics except in FLT3-ITD cases. Overall, we demonstrate RUNX1A overexpression in AML and its association with the FLT3-ITD molecular subtype. Our data shed light on the dark side of RUNX1 deregulation, paving the way for further investigations.

Introduction

RUNX1 is a transcription factor that plays the role of a “master regulator of hematopoiesis” through the antagonistic action of its three major protein isoforms, named RUNX1A, B, and C [1]. In adult hematopoiesis, RUNX1B/C (453 and 480 amino acids respectively, lacking exon 7a and including exons 7b and 8) are the predominantly expressed isoforms. They are transcribed from two alternative promoters and differ by 27 amino acids at their N-terminus. RUNX1A (250 amino acids, including exon 7a, which has a stop codon) accounts for only a minor fraction of the isoform pool. It is a C-terminally truncated isoform generated by alternative splicing (Fig. 1A) [1]. RUNX1A shows enhanced DNA binding [2], specific cofactors [3] and divergent effects on target gene transcription [4] as compared to RUNX1B/C.

Fig. 1. RUNX1 isoforms molecular quantification in AML.

A Schematic representation of RUNX1 three main transcript isoforms (top) and their respective protein products (bottom). B RUNX1 A, B, and C relative quantification (as the fraction of total RUNX1 transcripts) in HC and AML samples. In the 10 × 10 dot plot the relative abundance of each isoform was reported as “part of whole”; each dot corresponds to 1% of all RUNX1 transcripts. C Comparison of RUNX1 expression between healthy controls and AML cases at diagnosis (single isoforms and their sum are reported). D RUNX1 expression according to patient PLTs number. E Representatives immunoblot image of RUNX1A and beta-actin (top-left). As made explicit by the separating lines, the final image was obtained by splicing different sections of the same image (raw images are reported in Supplementary Fig. 2). HC#4: commercial bone marrow mononuclear cells. Densitometric analysis of RUNX1A bands expressed as relative optical density and normalized to beta-actin (bottom–left). Comparison between HC and AML RUNX1A protein levels (top–right). Simple linear regression between the RUNX1A transcript and protein levels (bottom-right). F RUNX1A expression according to the latest WHO AML classes. The comparisons BCR::ABL1 vs HC and CEBPA vs HC were not performed, given the rarity of cases among those enrolled (n = 2 and n = 2, respectively). G, H RUNX1A transcript levels according to the FAB classes. C-H Boxplots representing the distribution. The boxes extend from the 25th to 75th percentiles. The line in the box represents the median value. The whiskers range from the smallest value to the largest. P1: distal promoter; P2: proximal promoter; RHD: runt-homology domain; TAD: transactivation domain; HC: healthy controls; AML: acute myeloid leukemia; PLT-: thrombocytopenic cases ( <150.000 PLTs/ul); AML-MR: AML, myelodysplasia-related.

RUNX1A enhances self-renewal activity and hematopoietic stem cell (HSC) expansion ex vivo and in vivo [4]; furthermore, it suppresses myeloid differentiation [2]. In contrast, RUNX1B/C overexpression induces HSCs quiescence [5] and promotes differentiation [2] in murine models. For these reasons, RUNX1A overexpression can be leukemogenic, as suggested by pilot observations in acute leukemia [2, 6], and additional studies have shown the pathogenicity of RUNX1A in other blood diseases [3, 7–9].

Given the pivotal role of the RUNX1 gene in acute myeloid leukemia (AML), we evaluated the expression of its isoforms in our series, aiming to elucidate the role of RUNX1A in the AML pathogenesis, its possible associations with different disease subtypes, and its influence during the disease course.

Materials and methods

Patients

A total of 138 consecutive newly diagnosed AML patients, diagnosed and treated at the Hematology Section of the University of Bari from 2014 to 2023, were included in this study. The main patients clinical data are summarized in Supplementary Table 1 and a more detailed patients’ database is reported in Supplementary Table 2. Among them, the time of relapse of 21 cases was included in. Of the patients who underwent molecular disease monitoring (see paragraph 2.6), 11 were in complete remission (CR) and were considered. In total, 170 bone marrow (BM) samples collected and stored at the aforementioned institution were included in the study. The group of healthy controls (HC) consists of 11 samples: ten HC donors from the same institution and a commercial pool of 56 HC samples (Human Bone Marrow Total RNA-Clontech Laboratories, Inc., Mountainview, CA, USA).

RUNX1 isoforms absolute quantification

Mononuclear cells were isolated from BM samples by Ficoll separation, and total RNA was extracted using the RNeasy Mini Kit (Qiagen, Hilden, Germany); the RNA concentration and purity were checked using the Qubit 2.0 fluorometer and NanoDrop UV-Vis spectrometer (Thermo Fisher Scientific, Waltham, MA). One µg of total RNA was reverse transcribed into complementary DNA (cDNA) using the QuantiTect reverse transcription kit (Qiagen).

RUNX1A/B/C isoforms were quantified by three assays according to the droplet digital PCR (ddPCR) Gene Expression EvaGreen Assays Protocol (BioRad, Hercules, CA). For RUNX1A (NM_001122607) and RUNX1C (NM_001754) two distinct pairs of specific primers were used (RUNX1A_F: AACCCTCGTGCCTCCCTGAA and RUNX1A_R: AGCTCTATCCTGGCTGGGGA; RUNX1C_F: GGCTTCAGACAGCATATTTGAG and RUNX1C_R: AACGCCTCGCTCATCTTG). For the RUNX1B transcript isoform (NM_001001890), a primer pair quantifying RUNX1A + B was used (RUNX1A + B_F: CCGTCTGGTAGGAGCTGTTT and RUNX1B_R: AACGCCTCGCTCATCTTG) because no unique exon-usage exists unlike for the other isoforms. Therefore, to perform the absolute quantification of RUNX1B, the RUNX1A quantification was subtracted from the RUNX1A + B quantification. All assays were performed as already reported [10, 11]. The target quantification was calculated as the ratio between each RUNX1 isoform and the number of copies of GUSB (R/G); RUNX1 gene expression (RUNX1all) was calculated as the sum of its three isoforms.

RUNX1A protein quantification

Mononuclear cells isolated from BM samples by Ficoll separation or purchased from StemCell™ Technologies (Cologne, Germany) were mechanically lysed using the homogenizer T10 basic Ultra-Turrax1 (IKA, Staufen, Germany) in RIPA buffer (Thermo Fisher Scientific), supplemented with cOmplete™ Protease Inhibitor Cocktail and PhosSTOP™ phosphatase inhibitor (Roche, Basel, Switzerland) to extract total proteins. The lysates were centrifuged at 13,000 × g at 4 °C for 10 min to remove debris, and total protein levels were quantified using the BioRad Protein Assay. Equal amounts of protein were mixed with Laemmli sample buffer, resolved by SDS-PAGE, and blotted on nitrocellulose membranes using a Trans-Blot Turbo Transfer System (BioRad). Membranes were blocked with the EveryBlot Blocking Buffer (BioRad), incubated with primary antibodies (anti-RUNX1 antibody ab240639, Abcam, MA, USA; anti-beta-actin antibody C4, sc-47778, Santa Cruz Biotechnology, Inc., TX, USA; 1:1000 dilution) overnight at 4 °C, and washed with 1X TBS-Tween 0.1%. After incubation with a horseradish-peroxidase-conjugated secondary antibody (1:3000 dilution, BioRad) for 1 h at room temperature, Clarity Western ECL Substrate (BioRad) was used for protein visualization with a ChemiDoc Imaging System (Bio-Rad). The signal intensity of the protein bands was measured using Image Lab Software (BioRad). The target quantification was calculated as the ratio between RUNX1A and beta-actin (R/B). AML samples were compared with those from HC donors and with commercial human bone marrow mononuclear cells (BMMCs) StemCell™ Technologies (Cologne, Germany).

Targeted next-generation sequencing (NGS)

NGS analysis was performed on genomic DNA extracted from BM samples at diagnosis using an AmpliSeq customized panel (Thermo Fisher Scientific) encompassing the full coding regions or specific exons of 26 target genes involved in the pathogenesis of myeloid malignancies (ANKRD26, ASXL1, CALR, CBL, CEBPA, DDX41, DNMT3A, ETV6, EZH2, FLT3, GATA2, IDH1, IDH2, JAK2, KIT, KRAS, MPL, NPM1, NRAS, RUNX1, SF3B1, SRSF2, TET2, TP53, U2AF1, ZRSR2), as previously reported [12–14]. Torrent Suite Software (Thermo Fisher) was used for quality control, alignment to the human genome (hg19), and variant calling, using the somatic workflow for single samples and the default parameters. Variants were annotated with Ion Reporter Software (Thermo Fisher). Variants located in intronic regions, synonymous or those present with >1% global minor allele frequency in the normal population were filtered out. Selected variants were investigated for a potential pathogenic role using the SIFT and PolyPhen scores and the Catalogue of Somatic Mutations in Cancer database [13].

FLT3 mutational analysis

FLT3 internal tandem duplication (ITD) and D835/I836 variants were investigated on BM samples at disease onset and relapse, as reported [15]. The FLT3 ITD burden was reported as an allelic ratio (AR). Cases carrying an AR < 0.5 were defined as ITD-low, and patients with an AR ≥ 0.5 as ITD-high, according to the 2017 ELN recommendations [16]. ITD cases were also distinguished between single ITD (single duplication) and multiple ITD (≥2 different duplications), as reported [17]. According to the duplication length, patients were divided into ITD short and ITD long, using as cut-off value the median ITD length observed in our cohort (50 bps), as previously suggested [18]. Other tyrosine kinase domain (TKD) variants were identified by NGS analysis (see paragraph 2.4.).

Molecular monitoring of measurable residual disease

For AML cases carrying one of the following molecular markers: NPM1, PML::RARA, RUNX1::RUNX1T1, CBFB::MYH11, molecular monitoring was performed by RQ-PCR using the following kits (Ipsogen): NPM1 mutA MutaQuant Kit, PML-RARA bcr1/bcr3 Kit, RUNX1-RUNX1T1 Kit, CBFB-MYH11 A Kit.

Gene expression profiling

Gene expression profiling was performed by RNA-Seq. Twelve patients were sequenced using the massive sequencing Illumina NovaSeq6000 platform: seven cases with “high” RUNX1A expression [fold change (FC) ≥ 5] and five cases expressing “normal/low” levels (FC ≤ 3). In brief, paired-end reads obtained in FASTQ format were quality-checked using FastQC version 0.12.1. and the analysis results were summarized in a single HTML report using MultiQC [19]. Reads were trimmed using the Trimmomatic tool version 0.32 [20] to remove sequences and adapters and then mapped against the Homo sapiens reference genome (GRCh.38: https://ftp.ensembl.org/pub/release-109/fasta/homo_sapiens/dna/) using the STAR aligner [21].

Reads mapped to each gene were summarized based on the genome annotation (version 109) from the ESEMBL database (https://ftp.ensembl.org/pub/release-109/gtf/homo_sapiens/Homo_sapiens.GRCh38.109.gtf.gz) using the FeatureCounts algorithm [22]. Principal Component Analysis (PCA) was performed using the DESeq2 R package by means of regularized log transformation (rld) of read counts to illustrate the overall distance between “high” and “normal/low” RUNX1A expression groups.

RUNX1A reads for each sample were calculated using the TMM (Trimmed Mean of M-values), while DESeq2 was used to detect differentially expressed genes (DEGs) [23]. DEGs were selected by filtering them for Log2FC > 1.5 and < −1.5, and adjusted (adj) p value < 0.05. Data were visualized on a volcano plot.

Gene names and gene ontology annotations (BP Biological Process, CC Cellular Component, MF Molecular Function) of DEGs were retrieved from the Human Genome Annotation Resource (available via the org.Hs.eg.db R package) in combination with the AnnotationHub and AnnotationDbi R packages and analyzed using the Database for Annotation, Visualization and Integrated Discovery (DAVID) v2024q2 (July 2024) [24, 25]. DEGs were analyzed in QIAGEN Ingenuity Pathways Analysis (IPA) software (QIAGEN, Hilden, Germany) to categorize differentially expressed transcripts and identify significantly regulated functional pathways. The list of DEGs was used as a query in IPA. The ‘core analysis’ function was run to identify canonical pathways [26].

Methylation analysis

To investigate the possible role of epigenetic regulation in RUNX1 expression, genome-wide methylation data from the TF-1 cell line stably expressing DNMT3A isoform 1 were used to evaluate the methylation status of CpG sites mapped to the RUNX1 gene according to the DNMT3A mutation status. Specifically, methylation data from the TF-1 cell line stably expressing DNMT3A wild-type (WT) or AML-associated mutant hotspots at Arg822 (R882) [specifically R882C or R882H, which induce macro-oligomer formation, alone or in combination with R676K, a macro-oligomerization-reducing mutation (GEO Datasets ID: GSE226062)] were analyzed [27]. Data analysis of the IDAT raw methylation files from the Illumina Methylation EPIC BeadChip microarrays was performed using a dedicated R Bioconductor workflow that incorporated several packages for the analysis of methylation array data (https://www.bioconductor.org/packages/devel/workflows/vignettes/methylationArrayAnalysis/inst/doc/methylationArrayAnalysis.html). The IlluminaHumanMethylationEPICanno.ilm10b4.hg19 R package was used for the annotation of Illumina EPIC methylation arrays. Unsupervised hierarchical clustering of CpG sites mapping to the RUNX1 gene was performed to assess their correlation according to the methylation status in the aforementioned cell lines.

Statistical analysis

All statistical analyses were carried out using GraphPad Prism version 8.3.0 (GraphPad Software Inc., San Diego, CA, USA). Continuous variables were presented as median (minimum-maximum). The Shapiro-Wilk test was performed to check for the normal distribution of continuous variables. The differences in the distribution of continuous variables between categories were compared using either paired or unpiaired t test (for variables passing the normality test) or Mann–Whitney and Wilcoxon matched pairs tests (for variables not passing the normality test). One-way ANOVA and Kruskal-Wallis tests were used to compare the distributions of more than two groups of continuous variables. Multiple linear regression analysis was performed to test the influence of gene mutations (independent variables) on RUNX1A levels (dependent variable). Spearman’s rank-order correlation test was conducted to determine the relationship between the RUNX1A transcript and protein levels. Simple linear regression analysis was performed to predict RUNX1A protein levels from transcript levels. Univariate and multivariate Cox regression analyses were conducted to investigate the impact of RUNX1 isoform levels on patients’ overall survival (OS). Spearman’s rank-order correlation test was also performed to study the relationship between ddPCR and RNA-Seq RUNX1A quantifications. A p value < 0.05 was considered statistically significant.

Results

RUNX1A and RUNX1B isoforms are overexpressed in newly diagnosed AML, with higher levels in thrombocytopenic cases

As observed in the normal hematopoietic system [5], among AML patients, considering median values, RUNX1B/C were confirmed as the dominantly expressed isoforms: 73% and 25% of the total RUNX1 transcripts, respectively, whereas RUNX1A accounted for 2% of the isoforms pool (Fig. 1B). When we examined the absolute quantification of each isoform, comparing AML cases (n = 138) with HC (n = 11), we confirmed RUNX1 overexpression in AML (1.209 vs 0.747 R/G; p = 0.028), as already known [28]. Moreover, we observed overexpression of RUNX1A (0.027 vs 0.003 R/G; p < 0.0001) and RUNX1B (0.827 vs 0.433 R/G; p = 0.006), whereas no difference was observed for RUNX1C (Fig. 1C).

Interestingly, when comparing RUNX1 isoform expression with the patients’ main clinical and biological data (Supplementary Table 1), thrombocytopenic cases (n = 93, <150,000 PLTs/ul) exhibited higher levels of the RUNX1A and RUNX1B isoforms compared to AMLs with normal PLTs levels (n = 23) (0.030 vs 0.014 R/G; p = 0.002 and 0.876 vs 0.517 R/G; p = 0.007 respectively), confirming the previously established involvement of the RUNX1 gene in megakaryocytopoiesis (Fig. 1D) [29]. No other associations emerged from the comparisons between RUNX1 isoforms levels and variables reported in Supplementary Table 1.

Given the known leukemogenic role of RUNX1A and its more pronounced overexpression compared to the other RUNX1 isoforms (~10-fold increase between AML and HC median levels—Fig. 1C), we decided to focus on RUNX1A expression and its possible association with AML pathogenesis.

Firstly, we verified the RUNX1A overexpression at the protein level by immunoblot analysis (Fig. 1E). Increased RUNX1A protein levels were observed in AML patients compared to HC (2.018 vs 0.082 R/B, p = 0.0005), and a positive correlation was found between the RUNX1A transcript and protein levels (r_s=0.805, p = 0.025) (Fig. 1E). Using simple linear regression analysis to predict protein levels from transcript levels, a significant regression equation was identified [F(1,6) = 53.55, p = 0.0003] with R² = 0.899. Therefore, RUNX1A protein and transcript quantities are correlated and as expected, the former depends on the latter.

RUNX1A overexpression is independent of patient classification but related to the disease phenotype

The increased RUNX1A expression (in AMLs vs HC) was confirmed even when our AML cases were classified according to the latest edition of the WHO classification [30]. Considering the RUNX1A median value for each AML class (PML::RARA, n = 12; RUNX1::RUNX1T1, n = 8; CBFB::MYH11, n = 8; NPM1, n = 22; AML-MR, n = 46), the difference compared to HC remained statistically significant (Fig. 1F), whereas no changes were observed when comparing disease classes (p = 0.371). These observations suggested a possible RUNX1A involvement in AML pathogenesis independent of patient classification. The comparisons BCR::ABL1 vs HC and CEBPA vs HC were not performed, given the rarity of these cases among those enrolled (n = 2 and n = 2, respectively).

Because preliminary data on acute leukemia indicated higher expression levels of RUNX1A in acute lymphoblastic leukemia and in AML-M2 patients [6], we classified our cases according to the older FAB classification [31], eliciting a significant difference for all morphological subtypes (M0, n = 11; M1, n = 14; M2, n = 30; M3, n = 12; M4, n = 26; M5, n = 9; M6, n = 8; M7, n = 3) compared to HC (Fig. 1G). Interestingly, the comparison between classes highlighted significant differences (Fig. 1H). In particular, the RUNX1A levels were higher in AML patients with a more immature phenotype and characterized by impaired granulocytic differentiation (M0-M3) as compared to cases showing a monocytic morphology (M4-M5), in line with the known RUNX1A ability to suppress myeloid differentiation, enhancing the self-renewal HSCs capacity [2, 4, 32]. In this context, M6 and M7 classes appeared to follow other pathways, which we cannot address given the rarity of cases among those enrolled.

FLT3-ITD mutated patients show the highest median levels of the RUNX1A isoform

Then, we subdivided our patients according to their mutational profile. Overall, FLT3-ITD positive cases (n = 23) presented the highest median RUNX1A levels at the disease onset (0.071 R/G), as shown in Fig. 2A. A strong relationship between the FLT3-ITD mutations and RUNX1A transcript levels was confirmed by multiple linear regression analysis. Specifically, between all gene mutations detected in our cohort (TP53, n = 7; NPM1, n = 22; CEBPA, n = 2; RUNX1, n = 11; FLT3-ITD, n = 23; FLT3-TKD, n = 10; DNMT3A, n = 8; TET2, n = 17; IDH2, n = 8; EZH2, n = 3; ASXL1, n = 16; SRSF2, n = 6; SF3B1, n = 3; U2AF1, n = 5; ETV6, n = 3; CBL, n = 3; NRAS, n = 15; GATA2, n = 3), the presence of FLT3-ITD was the only variable able to influence the RUNX1A levels (p = 0.0005; Supplementary Table 3). On this basis, we explored the possible connection between FLT3 mutational status and RUNX1A expression (Fig. 2B). A significant difference was observed between FLT3-mutated (n = 33) and FLT3-wt AML (n = 82) (0.048 vs 0.020 R/G, p = 0.019), but this difference became more evident when we compared FLT3-ITD (n = 23) with FLT3-wt (n = 82) cases (0.071 vs 0.020 R/G, p < 0.0001). While FLT3-ITD AML exhibited higher RUNX1A levels than FLT3-wt cases, FLT3-TKD patients (n = 10) had lower RUNX1A levels than FLT3-wt ones (0.006 vs 0.020 R/G, p = 0.025). Again, this observation supports the connection between the ITD mutation and RUNX1A expression. Focusing on this aspect, we found that neither the ITD allelic ratio (low vs high) nor the ITD number (single vs multiple) seemed to influence RUNX1A transcript levels. However, RUNX1A overexpression was higher in long ITD cases (n = 11, characterized by more FLT3 auto-phosphorylation and poorer survival outcomes [18]) compared to short ITD ones (n = 8) (0.131 vs 0.037 R/G, p = 0.02) (Fig. 2C). The association between FLT3-ITD and RUNX1A overexpression indirectly influenced the median levels of the isoform in the three risk categories according to the ELN 2022 recommendations [33]. AML patients in the intermediate category (where all FLT3-ITD but not FLT3-TKD are included) showed higher RUNX1A levels than those in the favorable and adverse categories (0.070 vs 0.024 R/G, p = 0.003 and 0.070 vs 0.015 R/G, p = 0.0004 respectively), whereas no differences were observed between the latter two categories (Fig. 2D).

Fig. 2. Distribution of RUNX1A levels in AML.

A RUNX1A expression according to the patients’ mutational profile (from the highest to the lowest RUNX1A level). B RUNX1A transcript level in AML cases subdivided by FLT3 mutational status; C a detailed analysis of FLT3-ITD patients is reported. Short ITD: duplication length ≤ 50 bps, long ITD: duplication length > 50 bps. D The expression of RUNX1A in the AML cases subdivided into the three risk categories according to the ELN 2022 recommendations [33]. E RUNX1A abundance as a proportion of total RUNX1 transcripts in patients categorized by main RUNX1 gene alterations. F The RUNX1A/RUNX1C ratio in AML cases stratified according to the mutational status of key epigenetic modifiers. G Hierarchical clustering of methylation values of CpG sites mapping to RUNX1 based on Illumina methylation annotation in the TF-1 cell line stably expressing wild-type DNMT3A (isoform 1) or AML-associated hotspot mutants (either R882C or R882H) alone or in combination with R676K [27]. CK: R882C/R676K; HK: R882H/R676K. †Probe names.

Although RUNX1A overexpression is strongly associated with FLT3-ITD, isoform levels do not appear to influence patients’ overall survival (OS) (HR: 0.363, 95% CI: 0.031-4.291, p = 0.421) (Supplementary Fig. 3). Finally, the close relationship between FLT3-ITD and RUNX1A expression was observed again at disease relapse (DR), but this aspect will be discussed later.

DNMT3A-mutated AMLs present higher RUNX1A levels than DNMT3A-wt cases

Although there is a clear association between RUNX1A overexpression and the FLT3-ITD mutation, we considered other possible genetic mechanisms influencing RUNX1A levels in AML, namely RUNX1 gene alterations (SNVs and INDELs, RUNX1::RUNX1T1 and trisomy 21), spliceosome mutations (SRSF2, SF3B1, and U2AF1 variants) and epigenetic modifiers mutations (DNMT3A, TET2, IDH2, EZH2, ASXL1).

None of the explored RUNX1 gene alterations appeared to be associated with isoform A overexpression (absolute levels), including trisomy 21 (as somatic event), unlike what has been described in Down Syndrome—associated myeloid leukemia (ML-DS) [3] (Supplementary Fig. 1A). To better characterize the relationship between isoform A and RUNX1 gene alterations, we focused also on the RUNX1A quantification relative to all RUNX1 transcripts. Notably, lower RUNX1A/RUNX1all levels were observed in RUNX1-mutated cases (n = 11) compared to wild-type ones (n = 43)(0.010 vs 0.018, p = 0.015), suggesting a mutually exclusive relationship between gene mutations (loss of function and/or altered function [34]) and the expression of the leukemogenic isoform A (Fig. 2E).

Although RUNX1A is generated by alternative splicing [35, 36], no differences in RUNX1A levels were observed in AML patients carrying spliceosome mutations compared to wild-type cases, neither considering its absolute quantification nor the ratio between the two splicing isoforms RUNX1A and RUNX1B (Supplementary Figure 1B/C).

Last but not least, we investigated the potential role of mutations in epigenetic modifiers in contributing to the disequilibrium of RUNX1 isoforms, focusing on the ratio between RUNX1A and RUNX1C, transcribed from the P2 and P1 promoters, respectively. Interestingly, DNMT3A-mutated AML cases (n = 8) displayed a significantly higher RUNX1A/RUNX1C ratio compared to DNMT3A–wild-type cases (n = 46) (0.304 vs 0.045, p = 0.0029), suggesting a possible differential methylation activity of DNMT3A on the two promoters (Fig. 2F). This difference remained significant when analyzing RUNX1A absolute expression levels (see Supplementary Fig. 1D) so we decided to study the relationship between DNMT3A mutational status and the methylation profile of RUNX1 gene. Recently, TF-1 cell line was engineered to stably expressed DNMT3A (isoform 1), including WT and AML-associated hotspot mutants (either R882C or R882H) alone or in combination with R676K, an additional macro-oligomerization-decreasing mutation: R882C/R676K (DNMT3A_CK) and R882H/R676K (DNMT3A_HK) [27]. Comparing their methylation profile, global hypomethylation of RUNX1 CpG sites was observed in all DNMT3A-mutated TF-1 cells compared to DNMT3A_WT cells (Fig. 2G). Differential methylation analysis between WT and all DNMT3A mutant cell lines revealed statistically significant hypomethylation of almost all CpG sites of RUNX1, regardless of the genomic location of the CpG site (data not shown).

RUNX1A overexpression is disease-related and reappears at relapse, with no clear kinetics, except in FLT3-ITD cases

With the aim of studying RUNX1A kinetic expression during the disease course, we evaluated the levels at the CR stage and at DR.

When patients achieved CR, their RUNX1A levels reverted to normal values (Fig. 3A). In fact, no differences were found between patients at this stage and HC (0.002 vs 0.003, p = 0.298) (Fig. 3B).

Fig. 3. RUNX1A, kinetics and RNA-Seq data analysis.

A RUNX1A kinetics from diagnosis to CR. B RUNX1A expression during the disease course. C Comparison of RUNX1 expression between healthy controls and AML cases at disease relapse (single isoforms and their sum are reported). D–GRUNX1 isoforms kinetics from diagnosis to disease relapse for each patient. H RUNX1A kinetics from diagnosis to disease relapse, for four cases that were FLT3-ITD negative at disease onset but FLT3-ITD mutated at relapse. I Representative examples of RUNX1A kinetics (diagnosis, CR, DR) in three patients. L Principal Component Analysis (PCA) showing the separation between the two groups of samples (high and normal/low RUNX1A expression) and highlighting the homogeneity of samples within each group. M Top ten deregulated canonical pathways identified from IPA analysis. N Volcano plot visualizing differentially expressed protein-coding genes (with p value < 0.05) between the two groups of patients (high and normal/low RUNX1A expression). Upregulated (Log2FC > 1.5) and downregulated (Log2FC < −1.5) genes are marked in red and green, respectively. The top five deregulated protein-coding genes are reported in the table (top–right).

At DR, we again quantified the expression of all three isoforms. As observed at disease onset, when comparing relapsing AML cases (n = 21) with HC we confirmed the overexpression of RUNX1A (0.020 vs 0.003 R/G, p < 0.0001) and RUNX1B (1.167 vs 0.433 R/G, p = 0.0012), whereas no difference was observed for RUNX1C (0.376 vs 0.322 R/G, p = 0.341) (Fig. 3C).

Considering, for each patient, the evolution of RUNX1 isoforms expression between the diagnosis and DR, no clear kinetics emerged from the analyzed cases (Fig. 3D–G). Notably, when we focused on four cases that were FLT3-ITD negative at disease onset but FLT3-ITD mutated at DR, all of these cases exhibited an increased RUNX1A expression compared to the values at diagnosis (p = 0.033), confirming once again the close connection previously demonstrated between FLT3-ITD mutations and RUNX1A overexpression (Fig. 3H).

Overall, no differences emerged when comparing median RUNX1A expression levels between diagnosis and DR (0.027 vs 0.02 R/G, p = 0.728), as also when comparing patients in CR with HC (Fig. 3B). In short, RUNX1A overexpression was disease-related and this observation holds true not only when considering the median level of isoform A, but also its kinetics in the few cases for which diagnostic, CR and DR samples were available. Three representative examples are shown in Fig. 3I.

RUNX1A overexpression is associated with a specific transcriptional profile

To verify whether RUNX1A upregulation is linked to a specific transcriptional profile, high-throughput RNA sequencing was conducted. Interestingly, patients with “high” RUNX1A expression (FC ≥ 5) also produced a greater number of RUNX1A reads (≥18,000 reads) compared to patients with “normal/low” (FC ≤ 3) RUNX1A expression (<12,000 reads), demonstrating a positive correlation of about 90% (r_s = 0.89, p = 0.0002), indicating the reliability of the sequencing approach. PCA showed the separation between the two groups of samples (high and normal/low RUNX1A expression) and highlighted the homogeneity of samples within each group (Fig. 3L).

A total of 3622 protein-coding genes (1491 up and 2131 down) and 3108 elements belonging to other categories (lncRNAs, miRNAs, pseudogenes, etc.) were found to be differentially expressed between the two groups of patients. The complete list of differentially expressed elements is reported in Supplementary Table 4. Among these, 756 protein-coding genes (557 up and 199 down) and 1159 elements belonging to other categories (lncRNAs, miRNAs, pseudogenes, etc.) were differentially expressed, with a Log2FC > 1.5 or < –1.5 and with a p value and a padj value < 0.05 between the two groups of patients.

Statistically significant (padj < 0.05) DEGs were categorized into Gene Ontology (GO) categories using the online software DAVID (https://david.ncifcrf.gov/tools.jsp). “Signal Transduction” was identified as the TOP BP (p = 7.00E-07), “Plasma membrane” the TOP CC (p = 3.02E-23), and “serine-type endopeptidase activity” the TOP MF (p = 2.07E-6) items. The complete list of gene ontology categories (BP, CC, MF) and the categorized genes are reported in Supplementary Table 5.

Then, a “core analysis” was performed using IPA software, to investigate canonical pathways enriched in DEGs. “Neutrophil degranulation” (p = 5.85E-20) was identified as the TOP canonical pathway deregulated, with 56 DEGs; the majority of them being downregulated genes. The activation state of the pathway was identified as “decreased” (z-score = −6682) (Fig. 3M). Interestingly, this observation is in line with the known RUNX1A ability to suppress myeloid differentiation [2, 4, 32] and with the higher RUNX1A levels observed in our AML cases with a more immature phenotype and characterized by impaired granulocytic differentiation (M0-M3). The complete list of canonical pathways resulting from the analysis is reported in Supplementary Table 6.

Notably, when focusing on the TOP five DEGs (upregulated and downregulated) between the two groups (Fig. 3N), the POU4F1 (Log2FC = 6.72, padj = 0.009) and SFRP1 (Log2FC = −5.33, padj=0.005) genes stand out. In fact, POU4F1 overexpression has already been associated with RUNX1::RUNX1T1 AML and is known to contribute directly to its unique transcriptional signature; on the contrary, SFRP1 is a transcriptional repression target of the RUNX1::RUNX1T1 fusion protein in AML [37, 38].

Discussion

The pivotal role of RUNX1 in the control of hematopoiesis has been amply demonstrated [39]. Dysregulated RUNX1 can contribute to blood diseases in many ways, whereby either an excess or a deficiency of RUNX1, or an altered function, can promote leukemogenesis [34]. Among the main RUNX1 alterations, gene mutations (germline and somatic) and RUNX1::RUNX1T1 fusion are the most common events, widely studied in AML pathogenesis [34, 40, 41]. However, additional genetic aberrations (copy number events or further genomic rearrangements) can affect its function [42, 43]. Conversely, less attention has been paid to RUNX1 expression in AML, even if its overexpression has been associated with poorer outcomes in cytogenetically normal AML [28].

To the best of our knowledge, this is the first study to report the absolute quantification of the three main RUNX1 isoforms in an AML series. The observed RUNX1A and RUNX1B overexpression suggests a possible different epigenetic regulation of these two isoforms, transcribed from the P2 promoter (proximal), whereas the expression of RUNX1C (transcribed from the P1 promotor—distal) [44, 45] remains unchanged.

Among the three main RUNX1 isoforms, several studies have shown the leukemogenic role of RUNX1A in different hematological diseases. A study on myelodysplastic/myeloproliferative neoplasms (MDS/MPN) showed the involvement of RUNX1A in disease progression [7]. A recent study on ML-DS demonstrated the key role of a RUNX1 isoform disequilibrium favoring RUNX1A in the pathogenesis of this rare disease [3]. Even though based on a limited number of cases, RUNX1A overexpression was also observed in two previous studies on acute leukemia [2, 6].

In our study, we investigated RUNX1A levels in AML and linked its overexpression with key clinical and biological parameters and the disease course. Interestingly, our observations align with the already-known role of RUNX1 in hematopoiesis. Specifically, the higher levels of RUNX1A observed in thrombocytopenic cases confirm the widely demonstrated involvement of the RUNX1 gene in megakaryocytopoiesis [29]. Furthermore, RUNX1A overexpression was recently described in the familial platelet disorder with associated myeloid malignancy (FPDMM), suggesting a new potential role for RUNX1 isoforms disequilibrium in the development of myeloid malignancy in FPD [8, 9].

The ability of RUNX1A to suppress myeloid differentiation while enhancing the HSC self-renewal capacity [2, 4, 32] aligns with the higher RUNX1A overexpression observed in the more undifferentiated cases in our cohort. In murine models, enforced expression of RUNX1A expanded the immature hematopoietic cell population, conferring the potential to differentiate into multiple lineages ex vivo [4]. Other data showed that overexpressed RUNX1A inhibits myeloid cell differentiation and stimulates cell proliferation upon granulocyte colony-stimulating factor (G-CSF) treatment [2].

However, the highest RUNX1A levels were observed in FLT3-ITD AML patients. The RUNX1 cooperation with FLT3-ITD in leukemia induction is already known [46, 47]. FLT3-ITD AML patients express high levels of RUNX1; FLT3-ITD directly impacts RUNX1 activity, whereby upregulated and phosphorylated RUNX1 cooperates with FLT3-ITD to induce AML [46, 47]. Notably, no studies have specifically focused on the expression of the three RUNX1 isoforms. However, Cauchy et al. demonstrated that the RUNX1 upregulation in FLT3-ITD positive AML may partially result from the presence of an ITD-specific open region of chromatin (a DNase I hypersensitive site - DHS) within the RUNX1 gene, ~10 kb upstream of exon 6, whose alternative splicing generates the RUNX1A isoform (Fig. 4) [46]. Although we cannot verify this direct connection, it is conceivable that in FLT3-ITD AML, this open region at the 5’ end of exon 6 may facilitate regulator access, promoting alternative splicing in favor of RUNX1A production, as recently described [35]. Moreover, RUNX1A overexpression could also result partially from epigenetic mechanisms, as suggested by the higher levels detected in our DNMT3A-mutated cases. In fact, as we observed in TF-1 cells, DNMT3A dysfunction impacts the RUNX1 methylation profile.

Fig. 4.

Illustration of a possible molecular connection between FLT3-ITD and RUNX1A overexpression, based on recent literature data [35, 46]. CR complete remission, HC healthy controls, AML acute myeloid leukemia, DHS DNase I hypersensitive site.

By studying RUNX1A kinetics during follow-up, we highlighted its close association with the disease course. RUNX1A overexpression is disease-related and drives a specific transcriptional profile which, in some respects, overlaps the RUNX1::RUNX1T1 gene expression signature. Notably, RUNX1A overexpression is closely linked to POU4F1 upregulation and SFRP1 downregulation, two events already observed in RUNX1::RUNX1T1 AML [37, 38].

The main strength of this study is the strong association observed between RUNX1A deregulation and significant AML clinical and biological parameters, particularly the FLT3-ITD mutation. However, the hypotheses presented were not verified through functional models (the main limitation of our work); further studies are required to validate these findings.

Conclusions

The AML biological heterogeneity underscores the need for deeper insights into the molecular mechanisms underlying disease development. Our data aims to contribute a new piece to the jigsaw of the AML molecular pathogenesis, in which RUNX1 involvement is established. Our effort to shed light on the dark side of RUNX1 dysregulation may also offer the opportunity to look on RUNX1A as a new AML therapeutic target. In fact, restoring the RUNX1 isoform equilibrium could reverse the oncogenic potential of isoform A, as recently shown in ML-DS. In FLT3-ITD AML cases, for which targeted treatments are already available, clarifying unknown aspects of the FLT3 pathway could identify new druggable molecules whose synergistic action with FLT3 inhibitors warrants investigation. In the meantime, another piece has been added to the complex scenario of AML pathogenesis.

Author contributions

Conception and design of the study: CC and FA. Acquisition of data and/or analysis and interpretation of data: CC, FT, EP, LA, AZ, NC, GT, IR, MRC, AM, CFM, GS, PM, and FA. Clinical data acquisition: FT, VPG, and MD. Protein quantification: GB, AN, and FG. RNA-sequencing: MFC, FM, CT, SNC, BB and AT. Methylation analysis: PO and MG. Drafting of the manuscript: FA. All authors revised the manuscript for important intellectual content and approved the final version submitted for publication.

Funding

Open access funding provided by Università degli Studi di Roma Tor Vergata within the CRUI-CARE Agreement.

Data availability

The data generated in this study are available upon request from the corresponding author.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

The local Ethics Committee of “Azienda Ospedaliero Universitaria Policlinico di Bari” approved the study. Informed consent was obtained from all patients before study inclusion, in accordance with the Declaration of Helsinki. Patients’ records/information were anonymized and de-identified before analysis.

Supplementary information

The online version contains supplementary material available at 10.1038/s41417-025-00939-z.