Azra Krek ^{1, 2, 4}, Dominic Grün ^{1, 4}, Matthew N Poy ^{3, 4}, Rachel Wolf ¹, Lauren Rosenberg ¹, Eric J Epstein ³, Philip MacMenamin ¹, Isabelle da Piedade ¹, Kristin C Gunsalus ¹, Markus Stoffel ³, and Nikolaus Rajewsky ¹

¹ Center for Comparative Functional Genomics, Department of Biology, New York University, 100 Washington Square East, New York, New York 10003, USA.
² Department of Physics, New York University, New York, New York 10003, USA.
³ Laboratory of Metabolic Diseases, The Rockefeller University, New York, New York 10021, USA.
⁴ These authors contributed equally to this work.

Correspondence should be addressed to Nikolaus Rajewsky:  nikolaus.rajewsky@nyu.edu

MicroRNAs are small noncoding RNAs that recognize and bind to partially complementary sites in the 3' untranslated regions of target genes in animals and, by unknown mechanisms, regulate protein production of the target transcript [1, 2, 3]. Different combinations of microRNAs are expressed in different cell types and may coordinately regulate cell-specific target genes. Here, we present PicTar, a computational method for identifying common targets of microRNAs. Statistical tests using genome-wide alignments of eight vertebrate genomes, PicTar's ability to specifically recover published microRNA targets, and experimental validation of seven predicted targets suggest that PicTar has an excellent success rate in predicting targets for single microRNAs and for combinations of microRNAs. We find that vertebrate microRNAs target, on average, roughly 200 transcripts each. Furthermore, our results suggest widespread coordinate control executed by microRNAs. In particular, we experimentally validate common regulation of Mtpn by miR-375, miR-124 and let-7b and thus provide evidence for coordinate microRNA control in mammals.

Supplementary Files:
http://www.nature.com/ng/journal/v37/n5/suppinfo/ng1536_S1.html

Figure 1. The PicTar algorithm.

(a) Schematic overview. Input to PicTar consists of multiple alignments of RNA sequences (typically 3' UTRs) and a search set of mature (coexpressed) microRNA sequences. The program nuclMap locates all perfect nuclei (length 7, starting at position 1 or 2 of the 5' end of the microRNA) and imperfect nuclei in 3' UTR sequences. Highly probable nuclei that survive the optimal free energy filter and fall into overlapping positions in the alignments for all species under consideration are called anchors. If a 3' UTR multiple alignment has a minimal (user-defined) number of anchors, each UTR in the alignment will be scored by the central PicTar maximum likelihood procedure

(b). Scores for individual UTRs in an alignment are combined to obtain the final PicTar score, which can be used to obtain a ranked list of all sets of orthologous transcripts. (b) PicTar scoring of a single 3' UTR sequence. PicTar tallies all segmentations of the RNA sequence (3' UTR) into binding sites and background sequences [15]. PicTar computes the maximum likelihood score (PicTar score) that the RNA sequence is targeted by combinations of microRNAs from the search set when compared to background and the individual probability p i for each subsequence of the RNA sequence to be bound by a microRNA (only the nuclei for the binding sites are depicted). These posterior probabilities are different from the probability that a single subsequence is a microRNA binding site and reflect the competition of microRNAs and background for binding in the UTR.

Figure 2. Signal-to-noise ratio for vertebrate microRNA target site predictions.

(a) Signal-to-noise ratio for predicted single target sites. The number of predicted conserved target sites (anchors) for the set of 58 unique conserved human microRNAs versus the corresponding number for randomized microRNAs, requiring conservation of anchor sites between human, chimpanzee, mouse (first column), rat and dog (second column), chicken (third column) and pufferfish (last column). Inclusion of more distantly related species substantially boosts signal-to-noise ratio (indicated above black bars). For human, chimpanzee, mouse, rat and dog, we predict 17,542 conserved target sites with a signal-to-noise ratio of 2.3 and therefore approx 10,000 true target sites.

(b) Multiplicity of target sites boosts the signal-to-noise ratio. The ratio of the number of transcripts with at least n anchor sites per microRNA for real versus random microRNAs provides an estimate of the signal-to-noise ratio for sites conserved in human, chimpanzee, mouse (black bars), rat, dog (light gray bars) and chicken (dark gray bars). The multiplicity of target sites (scored by PicTar) helps to raise the signal-to-noise ratio.

(c) PicTar score-dependent sensitivity and specificity of single microRNA target site predictions. The average number of predicted targets of a single microRNA with at least one anchor site per transcript in human, chimpanzee, mouse, rat and dog (upper curve) or in human, chimpanzee, mouse, rat, dog and chicken (lower curve) is plotted as a function of a PicTar score cutoff (discarding transcripts with a score below cutoff). Signal-to-noise ratios are indicated above each curve.

(d) PicTar score-dependent sensitivity and specificity of target site predictions for four sets of coexpressed microRNAs [4] (Supplementary Table 4 online) and corresponding sets of randomized microRNAs, requiring two anchor sites for different microRNAs in human, chimpanzee, mouse, rat and dog. The plot shows the average number of targets per pair of microRNAs as a function of the PicTar score cutoff (signal-to-noise ratios above the curve).

Figure 3. Estimate of the number of coordinately regulated targets for sets of three microRNAs.

The probability p(n) that a set of three microRNAs hits at least n transcripts is plotted on a log scale for real (upper curve) and random (lower curve) microRNAs as a function of n. The criterion for a 'hit' was the presence of at least one anchor site, conserved in human, chimpanzee, mouse, rat and dog, for each microRNA in the triplet. The probability of obtaining not a single hit for a triple of real microRNAs is 1 - p(1) = 0.35. p(n) drops off exponentially and much more steeply for random microRNAs, indicating that PicTar runs with random microRNAs will typically yield a vastly reduced number of predictions. p(25) is approx 0.001 for real microRNAs, thereby indicating that only approx 30 of approx 30,000 possible sets of microRNA triples (sampled from our set of 58 microRNAs) are candidates to coordinately regulate at least 25 different transcripts each.

Figure 4. Regulation of Mtpn by miR-375, miR-124 and let-7b.

(a) Immunoblotting (IB). N2A cells were transiently transfected with siRNAs designed against eGFP (si-GFP), miR-124 (si-124) or let-7b (si-let-7b) and lysed after 48 h. Mtpn expression was assessed after SDS-PAGE and immunoblotting with antibodies to Mtpn. TBP (TATA-binding protein) expression was analyzed as a loading control.

(b) Dual luciferase assay of N2A cells transfected with a Renilla reniformis luciferase (Rr-luc) construct containing the full length 3' UTR of Mtpn and si-GFP, si-124, si-let-7b, si-375 or all three siRNAs for 48 h and lysed. A Photinus pyralis luciferase (Pp-luc) served as an internal transfection control. The ratios of Rr-luc to Pp-luc expression were normalized to the si-GFP transfections. Error bars represent the standard error (s.e.) from three independent experiments. * P 0.05; ** P 0.01.

The genes tested were selected from the single microRNA target prediction lists requiring different levels of conservation (human, chimpanzee and mouse or human, chimpanzee, mouse, rat and dog). Genes indicated in bold were validated by immunoblotting or luciferase reporter assay (+). -, no detectable decrease of endogenous target protein levels or luciferase reporter activity following microRNA overexpression; ND, not determined. Predictions to be tested were not selected by their PicTar score. The average score of the predicted targets for miR-375 and miR-124 is 2.66 and 2.04, respectively, which is low. Therefore, our number of false positives is comparable to the predicted signal-to-noise ratio (Fig. 2).

Row 1 enumerates the total number of nucleotides in our raw 3' UTR multiple alignments; row 2 lists the same quantities for unique and repeat masked sequences.

1. Johnson SM, Grosshans H, Shingara J, Byrom M, Jarvis R, Cheng A, Labourier E, Reinert KL, Brown D, and Slack FJ, "RAS Is Regulated by the let-7 MicroRNA Family".

2. Grosshans H, Johnson T, Reinert KL, Gerstein M, and Slack FJ, "The Temporal Patterning MicroRNA let-7 Regulates Several Transcription Factors at the Larval to Adult Transition in C. elegans".

3. Storz G, Altuvia S, and Wassarman KM, "An Abundance of RNA Regulators".

4. Iwama H, and Gojobori T, "Highly Conserved Upstream Sequences for Transcription Factor Genes and Implications for the Regulatory Network", OPEN  ACCESS ARTICLE.

5. Washietl S, Hofacker IL, and Stadler PF, "Fast and reliable prediction of noncoding RNAs".

6. Ling J, Ainol L, Zhang L, Yu X, Pi W, and Tuan D, "HS2 Enhancer Function Is Blocked by a Transcriptional Terminator Inserted between the Enhancer and the Promoter".

7. Cai X, Hagedorn CH, and Cullen BR, "Human microRNAs are processed from capped, polyadenylated transcripts that can also function as mRNAs".

8. Bracht J, Hunter S, Eachus R, Weeks P,  and Pasquinelli AE, "Trans-splicing and polyadenylation of let-7 microRNA primary transcripts".

Links to RNA and Biological Causality:

A Brief History of Activator RNA:

Links to Euchromatin Activator RNA Reviews:
Links to Euchromatin Activator RNA Research:
Links to Ultrastructural Probes of DNase I-Sensitive Sites:
Links to RNA as a Therapeutic Agent:
Links to Hodgkin Lymphoma Immuno-Pathology:
Links to Activated T-Lymphocyte Immunotherapy:
Links to Medical Systems Biology:
Links to Selective Gene Transcription:
Links to RNA-Induced Epigenetics:
Links to RNA-Induced Embryogenesis:
Links to RNA and Biological Causality:
Links to Reprogramming and Neoplasia:

A Brief History of Activator RNA:

"Ultrastructural Probes of Active DNA Sites, and the RNA Activators of DNA". (PowerPoint Presentation).