Nuclear Accumulation of an Uncapped RNA Produced by Drosha Cleavage of a Transcript Encoding miR-10b and HOXD4

Patterning of the animal embryo's antero-posterior (AP) axis is dependent on spatially and temporally regulated Hox gene expression. The murine Hoxd4 gene has been proposed to harbour two promoters, an upstream promoter P2, and a downstream promoter P1, that lie 5.2 and 1.1 kilobase pairs (kb) upstream of the coding region respectively. The evolutionarily conserved microRNA-10b (miR-10b) gene lies in the Hoxd4 genomic locus in the intron separating the non-coding exons 4 and 5 of the P2 transcript and directly adjacent to the proposed P1 promoter. Hoxd4 transcription is regulated by a 3′ neural enhancer that harbours a retinoic acid response element (RARE). Here, we show that the expression profiles of Hoxd4 and miR-10b transcripts during neural differentiation of mouse embryonal carcinoma (EC) P19 cells are co-ordinately regulated, suggesting that both Hoxd4 and miR-10b expression is governed by the neural enhancer. Our observation that P1 transcripts are uncapped, together with the mapping of their 5′ ends, strongly suggests that they are generated by Drosha cleavage of P2 transcripts rather than by transcriptional initiation. This is supported by the colocalization of P1 and P2 transcripts to the same posterior expression domain in the mouse embryo. These uncapped P1 transcripts do not appear to possess an Internal Ribosomal Entry Site (IRES), but accumulate within multiple punctate bodies within the nucleus suggesting that they play a functional role. Finally, similar uncapped Drosha-cleaved P1-like transcripts originating from the paralogous Hoxb4/miR-10a locus were also identified. We propose that these transcripts may belong to a novel class of regulatory RNAs.


Introduction
MicroRNAs are a class of highly conserved small noncoding RNAs (ncRNAs) expressed in a wide range of organisms [1,2]. Like proteins, many miRNAs are encoded by genes transcribed by RNA Polymerase II to give a long primary microRNA (pri-miRNA) transcript which is 59-capped and 39-polyadenylated. The pri-miRNA forms a hairpin-loop structure that is cleaved at its base by an RNAse III enzyme, Drosha, to form the precursor microRNA (pre-miRNA) which is exported out of the nucleus and then cleaved again on the loop side of the hairpin by another RNAse III enzyme, Dicer. This generates an miRNA: miRNA* duplex, one strand of which is preferentially selected and incorporated into the RNA-induced silencing complex (RISC). The single stranded mature miRNA is typically 21-23 nucleotides long and functions by base-pairing to target complementary mRNAs to regulate gene expression. In animals, this regulation occurs mostly, but not always, at the post-transcriptional level [3]. Recent evidence suggests that miRNAs are also able to epigenetically silence genes at the transcriptional level [4].
Homeobox (Hox) genes encode homeodomain-containing transcription factors that control segmental patterning and determine the identity of embryonic regions along the AP axis before and during gastrulation in the mouse [5,6]. They are highly conserved and found to be essential for normal development in all species where they have been tested [7]. Homeotic transformations and malformations in the embryo arise when Hox gene expression is deregulated by either a loss or gain of function, and the precise spatio-temporal control of their expression is therefore critical to normal development [6,8,9].
In mammals, there are 39 Hox genes organized in 4 paralogous Hox clusters, A-D (Fig 1A). There are three known microRNAs or miRNA families embedded in vertebrate Hox clusters: miR-10, miR-615 and miR-196 (Fig. 1A). The position of these miRNAs within the Hox clusters is highly conserved during evolution. For example, both the position and sequence of the miR-10 family are conserved in Drosophila, ancestral vertebrates, teleosts and mammals [10][11][12]. The Hoxb4 and Hoxd4 paralogs in mammals are orthologous to the Deformed Hox gene of flies. A miR-10 family member is embedded 59 to the coding region of each of these Hox genes. In mammals, the sequence of mature miR-10a and miR-10b differs by a single nucleotide. miR-10a, located upstream of Hoxb4, was reported to repress Hoxd4 transcription by targeting its promoter region in human breast cancer cells [13]. miR-10b is found 59 to Hoxd4 and regulates metastasis and cell migration in human breast cancer cells via suppression of Hoxd10 [14][15][16]. It is also known to be upregulated in malignant gliomas [17], pancreatic cancer [18] and chronic lymphocytic leukemia [19].
miR-10a and miR-10b are expressed in the central nervous system and trunk in a sub-domain of the Hoxb4 and Hoxd4 expression domains. This spatio-temporal restriction along the AP axis is reminiscent of Hox gene expression and is conserved in many animals such as Drosophila, zebrafish and mouse [20][21][22][23].
Hoxd4 patterns the anterior cervical skeleton of the mouse, and has been implicated in acute lymphoblastic leukemia [24][25][26][27]. Two Hoxd4 promoters, an upstream promoter termed P2 and a downstream promoter termed P1 (Fig. 1B) have been deduced from the results of S1 nuclease and RNase protection assays, and 59 Rapid Amplification of cDNA Ends (59RACE) [26]. While these are rigorous methods to map the 59 ends of RNA transcripts, they do not reveal whether these 59 ends are capped.
In the mouse, the full Hoxd4 expression domain in the developing central nervous system has an anterior limit at the boundary between rhombomeres 6 and 7 (r6/7) of the embryonic hindbrain. A region termed the 39 neural enhancer (39 NE) is essential for r6/7-restricted Hoxd4 expression in the central nervous system of the developing mouse embryo as well as in differentiating P19 cells [28][29][30][31][32][33][34]. However, a subset of Hoxd4 transcripts originating from an upstream promoter (P2) is expressed more posteriorly at a level just above the forelimb bud. miR-10b is expressed in a similar temporal and spatial pattern as the Hoxd4 P2 transcripts in the developing E9.5 mouse embryo, with an anterior expression border that is considerably posterior to the r6/7 anterior expression border [23,26]. The anterior expression border of miR-10a is likewise posterior to the r6/7 expression border of the full Hoxb4 domain [20]. In zebrafish, miR-10b is expressed in the spinal cord with an anterior boundary somewhat posterior to the r6/7 boundary [35]. The zebrafish miR-10 family is also found to repress hoxb1a and hoxb3a within the spinal cord in cooperation with hoxb4a [35]. The location of mouse miR-10b immediately adjacent to the presumptive P1 transcriptional start site and within the intron separating exons 4 and 5 of the P2 transcript ( Fig. 1B) raised several questions regarding Hoxd4 and the biogenesis of miR-10b. First, P2 may serve to encode both HOXD4 and miR-10b. In other words, the Hoxd4 P2 transcript could also serve as the primary miR-10b transcript that is the substrate for processing by Drosha. Second, expression of the Hoxd4/miR-10b transcript may be dependent on the neural enhancer located 39 to the Hoxd4 coding region. Third, the previously mapped 59 end of the Hoxd4 P1 transcript may not be generated by transcriptional initiation but by the cleavage of the Hoxd4 P2 transcript by Drosha (Fig. 1B).
Here, we show that these three hypotheses are correct. In addition, though P1 transcripts are not capped, they are abundant and largely localized to punctate sub-nuclear bodies.

Results
The expression of Hoxd4 and miR-10b is co-ordinately regulated in differentiating P19 cells To determine if the expression of pri-miR-10b is controlled by the Hoxd4 neural enhancer, the relative levels of Hoxd4 P1, P2 and miR-10b transcripts were measured in differentiating P19 mouse EC cells. P19 cells can be induced to undergo neuronal differentiation by treatment with all-trans retinoic acid (RA) and cell aggregation [36]. P19 neural differentiation recapitulates aspects of normal embryonic differentiation whereby Hox genes are expressed sequentially in a colinear manner [33,[37][38][39][40]. All Hoxd4 P1, P2 and miR-10b transcripts were barely detectable in undifferentiated P19 cells (Fig 2, P1, P2, miR-10b at 0 h). The accumulation of nestin transcripts, a neural lineage marker, served as a marker for the P19 differentiation process. Hoxd4 P1 and P2 transcripts were highly induced during P19 differentiation, peaking at maximum levels on day 3 of differentiation (Fig 2, P1, P2 at D3). In a fashion similar to the Hoxd4 transcripts, the miR-10b transcripts were first expressed at very low levels in undifferentiated cells and were then strongly induced upon RA treatment and aggregation, peaking at day 4 (Fig 2, miR-10b at D4). When compared to the Hoxd4 P1 and P2 transcripts, we observed that the miR-10b peaked a day later, on day 4. This delay in the expression peak may reflect differences in processing or stability; however, the overall expression profile is similar. This is in contrast to the expression profiles of other Hox genes such as Hoxa1 which is induced by 6 h and peaks as early as 48 h of neural differentiation [37,39].
In summary, miR-10b expression was shown to be induced together with Hoxd4 P1 and P2 transcripts during P19 differentiation. These data are consistent with a role for the Hoxd4 39 neural enhancer in directing expression of both Hoxd4 and miR-10b transcripts.
Hoxd4 transcripts originating from P1 are generated by Drosha cleavage To determine if Hoxd4 P1 transcripts are generated by Drosha cleavage or transcriptional initiation, an RNA-ligase mediated 59 Rapid Amplification of cDNA Ends (59 RLM-RACE) was performed. Transcripts whose 59 ends are generated by Drosha cleavage will bear a 59 phosphate and will be suitable substrates for RNA ligase, unlike intact transcripts synthesized by RNA pol II which will have a 59 7-methylguanosine (59 m 7 G) cap and therefore be unable to participate in RNA-ligase-mediated reactions.
Total RNA from P19 cells on day 3 of differentiation was collected. An aliquot was treated with calf intestinal phosphatase (CIP) to remove all free 59 phosphates in the total RNA. An aliquot of the CIP-treated RNA was subsequently treated with tobacco acid pyrophosphatase (TAP), which removes the 59 cap on mRNA transcripts to leave a free 59 phosphate. This was followed by ligation of all three RNA samples (untreated, CIP-treated and CIP/TAP-treated) to a synthetic RNA oligonucleotide and RT-PCR. Only transcripts bearing a 59 phosphate, such as produced by Drosha cleavage (Fig 1B), can be amplified from untreated RNA samples. The untreated RNA sample gave a band of approximately 190 bp after PCR, which corresponded to the 188 bp from the primer to the predicted Drosha cleavage site ( Fig  3A, arrowhead). Moreover, no comparably sized products were amplified from the CIP/TAP-treated RNA which would have allowed detection of capped transcripts ( Fig 3A). The mRNA for a b-actin control was amplified only following treatment with both CIP and TAP ( Fig 3A, arrow), as expected for a capped transcript.
Next, we confirmed the presence of these presumptive Droshacleaved P1 transcripts in the mouse embryo. The otic vesicle is a morphological landmark that lies just anterior to the Hoxd4 expression border at r6/7. Embryonic day (E) E9.5 mouse embryos were therefore bisected just anterior to the developing otic vesicle and RNA extracted from anterior (Hoxd4-negative) and posterior (Hoxd4-expressing) tissues followed by 59 RLM-RACE. The 188 bp band indicating the presence of Drosha-cleaved P1 transcripts was amplified only from posterior tissue ( Fig 3B). This demonstrates that the presumptive Drosha-cleaved P1 transcripts are generated during normal mouse embryonic development and only in embryonic regions expressing Hoxd4.
The approximately 188 bp band derived from P19 cell RNA was then cloned and sequenced. The 59 ends of all four clones began with 59-TATGG-39, mapping precisely to the predicted Drosha cleavage site on the pri-miR-10b transcript, 11 nt from the base of the pri-miRNA stem junction [41]. Importantly, this site is within one nucleotide of the previously deduced P1 start site cluster. Together, these observations suggest that P1 transcripts are generated by Drosha cleavage of the primary microRNA for miR-10b. The most likely origin for pri-miR-10b is a transcript initiating at P2 (Fig. 1B). In other words, pri-miR-10b and the Hoxd4 P2 transcript are one and the same.
Drosha cleavage of the Hoxb4/pri-miR-10a transcript generates similar uncapped Hoxb4 transcripts Similar to miR-10b and Hoxd4, miR-10a is located in a conserved position 59 to Hoxb4 (Fig. 1A). qPCR analysis showed that both miR-10a and Hoxb4 transcripts are induced in a similar manner to miR-10b and Hoxd4 transcripts during RA-induced P19 neural differentiation (unpublished observations). To determine if there was a Drosha-cleaved Hoxb4/pri-miR-10a transcript paralogous to Hoxd4 P1, we carried out 59-RLM-RACE with a Hoxb4-specific reverse primer (RACE-Hoxb4-R). Only the untreated RNA sample gave an amplification product of 300 bp corresponding to the distance between the reverse primer and the predicted Drosha cleavage site (Fig 3C, arrow). Sequencing of the Hoxb4 PCR product confirmed that it is indeed Hoxb4-specific, mapping it to the predicted Drosha cleavage site precisely 11 nt from the base of the pri-miRNA stem junction, just as for Hoxd4 (Fig 3D). This indicated the presence of similar uncapped Drosha-cleaved Hoxb4 transcripts in differentiating P19 cells, suggesting that they may belong to a novel class of RNA species.
Drosha-cleaved Hoxd4 P1 and P2/pri-miR-10b transcripts are all expressed posterior to the rhombomere 6/7 boundary in the mouse embryo Using a probe against the Hoxd4 59 coding and non-coding region (P1+P2 ex5, Fig. 4A), Hoxd4 transcripts have been shown to have an anterior expression boundary between r6 and r7 within the developing hindbrain of mice. This expression border is conserved in zebrafish [23,26]. By contrast, a probe that specifically detected P2 transcripts (probe name = P2 ex4, Fig. 4A) revealed a more posterior expression boundary in the anterior spinal cord above the forelimb bud [26]. Thus, the anterior part of the Hoxd4 expression domain up to r6/7 was attributed to the activity of a distinct P1 promoter. However, given our observation that P1 transcripts are likely generated by Drosha cleavage of P2 transcripts, and therefore do not originate from a transcriptional start site, we re-visited the spatial distribution of P1 and P2 transcripts in the mouse embryo.
We designed two new in situ hybridization probes to more precisely investigate the expression of these different transcripts in the embryo. First, a P1 probe was designed to be just downstream of miR-10b and spanning the 59 untranslated region (59 UTR) of the Drosha-cleaved P1 transcripts (probe name = P1, Fig. 4A). This region is spliced out in the P2 transcripts and thus the P1 probe will specifically detect P1 transcripts only. A second probe (probe name = P2 ex 1-3, Fig. 4A), spans the first 3 exons of the P2 transcript and detects P2 transcripts only. In situ hybridizations with both the P1 and P2 ex 1-3 probes showed an expression domain that is posterior to the r6/7 boundary, similar to the expression domain of mature miR-10b in the E9.5 embryo. Neither probe detected more anterior expression up to the r6/7 boundary (Fig 4B, C, D). We conclude that P1 transcripts are derived by Drosha-cleavage of transcripts originating at the posteriorly active P2 promoter. The more anterior expression (up to r6/7) previously detected with probe P1+P2 ex 5 may be due to the action of a more anteriorly active promoter (hypothetical transcript P3, Fig. 4A). Alternatively, P2 may be the only Hoxd4 promoter, but the presence or absence of destabilizing elements due to differential splicing determines whether P2-derived transcripts accumulate at anterior or posterior positions (hypothetical transcript P2.3, Fig. 4A).

Drosha-cleaved P1 transcripts are abundant in the nucleus
We determined that P1 transcripts are generated by Drosha cleavage. However, these uncapped P1 transcripts appeared to be abundant in differentiating P19 cells (Fig 2) and the mouse embryo ( Fig 3B) [26]. We therefore asked if these abundant, uncapped P1 transcripts may have a function. As a first indication, we determined whether the uncapped P1 transcripts were present in the cytoplasm or nucleus.
Nuclear-cytoplasmic fractionation of neurally differentiating P19 cells was performed, followed by RLM-PCR as described above. The controls -47S pre-rRNA and snoU6 RNA -were correctly enriched in the nuclear fraction while b-actin mRNA was more abundant in the cytoplasmic fraction as expected (Fig 5A). We found almost three quarters of Hoxd4 P1 transcripts in the nuclear fraction.

Absence of robust IRES activity within the 59 UTR of P1 transcripts
The presence of approximately one quarter of P1 transcripts in the cytoplasm suggested that they may be translated via an IRES. To determine whether the 59 UTR of the cleaved P1 transcript has IRES activity, an IRES reporter assay was designed. The IRES reporter consisted of a firefly luciferase cassette driven by a CMV promoter, followed by the putative IRES region driving lacZ gene expression and ending with an SV40 polyadenylation signal. The ''Inverted IRES'' construct had the putative IRES region inverted and cloned in the same position within the vector and served as a negative baseline control (Fig 5B). Any increase in b-galactosidase levels of the IRES construct compared to the Inverted IRES construct would thus reflect IRES activity.
The IRES test plasmids were co-transfected with control renilla luciferase expression vectors in either HEK293T cells or RAtreated P19 cells. Cell lysates were assayed at 48 h posttransfection and b-galactosidase was normalized to renilla luciferase activity to account for variations in transfection efficiency. No significant IRES activity was observed in the HEK293T cells (Fig 5C). There was also no significant b-galactosidase activity in transfected and RA-treated P19 cells compared to untransfected cells (data not shown). This suggests that the Drosha-cleaved P1 transcripts are not translated through an IRES. This result also indicates the absence of any additional promoter activity in the putative IRES region that stretches for about 1 kilobase pairs upstream from the Hoxd4 coding region. Efficient cleavage of the IRES reporter fusion transcript by Drosha was evident from the fact that firefly luciferase activity derived from this reporter was low by comparison to that obtained with the Inverted IRES reporter (data not shown).

Hoxd4 P1 transcripts localize to punctate nuclear bodies
We performed RNA fluorescence in situ hybridization (RNA FISH) to determine sub-localization of nascent Hoxd4 and P1 transcripts in day 3 differentiated P19 cells. A Cy3-labelled Hoxd4 P1 probe spanning 1.5 kb extending from the Drosha cleavage site to the Hoxd4 coding region (Fig 4A) yielded multiple signals corresponding to punctate nuclear bodies (Fig 5D, middle and  bottom rows). A negative control Cy3-labelled probe (pCR2.1 control) yielded no specific signal (Fig 5D, top row). Likewise, undifferentiated P19 cells which do not express Hoxd4 did not yield Black arrows show the Drosha cleavage sites on the pri-miR-10b/a hairpin. The cleavage site is exactly 11 bp from the bottom of both the pri-miR-10b and pri-miR-10a stem junction on the downstream side. This is within a single nucleotide of the previously mapped P1 start site (a cluster of 4 nt underlined and denoted ''P1'' on pri-miR-10b). doi:10.1371/journal.pone.0025689.g003

Discussion
We have established that the 59 ends of Hoxd4 P1 transcripts are not capped, bear a terminal phosphate and map to the predicted Drosha cleavage site at the base of the stem of the pri-miR-10b stem-loop. The 59 end of the Hoxd4 P1 transcript is precisely 11 nucleotides from the base of the pri-miRNA stem junction. A similar result was obtained for the Drosha cleavage product of miR-10a and Hoxb4. This result is consistent with a validated model in which Drosha cleaves at a position 11 nt from the stem-ssRNA junction [41]. We conclude that Hoxd4 P1 transcripts are indeed the result of Drosha cleavage of the pri-mir-10b transcript and, contrary to our previous interpretation [26], are not generated by transcriptional initiation from a distinct promoter, but by the action of Drosha on transcripts initiated at the P2 promoter.
A shared promoter and common regulatory elements for Hoxd4 and miR-10b A significant majority of human miRNAs resides in intronic regions and in the same orientation as the host coding genes [42]. The expression patterns of these intronic miRNAs frequently coincide with the genes in which they are embedded, indicating that they could be regulated by common cis-regulatory elements [43]. Expression of Hoxd4 transgenes in P19 cells and in transgenic mouse embryos is critically dependent on a 39 neural enhancer [28,[31][32][33][34]. The expression profiles of Hoxd4 and miR-10b during P19 differentiation as measured by qRT-PCR are similar, with low basal expression levels in undifferentiated P19 cells and strong induction of their expression upon neural differentiation, peaking at day 3 or 4 and declining thereafter. This is in contrast to the expression profiles of other Hox genes in P19 cells such as Hoxa1 which has been shown to be induced as early as 6 hours of neural differentiation [37]. Our data are therefore consistent with both Hoxd4 and miR-10b transcripts coming under the control of this same regulatory region.
Primary miRNAs transcribed by RNA polymerase II are 59 capped and 39 polyadenylated, making them structurally identical to messenger RNAs. It has been reported that human pri-miRNA transcripts also give rise to mRNA coding for a protein [44]. This is consistent with our observation that transcripts initiating at Hoxd4 P2 have the potential to encode both miR-10b and HOXD4. Such an arrangement would facilitate the co-regulated expression of both Hoxd4 and miR-10b during development, as activation of the P2 promoter would lead to production of both miR-10b and HOXD4. This is supported by in situ data that showed an extensive overlap of mature miR-10b and Hoxd4 expression both spatially and temporally [23,26]. The coordinated regulation of both genes suggests that they may have shared functions during early development such as has been described for the shared repressive functions of the miR-10 family and hoxb4 in zebrafish [35].

Expression along the antero-posterior axis
The full Hoxd4 expression domain extends anteriorly to the boundary between r6 and r7 in the developing hindbrain (Fig. 4D, combined blue plus red shading) as revealed by in situ probes overlapping the 59 or 39 ends of the coding region [26,34,40,45]. However, miR-10b and Hoxd4 P2 and P1 transcripts are not detected in the anterior-most Hoxd4 expression domain up to r6/7 [26]. If a single P2 promoter drives expression of all transcripts derived from the Hoxd4 locus, then there must be posttranscriptional controls which prevent some RNAs from accumulating in anterior tissues up to r6/7. In one possible mechanism, P2 transcripts that include a hypothetical destabilizing element (grey triangle in Fig. 4A) located in the Hoxd4 59 UTR are unstable and degraded in the anterior-most part of the Hoxd4 expression domain up to r6/7 (Fig 4D, shaded red). In this scenario, only alternatively spliced transcripts that lack this destabilizing element, such as the hypothetical P2.3 transcript shown in figure 4A, accumulate in the anterior-most domain.
Alternatively, an as-yet uncharacterized promoter (hypothetical promoter P3, Fig. 4A) is active in anterior neural tissue and is responsible for expression in the anterior-most portion of the Hoxd4 expression domain. This is supported by the presence of cDNA and Expressed Sequence Tag (EST) clones (FANTOM) whose 59 ends map immediately upstream of the Hoxd4 coding region. In addition, on the basis of primer extension results, a putative human HOXD4 promoter has been mapped 21 bp 59 of the ATG start codon [46]. However, such Hoxd4 transcripts may not be expressed in the neural lineage, and we have been unable to detect Hoxd4 59 ends in the P3 region despite extensive S1 nuclease and RNase protection assays [26] and 59RLM-RACE (unpublished data). In addition, the results presented in figure 5C fail to reveal promoter activity in the 1 kb sequence upstream of the Hoxd4 ATG start codon. The lack of a promoter at presumptive P3 is further substantiated by a low density of elongating RNA pol II, p300, TBP and H3K4me3 adjacent to the Hoxd4 coding region. Supporting one or more far upstream promoters, the density of these four factors is high in a broad region spanning P2 (ChIP-seq data from CH12 cells from ENCODE/Stanford/Yale displayed on the UCSC Genome Browser) [47].
The relationship between the miR-10 family and Hox4 genes is surprisingly well conserved through evolution. In the zebrafish genome, three miR-10 members, miR-10b-1, miR-10b-2, and miR-10c, are positioned upstream of the Hox group 4 paralogs hoxd4a, hoxc4a and hoxb4a, respectively [48]. The remaining two miR-10 family members, miR-10c and miR-10d, are located at homologous positions near sites from which 4 th group paralogs have been lost in the HoxBa and (vestigial) HoxDb clusters [48]. The sequences of mature miR-10b-1 and miR-10b-2 are identical and they are expressed slightly posterior to the r6/7 boundary, reminiscent of the situation in the mouse embryo [23,26,35].
IRES region in front of the lacZ gene and SV40 polyadenylation signal. The inverted IRES construct consists of the same components except that the putative IRES region, which is about 1 kb in length, is inverted. Boxes indicate exons as per figure 1B. (D) RNA FISH of P19 cells on day 3 of neural differentiation. Vertical columns show images obtained under excitation for Cy3, GFP and DAPI. Images obtained with excitation for GFP control for autofluorescence. The upper row shows a negative control Cy3-labelled probe (pCR2.1 control) and the middle and lower rows show the localization of the Cy3-labelled Hoxd4 P1 probe (Hoxd4 P1). Hoxd4 transcripts are specifically detected as speckles in the nucleus of neurally differentiated P19 cells. Images in the bottom row have been digitally enlarged in order to present the punctate bodies more clearly. The Adobe Photoshop application was used to enhance the contrast simultaneously to pairs of images in the Cy3 and GFP columns. Arrows denote punctate bodies visible with the Cy3labelled Hoxd4 P1 probe. These bodies are not visible when the sample is excited for GFP and do not correspond to chromatin structures detected by DAPI. doi:10.1371/journal.pone.0025689.g005 Long intergenic primary Hox transcripts have been documented from the earliest days of the field [49]. More recently, highresolution transcriptional profiling of the Hox clusters has revealed extensive polycistronic transcripts and a high degree of transcriptional complexity within the mammalian Hox clusters [50]. This increases the possible sources of transcripts acting as primary microRNAs and/or Hox messenger RNAs. A long primary transcript (Genbank BK005082) originating downstream of zebrafish hoxb5a spans miR-10c, hoxb4a and hoxb3a and is expressed in a domain posterior to hoxb4a [35,51]. A similar situation exists in mouse where a long-range Hoxd3 transcript initiated from the Hoxd4 P2 promoter has the potential to code for miR-10b as well (NM_010468). Thus, it may be generally true of Hox complexes that long intergenic transcripts governed by a variety of enhancers contribute to miRNA accumulation at different points along the antero-posterior axis.

Putative function of Drosha-cleaved P1 transcripts
Although uncapped transcripts are typically degraded rapidly [52], the Hoxd4 P1 product of Drosha cleavage appears to be both stable and abundant in the mouse embryo and neurally differentiating P19 cells. We have also shown the presence of uncapped Drosha-cleaved Hoxb4 transcripts (Fig 3C), leading us to speculate that this may be a new class of RNA species with novel functions.
While P1 transcripts might be translated, they do not appear to have an IRES necessary for cap-independent translation (Fig 5C). By contrast, however, P1 transcripts are abundantly localized to punctate nuclear bodies in differentiating P19 cells. There is now widespread evidence for the prevalence and importance of actively transcribed ncRNAs in mammalian cells [53][54][55]. The highly regulated expression of these ncRNAs in a temporal or spatial manner is suggestive of functional significance. Some Hoxassociated ncRNAs which are 59 capped, spliced and polyadenylated, such as HOTAIR and HOTTIP were found to interact with Polycomb/PRC2 and Trithorax/MLL complexes, respectively, controlling histone methylation and chromatin remodeling and thereby silencing or activating transcription [56,57]. Intriguingly, the HOTAIR ncRNA employs a trans mode of action by silencing transcription at a distant HOXD locus while the HOTTIP ncRNA activates gene expression in cis, on the proximal HOXA genes. It is thus conceivable that the Hoxd4 P1 transcripts may perform similar roles by binding chromatin remodeling complexes to direct transcription of genes in trans or cis. Further insight will require the use of appropriate probes to co-localize P1 transcripts to one of the known sub-nuclear structures such as paraspeckles [58].

Ethics statement
All animal procedures were performed according to the Singapore A*STAR Biopolis Biological Resource Center (BRC) Institutional Animal Care and Use Committee (IACUC) guidelines and the IACUC protocols employed were reviewed and approved by the aforementioned committee before any animal procedures were undertaken for this study described here (IACUC Protocol No: 080348 and 080377).

Cell culture and differentiation
P19 cells (American Type Culture Collection #CRL-1825) were cultured in DMEM (Gibco) supplemented with 10% fetal bovine serum. For differentiation, the P19 cells were seeded at a density of 1610 5 cells/ml in a 10 cm untreated polystyrene bacterial petri dish (Greiner Bio-One, Germany), treated with 0.3 mM of all-trans retinoic acid and allowed to aggregate. This was designated as day 0. Fresh media with 0.3 mM RA was replaced after 2 days. After 4 days, fresh media without RA was replaced and the cells plated onto tissue culture plates (Corning, USA).

Quantitative reverse transcriptase PCR
RNA was extracted using the PureLink TM Micro-to-Midi TM Total RNA Purification System (Invitrogen). 1 mg total RNA was first treated with 1 U DNase I (Fermentas) at 37 uC for 30 min. Reverse transcription of mRNA was carried out with SuperScriptH III First-Strand Synthesis (Invitrogen) and reverse transcription of miRNA was carried out with NCode TM miRNA First-Strand cDNA Synthesis (Invitrogen) as per the manufacturer's instructions. The cDNA was then used for quantitative PCR (qPCR) together with SYBR GreenER TM qPCR SuperMix (Invitrogen) on a BioRad iCycler iQ5. 18S RNA was used as internal control for the mRNAs while U6 RNA was used as an internal control for the microRNAs.
RNA ligation mediated 59 rapid amplification of cDNA ends RNA was extracted from P19 cells on day 3 of differentiation using the PureLink TM Micro-to-Midi TM Total RNA Purification System (Invitrogen) and 59 RACE performed as per manufacturer's instructions (GeneRacerH Kit, Invitrogen).

RNA in situ hybridization
Embryos were fixed overnight at 4uC in 4% paraformaldehyde and subsequently processed for 10 mm paraffin-embedded sections as described [59]. RNA in situ hybridization with digoxigenin (DIG)-labeled probes was performed as previously described [60]. The following plasmids were used as templates for synthesizing antisense DIG-labeled RNA probes: pGEMT Hoxd4 uP1 and pCR4-TOPO Hoxd4 P2 exon1-3. Following hybridization and washing, sections were stained with NBT/BCIP and exposed overnight at 4uC in dark according to manufacturer's instructions (Roche). Sections were subsequently washed in PBS and mounted with glycerol gelatin. All sections were photographed using a Zeiss Axio Imager Z1.

RNA FISH
Day 3 neurally differentiated P19 cells were typsinized and diluted to a concentration of 7610 5 cells/ml before being cytospun onto glass slides using a cytocentrifuge, CytoSpin 4 (Thermo Scientific). Cells were then washed with ice-cold PBS for 5 min followed by fixation at room temperature in 4% paraformaldehyde (PFA) for 10 min. The slides were stored in 70% ethanol at 4uC. Before use, the slides were sequentially dehydrated through 80, 90 and 100% ethanol for 2 min each. The RNA FISH probes were labelled with the Nick translation kit and Cy3-dUTP according to the manufacturer's instructions (Roche). Cot-1 mouse DNA was added to the Cy3-labelled probe and stored in hybridization buffer (50% formamide, 2x SSC pH 7.4, 2 mg/ml BSA, 10% Dextran Sulfate-500K) with a final concentration of 500 ng/ml and 50 ng/ ml respectively. Hybridization was carried out at 42uC for 3 hours, followed by three washes of 5 min each in 50% (v/v) formamide in 2x SSC and another three washes of 5 min each in 2x SSC at 45uC. All cells were imaged using a Nikon Eclipse Ti.

Nuclear-cytoplasmic fractionation
Day 3 neurally differentiated P19 cells (,10 6 cells) were pelleted and washed twice in PBS. Cells were then resuspended in 0.5 ml chilled lysis buffer (10 mM Tris pH 7.4, 3 mM MgCl 2 , 10 mM NaCl, 150 mM sucrose, 0.5% NP-40) with 10-20 U of RNase-OUT (Invitrogen) and kept on ice for 5-10 min with gentle mixing every minute. The cell lysate was centrifuged at 250 g in a microcentrifuge at 4uC for 5 min. Supernatant containing the cytoplasmic fraction was collected. The nuclear pellet was washed twice with 1 ml lysis buffer without NP-40 and resuspended in 100 ml of lysis buffer. RNA extraction of both nuclear and cytoplasmic fractions was performed with RNA mini kit (Invitrogen) followed by phenol-chloroform (pH 4.7) extraction and ethanol precipitation. RNA was then resuspended in equal volumes of DEPC-treated water. Equal volumes (1-2 ml) of the total RNA obtained from the nuclear and cytoplasmic fractions was then individually ligated to the RNA oligonucleotide adaptor using T4 RNA ligase (Fermentas). Reverse transcription of mRNA was carried out with SuperScriptH III First-Strand Synthesis (Invitrogen). The cDNA was then used for qPCR together with SYBR GreenER TM qPCR SuperMix (Invitrogen) on a BioRad iCycler iQ5.

IRES plasmid transfection
293T cells (American Type Culture Collection #CRL-11268) were seeded at a density of 1610 5 cells/ml in a 12-well plate and allowed to attach overnight. Transfection was performed the next day with Lipofectamine 2000 (Invitrogen) and 1.2 mg of IRES or inverted IRES plasmids together with 50 ng of renilla reporter plasmid. Cell lysate was extracted 2 days after transfection. b-gal activity was measured using the Dual-LightH Combined Reporter Gene Assay System (Applied Biosystems) and normalized to renilla activity which was measured using the Dual-Luciferase TM Reporter Assay System (Promega) on a Fluoroskan Ascent FL (Thermo). Experimental triplicates were performed.

Plasmid constructs
The pGEMT Hoxd4 uP1 plasmid used for mouse in situ hybridizations was made by T-tailed cloning of an approximately 700 bp PCR fragment amplified from pSNlacZpA (Zhang et al. 2000) starting from the P1 start site to the 39 splice acceptor of Hoxd4 intron 4 into the pGEMT vector (Promega).
The pCR4-TOPO Hoxd4 P2 exon 1-3 plasmid used for mouse in situ hybridizations was made by T-tailed cloning of an 287 bp PCR fragment amplified from cDNA of P19 day 4 differentiated cells, comprising of the first three Hoxd4 exons into pCR4-TOPO vector using the TOPOH TA CloningH Kit (Invitrogen).
The non IRES Reporter plasmid was made by cloning a 3.2 kb fragment amplified from PSNlacZpA (Zhang et al. 2000) starting from the Hoxd4 ATG start codon and extending throughout the entire lacZ coding sequence into SpeI and PmeI sites in the pMIR-REPORT Luciferase plasmid from the PSNlacZpA (Zhang et al. 2000) starting from approximately 140 bp upstream from the Drosha cleavage site of miR-10b to the sequence just 59 of the Hoxd4 ATG start codon into SpeI sites in the non IRES Reporter plasmid.