Co-transcriptional recruitment of Drosha

We carried out chromatin immunoprecipitation (ChIP) assays using a Drosha-specific antibody on HeLa cell chromatin of endogenous, intergenic or intron-encoded miRNA genes. Intergenic miR-7, miR-223, miR-23a and let-7a-1 miRNA genes and two host genes, the echinoderm microtubule-associated protein–like 2 (EML2) and minichromosome-maintenance complex component 7 (MCM7) genes, which contain miR-330 and miR-25 located in the first intron of EML2 and the thirteenth intron of MCM7, respectively, were tested (Fig. 1). Drosha specifically associates with chromatin of the miR-23a and let-7a1 genes (Fig. 1a, Drosha ChIP) but not with chromatin of the miR-223 and mir-7 genes, which are not expressed in HeLa cells^22,23. Similarly, Drosha recruitment was also observed on both intronic miRNA loci (Fig. 1b, Drosha ChIP) but not on GAPDH and actin genes, used as negative controls (Fig. 1a,b). We tested the expression profile of all miRNA loci and that of GAPDH and actin genes using Pol II immunoprecipitation (Fig. 1a,b, Pol II ChIP). Notably, the binding of Drosha correlates with the presence of Pol II on the same miRNA loci, suggesting that its recruitment is associated with transcription. We also analyzed the chromatin association of Drosha for the tissue-specific miR-223 gene, a known modulator of human myeloid differentiation, which is activated during retinoic acid–induced differentiation of NB4 cells²² (Fig. 1c, northern blot). ChIP analysis showed that the recruitment of Drosha on miR-223 in NB4 cells (Drosha ChIP) follows its activation (Pol II ChIP), establishing that Drosha recruitment requires ongoing transcription of the miR-223 gene (Fig. 1c).

Next, we generated two different constructs in which the pri–miR-223 coding sequence, with (wild type) or without (Δ) the SD junction required for efficient Drosha cleavage⁵, was positioned downstream of the ubiquitous miR-23a~27a~24-2 promoter²⁴. After transfecting the constructs into HeLa cells, we isolated the total RNA and analyzed the nuclear chromatin (Fig. 1d). The wild-type but not the Δ construct generated mature miR-223 and yielded a positive Drosha ChIP signal. As positive controls, U2 small nuclear RNA (snRNA) for the northern blot and endogenous miR-23a for the Drosha ChIP gave similar signals in both wild-type and Δ samples (Fig. 1d).

To validate the specificity of Microprocessor complex recruitment, we performed ChIP analysis across the host EML2 gene and used oligonucleotide pairs to detect the promoter and coding region of the miR-23a gene (Fig. 2a,b). Notably, we observed only Drosha over the miR-330 and miR-23a sequences. Acting as a further control, we detected Pol II ChIP signals across the EML2 gene, even though a two-fold higher signal was observed nearer the 5′ end, presumably related to promoter-proximal pausing²⁵. We furthermore determined whether the interaction of the Microprocessor with chromatin occurs through association with the nascent transcript by testing the effect of RNase treatment before the immunoprecipitation step in the ChIP analysis. These results (Fig. 2b,c) show that Drosha association with the chromatin both at miR-23a and miR-330 sequences is RNase sensitive, whereas the Pol II association is unaffected (Fig. 2c, right). Previous in vitro analysis has shown that the Microprocessor complex interacts with pri-miRNAs through DGCR8, which possesses tandem dsRBDs^5,26. We therefore tested whether Drosha association with mir-23a and mir-330–containing chromatin depends on DGCR8 by depleting DGCR8 levels using RNA interference (RNAi). miR-23a processing is indeed impaired (Supplementary Fig. 1 online), and we also observed a substantial reduction in Drosha levels at the miRNA loci tested (Fig. 2d). These results suggest that, in vivo, DGCR8 may bind pri-miRNA hairpins before Drosha.

Microprocessor cleaves pre-miRNA harpin before splicing

We previously showed that co-transcriptional cleavage of nascent intronic transcripts does not affect the efficiency of splicing, predicting that exons of pre-mRNA are tethered to the elongating RNA Pol II either directly or indirectly^27,28. Our results demonstrate that the Microprocessor complex is co-transcriptionally recruited to chromatin of Pol II host miRNA genes (Figs. 1 and 2), and for this reason, we predict that the pre-miRNA sequence is excised from introns during transcription without affecting the splicing of the host gene.

We generated two constructs, βInt2-miR-330 and βInt2-miR-let-7a3, containing miR-330 and let-7a3 pre-miRNA hairpins in the middle of the second intron of the β-globin gene transcribed from the HIV-1 promoter (Fig. 3a). After transfection into HeLa cells, we carried out northern blotting of the cytoplasmic RNA. Both miR-330 and let-7a3 mature miRNAs were readily detected by northern blot analysis, indicating their efficient processing and proper localization to the cytoplasmic fraction (Fig. 3a, right). Note that in HeLa cells, endogenous miR-330 is expressed at low levels (Supplementary Fig. 2a online), even though the miR-330 host gene EML2 is expressed in HeLa cells¹⁴ and both Pol II and Drosha associate with the miR-330–containing intron 1 (Fig. 1b). It is possible that down-regulation of miR-330 processing occurs as described for other miRNA-containing genes^29,30.

The possibility that these intronically positioned miRNA sequences are co-transcriptionally cleaved was analyzed by a refinement of the nuclear run-on (NRO) protocol previously used to characterize intron co-transcriptional cleavage events²⁷. We transiently transfected HeLa cells with βInt2-miR-330, isolated the cell nuclei and subjected them to run-on analysis. An antisense biotinylated probe complementary to the intronic sequence, upstream of the Drosha cleavage site was then used to select these nascent transcripts (Fig. 3b, diagram) and the selected and unselected fractions were hybridized to separate filters containing single-stranded M13 DNA probes (Fig. 3b, Select and Unselect). In parallel, as a control, we carried out a regular NRO experiment on the same construct, and this produced strong hybridization signals over all gene probes (Fig. 3b, NRO). The selected and unselected fractions showed differences in hybridization signal beyond probe a, located upstream of the Drosha cleavage site (B3/a ratio, Fig. 3b, histogram). Thus, the unselected fraction gave a higher B3/a value than the selected fraction. These results suggest that co-transcriptional cleavage of the nascent transcript occurs over the pre-miRNA sequence and consequently confirms that Drosha cleavage is a co-transcriptional event. However, as there is partial selection of transcripts upstream of the Drosha cleavage site, these data also indicate that Drosha cleavage is a relatively slow process.

To compare Drosha cleavage with another well-characterized co-transcriptional cleavage event, we cloned miR-330 and let-7a3 hairpins downstream of the poly(A) site of the β-globin gene in place of its co-transcriptional cleavage (CoTC) termination sequence (Fig. 3c, diagram). Transcriptional termination relies on the specific CoTC sequence, which is cleaved co-transcriptionally, resulting in the recruitment of the nuclear 5′-3′ exonuclease XRN2 (ref. 31). We carried out NRO analysis (NRO, Fig. 3c, left) on these constructs using single-stranded M13 probes spanning the β-globin gene and an M13 (M) alone control. Whereas lower hybridization signals were observed over the probes A and U3 in the β-globin construct owing to efficient transcriptional termination, stronger signals were observed for the termination-deficient ΔCoTC construct and those containing miRNA hairpins in the 3′ region. This suggests that Drosha co-transcriptional cleavage does not promote efficient transcription termination. Together, these results (Fig. 3b,c) suggest that Drosha pre-miRNA processing, although a co-transcriptional mechanism, is kinetically slower than CoTC cleavage and so is unable to promote efficient transcriptional termination downstream of the β-globin poly(A) site.

We next determined whether pre-miRNA maturation occurs from β3′-miR-330 and β3′-let-7a3 gene constructs by northern blot analysis (Fig. 3c, right). We observed accumulation of mature miRNAs from both constructs at comparable levels to the intronic constructs βInt2-miR-330 and βInt2-let-7a3. These results demonstrate that pre-miRNAs may be processed from the 3′ flanking regions of a gene if extended 3′ transcription occurs. As these transcripts are highly unstable and remain tethered to the elongating Pol II complex³², this confirms that pre-miRNA maturation occurs co-transcriptionally.

We next used hybrid selection circular rapid amplification of cDNA ends (hscRACE³³) to establish that Drosha cleavage occurs on nascent transcripts. βInt2-miR-330 and βInt2-let-7a3 nuclear RNA were hybridized to a biotinylated antisense RNA probe complementary to the second exon of the β-globin gene to select nascent transcripts. The hscRACE procedure is described in Figure 4a. For both selected RNA samples, we observed one major hscRACE DNA product, which, based on sequence analysis (data not shown), corresponds to unspliced RNA up to the Drosha cleavage site; we also observed some minor bands that probably correspond to degradation intermediates (Fig. 4a). We also included controls to show that the hscRACE products correspond to in vivo Drosha cleavage products rather than nonspecific RNAs generated by the hscRACE technique (Supplementary Fig. 2b,c)

Overall, these data confirm that the RNA 3′ ends observed by hscRACE analysis were authentic in vivo products of Drosha cleavage. In addition, sequence analysis of the main PCR products revealed that the 3′ processed molecules still contained the exon 2–intron 2 junction region. Consequently, we provide direct evidence that unspliced pre-mRNA transcripts are major substrates of Drosha processing.

Co-transcriptional Drosha cleavage of intronic microRNAs

Our above results demonstrate that the Microprocessor complex binds miRNA sequences during transcription and that Drosha processing occurs before the host intron is spliced out. However, to determine whether nascent RNA, rather than an unprocessed, released transcript, is the actual substrate for Drosha cleavage, we used a fractionation procedure to separate chromatin-associated nuclear RNA from released nucleoplasmic transcripts^27,34.

We transfected HeLa cells with the βInt2-miR-330 plasmid or a mutant version containing altered sequences in the miR-330 stem region that are known to strongly inhibit Drosha processing¹⁴. We confirmed that mature miRNA was not detectable by northern blot analysis for mutant miR-330 (Fig. 4b). Following transfection, equal amounts of RNA extracted from chromatin-associated (Pellet) and nucleoplasmic fractions were subjected to reverse-transcription PCR (RT-PCR) analysis to detect the distribution of β-globin transcripts containing unspliced introns 1 and 2 (primers Ex1-Int2a and 330β-F-Int2b). Intron 1 and/or the miRNA harboring intron 2 were predominantly localized in the chromatin-associated (greater than 90%) as compared to the nucleoplasmic fraction (Fig. 4c, above and middle, Pellet and SN). Note that all these RT-PCR data were also analyzed by quantitative real-time PCR analysis, as shown below the gel images. Little β-globin transcript was released from the site of transcription before all the introns were spliced out. Both the wild-type and mutant constructs showed similar distributions, indicating that a functional intronic miRNA does not change the chromatin-nucleoplasmic localization of β-globin transcripts. Note that a higher pellet signal was observed for mutant RNA across intron 2 (Fig. 4c, middle), presumably because of the lack of Drosha cleavage. As a control, the same RNA samples were reverse transcribed in the presence of oligodT, and the cDNA was amplified using Ex1 and Ex3 PCR primers (Fig. 4c, below). A band corresponding to the processed β-globin mRNA was present in only the supernatant fraction (SN), showing that spliced transcripts are readily detected in this fraction. As miRNA-harboring transcripts are preferentially detected in the chromatin-associated fraction, both pre-miRNA cleavage and intronic splicing must occur on the same nascent transcripts.

Drosha cleavage promotes intron degradation by exonucleases

We reasoned that co-transcriptional cleavage of intronic miRNAs by Drosha might be associated with recruitment of 5′-3′ and 3′-5′ exonucleases to the newly generated 5′ and 3′ ends of the host intron transcript. Therefore, we performed ChIP analysis of the EML2 and MCM7 genes using antibodies against both the XRN2 5′-3′ exonuclease (human homolog of the yeast Rat1) and the PMscl100 component of the nuclear Exosome (Fig. 5a,b). When compared to the actin gene, used as a negative control, enrichment was observed for XRN2 exonuclease on the intronic regions containing miR-330 and miR-25 (Fig. 5b). Similarly, an enrichment of PMscl100 was detected over both introns containing miRNA sequences as compared to other gene sequences (Fig. 5b). Co-localization of the XRN2 and Exosome exonuclease activities as well as Drosha over host introns containing miRNA sequences strongly supports the co-transcriptional miRNA-processing model and further suggests that after Drosha cleavage the intronic sequences are co-transcriptionally degraded.

To determine the effect of both exonucleases and Drosha on the splicing efficiency of miRNA-containing introns, we performed RNAi-mediated knockdown in HeLa cells of XRN2, Exosome (PMscl100) and Drosha. In each case, protein levels were substantially reduced compared to mock depleted cells (Supplementary Fig. 1a). We then measured the accumulation of unspliced versus spliced forms of endogenous gene transcripts hosting intronic miRNAs, following knockdown of exonucleases. We selected two specific miRNA-hosting genes, MCM7 and CTDSPL, which contain miRNA sequences in a short and a long intron, respectively (Fig. 5c). Total RNA was extracted from both mock and siRNA-treated cells and reverse transcribed with a gene-specific oligonucleotide. Primers complementary to the miRNA-harboring intron and to the neighboring exons were used to quantitatively detect unspliced and spliced transcripts, respectively (Fig. 5c). In both miRNA-hosting genes, the splicing efficiency of the miRNA-containing intron was reduced with either XRN2 or PMscl100 knockdown, regardless of intron length (Fig. 5c, unspliced/spliced ratio), whereas a weaker effect was observed in Drosha-depleted cells. These results confirm that, although co-transcriptional cleavage of the intronic nascent transcript does not affect splicing efficiency²⁷, in introns containing miRNAs, clearance of intronic sequences following Drosha cleavage may act to enhance the splicing efficiency.

Some intronic sequences are known to contain regulatory signals that are important to enhance or prevent the inclusion of an alternative exon in the mature transcript. We therefore reasoned that co-transcriptional degradation of such introns could detrimentally affect the splicing pattern of specific transcripts. We used a bioinformatics approach to analyze the distribution of intronic miRNA hairpins in the human genome, matching the miRNA location with protein-coding genes. Among 344 intronic miRNAs analyzed, 243 (70%) are located within introns of characterized protein genes (miRNA introns, Fig. 5d), and they are furthermore preferentially located near the middle of the intron (data not shown). Notably, recent analysis of intron size in the human and mouse genomes has shown that miRNA-harboring introns are generally larger than other introns, even comparing introns of the same host gene³⁵. Finally, when refining the bioinformatic analysis for alternatively spliced transcripts, we observe that, even though 70–80% of human genes are alternatively spliced^36,37, only 2% of intron-encoded miRNAs are located in introns adjacent to alternatively spliced exons. Consequently, our data suggest that during evolution, the presence of co-transcriptionally cleaved sequences within introns and subsequent degradation events may have acted as a driving force for the intron size by excluding the presence of miRNAs near key sites of splicing regulation.

Primers

A list of oligonucleotides sequences is given in Supplementary Table 1 online.

Plasmid constructions

The miR-330 and let-7a3 sequences were PCR amplified using HeLa cell genomic DNA with the primer pairs 330β-F/330β-Rev and let7β-F/let7β-R, respectively. To generate the βInt2-miR-330 and βInt2-let-7a3 reporter constructs, PCR products were 5′ phosphorylated with polynucleotide kinase (Roche) in the presence of 1 mM ATP and blunt-end ligated to the βΔ5-8 vector backbone generated by long-range PCR with the Int2F/Int2a primer pair. We also carried out blunt-end ligation of the same miRNA fragments with different long-range PCR products resulting from PCR amplification of βΔ5-8 with the 3′B4/5′B10 primer pair to obtain the β3′-miR-330 and β3′-let-7a3 constructs. The mutant version of the miR-330-β construct was generated by blunt-end ligation of PCR product from amplification of the βInt2-miR-330 construct with 330m-fw/330m-rev oligonucleotides. To obtain wild-type and Δ hybrid-miR-223 constructs, a fragment (referred to as the miR-23a promoter) from −603 bp to +36 bp relative to the transcription-initiation site of pri-miR-23a~27a~24-2 was PCR amplified using miR-23-603fw/miR-23+36rev oligonucleotides and inserted into pcDNA 3.1 (+) plasmid (Invitrogen), lacking the T7 promoter and BGH poly(A) site, using BglII and NheI sites. DNA fragments from −92 (−50) and −42 (Δ) to +242 bp relative to the mature miR-223 sequence, obtained from genomic HeLa DNA by PCR amplification with mi-R223SL-50/ and miR223SL Δ/ miR-223 +200 oligonucleotides, respectively, were inserted between the NheI and EcoRI sites of pcDNA 3.1, containing the promoter region of pri-miR-23a~27a~24-2.

Cell culture and transfection procedure

HeLa cells were maintained in DMEM medium supplemented with 50 IU ml⁻¹ of penicillin, 50 μg ml⁻¹ streptomycin and 10% (v/v) FCS. Cells were cultured overnight at 37 °C in 5% CO₂ to a confluency of 80–90% and then transiently transfected with test plasmid (10 μg) and Tat expression plasmid (1.5 μg) to allow activation of the HIV-1 promoter, using Lipofectamine 2000 (Invitrogen) reagent following the manufacturer’s instructions. RNA was extracted 24 h after transfection. RNAi procedures for XRN2, PMscl100, Drosha and DGCR8 knockdown have been described previously^14,33,37,47.

The NB4 cell line was maintained in RPMI 1640 medium supplemented with 50 IU ml⁻¹ of penicillin, 50 μg ml⁻¹ streptomycin and 10% (v/v) FCS. Retinoic acid was purchased from Sigma and used at a concentration of 1 μM for the indicated times²².

Chromatin immunoprecipitation assays

We carried out ChIP assays as described previously⁴⁸ with minor modifications. A total of 1 × 10⁷ HeLa cells were fixed in 1% (v/v) formaldehyde for 10 min at room temperature (25–30 °C). Chromatin was sheared to an average size of 0.3–0.8 kb and immunoprecipitated with anti-Drosha (Abcam), anti–Pol II (N-20, Santa Cruz), anti-DGCR8 (Protein Tech Group), anti-PMscl100 (E. DijK, Radbound University) and rabbit polyclonal XRN2 antibody⁴¹.

We carried out ChIP analysis of transfected plasmids with the following modifications: HeLa cells were washed in PBS and resuspended in 4 ml of HLB/Nonidet P-40 buffer (0.14 M NaCl, 1.5 mM MgCl₂, 10mM Tris-HCl, pH7.5, 0.5% (v/v) Nonidet P-40, 0.5 mM PMSF, 0.8μg ml⁻¹ pepstatinA, 1μg ml⁻¹ leupeptin). After incubating on ice for 5 min, nuclei were spun through 1 ml of underlayered HLB/Nonidet P-40 24% (w/v) sucrose buffer. Isolated nuclei were resuspended in 400 μl of nuclear lysis buffer (1% (w/v) SDS, 10mM EDTA and 50mM Tris, pH 8.1, 0.5 mM PMSF, 0.8μg ml⁻¹ pepstatinA, 1μg ml⁻¹ leupeptin) and incubated on ice for 10 min, and then the extracted chromatin was sonicated.

The immunoprecipitated DNA was recovered using the Qiaquick PCR purification kit (Qiagen). Quantitative PCR was performed with either a Rotorgene 3000 real-time PCR machine (Corbett Research) using Quantitec SYBR green (Qiagen) or in the presence of [α-³²P]dATP as described previously⁴⁹.

Northern blot analysis

Nuclear-cytoplasmic fractionation of HeLa RNA was carried out as previously described⁵⁰. 30 μg of cytoplasmic HeLa and total NB4 (extracted by Trizol reagent) RNA was fractionated on 10% urea-polyacrylamide gel and transferred to a GeneScreen Plus membrane (Perkin Elmer). All the antisense oligonucletide probes (listed in Supplementary Table 1) were end-labeled using T4 polynucleotides Kinase (Bio Labs) following the manufacturer’s guidelines.

Biotinylated selection probes

EX2 and Bio-int2 riboprobe templates were generated by PCR using βInt2-miR-330 construct and either T7-EX2R/EX2F or T7-b2R/b2F oligo pairs respectively. Both riboprobe templates were then in vitro transcribed in the presence of 0.35 mM Biotin-16-UTP (Roche) as described previously⁵¹.

Hybrid selection circular rapid amplification of cDNA ends

HscRACE was performed as described³³. EX2-H oligonucleotide was used to release the selected RNA from the streptavidin beads by RNase H (Roche) digestion. Reverse transcription of 5 μl ligated RNA mix was carried out using Superscript III reverse transcriptase (Invitrogen) following the manufacturer’s guidelines in the presence of βEX2rev oligonucleotide. We PCR amplified 2 μl cDNA with βEX2rev/EX2fw divergent primers using Taq Polymerase (Invitrogen). Bands were excised from an agarose gel, cloned into the TOPO vector (Invitrogen) and then sequenced.

Reverse-transcription polymerase chain reaction

1 μg of RNA was reverse transcribed using Superscript III reverse transcriptase (Invitrogen) in a 20-μl reaction. Taq polymerase (Invitrogen) was used under standard conditions to amplify 2 μl of cDNA (20-25 cycles). For quantitative real-time PCR, 2 μl of cDNA was analyzed using a Rotorgene 3000 real-time PCR machine (Corbett Research) in the presence of Quantitec SYBR green (Qiagen).

Fractionation of chromatin-associated and nucleoplasmic transcripts

Nuclear fractionation and RNA extraction from both the nucleoplasmic and chromatin-associated pellet fractions were performed as previously described^27,34.

Nuclear run on and hybrid selection nuclear run on

NRO and hybrid selection NRO analyses were carried out as described previously⁵¹.

Databases and EST analysis

The human miRNA list was retrieved from the miRBase registry release 10.1 (ref. 52). A BLAST search through the University of Calfornia in Santa Cruz (UCSC) web server⁵³ and a miRGen query⁵⁴ was initially carried out to identify the position of miRNA loci. After identifying the miRNA site, its position was compared with gene loci using RefSeq mRNA to identify miRNA matched with a protein-coding gene. For each sefseq entry, the UCSC database was used to indentify the features of all EST and splicing isoforms mapped to that region.