Alu-directed transcriptional regulation of some novel miRNAs

Process of prediction

We investigated the number of miRNAs that may reside within the region downstream of Alu elements in the human genome using a newly-developed approach [28]. First, we downloaded all the repeat sequences annotated by the Repeatmasker from UCSC http://genome.ucsc.edu/, and from this we extracted about 1,180,972 Alu sequences and their extensions (200 bp). It is well known that the full length of Alu is about 300 bp, while sequences that can be transcribed by Pol-III are about 500 bp [8]. So, a 200 bp sequence downstream of Alu may be transcribed by Pol-III. Secondly, all the Alu elements and extended sequences not collected in EST were filtered. The resulting 86,966 sequences were found to be transcribed. Thirdly, we predicted 259 miRNAs at the threshold value of 7 using the PriMir software described briefly in the methods section [28]. Two hundred and nineteen sequences remained after the elimination of one nucleotide overlapping with the exon. Alignment of sequences from phylogenetically related species showed that 60 miRNAs and their upstream Alu sequences were highly conserved in primates (Additional file 1). Since Alu is conserved only in primates, those sequences that are conserved between Pan troglodytes and humans were selected. When sequences with more than four consecutive Ts (generally known as the Pol-III terminator) between the Alu elements and miRNA genes were eliminated, 60 sequences remained, as shown in the Additional file 1. The entire process is shown in Figure 1A. The secondary structures of some pre-miRNA candidates are illustrated in Figure 1B. Mature miRNA can be generated from either or both arms of a pre-miRNA.

Analysis of prediction results

Analysis of these 60 sequences indicated that the distribution of Alu length had two peaks, one around 120 nt and the other around 290 nt (Figure 2B). These sequences were allocated to two classes: >200 nt, considered the full length of Alu (Additional file 2); and <200 nt, considered to be stable Alu segments inserted in the genome (Additional file 3). The number (n = 28) of >200 nt Alu sequences proved similar to that (n = 32) of the <200 nt Alu sequences. All the AluSc, AluSg and AluY family sequences are longer than 200 nt while those of the FLAM and FRAM families are shorter (Additional file 2 and 3). AluY is the youngest class of the Alu family; AluSg and AluSc are the two youngest classes of the AluS family and are second only to AluY in evolutionary age. FLAM and FRAM are the oldest Alu families [29]. It has been shown that in the young families, the sequences and biological functions of Alu remain comparatively intact, whereas the sequences of the old families such as FAM and AluJo are no longer intact and some have even lost their functions owing to continual mutation during evolutionary history. In contrast, our observation suggested that both young and old members of Alu families may be functional (Additional file 2 and 3). Sequence alignment of the two groups of data showed that most Alu sequences longer than 200 nt basically preserved the A box and B box required for transcription. Only three sequences lacked the B box (Additional file 2). The sequences shorter than 200 bp were all mapped to the left arm of the consensus Alu sequence (Additional file 3). Although sequences shorter than 200 bp no longer had intact A boxes, they basically retained the B box necessary for transcription, as shown in Additional file 3. In contrast, both AluS-126274 (id) and AluJ/FLAM-741298 (id) lacked the B box but retained the A box. It is well known that the A and B boxes, especially the latter, are the internal promoters for Alu transcription [20, 21]. This is consistent with our prediction results, showing that both short and long sequences retain the B box. Most of the 60 sequences retained the A box and B box and some sequences around those boxes were also highly conserved. Weblogo analysis (Figure 2C) indicated that the first two nucleotides of the B box (positions -1 and -2) were highly conserved, and G in position -2 was observed in almost all sequences. The sequences following the A box were also found to be similar. For example, positions 11-14 were CTCA in almost all the Alu >200 nt sequences, suggesting that the motif around the A box may be important for Alu transcription.

Of the 60 sequences, 14 were located between protein-coding genes, 19 within the antisense strands of introns and 27 within introns, as shown in Figure 2D. Twenty-seven sequences derived from introns could be mapped to 65 different introns, suggesting that their sources are multiple loci.

miRNA verification

The same method was also used to explore other repeats. MIR belongs to the same SINE family, and was closely followed by candidate miRNA sequences. Besides the four known miRNAs, a further nine sequences downstream of MIR were verified (Additional file 6). Although the functions of MIR are not yet established, it may have some transcriptional activities. Whether MIR can direct the transcription of its downstream miRNAs awaits further study.

Verification of expression

From the sequencing results, three representative sequences were randomly selected for serial experiments: AluJb-641, AluJo-576611 and AluJo-135090. AluJo-576611 was found to be located in the antisense strand of TRIM14, AluJo-135090 between protein-encoding genes, and AluJb-641 within the seventh intron of the gene encoding mitochondrial membrane protein ATAD3B. To study the relationship between Alu and miRNA expression, we used cycloheximide and heat shock to induce high expression of full-length Alu RNA [32] and then determined whether its downstream miRNAs were also highly expressed. First, Northern blotting was employed to detect changes in the amount of Alu transcripts (Figure 3A and 3B). The results showed that after cycloheximide induction, the levels of full-length Alu transcripts in HeLa and HEK293 cells were elevated 4.7- and 5.1-fold respectively. This observation obtained further support. After heat shock induction, the content of full length Alu transcripts in HeLa and HEK293 cells were elevated 4.2- and 4.8-fold respectively. Real-time TaqMan PCR was then used to detect dynamic changes in miRNAs. After cycloheximide and heat shock induction, small RNAs were reverse-transcribed with a specific Stem-loop RT-primer and quantified by real-time TaqMan PCR. The results showed that in HEK293 cells, AluJb-641, AluJo-576611 and AluJo-135090 were respectively up-regulated 2.19-, 3.43- and 3.66-fold after cycloheximide induction (Figure 3D). Similar results were obtained using HeLa cells (Figure 3E). In contrast, no change in expression of the control miR-22 was detected, strongly suggesting that the expression of these miRNAs is closely related to the expression of their upstream Alu.

To confirm the involvement of Pol-III in miRNA transcription further and exclude the possibility of Pol-II-mediated miRNA generation, HeLa cells were treated with tagetitoxin at a concentration (4 U/ml) that specifically inhibits Pol-III transcription [33]. As shown in Figure 3F, the levels of mature AluJb-641, AluJo-576611 and AluJo-135090 was dramatically enhanced in heat-treated cells, while the level of let-7a, used as a Pol-III negative control, remained unchanged. More importantly, tagetitoxin specifically inhibited Pol-III and resulted in failure to detect mature AluJb-641, AluJo-576611 and AluJo-135090 bands in the 22 nt marker region. However, the transcription of let-7a, which is driven by Pol-II, was not affected by tagetitoxin. This suggests that the transcription of some miRNAs can be regulated by Pol-III and supports our hypothesis that Alu elements may be the promoter specific for nearby downstream ncRNAs.

To clarify whether such intron-derived miRNAs can be co-expressed when their host genes are transcribed, we investigated the expression profiles of AluJb-641 and its host gene. The former was found to reside within the seventh intron of ATAD3B, which encodes a mitochondrial membrane protein. This intron was 2,401 bp in full length and AluJb-641 was located at locus 192. RT-PCR showed that expression of the host gene was not enhanced by heat shock or cycloheximide stimulation (Figure 3C), which generally induce high expression of Alu and AluJb-641 (Figure 3A). Thus, AluJb-641 may not be co-expressed with its host gene, but along with the expression of its upstream Alu. Our results also showed that ATAD3B was not co-expressed with Alu.

Identification of a conserved element

Multiple sequence comparisons among three Alus used in this study and AluSg/x-1110454, which resides upstream of miR-517a, revealed that the sequence GAGGCTGAGG was very highly conserved (Figure 4A). Further careful comparison between the sequence GAGGCTGAGG and the consensus sequences of various Alu families revealed that almost all Alu sequences, long or short, contained this sequence. The alignment of Alu sequences (Additional file 2) showed that this motif existed in all the predicted sequences of more than 200 nt (Figure 4B). This suggests that it is highly conserved and may play an important role in Alu transcription or enhance its functions, especially when the A box and/or B box is not intact or deficient.

Datasets

Most of our data were downloaded from the UCSC table http://genome.ucsc.edu/ in May 2006: repeats from Variation and Repeats, miRNAs and exons from Genes and Gene Prediction Tracks, EST from mRNA and EST tracks. All data were saved in the bed format and uploaded to Galaxy http://main.g2.bx.psu.edu/ for analysis. In view of our knowledge that pre-miRNAs may be located immediately downstream of SINEs, Pol-III-transcribed sequences are usually less than 500 bp in length and most SINE elements are less than 300 bp, we extended 200 bp downstream of SINEs as our initial sequences for predicting candidate miRNA genes.

Human Genome (hg18) and Primate Genome (panTro2) were also downloaded from UCSC using FTP to prepare for conservation analysis.

The consensus sequences of different primate SINE families were downloaded from giri http://www.girinst.org/.

Filtering by ESTs

To ensure the SINEs were expressed, we compared all SINEs and extended sequences with ESTs. Although most ESTs are transcribed by Pol-II [39], the discovery of non-coding RNA (ncRNA) has led to the inclusion of expressed sequences transcribed by other polymerases in the EST database. Umylny et al. found that 452 Alu ncRNAs longer than 200 bp and probably transcribed by Pol-III are also listed in dbEST [40]. We used a similar method to filter our data except for the differences in SINE lengths. If both SINEs and extended sequences were collected among ESTs, we took them as sequences that are potentially still active.

Prediction of pre-miRNAs

The software PriMir, developed by us [28], was used to predict miRNAs by the following procedure. First, we created a training data set using known miRNAs and a background data set, using hairpins, extracted from 200 bp extended sequences. Secondly, 11 secondary structure features of known pre-miRNAs were used as screening criteria: (1) the total number of paired bases in the 10-bp up- and down-stream extensions of the pre-miRNA; (2) its total bulge size; (3) the total number of paired bases in it; (4) the length of its loop; (5) the distance between the mature miRNA and the terminal loop; (6) the sequence bias of the first five bases in the mature miRNA; (7) the total number of paired bases in the mature miRNA portion of the pre-miRNA; (8) the minimum free energy (mfe) of the pre-miRNA stem-loop calculated using the RNAfold program; (9) the length of the pre-miRNA; (10) its GC content; and (11) the GC content of the mature miRNA. Using these 11 features, we constructed a PMS matrix. For each entry in the matrix, we calculated the ratio between the frequencies of each feature value in the training and background sets. All entries were calculated according to the definition of (A_ij). For a given feature i with value j, f_i (j) and h_i (j) are the frequencies of this feature in the training and background sets, respectively. X_i is the feature value set of feature i. If a given j value belongs to X_i, A_ij is defined as the log₂ of f_i (j)/h_i (j). Otherwise, A_ij is assigned the minimum value of A_i x_i. Then we used PriMir to predict a potential pre-miRNA candidate and calculate a PriMir score from the PMS matrix. The PriMir score value S is defined as the sum of the scores of all features for a given hairpin: Here x_i is the value of feature i.

Thirdly, we folded the 200 bp extended sequences and extracted pre-miRNA candidates from them on the basis of the 11 features. The reliability of PriMir prediction was evaluated by cross-validation. The training and background sets used to establish the PMS Matrix were divided into five equal parts. Four parts were selected to establish the PMS Matrix, and the remaining part (from both training and background sets) was used to test the performance of PriMir by the ROC curve. The above analysis was repeated five times, each time using a different portion of the data as test data set. A PriMir score of "7" was used as cutoff value. This is a stringent criterion, as ROC curve analysis of the performance of PriMir indicates that the AUC (area under curve) is approximately 0.99, and that the false positive rate is 0 at a PriMir score of 7. Comparisons between PriMir and three other algorithms [41–43] suggested that PriMir was at least equal to and in some respects outperformed those three methods. In addition, we used EST and conservation to filter the initial results, which made our results more convincing. The Accession Numbers for the 24 Alu-related miRNAs deposited at Gene Bank are FJ601661 to FJ601684.

Conservation analysis

Because Alus are primate-specific repeats and there is only one primate genome, panTro2, in UCSC, we only examined conservation between Pan troglodytes and human. In order to find highly conserved miRNAs, we only considered the miRNA candidates with more than 96% similarity to those in Pan troglodytes. The lengths of these pre-miRNA candidates in human should be more than 90% of those in Pan troglodytes.

We used a similar criterion to find other conserved sequences in SINEs located upstream of pre-miRNAs. The criteria for selection were that the lengths of conserved repeat sequences in Pan troglodytes must be more than 90% of those in human and the corresponding similarity must also exceed 90%. If both SINEs and the corresponding pre-miRNAs were conserved and the distance between them was less than 200 nt, we took the pre-miRNAs as our candidates.

Filtering the candidates

If repeats are expressed in exons of protein-coding genes, they probably do not function as pre-miRNA promoters. Few miRNAs so far known are present in exons, so we filtered out those predicted miRNAs and SINEs that overlapped with exons.

Since more than four Ts between a predicted pre-miRNA and SINE are highly likely to intervene in the transcription of long sequences, we deleted each predicted sequence with more than four Ts between the pre-miRNA and SINE.

Cell culture

HeLa and HEK293 cells were obtained from ATCC (American Type Culture Collection). The cells were cultured in DMEM (Dulbecco's Modified Eagle Medium) supplemented with 100 U/ml penicillin, 100 U/ml streptomycin and 10% FBS (fetal bovine serum) at 37°C in a humidified atmosphere containing 5% CO₂. For heat shock experiments, cells were heated in 75 cm² flasks containing 10 ml medium in a 45°C water bath. After 30 min heat shock, they were incubated at 37°C for 1 h before total RNA was extracted. For cycloheximide treatment, the cells were grown in the presence of 100 μg/ml cycloheximide. The incubation time was 3 h for HeLa cells and 6 h for HEK293 cells prior to total RNA extraction.

Inhibitors of transcription

Tagetitoxin, a specific inhibitor of RNA Pol-III, was obtained from Epicentre Technologies (Madison, WI). HeLa cells were treated with 120 μmol/l tagetitoxin for one day and then by heat shock as described above. Total RNA was extracted from tagetitoxin-treated and untreated cells. Mature miRNA was isolated and amplified by real-time PCR as described below. The effects of tagetitoxin on in vivo transcription were assessed by Northern blotting. Inhibition of RNA Pol-III was evaluated by the levels of transcription of three Alu-associated miRNAs, with Let-7a as a negative control.

Small RNA cloning and sequencing

We employed Illumina sequencing methods as described previously. Briefly, total RNA was extracted from human cells with Trizol, and the small RNAs were size-fractionated, purified by polyacrylamide gel electrophoresis (PAGE) to enrich for molecules in the range 18-30 nt, and ligated sequentially to 5'- and 3'-end RNA oligonucleotide adapters. The samples were used as templates for cDNA synthesis. The cDNA was amplified over 15-18 PCR cycles to produce sequencing libraries using Illumina's small RNA primer set. The purified PCR products were subjected to Solexa's proprietary sequencing-by-synthesis method on the Illumina 1G Genome Analyzer. All sequencing services were provided by the Beijing Genomics Institute, Shenzhen http://www.genomics.org.cn and the Genome Sciences Centre, Vancouver. Raw data from IlluminaGA were processed using the initial stages of the Solexa software pipeline (Illumina). Low quality reads were trimmed using perl script. Adaptor sequences were accurately clipped using a dynamic programming algorithm. After redundancy was removed, sequences ≥ 18 nt were mapped to the human genome (UCSC hg18) using SOAP. Small RNA sequences were mapped to human shRNA hairpins by BLAST. The count sum of all isoforms of a shRNA gene was used to measure its expression level.

Reverse transcriptase reactions

Total cellular RNA was isolated with TRIzol reagent (Invitrogen) according to the manufacturer's instructions. Reverse transcriptase reactions contained 1 μg purified total RNA, 50 nM stem-loop RT primer (Applied Biosystems), 1× RT Buffer, 0.25 mM each dNTP, 5 U/μl M-MLV reverse transcriptase (Promega) and 0.25 U/μl RNase inhibitor (Promega). The mixtures were incubated in a thermocycler for 30 min at 16°C, 30 min at 42°C and 10 min at 85°C, then held at 4°C. All reverse transcriptase reactions, including controls, were run in triplicate.

Real-time PCR

Real-time PCR was performed using a standard TaqMan MicroRNA Assay kit protocol on a Roter-Gene 6000 (Corbett) Sequence Detection System. The 20 μl PCR reactions included 1 μl RT product, 1× Real-time PCR Master Mix including TaqMan probe, 0.15 μM miRNA specific primer set. The mixtures were incubated at 95°C for 3 min, followed by 40 cycles at 95°C for 15 s and 60°C for 40 s. All reactions were run in triplicate. The threshold cycle (CT) was defined as the fractional cycle number at which the fluorescence passed the fixed threshold. TaqMan CT values were converted to absolute copy numbers using a standard curve from synthetic miRNA Standard.

Northern blotting