Stephane Flamant

Predicting microRNA targets and functions: traps for the unwary.

To the Editor: microRNAs (miRNAs) are short RNAs that are important in gene regulation in many organisms. In mammals, they guide the RNA-induced silencing complex to target sites typically located in the 3′ untranslated regions (UTR) of mRNAs, causing translational repression and/or mRNA degradation1. Despite recent advances in large-scale target screening techniques, experimental validation of miRNA targets remains cumbersome, and computational approaches remain the most commonly used method to identify putative target genes. Most algorithms use sequences collected from either the Ensembl2 or University of California Santa Cruz (UCSC)3 databases and then preprocess these sequences to correct for genes that do not have a predicted 3′ UTR of sufficient length. However, these databases use different criteria to define 3′ UTR boundaries (Supplementary Fig. 1 online) and these boundaries will change as more sequences from varied tissue types are added to them4. To determine whether these differences can alter target prediction, we applied four popular algorithms, miRanda5, TargetScan6, RNA22 (ref. 7) and PITA8 (Supplementary Table 1 online), to two 3′ UTR databases obtained from the Ensembl and the UCSC servers (Table 1 and Supplementary Methods online). When we ran TargetScan and Miranda on sequences from the same database, the results overlapped by 39.5%, whereas when we ran the same two algorithms on two different databases, the overlap was 11.5% (Supplementary Fig. 2 online). This result demonstrates the importance of using 3′ UTR sequences from both sources when comparing results between algorithms as optimally the overlap should approach 100%. Notably, MiRBase9 uses a modified version of miRanda on Ensembl data and TargetScan uses UCSC data. Current databases of miRNA target genes provide a list of hundreds of predictions for each miRNA. To reduce the number of predictions, investigators often consider only those targets that are predicted by multiple algorithms and consider this overlap as a higher-quality subset of predictions. To test the efficiency of intersecting multiple predictions, we performed an enrichment analysis on predictions from five commonly used databases: picTar10, TargetScan, RNA22, miRBase and PITA. The rationale of our analysis was the following: if the intersection of two of these datasets is enriched in true targets, then predictions taken from this intersection are more likely to be present in the intersection of two other datasets than in their exclusion (Supplementary Fig. 3 online). We found that this relation was not significantly true in 29/30 of the permutations tested and was of borderline significance in one (P = 0.03). From this analysis we concluded that the enrichment in true targets of overlapping predictions was weak at best. Admittedly, our test would be more reliable if we compared each overlap with a dataset of experimentally validated targets. However, such a set of quality verified targets, in which target site mutagenesis is shown to reduce the efficiency of miRNA regulation, is still too small to be used as a benchmark dataset (only 48 such targets are reported in the miRecords database; http://mirecords.umn.edu/miRecords/). We suggest that the routine identification of an overlap between miRNA target prediction algorithms should be discouraged owing to a lack of utility and rationale. Two more-logical approaches to filtering miRNA target predictions would be to examine coexpression of miRNA-target pairs (Supplementary Table 2 online) or to identify instances of multitargeting: genes that are targeted multiple times by the same miRNA. Four commonly used algorithms (picTar, TargetScan, miRanda and RNA22) can detect a number of multitargeting occurrences that is much higher than the number expected by chance (Supplementary Fig. 4 online). This demonstrates that multitargeting is widespread and nonrandom. In practical terms, picTAR and RNA22 are likely to reliably predict target genes when a given miRNA target site appears three or more times in the same 3′ UTR even though this method may omit many true targets. Because miRNAs can repress many genes, it is of interest to identify groups of genes targeted by the same miRNA that share a biological function or localization. This functional profiling of miRNA targets can be performed through the Gene Ontology (GO) website (http://www.geneontology.org/) and its associated tools. In addition to the known biases inherent in GO enrichment analyses, the choice of algorithm used to predict miRNA targets can be decisive in the outcome of functional profiling (Supplementary Fig. 5 online). To test how often different target prediction algorithms will predict a different enriched GO function, we created a program called MirGO (Supplementary Data online). This program uses publicly available data to discover the GO enriched function of a set of predicted targets of a given miRNA predicted by miRanda, TargetScan or picTar. Running MirGO 1,000 times on randomly selected miRNAs (Supplementary Fig. 6 online) showed that these three popular algorithms predicted different and unrelated functions in 94% (942/1,000) of cases tested. Notably, when we ran the same experiment considering only enriched GO functions predicted with very low P values (<0.001), their predicted function was discordant in only 4% (42/1,000) of the runs, even though the predicted genes were different for each of the algorithms. The approach used in MirGO has been implemented in many online resources such as

Micro-RNA response to imatinib mesylate in patients with chronic myeloid leukemia

Background Micro-RNAs (miRNAs) control gene expression by destabilizing targeted transcripts and inhibiting their translation. Aberrant expression of miRNAs has been described in many human cancers, including chronic myeloid leukemia. Current first-line therapy for newly diagnosed chronic myeloid leukemia is imatinib mesylate, which typically produces a rapid hematologic response. However the effect of imatinib on miRNA expression in vivo has not been thoroughly examined. Design and Methods Using a TaqMan Low-Density Array system, we analyzed miRNA expression in blood samples from newly diagnosed chronic myeloid leukemia patients before and within the first two weeks of imatinib therapy. Quantitative real-time PCR was used to validate imatinib-modulated miRNAs in sequential primary chronic myeloid leukemia samples (n=11, plus 12 additional validation patients). Bioinformatic target gene prediction analysis was performed based on changes in miRNA expression. Results We observed increased expression of miR-150 and miR-146a, and reduced expression of miR-142-3p and miR-199b-5p (3-fold median change) after two weeks of imatinib therapy. A significant correlation (P<0.05) between the Sokal score and pre-treatment miR-142-3p levels was noted. Expression changes in the same miRNAs were consistently found in an additional cohort of chronic myeloid leukemia patients, as compared to healthy subjects. Peripheral blood cells from chronic phase and blast crisis patients displayed a 30-fold lower expression of miR-150 compared to normal samples, which is of particular interest since c-Myb, a known target of miR-150, was recently shown to be necessary for Bcr-Abl-mediated transformation. Conclusions We found that imatinib treatment of chronic myeloid leukemia patients rapidly normalizes the characteristic miRNA expression profile, suggesting that miRNAs may serve as a novel clinically useful biomarker in this disease.

mimiRNA: a microRNA expression profiler and classification resource designed to identify functional correlations between microRNAs and their targets

MOTIVATION microRNAs (miRNAs) are short non-coding RNAs that regulate gene expression by inhibiting target mRNA genes. Their tissue- and disease-specific expression patterns have immense therapeutic and diagnostic potential. To understand these patterns, a reliable compilation of miRNA and mRNA expression data is required to compare multiple tissue types. Moreover, with the appropriate statistical tools, such a resource could be interrogated to discover functionally related miRNA-mRNA pairs. RESULTS We have developed mimiRNA, an online resource that integrates expression data from 1483 samples and permits visualization of the expression of 635 human miRNAs across 188 different tissues or cell types. mimiRNA incorporates a novel sample classification algorithm, ExParser, that groups identical miRNA or mRNA experiments from separate sources. This enables mimiRNA to provide reliable expression profiles and to discover functional relations between miRNAs and mRNAs such as miRNA targets. Additionally, mimiRNA incorporates a decision tree algorithm to discover distinguishing miRNA features between two tissue or cell types. We validate the efficacy of our resource on independent experimental data and through biologically relevant analyses. AVAILABILITY http://mimirna.centenary.org.au. SUPPLEMENTARY INFORMATION Supplementary data are available at Bioinformatics online.

Mesenchymal cells generated from patients with myelodysplastic syndromes are devoid of chromosomal clonal markers and support short- and long-term hematopoiesis in vitro

Myelodysplastic syndromes (MDS) are clonal malignant stem cell disorders characterized by inefficient hematopoiesis. The role of the marrow microenvironment in the pathogenesis of the disease has been controversial and no study has been performed so far to characterize mesenchymal cells (MC) from MDS patients and to analyse their ability to support hematopoiesis. To this end, we have isolated and characterized MC at diagnostic marrow samples (n=12) and have purified their CD34+CD38− and CD34+CD38+ counterparts (n=7) before using MC as a short- and long-term hematopoietic support. We show that MC can be readily isolated from MDS marrow and exhibit a major expansion potential as well as an intact osteoblastic differentiation ability. They do not harbor the abnormal marker identified by FISH in the hematopoietic cells and they stimulate the growth of autologous clonogenic cells. Conversely, highly purified stem cells and their cytokine-expanded progeny harbor the clonal marker with variable frequencies, and both normal and abnormal long-term culture-initiating cell-derived progeny can be effectively supported by autologous MC. Thus, we demonstrate that MDS marrow is an abundant source of MC appearing both cytogenetically and functionally noninvolved by the malignant process and able to support hematopoiesis, suggesting their possible usefulness in future cell therapy approaches.

Interaction between AIF and CHCHD4 Regulates Respiratory Chain Biogenesis.

Apoptosis-inducing factor (AIF) is a mitochondrial flavoprotein that, beyond its apoptotic function, is required for the normal expression of major respiratory chain complexes. Here we identified an AIF-interacting protein, CHCHD4, which is the central component of a redox-sensitive mitochondrial intermembrane space import machinery. Depletion or hypomorphic mutation of AIF caused a downregulation of CHCHD4 protein by diminishing its mitochondrial import. CHCHD4 depletion sufficed to induce a respiratory defect that mimicked that observed in AIF-deficient cells. CHCHD4 levels could be restored in AIF-deficient cells by enforcing its AIF-independent mitochondrial localization. This modified CHCHD4 protein reestablished respiratory function in AIF-deficient cells and enabled AIF-deficient embryoid bodies to undergo cavitation, a process of programmed cell death required for embryonic morphogenesis. These findings explain how AIF contributes to the biogenesis of respiratory chain complexes, and they establish an unexpected link between the vital function of AIF and the propensity of cells to undergo apoptosis.

Conserved expression patterns predict microRNA targets.

microRNAs (miRNAs) are major regulators of gene expression and thereby modulate many biological processes. Computational methods have been instrumental in understanding how miRNAs bind to mRNAs to induce their repression but have proven inaccurate. Here we describe a novel method that combines expression data from human and mouse to discover conserved patterns of expression between orthologous miRNAs and mRNA genes. This method allowed us to predict thousands of putative miRNA targets. Using the luciferase reporter assay, we confirmed 4 out of 6 of our predictions. In addition, this method predicted many miRNAs that act as expression enhancers. We show that many miRNA enhancer effects are mediated through the repression of negative transcriptional regulators and that this effect could be as common as the widely reported repression activity of miRNAs. Our findings suggest that the indirect enhancement of gene expression by miRNAs could be an important component of miRNA regulation that has been widely neglected to date.

MicroRNA Target Prediction and Validation

The accurate prediction and validation of microRNA targets is essential to understanding the function of microRNAs. Computational predictions indicate that all human genes may be regulated by microRNAs, with each microRNA possibly targeting thousands of genes. Here we discuss computational and experimental methods for identifying mammalian microRNA targets. We describe microRNA target prediction resources and procedures that are suitable for experiments where more accurate prediction of microRNA targets is more important than detecting all putative targets. We then discuss experimental methods for identifying and validating microRNA target genes, with an emphasis on the target reporter assay as the method of choice for specifically testing functional microRNA target sites.

MicroRNA in acute myeloid leukemia.

n engl j med 359;6 www.nejm.org august 7, 2008 653 Two comments relate to the technique of donor versus no-donor comparison that was applied in our study and in a recent meta-analysis by Cornelissen et al.2 We concede that from a statistical point of view, patients without siblings should be excluded from such an analysis. However, this does not reflect clinical reality and — most important — the decision-making algorithms within the clinical trials that provided the basis for our analysis. Matched unrelated donors have now become available for the majority of patients, and there no longer appears to be a difference in outcome between matched related and unrelated donors. As a consequence, in future clinical trials, true randomizations can be made between allogeneic transplantation and conventional intensive consolidation, thereby avoiding potential selection bias due to the procedure of genetic randomization. Saber and Williams raise the question of a potential imbalance in favor of the donor group in terms of early events. The composite end point in their comment is not well defined. The category “no consolidation” did not automatically translate into an event that was relevant for survival analysis; furthermore, frequency tables cannot substitute for correct assessment of the time dependence of events. In response to Narimatsu’s question: the 3-year survival probabilities for patients who underwent allogeneic transplantation were 60% (95% CI, 36 to 99), 56% (95% CI, 31 to 99), and 43% (95% CI, 29 to 64) for 10 patients with mutant NPM1 without FLT3-ITD, 9 patients with mutant CEBPA, and 38 patients with other genotypes, respectively.

Identifying microRNA determinants of human myelopoiesis

Myelopoiesis involves differentiation of hematopoietic stem cells to cellular populations that are restricted in their self-renewal capacity, beginning with the common myeloid progenitor (CMP) and leading to mature cells including monocytes and granulocytes. This complex process is regulated by various extracellular and intracellular signals including microRNAs (miRNAs). We characterised the miRNA profile of human CD34+CD38+ myeloid progenitor cells, and mature monocytes and granulocytes isolated from cord blood using TaqMan Low Density Arrays. We identified 19 miRNAs that increased in both cell types relative to the CMP and 27 that decreased. miR-125b and miR-10a were decreased by 10-fold and 100-fold respectively in the mature cells. Using in vitro granulopoietic differentiation of human CD34+ cells we show that decreases in both miR-125b and miR-10a correlate with a loss of CD34 expression and gain of CD11b and CD15 expression. Candidate target mRNAs were identified by co-incident predictions between the miRanda algorithm and genes with increased expression during differentiation. Using luciferase assays we confirmed MCL1 and FUT4 as targets of miR-125b and the transcription factor KLF4 as a target of miR-10a. Together, our data identify miRNAs with differential expression during myeloid development and reveal some relevant miRNA-target pairs that may contribute to physiological differentiation.