Computational prediction of miRNAs in Arabidopsis thaliana

Abstract

MicroRNAs (miRNAs) are post-transcriptional regulators of gene expression in animals and plants. Comparative genomic computational methods have been developed to predict new miRNAs in worms, flies, and humans. Here, we present a novel single genome approach for the detection of miRNAs in Arabidopsis thaliana. This was initiated by producing a candidate miRNA-target data set using an algorithm called findMiRNA, which predicts potential miRNAs within candidate precursor sequences that have corresponding target sites within transcripts. From this data set, we used a characteristic divergence pattern of miRNA precursor families to select 13 potential new miRNAs for experimental verification, and found that corresponding small RNAs could be detected for at least eight of the candidate miRNAs. Expression of some of these miRNAs appears to be under developmental control. Our results are consistent with the idea that targets of miRNAs encompass a wide range of transcripts, including those for F-box factors, ubiquitin conjugases, Leucine-rich repeat proteins, and metabolic enzymes, and that regulation by miRNAs might be widespread in the genome. The entire set of annotated transcripts in the Arabidopsis genome has been run through find MiRNA to yield a data set that will enable identification of potential miRNAs directed against any target gene.

Figure 1.

Ranked clusters of predicted candidate miRNA precursor sequences derived from two different findMiRNA experiments. Crosses (×) mark clusters containing previously reported miRNA precursors, filled triangles (▴) show clusters containing candidate miRNA precursors that were analyzed further, dots (·) indicate clusters of other candidate sequences. Clusters containing antisense sequences of previously reported miRNA precursors, for which the predicted target is erroneous, are shown in gray type. (A) Preliminary scan of 5701 transcripts. (Inset) All 1599 ranked clusters, with the magnified region delineated by gray borders. Clusters containing aligned sequences with 100% sequence identity across the entire alignment have a score of 0.0. (Magnified) Shows top-ranking clusters down to the cluster containing the cmr214 candidate precursor sequences. (B) Whole-genome scan, showing only ranked clusters that contain exact matches for the predicted miRNA sequence in rice. (Inset) All 236 ranked clusters, with the magnified region delineated by gray borders. (Magnified) Open circles (○) show candidates additional to those shown for preliminary scan. Small crosses (^×) represent clusters containing some sequences that are also present in superior clusters (×) of previously reported miRNA precursors. Clusters containing aligned sequences with 100% sequence identity across the entire alignment have a score of 4.0.

Figure 2.

RNA gel-blot analysis of size-selected small RNAs from three tissues, leaves, stems, and flowers, from both wild-type Columbia plants and transgenic Columbia plants expressing the HC-Pro gene under the control of the 35S promoter. Signals were detected for 8 of the 13 selected miRNA candidates following hybridization of specific end-labeled DNA oligonucleotides. (Bottom) (EtBr) show representative RNA loadings for the set of replicate blots above.(A) Signals corresponding to ∼21 nt small RNAs. The candidates, cand1 and cmr214, show the expected increase in expression in the HC-Pro samples. Expression is detected only in HC-Pro lanes for cmr6 and cmr3, with higher expression in RNA isolated from leaf and stem tissue, respectively. For cand2, lower expression was observed in HC-Pro than the Columbia wild-type samples. Arrows indicate the position of a 23-nt DNA size marker that corresponds to RNAs of ∼20 nt. (B) StarFire probes were used to improve weaker signals corresponding to ∼24 nt small RNAs. Probes to the antisense sequences were also used to help identify whether the correct genomic strand was identified by findMiRNA (prefix `as-'). (C) Signals corresponding to end-labeled probes specific for miR167 and the cand1 candidate miRNA for RNA isolated from HC-Pro samples. An RNA size marker shown on the left edge has values shown in nucleotides. Larger bands that may represent precursor intermediates are evident for cand1. Two size classes of small RNA, ∼22 nt and ∼24 nt, are observed for the control, miR167 (arrowheads).

Figure 3.

(A) Web site entry panel for query transcript At1g10200, with default precursor and miRNA parameter options. (B) Output from above query. (Graphic) Shows a map of the query transcript, below which is marked the positions of potential target sites (filled squares). These are labeled with the intergenic regions containing the precursor candidate loci from which the predicted miRNA candidates originate. The two labels on the right of this map, At1g20380 and At1g76140, correspond to the two cand2 loci that have Rfam names MIR394a and MIR394b, respectively. (Text) The predicted miRNA sequences are shown (white text on black), and, for comparison in this figure, the known mature miRNAs are indicated above the predicted precursors. When the `predicted orthologs in rice' box for the query options is also checked, only the At1g76140 cand2 candidate (MIR394b) record is returned. The At1g20380 cand2 candidate (MIR394a) record is not returned because the predicted miRNA for this record is 2 nt longer and does not have an exact match in rice. Boxes on the left can be checked and the precursor candidates aligned with ClustalW or downloaded in fasta format using the option shown below the listed candidate records.