Introduction
Acute myeloid leukemia (AML) is a relatively rare malignancy among childhood malignancies, accounting for less than one-fifth of acute leukemia cases diagnosed in childhood. However, it remains a challenging disease with a poor outcome compared with that of pediatric acute lymphoblastic leukemia (ALL) [
1]. Despite enormous progress in AML diagnosis and treatment during the past decades, approximately 40% of children with AML experience short-term relapse [
2,
3]. In AML, some risk factors, such as age, cytogenetic characteristics, the white cell count (WBC), and minimal residual disease (MRD), are currently used for prognosis [
4‐
7]. However, these assessed covariates are of limited predictive value and have been commonly reported in adults. Few have been assessed in pediatric populations of patients [
8,
9]. Therefore, not only accurate and timely diagnosis but also valuable prognostic biomarkers and novel therapeutic algorithms are critical for the clinical management of pediatric AML [
10,
11].
Circular RNAs (circRNAs) are covalently closed nonlinear RNA molecules that are produced from pre-mRNA backsplicing and have been considered faulty splicing products for several decades [
12]. However, it has become clear that circRNAs are functional entities [
13]. Notably, the majority of circRNAs consist of varying numbers of constitutive exons and are extremely stable RNA molecules with cell type- or tissue type-specific expression patterns [
14]. CircRNAs play a potential role as biomarkers in a variety of cancers owing to their biological characteristics as well as their involvement in tumorigenesis through diverse mechanisms, such as microRNA (miRNA) sponging, RNP binding, or acting as templates for translation [
15‐
18].
Emerging evidence has shown that the roles of a few circRNAs are well established in AML [
19]. For instance, circ-ANXA2 is upregulated in AML, and its knockdown (KD) suppresses the proliferation, enhances the apoptosis and increases the chemosensitivity of THP-1 and KG-1 cells by acting as an effective microRNA sponge of miR-23a-5p and miR-503-3p [
20]. CircPAN3 was validated to mediate doxorubicin resistance in AML cells by targeting the miR-153-5p/miR-183-5p/XIAP axis and enhancing autophagy activity [
21]. More recently, Sun et al. reported that circMYBL2 promotes the proliferation of FLT3-ITD–positive AML cells by directly interacting with the PTBP1 protein
in vitro and
in vivo [
22]. It is worth noting, however, that much less is known about the regulatory influence of circRNAs in pediatric AML.
Here, we sought to investigate the regulatory roles played by circRNAs in pediatric AML. Our screening strategy identified that circRNF220, a circRNA derived from the RING domain E3 ubiquitin ligase gene RNF220, was specifically upregulated in pediatric AML and was required for AML progression. CircRNF220 KD significantly impeded the activities of primary pediatric AML cells. Furthermore, we revealed that circRNF220 performed its regulatory functions as a competing endogenous RNA by binding miR-30a and then regulating the expression of downstream genes Myb like, SWIRM and MPN domains 1(MYSM1) and immediate early response 2(IER2). Importantly, we showed that circRNF220 not only outperformed current markers in terms of its high sensitivity, specificity and diagnostic simplicity but also demonstrated prognostic value for early recurrence by evaluation in bone marrow (BM) and even peripheral blood (PB) samples.
Methods
Patients and samples
Diagnostic BM or PB samples were collected from 149 pediatric patients (1 to 14 years of age) with de novo AML of French-American-British (FAB) classification M1-M7 enrolled in this study between January 2015 and May 2020 at the Guangzhou Women and Children’s Medical Center of Guangzhou Medical University who had available pretreatment and post chemotherapy cell samples. The demographic characteristics of the AML patients are summarized in supplementary Table S
1. All patients received intensive, response-adapted double induction and consolidation therapy, and the details of the treatment regimen have been reported previously [
23]. In addition, BM samples from 86 patients with ALL, 28 patients with idiopathic thrombocytopenic purpura (ITP), and 33 patients with other hematological diseases, including myelodysplastic syndrome (MDS), chronic myeloid leukemia (CML), juvenile myelomonocytic leukemia (JMML), anemia, lymphoma, and thrombocytopenia, were also analyzed. Detailed information on these patients is available in supplementary Table S
2. Informed patient consent for the study and accompanying scientific investigations was obtained from the participants’ parents or guardians. The research was conducted after approval by the Ethics Committee of Guangzhou Women and Children’s Medical Center according to the principles of the Declaration of Helsinki.
CircRNA microarray
A circRNA microarray (Arraystar Human circRNAs chip, ArrayStar) containing more than 5000 probes specific for splice sites in human circRNAs was used in this study. After hybridization, 5 pediatric AML samples (pooled) and 5 healthy donor BM samples (pooled) were examined using the circRNA microarray provided by Kangcheng Bio-Tech Inc. R software was used to process the subsequent data after quantile normalization. Differentially expressed circRNAs were identified through volcano plot filtering and fold change filtering. CircRNAs with a fold change ≥ 2.0 or ≤ 0.5 and a P-value < 0.05 were identified as significantly differentially expressed circRNAs. We further confirmed the microarray results for 10 randomly selected circRNAs by qRT-PCR in samples from 10 newly diagnosed AML patients and healthy controls.
RNA isolation, reverse transcription, and real-time quantitative polymerase chain reaction (qRT-PCR)
Total RNA from whole-cell lysates or the nuclear and cytoplasmic fractions was extracted from BM and PB samples using TRIzol® reagent (Life Technologies). To quantify the amounts of mRNAs and circRNAs, cDNA was synthesized from 500 ng of RNA with PrimeScript RT Master Mix (Takara, Dalian, China). QPCR analyses were performed using SYBR Premix Ex Taq II (Takara) and a LightCycler® 480 Real-Time PCR System (Roche, Basel, Switzerland). Specifically, divergent primers annealing at each side of the backsplice junction were used to determine the abundances of circRNAs. The expression levels were calculated using the 2
−△△Ct method, and GAPDH was used as the internal standard. All assays were performed in triplicate. The sequences of all primers used are available (Supplementary Table S
3).
RNA fluorescence in situ hybridization (RNA-FISH)
RNA FISH was performed on AML cells and 293 T cells using an RNA Fluorescence In Situ Hybridization Kit (Exonbio Lab, Guangzhou, China) in accordance with the manufacturer’s protocol. Probe sequences designed by Exonbio Lab are listed in Supplementary Table S
3. AML cells or 293 T cells at 85–95% confluence were fixed with 4% paraformaldehyde for 12 min. After prehybridization, cells were incubated with probes in hybridization buffer at 37 °C overnight. After hybridization, slides were washed in 2 × SSC with 0.5% Tween 20 two tough times for 15 min each at room temperature. Nuclei were counterstained with DAPI (Invitrogen, Carlsbad, CA, USA) for 10 min. Images were acquired on a Leica TCS SP8 confocal microscope (Leica Microsystems, Mannheim, Germany).
Cell culture and treatments
Human leukemia cell lines (HL-60, THP-1, and K562) and the human embryonic kidney cell line HEK-293 T were purchased from the American Type Culture Collection and maintained in RPMI 1640 medium or DMEM (Invitrogen) supplemented with 10% fetal bovine serum (FBS). Cells were cultured at 37 °C in 5.0% CO2, and all cell lines were routinely tested for mycoplasma contamination. AML BM mononuclear cells were prepared by Ficoll density centrifugation at 600 × g for 30 min, and then were thawed and plated in MEMα (Gibco) supplemented with 10% FBS, 20 ng/ml hTPO, 20 ng/ml hIL-3, 20 ng/ml hG-CSF, and 1 × penicillin/streptomycin.
AML cells were exposed to 2 μg/mL actinomycin D (Sigma-Aldrich, Saint Louis, MO, USA) to block transcription for 4, 8, 12, 24, and 48 h. Then, the cells were harvested, and the stability of the circRNF220 and RNF220 mRNAs was analyzed using qRT-PCR.
All in vitro studies were repeated three times in AML cell lines, or performed in more than three primary AML cases BM cells.
RNase R assay
Total RNA was treated with RNase R. Briefly, 5 μg of total RNA was exposed to 2 U/μg RNase R (Epicenter Biotechnologies) at 37 °C for 30 min, and RNase-free water was used as a control (Mock). Digested RNA was subsequently purified using an RNeasy MinElute Cleanup Kit (Qiagen). The RNA concentrations of the treated samples were determined, and 1 μg of treated RNA was used for qRT-PCR.
Nuclear and cytoplasmic fractionation
The nuclear and cytoplasmic fractions were extracted according to the manufacturer’s protocol using NE-PER Nuclear and Cytoplasmic Extraction Reagents (Thermo Scientific). The effectiveness of nuclear and cytoplasmic separation was assessed by determining the protein levels of histone H3 and GAPDH, which are specifically expressed in the nucleus and cytoplasm, respectively.
Small interfering RNAs (siRNAs), vector construction, and cell electroporation
Small interfering RNAs (siRNAs) against circRNF220 (siRNA-circRNF220) and the negative control RNA duplex (siRNA-NC) were purchased from RiboBio, Guangzhou, China. The miR-30a mimics (miR-30a) or its inhibitor (inhibitor-miR-30a) and scrambled oligonucleotides (miR-NC or inhibitor-NC) were purchased from GenePharma Biotech (Shanghai, China).
The plasmid for circRNA overexpression (OE) was constructed using the pcDNA3.1( +) circRNA Mini Vector, which was produced by the Forevergen Company (Guangzhou, China). The region spanning the second exon of the RNF220 gene that constitutes the circRNA was PCR amplified from cDNA and seamlessly cloned between the splicing signals AG and GT, which are surrounded by the minimal introns that facilitate the generation of circRNF220 (pcDNA3.1-circRNF220). The construct was verified by sequencing.
In addition, based on the expression levels of circRNF220 in AML cells, we selected the AML cell lines with comparatively low circRNF220 expression (HL-60, THP-1 and K562) to perform gain-of-function experiments. Primary BM cells expressing comparatively high levels of circRNF220 were used to perform loss-of-function experiments. Unless otherwise noted, the plasmids and siRNAs were delivered into cultured AML cells by electroporation as we reported before [
24]. The exception was primary AML cells (3 × 10
5 cells/sample), which were transfected with 100 pmol of the siRNAs in T buffer using the Neon® Transfection System (Invitrogen) with settings of 1400 V, 10 ms and 1 pulse.
The transfected cells were collected at different time points after electroporation for biological functional analyses and for RNA extraction.
Lentiviral constructs and lentiviral transduction of patient AML samples
The circRNF220 KD lentivirus (LV-sh-circRNF220) was produced by GeneChem Company (Shanghai, China). The shRNAs only specifically targeted the head-to-head splicing junction of circRNF220. The particular sequences were synthesized and cloned into the AgeI and EcoRI sites in the 5’LTR of MCS-CBh-gcGFP, and the integrity of the insert was confirmed by sequencing.
AML BM cells were transduced with retroviral vectors according to the protocol. Briefly, LV-sh-circRNF220 lentiviral particles were centrifuged for 2 h at 1,200 × g in 24-well plates precoated with 100 μg/ml RetroNectin (Takara, Dalian, China). Three days later, the cells were sorted to separate GFP + populations for subsequent experimental studies.
CCK-8 assay
Cell proliferation was assessed using a Cell Counting Kit-8 (Dojindo Molecular Technologies, China), and samples were evaluated in a microplate reader (Varioskan Flash, Thermo Scientific) as described previously [
24].
Cell cycle analysis
Treated AML cells were stained with PI (KeyGEN, Nanjing, China) containing RNase A for 30 min at 37 °C. Cells were analyzed for DNA content by fluorescence activated cell sorting (FACS) (BD Bioscience, San Jose, CA, USA) with ModFit software used for data analysis, as described previously [
24].
AML cell differentiation
After treatment with 25 ng/mL phorbol-12-myristate 13-acetate (PMA, Sigma-Aldrich) for 2 days, THP-1 cells were reseeded in fresh medium without PMA for 2 more days to allow cell recovery. Then, monocytic differentiation was evaluated based on the cell morphology on Wright-Giemsa-stained slides and by FACS analysis. For FACS analysis, cells were stained with anti-human CD14-PE (eBioscience, San Diego, CA, USA) and anti-human CD11b-APC (Invitrogen) monoclonal antibodies. The data were analyzed using FlowJo software (BD Biosciences).
Apoptosis assay
Treated AML cells were stained with Annexin V/PI (KeyGEN) and subsequently analyzed by FACS. We defined all Annexin V+ cells as apoptotic cells, including the early apoptotic cells and the late apoptotic cells. Additionally, we performed terminal deoxynucleotidyl transferase (TdT)-mediated dUTP nick-end labeling (TUNEL) to detect apoptotic K562 cells using the corresponding kit (Beyotime).
RNA sequencing (RNA-seq) analysis
RNA-seq analysis was performed by Capital Bio Technology (Beijing, China). Before preparation of the RNA-seq libraries, total RNA (1 μg) was treated with a RiboMinus Eukaryote Kit (Qiagen, Valencia, CA) to remove rRNA. Strand-specific RNA-seq libraries were prepared using a NEBNext Ultra RNA Library Prep Kit for Illumina (NEB, Beverly, MA) following the manufacturer’s instructions. The libraries were quality controlled with a 2100 Bioanalyzer chip (Agilent, Santa Clara, CA) and sequenced on the HiSeq 2000 platform (Illumina, San Diego, CA) with 100 bp paired-end reads.
RNA sequencing reads were mapped to the GRCh38.p7 assembly using STAR aligner version 2.5.2b with Ensembl annotations. Data analysis was performed with R software with a low-stringency filtering scheme of a 1.5-fold change in the expression level after MAS5 normalization of all datasets. To investigate the global effect of circRNF220 on AML cells, we used KOBAS 3.0 (
http://kobas.cbi.pku.edu.cn/index.php) for pathway analysis of the gene expression profile data for AML cells transfected with siRNA-circRNF220. Finally, Gene Set Enrichment Analysis (GSEA) v.2.0 was used to analyze a preranked list based on the DESeq2
t-statistic using preassembled gene sets from MSigDB v5.2.
Circular RNA–microRNA analysis
To predict the miRNA and AGO protein binding sites in circRNF220, we used the bioinformatics databases CircNet [
25] and Circular RNA Interactome [
26]. By combining these data with GSEA data, five miRNAs whose expression was significantly correlated with that of circRNF220 and were predicted to bind within the AGO sites were selected for downstream investigation.
RNA pulldown assay
A pulldown assay was performed as previously reported [
27]. The biotin-labeled circRNF220 probe was synthesized by Exonbio Lab (Guangzhou, China). A total of 1 × 10
7 circRNF220-overexpressing AML cells were harvested and lysed. The circRNF220 or oligo probe was incubated with streptavidin-coupled dynabeads at room temperature for more than 30 min to generate probe-bound dynabeads. After the treated beads were washed, the RNA complexes bound to the beads were eluted and disrupted with lysis buffer and proteinase K before analysis by qRT-qPCR. The biotinylated probe sequences used in this study are listed in Supplementary Table S
3.
Statistical analysis
Differences in mRNA, circRNA, or miRNA expression between two groups were analyzed using the Mann–Whitney U test for independent unpaired samples and the Wilcoxon test for paired samples. For comparisons among more than two groups, the Kruskal–Wallis test was performed first, and the Bonferroni correction for multiple comparisons was then applied. The χ2 or Fisher’s exact test was used for analysis of categorical variables, and the Spearman correlation coefficient (r) was determined to analyze correlations. Receiver operating characteristic (ROC) analysis was applied to evaluate the power of circRNF220 expression as a prognostic marker in AML, and the value of (CtcircRNF220-CtGAPDH) served as the estimated optimal diagnostic threshold. The area under the curve (AUC), sensitivity, and specificity were also calculated. The Kaplan–Meier method was used to analyze relapse-free survival (RFS), and the log-rank test was then performed to compare the outcomes among subgroups. Finally, a Cox proportional hazard regression model was used for univariate and multivariate analysis of prognostic factors. All calculations were performed using GraphPad Prism software and R software with a two-sided P < 0.05 considered to indicate a statistically significant difference.
Discussion
Tumor-specific gene expression patterns have been widely researched for their potential roles in cancer diagnosis and prognosis [
39‐
43]. Relapses occur in exceeding30% of children with AML, and the molecular mechanisms underlying AML relapse are still not fully understood [
44,
45]. In the current study, we demonstrated that the differential expression patterns of circRNF220 in PB or BM were able to distinguish pediatric patients with AML from individuals with other hematological diseases with extremely high specificity and sensitivity. Furthermore, the circRNF220 expression level at diagnosis could predict relapse in pediatric AML patients. Importantly, depletion of circRNF220 significantly suppressed cell cycle progression and increased apoptosis in AML cell lines and primary pediatric AML BM cells. The combined results of transcriptome analysis and molecular experiments substantiated that circRNF220 activated apoptotic signaling pathways by regulating miR-30a and its targets. Our inspiring findings highlighted the importance of circRNF220 in regulating the activities of leukemic cells, suggesting that accumulation of circRNF220 might be a potential biomarker for diagnosis and predicting recurrence in pediatric AML.
CircRNAs constitute a novel class of exceptionally stable RNA species in eukaryotes. Some are abundant and evolutionarily conserved and are anticipated to be used as biomarkers in diseases [
46,
47]. For example, circ-ANXA2 is upregulated in AML cell lines and may be a potential biomarker and therapeutic target in AML [
20]. Lei et al
. found that circ_0009910 is significantly upregulated in AML BM samples and might be a novel prognostic biomarker in AML [
48]. A new study reported that hsa_circ_0012152 can discriminate ALL from AML in adult patients, consistent with our results in the pediatric population [
49]. Moreover, we found that circRNF220 not only showed excellent performance in distinguishing AML from ALL but also allowed clear separation of AML from other hematological diseases, including MDS and CML. Of note, the level of circRNF220 in PB samples also had outstanding specificity and sensitivity for the differential diagnosis of pediatric AML. More importantly, the research indicated that the relative circRNF220 expression level at diagnosis could be an independent prognostic indicator for childhood AML relapse.
Strikingly, these findings have crucial implications for improving the diagnostic yield of biopsies from patients whose BM sample yield or quality is inadequate for accurate histological diagnosis [
50‐
52]. BM aspiration and biopsy are often relatively time-consuming and unpleasant experiences, especially for pediatric patients. In contrast, only a small amount of PB is needed to obtain an adequate amount of RNA for detecting circRNF220, possibly because of its high stability and remarkable abundance. However, the sample size was still limited, the results need to be interpreted with caution and more validation.
We noted that circRNF220 can radically promote proliferation, accelerate cell cycle progression and suppress apoptosis in AML cell lines and primary pediatric AML BM cells. Gene expression analysis of primary AML BM cells treated with LV-siRNA-circRNF220 indicated a complex phenotype including genes related to the terms hematopoiesis and apoptotic process involved in development. These phenotypes were in accordance with those observed in in vitro experiments.
CircRNAs are known to act as miRNA sponges [
53‐
56]. Guo et al. predicted that hsa_circ_0012152 might be involved in AML through the miR-491-5p/EGFR/MAPK1 axis or the miR-512-3p/EGFR/MAPK1 axis [
49]. Because internal ribosome entry sites were not found in the region upstream of a putative open reading frame in circRNF220, we first considered that this circRNA functions as a competing endogenous RNA. By combining transcriptomic and experimental data, we provide evidence that miR-30a is a key downstream effector of circRNF220 in our models. Although miR-30a is a crucial regulator of human cancer progression, few studies on the function and mechanism of AML have been conducted to date [
57]. Our results revealed that miR-30a can partially offset the functions of circRNF220 through targets MYSM1 and IER2. Many lines of evidence support the essential role of MYSM1 in hematopoiesis and hematopoietic stem cells [
58], and suggested that it might be related to the recurrence of AML. But little is known about the function of IER2 in AML, Neeb et al. showed IER2 is strongly upregulated in a wide variety of human tumors, and is a new player in the regulation of tumor progression and metastasis [
59]. Either way, these data indicated that circRNF220 regulating cell growth and associated with relapse partly relied on the miR-30a/MYSM1 and miR-30a/IER2 pathway.
Concurrent to miRNA sponge, regulation of RBPs is a plausible alternative role for circRNAs, and interactions of circRNAs with RBPs have been reported [
18,
60]. This may be a possible reason behind only a partial rescue by miR-30a and not complete remission of circRNF220 effects in Fig.
6I-K and Supplementary Fig S
4G-J. Thus, we utilized a web tool, CircInteractome, to explore potential interactions of circRNF220 with RBPs [
26,
61,
62]. Except Ago2 protein, EIF4A3, PTB, and TDP43 are predicted to bind circRNF220 based on sequence matches. The impact of potential secondary or tertiary structures on circRNA sequence available for interaction with RBPs cannot be considered systematically, hence experimental validation is essential to verify the predicted RBPs in the future.
Given that circRNAs are resistant to RNA exonucleases, they have a sufficiently long half-life within cells [
12]. Although 4 circRNAs can be generated from the
RNF220 mRNA by variable cyclization, only circRNF220 was highly expressed in childhood AML. RNF220 is a RING domain E3 ubiquitin ligase that was reported to mediate the ubiquitination of multiple targets and is involved in various developmental and disease progression [
63‐
65]. It is intriguing that circRNF220 may also play a vital role in the regulation of cellular protein metabolic processes and polyubiquitination (GO analysis and GSEA results), although circRNF220 had a unique expression profile distinct from that of the
RNF220 gene. We observed only a weak correlation between circRNF220 and RNF220, and it is unclear whether competition between canonical splicing and back splicing is likely to exist for the majority of loci that generate circRNF220.
In summary, circRNF220 was differentially expressed within distinct subtypes that compose the pediatric AML hierarchy, with definite prognostic value in relapse. The circRNF220-miR-30a axis operates in the progression of AML by facilitating the expression of its targets related to apoptosis. Treatment with siRNA targeting circRNF220 suppressed the growth and survival of AML cells. Collectively, these data suggest that circRNF220 should be investigated as a potentially effective biomarker and therapeutic target in pediatric AML.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.