Background
Acute myeloid leukemia (AML) is a frequently fatal malignant disease of hematopoietic stem and progenitor cells (HSPCs). Prognostic factors include patient age, antecedent hematological disease, preceding cytotoxic treatments for a primary disorder, and the presence of specific cytogenetic, molecular, and epigenetic aberrations [
1‐
7]. Identification and investigation of such somatic genetic alterations has enhanced our understanding of disease biology, augments diagnosis and prognostication, and may aid monitoring of the course of disease [
1,
3,
7,
8]. Although cytogenetics and molecular genetics have already facilitated great progress in these respects, novel technologies like next generation sequencing and biological breakthroughs like the identification of microRNAs (miRNAs) and long noncoding RNAs (lncRNAs) as entirely novel gene classes still provide fundamental additional insights. In this context, the expression, functions, and potential prognostic value of miRNAs in AML have been studied by a number of research groups over the past few years [
9‐
15]. In contrast, little is known about the expression and roles of lncRNAs in this disease.
miRNAs are small (~22 nucleotide, nt) RNA molecules which typically regulate multiple target genes at the levels of mRNA stability and translation efficiency [
16,
17]. They are excised through sequential processing steps from larger, usually polymerase II transcribed, RNA molecules termed primary miRNAs (pri-miRNAs). One pri-miRNA molecule may give rise to one or several miRNA species, with the respective genomic region referred to as a miRNA cluster in the latter case [
18]. Classical pri-miRNA processing consists of two endonucleolytic cleavage steps: the first one is carried out in the nucleus by the RNase Drosha and gives rise to ~70 nt long, hairpin shaped precursor (pre-) miRNA molecules. These are exported to the cytoplasm and further processed by another RNase, Dicer, to yield mature miRNAs [
18‐
21]. After incorporation into the RNA induced silencing complex (RISC), miRNAs recognize their target mRNAs through a 7–8 nt seed sequence and repress their translation and stability [
16,
17,
20,
21]. In addition to this well described way of action of mature miRNAs, biochemical and biological functions have recently been ascribed to some pri-miRNAs [
22,
23], thus greatly expanding the potential biological relevance of miRNA genes.
Large scale expression analyses of mature miRNAs revealed specific miRNA patterns that were associated with recurrent genetic abnormalities in AML [
10‐
12,
24]. Certain miRNAs were deregulated in AML compared to healthy bone marrow (BM), peripheral blood (PB), or CD34 positive (CD34+) HSPCs [
10‐
12,
24,
25], and miRNA signatures predictive of survival have been identified [
9,
10]. However, results from different research groups exhibited only limited concordance, most likely due to differences between patient collectives, types of healthy control, and methodologies [
13].
In this study, we re-addressed the question which miRNAs are differentially expressed between AML and healthy controls, and therefore might contribute to leukemogenesis. Using highly specific locked nucleic acid (LNA) oligonucleotide arrays capable of detecting 559 annotated and 77 proprietary human miRNAs, 64 miRNAs were found to be significantly misexpressed in AML. Further studies on the clustered miRNAs 221 and 222 revealed that pri-miR221/222 was overexpressed to a much higher extent than its mature products, which is best explained by an increased rate of transcription on the background of a limited processing capacity. Because pri-miR-221/222 is strongly overexpressed in a large proportion of patients with AML, it may represent a novel molecular marker and a putative oncogene in this disease.
Methods
Primary samples from patients and controls
This study was approved by the Ethics Committee of the Medical University of Vienna (EK no 609/2011). In some cases, archival samples were used. Written informed consent was obtained prior to sample collection from all other subjects. Data were analyzed anonymously.
For miRNA microarray hybridization, 52 AML PB samples and 13 control samples were used (Additional file
1: Tables S1A and B). The AML samples had been collected at the time of diagnosis, strongly enriched for leukemic blasts using Ficoll, and vitally frozen. Of the healthy control samples, three were CD34+ HPSCs enriched from bone marrow (BM) and five were BM mononuclear cells (MNCs) (all purchased from Lonza, Basel, Switzerland). Another five were PB samples from healthy volunteers, and were subjected to erythrocyte lysis prior to RNA isolation (Additional file
1: Table S1A). Twenty-two and 9 of these AML and control samples, respectively, were also used for qRT-PCR for mature miR-221 and 222. Characteristics of 27 additional diagnostic AML samples used to determine the ratio between pri-miR-221/222 and mature miR-221, as well as the expression of pri-miR-221/222 and other pri-miRNAs are summarized in Additional file
1: Table S1C. Eight additional control samples used for qRT-PCR experiments are included in Additional file
1: Table S1A. To investigate pri-miR-221/222 expression during the course of disease, four paired samples from the time of diagnosis and remission and three paired samples from the time of diagnosis and relapse were analyzed (Additional file
1: Table S1D).
Production and hybridization of miRNA microarrays, and primary data analysis
miRNA microarrays were produced as previously described [
26,
27]. Briefly, LNA modified oligonucleotide probes specific for the 559 human miRNAs contained in miRBase version 9.2 (
http://www.mirbase.org/) as well as probes for 77 proprietary miRPlus sequences (Exiqon, Vedbaek, Denmark) were spotted onto Hisens epoxy-coated glass slides (Schott Nexterion, Louisville, KY, USA) in eight replicates.
RNA was extracted with Trizol (Life Technologies, Carlsbad, CA, USA), subjected to quality control using a Bioanalyzer 2100 (Agilent Technologies, Santa Clara, CA, USA), and labelled with Hy3 using the miRCURY™ LNA microRNA array labelling kit (Exiqon). A Hy5 labelled mix of all samples was included in each hybridization as a common reference. Arrays were hybridized overnight at 60°C in microarray hybridization chambers (Corning, Corning, NY, USA). Hybridization and wash buffers from the miRCURY LNA microRNA Array Kit (Exiqon) were used according to the manufacturer’s instructions. Arrays were scanned with a GenePix 4100A Microarray Scanner and evaluated with GenpixPro 5.1 software (Molecular Devices, Sunnyvale, CA, USA). Primary data analysis was performed using ArrayNorm [
28]. Features were filtered for low quality spots, the local background was subtracted, spots were normalized to the global mean, and the ratios between the sample of interest and the common reference were log2 transformed.
Cell lines, miRNA expression vectors, stable transfections, and infections
KG1 and KG1a cells [
29] were obtained from the German Collection of Microorganisms and Cell Cultures, Braunschweig, Germany. HNT34 [
30] and MPD [
31] cells were kindly provided by Dr. Hiroyuki Hamaguchi, Musashino Red Cross Hospital, Tokyo, Japan, and Dr. Cassandra Paul, Wright State University, Dayton. Ohio, USA, respectively. HL60 [
32], HEL [
33], U937 [
34], HeLa [
35], MCF7 [
36], and 293 T [
37] cells were obtained from cell culture collections of the Medical University of Vienna. All cell lines were regularly tested for mycoplasma contamination.
The human myeloid cell lines KG1, HL60, HEL, U937, MPD, and HNT34 were cultured in Roswell Park Memorial Institute (RPMI) 1640 medium supplemented with 10%Fetal Bovine Serum (FBS) and 1%Penicillin-Streptomycin-Glutamine (PSG; all from Life Technologies, Carlsbad, CA, USA) in a humidified incubator at 37°C and 5%CO2. The same media were used for KG1a cells, except that they contained 20%FBS. The adherent cell lines HeLa, MCF7, and 293 T were maintained in Dulbecco’s Modified Eagle Medium (DMEM; Life Technologies) with 10%FBS and 1%PSG.
Plasmid miR-Vec-221/222, which contains the miR-221/222 cluster along with a blasticidin resistence gene in the pMSCV backbone, as well as the corresponding empty vector (miR-Vec) were kindly provided by the Agami lab [
38]. They were stably transfected into HL60 cells by electroporation and transfectants were selected using 2 μg/ml Blasticidin (Invivogen, San Diego, CA, USA).
pEZX-MR03-miR-221 (Homo sapiens microRNA miR-221 stem-loop expression clone, #HMIR0369), an HIV based lentiviral vector containing the miR-221 precursor, and pEZX-MR01-control (#CMIR0001-MR01), which contains a scrambled sequence instead, were obtained from GeneCopoeia (Rockville, Maryland, USA). They were infected into the myeloid cell lines HL60, KG-1, and KG-1a using standard procedures [
39]. Infected cells were sorted for GFP positivity on a FACS Aria (BD Biosciences, NJ, USA).
Total RNA was isolated from primary samples and cell lines using Trizol (Life Technologies). For detection of pri-miRNAs, RNA was treated with DNase I and converted to cDNA with M-MLV reverse transcriptase primed by random hexamer oligonucleotides (all reagents from Life Technologies). qRT-PCR was performed in an ABI Step One Plus sequence detection system (Applied Biosystems, Life Technologies) using the Mesa Green qPCR Master Mix Plus (Eurogentec, Liège, Belgium) and the primers listed in Additional file
2: Table S2A (synthesized by MWG Eurofins, Ebersberg, Germany). All primer pairs were subjected to standard curve analysis and yielded slopes between -3.0 and -3.5, indicating optimal or near optimal amplification efficiencies. Levels of mature miR-221 and miR-222 and of RNU6B were measured using Taqman assays (ID000524, hsa-miR-221 TaqMan Assay; ID002276, hsa-miR-222 TaqMan Assay; ID001093, mature miR control-RNU6B TaqMan Assay; Applied Biosystems). qRT-PCR reactions were performed in triplicate. The relative expression of pri-miRNAs compared to the housekeeping gene beta-2-microglobulin, and of mature miRNAs relative to RNU6B, were calculated according to the ΔΔCt method [
40].
RNA deep sequencing analysis and prediction of putative pri-miR 221/222 transcripts
RNA deep sequencing optimized for detection of lncRNAs was performed as described [
41]. In short, total RNA was extracted from human Hs27 foreskin fibroblasts with TRIreagent (Sigma-Aldrich, Seelze, Germany), treated with DNaseI (DNA-free kit, Ambion, Life Technologies), depleted of ribosomal RNA using RiboMinus Transcriptome Isolation Kit Human/Mouse (Life Technologies) and Ribo-Zero rRNA Removal Kit Human/Mouse/Rat (Epicentre Biotechnologies, Madison, WI, USA), and fragmented by hydrolysis. Double stranded cDNA was generated with SuperScript II Reverse Transcriptase (Life Technologies). The RNA-Seq library was prepared using the Chip-Seq DNA Sample Prep Kit (Illumina, San Diego, CA, USA), and sequenced using an Illumina Genome Analyzer II and Illumina HiSeq 2000 systems. 36 bp and 51 bp single end reads were aligned to human genome build hg18 using Bowtie (
http://bowtie-bio.sourceforge.net/index.shtml) and visualized on the University of California Santa Cruz (UCSC) genome browser (
http://genome.ucsc.edu/) (Vlatkovic IM, manuscript in preparation), together with publically available global run-on sequencing (GRO-Seq) data [
42], ENCODE histone modification Chip-Seq data generated by the BROAD Institute, and RefSeq Genes.
Cloning of reporter vectors, transient transfections, and luciferase reporter assays
Two fragments surrounding the transcription start site predicted for the putative 28.2 kb pri-miR-221/222 transcript were amplified from PB leukocyte genomic DNA using the primers shown in Additional file
2: Table S2B and Phusion High Fidelity Polymerase (New England Biolabs, Ipswich, MA, USA). The resulting PCR products were ligated into the pGL3-Promotor (pGL3-P) vector (Promega, Madison, WI, USA) using the SacI and XhoI restriction sites engineered onto the PCR primers to generate reporter vectors pGL3-P(-1874/+45) and pGL3-P(+17/+1952).
Subconfluent cultures of 293 T, HeLa and MCF7 cells growing in 24-well plates were transfected with 1 μg of the indicated pGL3-P vector derivative and 30 ng of the renilla luciferase vector pGL 4.70 (Promega, Madison, WI, USA), using JetPei Transfection Reagent (Polyplus Transfections, Illkirch, France) according to the manufacturer’s instructions. 48 h later, luciferase activities were measured using a Tristar LB941 (Berthold Technologies, Bad Wildbad, Germany) and the Dual-Luciferase Reporter Assay (Promega, Madison, WI, USA). Firefly luciferase activity was normalized to Renilla luciferase activity to control for transfection efficiency.
Statistical analyses
For statistical analyses of miRNA microarray data, only features with a valid human miRNA annotation, and which were detectable in at least half of the relevant samples (i.e., 27, 2, 3, and 3 of the AML, control CD34+, control BM, and control PB samples, respectively) were considered. One-way ANOVA was performed to test if a miRNA was differentially expressed in one of the classes, and a moderated t-test (R/Biconductor package limma) was used to test if a miRNA was differentially expressed between two respective classes. Resulting p-values were adjusted for multiple hypothesis testing based on the false discovery rate (FDR) [
43].
Spearman’s rank correlation was used to determine associations between miRNA expression as a continuous variable and the clinical parameters age, white blood cell count, percentage of blasts, lactate dehydrogenase (LDH), and cytogenetic risk. The point biserial correlation coefficient was used to show associations between miRNA expression and the dichotomous parameters gender and achievement of complete remission (CR), and significance was probed using the Wilcoxon rank-sum test. Associations between miRNA expression levels and the FAB type were given by the eta coefficient and significance was calculated by one-way ANOVA. All p-values were adjusted for multiple hypothesis testing based on the FDR [
43]. All calculations were performed using R.
For experiments using cultured cells at least three independent biological replicates were performed. Results are expressed as means ± standard error of the mean (SEM), or, where indicated, one representative experiment with standard deviations (SDs) from technical replicates is shown. Student’s two-tailed t-test at a significance level of 0.05 was employed to probe differences between groups for statistical significance.
Discussion
To identify miRNAs potentially contributing to leukemogenesis, the miRNA expression profiles of 52 AML samples were compared to those of hematopoietic cells from 13 healthy donors. Using microarrays containing highly specific LNA probes for over 600 human miRNAs - a substantially larger number than in most previous studies - 64 miRNAs were found to be misexpressed in AML in a statistically significant manner. Interestingly, subsequent studies on miR-221, one of the miRNAs most consistently overexpressed in AML, revealed a limited capacity of malignant myeloid cells to process its precursor forms. Firstly, only precursor, but not mature miR-221 levels were increased after introduction of appropriate expression vectors into several different human hematopoietic cell lines, and secondly, primary AML cells overexpressed pri-miR-221/222 to a much higher extent than its mature products. Because processing of vector-borne miRNA precursors, like that of endogenous pri-miRNAs, requires consecutive cleavage by Drosha and Dicer, the most plausible explanation of these results is a limited pri-miR-221/222 processing capacity at the Drosha level. pri-miR-221/222 overexpression was present in a high proportion of the patients investigated in this study (84%; 31/37 patients in the experiment shown in Figure
3, using the mean plus 3 standard deviations of the respective healthy controls as a cutoff for overexpression), so that the conclusion appears justified that it is a common phenomenon in AML. Underscoring the association between pri-miR-221/222 overexpression and AML, analysis of samples from the times of diagnosis, remission, and relapse showed that pri-miR-221/222 levels faithfully reflected the stage of disease.
As explanations for the elevated pri-miR-221/222 levels present in AML, pathologically increased transcription on the background of a physiologically limited processing capacity, or an acquired processing deficiency on the background of unaltered transcriptional activity may be considered. In disfavour of the latter possibility, a processing defect at the Drosha level does not necessarily lead to an accumulation of pri-miR molecules [
52‐
56]. Also, it could not easily be reconciled with our own and published [
12,
24,
25,
44] observations that mature miR-221 and 222 are not down-, but upregulated in AML. Increased transcription of pri-miR-221/222, in concert with a limited capacity of hematopoietic cells to process this molecule, therefore likely accounts for its induction in AML. This has not been proven, however, and will be the subject of future investigations.
A role for restricted miRNA processing capacities in tumor cells as well as in developmentally immature cells has been previously suggested by Thomson et al. [
53]. These authors found that mature let-7 g was almost undetectable in embryonic stem (ES) cells and P19 teratocarinoma cells, although the corresponding pri-miRNA was as abundant as in E14.5 embryos and differentiated P19 cells, which expressed high levels of let-7 g. Similarly, decreased expression of 22 miRNAs in a panel of human tumors of various origins was not accompanied by a reduction in the corresponding pri-miRNAs, indicating that it was due to an acquired processing defect rather than to transcriptional downregulation [
53]. In contrast, our data show that in AML transcriptional upregulation on the background of a probably a priori limited processing capacity causes an elevation of pri-miR-221/222 levels. Despite of these differences, both studies concur to suggest a role for restricted miRNA processing in tumorigenesis.
Since pri-miR-221/222 is upregulated in a large majority of AML, either this overexpression per se or an event leading to it should be essential for leukemogenesis. In principle, elevation of pri-miR-221/222 levels could be a bystander effect of the induction of a nearby oncogene. However, according to human genome assembly 19 (
http://genome.ucsc.edu/) no known genes are located within a genomic region that extends from 0.7 Mb centromeric to 0.5 Mb telomeric of the mature miR-221 sequence. Also, co-activation of pri-miR-221/222 with an adjacent oncogene would not be able to explain why other pri-miRNAs are also overexpressed in AML. Alternatively, a transcription factor whose deregulation contributes to misexpression of cancer genes could also activate pri-miR-221/222 and other pri-miRNAs in trans. The possible identity of such a transcription factor is presently obscure. Thirdly, an acquired pri-miRNA processing deficiency might contribute to leukemogenesis by diminishing the production of mature miRNAs. This appears unlikely for the reasons outlined above. Fourth, inhibition of pri-miRNA processing could be a side effect of mutations whose primary oncogenic effect is unrelated to miRNA biogenesis. For example, mutations inactivating the transcriptional competence of p53 also interfered with miRNA processing [
57]. However, this defect was not accompanied by increased pri-miRNA levels [
57], and furthermore p53 mutations are infrequent in AML [
58], so that such a mechanism remains speculative at present.
Thus, even though a number of possibilities can be conceived to explain pri-miR-221/222 overexpression as a bystander effect of other oncogenic events, at present none of these is sufficiently plausible to discredit the possibility that pri-miR-221/222 may itself act as an oncogene. lncRNAs have recently received great attention as possible cancer genes with diverse ways of action, e.g. the regulation of chromatin and transcription [
59], and pri-miRNAs may represent a novel type of lncRNA. In a recent study, the primary transcripts of the let-7 family of miRNAs (pri-let-7) were reported to be able to regulate gene expression [
22], thus ascribing a biochemical function to this type of molecule. Furthermore, some lncRNAs have been reported to serve as precursors for small RNA species in addition to their well described functions as macro ncRNAs [
23,
60]. Of particular interest, H19, an imprinted RNA that is overexpressed in many tumors and displays oncogenic functions [
61,
62], has been found to host a miRNA [
63]. It can therefore be considered as an example of a pri-miRNA that at the same time acts as an oncogenic lncRNA.
Conclusion
In this study we identified 64 miRNAs that were differentially expressed between AML and healthy controls, and may therefore contribute to leukemogenesis. Moreover, we found that pri-miR-221/222 was strongly overexpressed, but inefficiently processed, in a large proportion of AML, indicating that it may represent a novel molecular marker and a putative oncogene in this disease.
Acknowledgements
This work was funded by the Austrian National Bank (OeNB), grant no 12505, the Austrian Science Foundation (FWF), grants no P19795 and P20920 to R.W., and a Clinical Research Grant of the Austrian Society of Hematology and Oncology (OeGHO). None of these funding bodies had any role in the design of the study, the collection, analysis, and interpretation of data, the writing of the manuscript, or the decision to submit it for publication.
miR-Vec-221 and empty miR-Vec were kindly provided by the Agami lab [
38]. Michael Wagner and Holger Daims of the University of Vienna kindly allowed access to their microarray scanner. Karin Lind and Trang Le are gratefully acknowledged for administrating patient samples and data. Finally, we thank Denise Barlow (CeMM, Vienna, Austria) for the unpublished RNA-seq data.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
AR, KS, CS, ME, MS, and IV designed and performed experiments and interpreted their results. HH carried out statistical and bioinformatic analyses. RK, SCR, PV, and HS provided patient samples and clinical information. RW designed the study, interpreted data, and wrote the manuscript. All other authors read and approved the final manuscript.