Introduction

Acute myeloid leukemia (AML) is an aggressive hematopoietic malignancy characterized by excessive proliferation of immature leukemic blasts. Approximately 21,000 patients are diagnosed with AML each year in the United States, and the latest 5-year overall survival rate remains at only 28% (https://seer.cancer.gov)1,2. AML is maintained and propagated by a population of leukemia-initiating cells or leukemia stem cells (LSCs), which feature quiescence and therapy resistance, and contribute to subsequent clonal evolution of the disease and relapse3,4.

AML comprises multiple entities characterized by specific gene mutations and chromosomal abnormalities that drive leukemogenesis and can be used as prognosticators5,6. One of the most frequent chromosomal translocations detected in 5–12% AML patients is chromosome 16 inversion, inv(16)(p13q22) or t(16;16)(p13.1;q22) [henceforth inv(16)], which is associated with the FAB M4Eo AML subtype7,8. At the molecular level, inv(16) disrupts the CBFB gene encoding the CBFβ subunit of the core-binding factor (CBF) transcription factor complex that acts as a master regulator of hematopoietic development and lineage specification9,10. Specifically, inv(16) breaks and joins CBFB and the smooth-muscle myosin heavy chain (MYH11) gene, creating a fusion gene CBFB-MYH11 (CM) which encodes a fusion protein CBFβ-SMMHC11. Although inv(16) AML patients have a relatively favorable prognosis, only approximately 50–60% eventually achieve long-term survival with standard chemotherapy12,13. Therefore, the development of therapies capable of targeting LSCs is necessary to achieve a cure in the vast majority of the patients.

MicroRNAs (miRNAs) are small non-coding RNA molecules that modulate multiple targets by promoting mRNA degradation or repressing translation14,15. MiRNAs are transcribed as primary (pri)-miRNAs, processed into precursor (pre)-miRNAs by DROSHA, exported into the cytoplasm via the RAN (a small GTP-binding RAS-related nuclear protein)-Exportin-5 (XPO5) complex16,17, and processed by Dicer into mature miRNAs18. In hematopoiesis, miRNAs play critical roles in coordinating context-dependent differentiation programs and cell fate decisions19. MiR-126 is highly expressed in endothelial cells and plays a key role in angiogenesis, vascular development and homeostasis20,21. Within the hematopoietic compartment, miR-126 expression is enriched in long-term hematopoietic stem cells (HSCs) and plays a pivotal role in restraining cell cycle progression and maintaining HSC quiescence22. Aberrant miRNA expression profiles have been associated with malignant transformation and prognosis in leukemia and other hematologic malignancies23,24,25,26,27. Notably, miR-126 is aberrantly expressed in CBF leukemias including t(8;21) and inv(16)25,26. We and others have reported that miR-126 contributes to LSC activity, maintenance, and drug resistance in myeloid leukemias28,29,30. Furthermore, we recently reported that activating tyrosine kinase mutations (e.g., BCR-ABL, FLT3-ITD) deregulate intronic miRNA (e.g., miR-126) biogenesis by interfering with RAN-XPO5 mediated pre-miRNA processing via phosphorylation of SPRED1, a member of the Sprouty family of proteins30,31.

Here, we investigated the mechanism and functional impact of miR-126 dysregulation in inv(16) AML. We utilized the conditional Cbfb-MYH11 knock-in mouse model32,33 combined with a miR-126 floxed allele34 to determine the function and regulatory mechanism of miR-126 during inv(16)-induced AML development and evaluated the efficacy of targeting LSCs using miRisten, an anti-miR-126 oligonucleotide therapeutic.

Results

Cbfb-MYH11 (CM) upregulates miR-126 expression in hematopoietic stem and progenitor cells (HSPCs)

Previous studies have shown that miR-126 is highly expressed and contributes to the relative quiescence of normal CD34+ HSCs22,28,29,30. Herein we show that CD34+ cells from inv(16) AML patients have even higher miR-126 levels than CD34+ cells from normal healthy donors (Fig. 1a). MiR-126 is located within intron 7 of the EGFL7 gene35, which was highly co-expressed with miR-126 (Fig. 1b).

Fig. 1: Cbfb-MYH11 (CM) upregulates miR-126 expression in HSPC.
figure 1

a Relative levels of miR-126 in CD34+ and CD34− cells from healthy (HL; black circle; n = 6; HL CD34+ vs. HL CD34− p = 0.0036) donors and inv(16) AML patients (red square; n = 10; inv(16) CD34+ vs. HL CD34+ p = 0.024; inv(16) CD34− vs. HL CD34− p = 0.0053; inv(16) CD34+ vs. inv(16) CD34− p = 0.0002). b Relative levels of EGFL7 in CD34+ and CD34− cells from HL donors (black circle; n = 6; HL CD34+ vs. HL CD34− p = 0.016) and inv(16) AML patients (red square; n = 10; inv(16) CD34+ vs. HL CD34+ p = 0.006; inv(16) CD34+ vs. inv(16) CD34− p < 0.0001). c Counts per million (CPM) reads for miR-126-3p in PB based on miRNA-seq of control (Ctrl; black line; n = 7) and CM (red line; n = 6) mice over time until moribund with leukemia (two-way ANOVA analysis showed CM vs. Ctrl p < 0.0001). d Normalized reads per kilobase per million of transcript (RPKM) for Egfl7 in PB based on RNA-seq of Ctrl (black line; n = 7) and CM (red line; n = 6) mice over time (two-way ANOVA analysis showed Ctrl vs. CM p = 0.0003). e Relative levels of miR-126 in CM (red square) preleukemic (6 weeks after induction; left; n = 6) or leukemic (right; n = 6) vs. Ctrl (black circle; n = 6) HSPC populations, including LSK (preleukemic p = 0.0074; leukemic p = 0.049), GMP (preleukemic p = 0.0064; leukemic p = 0.043), Pre-Meg/E (preleukemic p = 0.00015; leukemic p = 0.0012), Pre-GM (preleukemic p < 0.0001; leukemic p = 0.0002), and EP (leukemic p < 0.0001). f Western blot for CBFβ and CM using anti-CBFβ in 32D cells transduced with CBFB vs. CM (left). Relative levels of primary pri-miR-126, precursor pre-miR-126, mature miR-126, and Egfl7 in 32D cells expressing CM (red dots/bars) vs. CBFβ (black dots, gray bars). g Western blot analysis for CM in 32D-CM non-silencing (NS) control vs. CM shRNA (A3, D4) cells (left). Relative levels of mature miR-126, pri-miR-126, pre-miR-126 and Egfl7 expression in NS (red), CM shRNA-A3 (blue) or CM shRNA-D4 (gray) cells (right). h Relative Gata2 expression in BM samples of control (black/gray; n = 5) vs. CM leukemia (red; n = 4; p = 0.0002) mice. i Relative level of Gata2 in leukemia BM cells transduced with NS control (red), Gata2 shRNA #2 (dark blue), Gata2 shRNA #4 (light blue). j Relative expression of miR-126 in leukemia BM cells transduced with NS control (red), Gata2 shRNA #2 (dark blue), Gata2 shRNA #4 (light blue). k EGFL7 promoter-firefly luciferase reporter activity normalized with renilla luciferase as internal control in 293T cells co-transfected with or without GATA2, RUNX1, CBFβ or CM expression vectors as indicated (gray: none; green: GATA2; light blue: RUNX1 + CBFβ; dark blue: RUNX1 + CBFβ + GATA2; orange: RUNX1 + CM; Red: RUNX1 + CM + GATA2). Each dot represents results from an individual sample and data are presented as the mean ± SEM in a, b, e, h; each line represents the trajectory of an individual mouse in c, d. Representative data of at least three independent experiments are shown in f, g, i, j and data are presented as the mean ± SD. Q-PCR data in a, b, e, f for pri-miR-126, pre-miR-126, Egfl7, and Gata2 were normalized to Beta-2-microglobulin (B2M) and miR-126 data were normalized to RNU44 (human) or snoRNA234 (mouse). The statistical significance for all comparisons shown was determined using two-tailed Student’s T tests (*p < 0.05; **p < 0.01; ***p < 0.001).

Full size image

We previously generated a conditional CM knock-in (Cbfb56M/+/Mx1-Cre) mouse model32,33, which recapitulates human inv(16) AML development in an average of 3–4 months after induction of CM expression by poly(I:C) treatment. In a time-sequential analysis of miR-126 levels in mononuclear cells from peripheral blood (PB) of CM knock-in mice, we observed a progressive increase of both miR-126 and Egfl7 levels over time (Fig. 1c, d) and in multiple CM HSPC populations33,36,37 (Fig. 1e). Significant upregulation of miR-126 was detected as early as 1 month after CM induction (Supplementary Fig. 1a).

CM upregulates Egfl7/miR-126 transcription in concert with Gata2

To determine whether CM directly dysregulates miR-126 expression, we forced CM or wild-type CBFβ expression in 32D myeloblast cells via MSCV-IRES-GFP (MIG) retroviral transduction. We observed upregulated miR-126, pri-miR-126 and pre-miR-126 as well as Egfl7 in CM-transduced GFP+ cells compared to CBFB-transduced GFP+ cells (Fig. 1f). Conversely, shRNA-mediated CM knockdown resulted in significant downregulation of miR-126, pri-miR126, pre-miR-126 and Egfl7 in 32D-CM cells (Fig. 1g). We and others have shown that CM induces upregulation of the Gata2 transcription factor33,38, which reportedly transactivates the EGFL7/miR-126 promoter39,40. Accordingly, we observed increased Gata2 levels upon CM transduction in 32D cells (Supplementary Fig. 1b), as well as in CM knock-in PB (Supplementary Fig. 1c) along with miR-126/Egfl7 levels (Fig. 1c, d), and AML cells from CM knock-in mice (Fig. 1h). Conversely, Gata2 knockdown in CM knock-in AML cells resulted in downregulation of miR-126 (Fig. 1I, j) and Egfl7 (Supplementary Fig. 1d). Analysis of the promoter regions of both murine and human EGFL7 identified several putative CBF binding sites (TRANSFAC database41) nearby the GATA-binding sites (Supplementary Fig. 1e). Chromatin immunoprecipitation (ChIP)-qPCR showed that both CBFβ and CM protein indeed could occupy these CBF sites within the Egfl7 promoter region and that some of the nearby GATA sites were also occupied by GATA2 (Supplementary Fig. 1e). In addition, we confirmed RUNX1, CBFβ/CM occupancy on the reported RUNX1 site42,43 as well as GATA2 binding to the reported GATA2 site42,43 in the human EGFL7 promoter by ChIP-qPCR using primary inv(16) AML CD34+ cells (Supplementary Fig. 1f). We then performed luciferase reporter assays to evaluate transactivation of the human EGFL7 promoter44 by GATA2 and the CBF complex, which is consisted of RUNX1/CBFβ or RUNX1/CM. A significant increase of EGFL7-luciferase transactivation was observed with forced expression of GATA2, RUNX1/CBFβ and RUNX1/CM (Fig. 1k). Co-transfection of RUNX1/CBFβ or RUNX1/CM with GATA2 further increased EGFL7 promoter transactivation, suggesting a co-regulatory activity of GATA2 and RUNX1/CBF on EGFL7/miR-126 transcription.

CM increases biogenesis of mature miR-126 through the HDAC8/RAN–XPO5–RCC1 axis

Next, we tested whether additional mechanisms other than transcriptional regulation of EGFL7/miR-126 were active in inducing high levels of mature miR-126 in CM cells. To this end, we previously reported and confirmed that CM interacts and enhances the activity of a class I histone deacetylase (HDAC), HDAC845 (Supplementary Fig. 2a). Notably, HDAC8 overexpression (OE) in 32D cells significantly increased mature miR-126 levels while reducing the pre-miR-126 precursor levels compared to vector alone (Supplementary Fig. 2b). Conversely, Hdac8 knockdown or treatment with an HDAC8-selective inhibitor (HDAC8i; 22d) significantly downregulated the levels of mature miR-126 and increased those of pre-miR-126 (Fig. 2a, b). These results led us to postulate a possible regulatory role of CM in enhancing the miR-126 biogenesis via HDAC8.

Fig. 2: CM upregulates miR-126 biogenesis through HDAC8/RAN-XPO5 axis.
figure 2

a Western blot analysis of HDAC8 in 32D-CM cells with NS or Hdac8 shRNA (#1, #2) (left). Relative expression of pre-miR-126 and mature miR-126 in 32D-CM cells with NS (red) vs. Hdac8 shRNA#1 (light blue), Hdac8 shRNA#2 (dark blue) (right). b Relative expression in 32D-CM cells treated with vehicle (red) vs. HDAC8i (22d 2.5 μM or 5 μM; purple) for pre-miR-126 (left; 2.5 μM p < 0.0001; 5 μM p < 0.0001) and miR-126 (right; 2.5 μM p = 0.012; 5 μM p = 0.0055). c Representative image of IF co-staining of RAN (green) and HDAC8 (red) in 32D cells (left; scale bar 10 μm). IP with anti-IgG control or anti-HDAC8 followed by immunoblotting (IB) with anti-RAN antibodies (right). Representative results of two independent experiments with similar results are shown. d IP with anti-RAN and IB with antibodies for Ac-Lys, RCC1, XPO5, or RAN in 32D-CM cells not-treated (NT) or treated with DMSO vehicle or 22d (2 µM) (left). The levels of RAN-GTP determined by RAN activation assay (right). The densitometry of RAN-GTP levels measured from three assays are shown on the bottom. e IP with anti-RAN and IB with antibodies for Ac-Lys, RCC1, XPO5, or RAN (left), and RAN activation assay (right) in 32D-CM cells not-treated (NT) or transduced with shCtrl control or shCM for 24 h. f IP with anti-RAN and IB with antibodies for Ac-Lys, RCC1, XPO5, or RAN (left), and RAN activation assay (right) of BM cells isolated from Ctrl or CM mice. g Schematic model of the mechanism by which CM regulates miR-126/EGFL7 transcription as well as miR-126 biogenesis through enhancing HDAC8 activity which in turn promotes RAN-XPO5 mediated transportation. Pre-miR-126 was normalized to B2m and mature miR-126 was normalized to snoRNA234. Data are presented as the mean ± SD and statistical significance shown was determined using two-tailed Student’s T tests (*p < 0.05; **p < 0.01; ***p < 0.001).

Full size image

First, we tested the impact of CM on HDAC8 protein stability. We found that CM expression enhanced the half-life of HDAC8 protein following cycloheximide treatment (Supplementary Fig. 2c). HDAC8 protein is subjected to MDM2-mediated ubiquitination (Supplementary Fig. 2d), which was reduced by CM expression (Supplementary Fig. 2e, left) and increased by CM knockdown (Supplementary Fig. 2e, right). ErbB3-binding protein 1 (EBP1) reportedly interacts with another HDAC family member HDAC2Small RNA library preparation and next-generation sequencing (NGS)

All libraries were prepared using the Illumina TruSeq Small RNA protocol with minor modification following the manufacturer’s instructions. Briefly, 280 ng of pooled total RNA was ligated to the sRNA 3′ adaptor (TCT GGA ATT CTC GGG TGC CAA GGA ACT CC) with T4 RNA Ligase 2, truncated (New England BioLabs) for 1 h at 22 °C, and subsequently ligated to a 5′ adaptor (0.5 μL of 5 μM per reaction): GUUCAGAGUUCUACAGUCCGACGAUCNNN with T4 RNA ligase1 (New England BioLabs) for 1 h at 20 °C. The constructed smRNA library was first reverse-transcribed using GX1 (5′-GGAGTTCCTTGGCACCCGAGA) as the RT primer and then subjected to PCR amplification for 13 cycles, using primers GX1 (5′-CAAGCAGAAGACGGCATACGAGAT[NNNNNN]GTGACTGGAGTTCCTTGGCACCCGAGAATTCCA) and GX2 (5′-AATGATACGGCGACCACCGAGATCTACACTCTTTCCCTACACGACGCT CTTCCGATCT) then followed by 6% TBE PAGE gel purification with size selection (for targeted small RNAs of 17–35 nt). The purified library was quantified using qPCR with a forward primer (5′-CAAGCAGAAGACGGCATACG) and a reverse primer (5′-AATGATACGGCGACCACCGA). Individual libraries were prepared using a unique index primer in order to allow for pooling of multiple samples prior to sequencing. The library was quantified using qPCR. Sequencing of 50 cycles was performed on a HiSeq 2500 (Illumina Inc., San Diego, CA), and image processing and base calling were conducted using Illumina’s pipeline.

miRNA-seq data analysis

Raw small RNA sequences were trimmed to remove the 3′-adapter (TCTGGAATTCTCGGGTGCCAAGGAACTCC) using cutadapt v0.9.3. Reads longer than 16 bp after trimming were aligned to mouse genome assembly mm9 using Bowtie v 0.12.7 with default settings. The expression level of mouse mature miRNAs from Sanger mirBase v18 were counted as previously described68. The counts of miRNAs were then normalized by the trimmed mean of M-values (TMM) method and counts per million (CPM) values were calculated by Bioconductor package “edgeR” v3.4.2.

RNA-seq and bioinformatics

Total RNA was extracted using the AllPrep DNA/RNA/miRNA Universal Kit (Qiagen, 80224); quality and quantity were estimated using BioAnalyser Systems (Agilent Technologies). Samples with a RIN >8.0 were included. External RNA Controls Consortium (ERCC) Spike-In Control Mix (Thermo Fisher Scientific, 4456740) was added to all samples per the manufacturer’s recommendations, although these were not used for downstream analyses. Sequencing libraries were constructed using the KAPA RNA HyperPrep Kit with RiboErase (HMR) (Roche, KK8560), loaded on to a cBot system for cluster generation, and sequenced on a Hiseq 2500 System (Illumina) with single end 51-bp for mRNA-seq to a nominal depth of 50 million reads. Raw RNA-seq sequences were subjected to adapter trimming using Trimmomatic v0.38 and poly(A) tails were removed using FASTP v0.19.4. The trimmed reads were aligned to mouse genome mm10 using Tophat v2.0.8 with default settings. Expression level of RefSeq gene (downloaded on 02/13/2020) were counted using HTSeq count v0.6.1. The raw count data were normalized by the TMM method using Bioconductor package “edgeR” v3.20.9. Differential expression analysis was carried out using the quasi-likelihood (QL) F-test implemented in “edgeR” to determine the differentially expressed genes, with the cutoff of average RPKM in one group ≥1, p values ≤0.01 and fold change ≥1.5. The gene set enrichment analysis (GSEA69) was performed by ClusterProfiler v3.10.1 to identify the affected GO, hallmark and KEGG pathways from MSigDB v770, using the pre-ranked gene list sorted by the −log10(p value) with a sign determined by the fold change direction.

Statistics and reproducibility

Comparison between groups was performed by a two-tailed, paired or unpaired Student’s T-test, one-way or two-way ANOVA for normal distributions. The log-rank test was used to assess significant differences between survival curves. All statistical analyses were performed using Prism version 8.0 software (GraphPad Software). Sample sizes chosen are indicated in the individual figure legends and were not based on formal power calculations to detect prespecified effect sizes. A p value less than 0.05 was considered statistically significant. All experiments are repeated independently with similar results at least two times.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.