Certain Adenylated Non-Coding RNAs, Including 5′ Leader Sequences of Primary MicroRNA Transcripts, Accumulate in Mouse Cells following Depletion of the RNA Helicase MTR4

RNA surveillance plays an important role in posttranscriptional regulation. Seminal work in this field has largely focused on yeast as a model system, whereas exploration of RNA surveillance in mammals is only recently begun. The increased transcriptional complexity of mammalian systems provides a wider array of targets for RNA surveillance, and, while many questions remain unanswered, emerging data suggest the nuclear RNA surveillance machinery exhibits increased complexity as well. We have used a small interfering RNA in mouse N2A cells to target the homolog of a yeast protein that functions in RNA surveillance (Mtr4p). We used high-throughput sequencing of polyadenylated RNAs (PA-seq) to quantify the effects of the mMtr4 knockdown (KD) on RNA surveillance. We demonstrate that overall abundance of polyadenylated protein coding mRNAs is not affected, but several targets of RNA surveillance predicted from work in yeast accumulate as adenylated RNAs in the mMtr4KD. microRNAs are an added layer of transcriptional complexity not found in yeast. After Drosha cleavage separates the pre-miRNA from the microRNA's primary transcript, the byproducts of that transcript are generally thought to be degraded. We have identified the 5′ leading segments of pri-miRNAs as novel targets of mMtr4 dependent RNA surveillance.


Introduction
Transcriptional activity in mammalian genomes, once considered limited, is now being viewed differently as more sophisticated techniques are developed and used to interrogate transcriptomes. The sheer abundance of regulatory RNAs (rRNA, snRNA, snoRNA and tRNA) fostered their early discovery and characterization [1][2][3][4][5][6], whereas high-throughput sequencing (HTS) methods continue to uncover an ever increasing population of rare RNAs. The use of HTS to survey mRNA reveals extensive transcript diversity due to alternative splicing, as well as use of alternative transcription start and polyadenylation sites [7][8][9]. These methods have also demonstrated the vast extent to which sequences previously characterized as 'junk DNA' are transcribed, accelerating the discovery, classification and characterization of a wide array of non-coding RNAs (ncRNAs) [10][11][12][13][14][15][16][17][18][19].
Following transcription, RNAs are routinely subject to a series of processing steps ranging from internal cleavage, trimming of ends, RNA editing, and in some cases, the introduction of covalent modifications into the nascent transcript. The production or processing of an RNA occasionally results in the introduction of errors, whereby important features critical to RNA function are disrupted. These defective RNAs can interfere with normal cellular functions, and have been linked to a host of diseases, most prominently neurodegenerative diseases and cancer [20][21][22][23][24][25].
To prevent accumulation or deployment of defective RNAs, eukaryotes invoke RNA surveillance as a ''quality control'' step that can identify and destroy aberrant RNAs [26][27][28][29][30]. Surveillance also recycles processed RNA intermediates for use in de novo rounds of transcription. Once mRNA leaves the nucleus, additional quality control occurs in the cytoplasm (Reviewed in [31]).
The surveillance and degradation machinery appears to be well conserved. Exosome complexes are found from archaea to humans [44][45][46], and homologs of the proteins that comprise the TRAMP complex are widely conserved among eukaryotes [47][48][49][50][51][52]. Experimental exploration of RNA surveillance has expanded to mammalian systems relatively recently [50,51]. One study in humans suggests a strong division of labor in targeting various RNAs to the exosome. Localization analyses suggest that the TRAMP complex may be restricted to the nucleolus, whereas the Nuclear Exosome Targeting (NEXT) complex is excluded from the nucleolus [49]. The common component in both of these complexes is the hMTR4 protein, which exhibits a strong interaction with the hRRP6 protein of the exosome [49]. While RNA targets of TRAMP and the exosome have been extensively characterized in yeast, similar knowledge is relatively scarce in mammalian systems. The transcriptome in mammals is more complex than yeast, both within protein coding genes as well as intergenic regions, now more fully appreciated as transcriptionally active regions producing numerous ncRNAs of mostly unknown function [10,15,19]. Thus, the identification and characterization of poly(A)+ RNAs that accumulate upon depletion of a TRAMP homolog subunit, or protein(s) required for RNA surveillance, is useful as an initial exploration of post transcriptional control of RNA expression in mammals.
To identify RNA targets of Mus musculus nuclear RNA surveillance, we used a small interfering RNA (siRNA) to deplete the RNA dependent ATPase component of the TRAMP and NEXT complexes, designated as Skiv2l2 (mMtr4) in mouse N2A cells grown in culture. We hypothesized that targets of mammalian TRAMP or NEXT complexes would accumulate in the mMtr4knockdown (mMtr4KD) as depletion of mMTR4 would impede degradation of these targets by the exosome. However, such targets should be successfully adenylated by the mTrf4 homolog, PAPD5, such that they could be identified based upon their poly(A) tails.
We used an RNA-seq strategy specifically designed to capture adenylated transcripts and precisely map the 39 end of the genomic template (PA-Seq; [53]). Our analyses identified poly(A)+ RNAs that accumulate significantly more in the mMtr4KD than in the mControlKD. Our data support conserved roles for mMtr4 in processing rRNA 59ETS for degradation, and in snoRNA surveillance. We identify a novel role for mMtr4 in targeting the 59 leader sequences of microRNAs for degradation.

Materials and Methods
Tissue culture and siRNA knockdown N2A cells (ATCC CCL-131; [54,55]) were cultured in DMEM with 10% fetal bovine serum at 37C. Freshly passaged N2A cells were plated in 5 ml of serum containing medium at 5610 5 cells per 60 mm tissue culture dish and incubated 24 hours prior to transfection with siRNA. Independent culture dishes were transfected in triplicate with a 1 ml suspension of siRNA in Opti MEM medium and Lipofectamine 2000 (Invitrogen) according the manufacturer's recommendations to a final concentration of 67 nM. Pre-designed siRNAs were obtained from Ambion: skiv2l2 (Cat#177475) targeted to mouse Mtr4, and ''Negative control #1'' (Cat #AM4611). After 48 hr incubation with the siRNA, medium was removed, and cells from individual plates were harvested. For western blot analyses, cell pellets were resuspended in lysis buffer and quantified using a Bradford assay. For RNA experiments, cells were directly resuspended with TRIzol Reagent, total RNA was isolated according to manufacturer's recommendations, and quantified using a Nanodrop ND-1000 instrument.

Determination of knockdown efficiency
Efficiency of the siRNA knockdown was confirmed at both RNA and protein levels. Reverse transcription reactions were completed using SuperScript II (Invitrogen) according to manufacturer's recommendations with an oligo-dT 18 primer. Primer sets were designed to span an intron to minimize the risk that genomic DNA contamination could result in PCR amplification. Moreover, all primer sets were subjected to melt-curve analysis to confirm single amplicon yield, and tested on Mock-RT reactions (no Reverse Transcriptase added) generated for each RNA sample to further demonstrate amplicons were specifically derived from an RNA template. Table S4 reports all primers used in this study, and includes information regarding amplicon size and intron(s) spanned.
Real-Time PCR reactions were performed in triplicate using iQ SYBR green master mix (BioRad) and equivalent aliquots of cDNA, with cycling reactions carried out in a My iQ thermocycler (BioRad). Real-Time PCR was used to quantify the level of Mtr4 mRNA relative to CyclophilinB in the control versus knockdown cell lines, thereby determining the relative extent to which the target mRNA had been knocked down. Upon confirmation of efficient knockdown of the target mRNA, these total RNAs were further purified for deep sequencing using an RNeasy kit (QIAGEN) with on-column DNaseI digestion.
Western blot analyses were performed using three independent siRNA treated cell cultures to demonstrate that Mtr4 protein levels were successfully knocked down. Anti-Mtr4 (Skiv2l2) was obtained from LSBio and anti-b actin from Thermo Scientific was used as a loading control.

PA-seq
Deep sequencing of the control vs. Mtr4 knockdown RNAs was performed largely as described [53,56] with the notable exception that (5 ug of each) RNA were first fragmented with magnesium, and reverse transcribed using a biotinylated oligo dT (16) /deoxyU/ TTTVN-39 primer to selectively identify adenylated RNAs. cDNAs were selected using Invitrogen's MyOne Streptavidin Dynabeads, dephosphorylated using APex Heat-Labile Alkaline Phosphatase (Epicentre), and digested with the USER enzyme (NEB) to selectively cut the RT primer at the Uracil, which released the dephosphorylated cDNAs from the Dynabeads, leaving a three T tag immediately adjacent to the complementary nucleotide representing the 39 terminal residue of the adenylated RNA. Following this step, and between each of the subsequent modification steps, the released cDNAs were purified using the DNA clean and concentrator-5 kit (ZYMO Research). Modifications included removal/blunt ending of the 39 poly(A) overhang remaining after USER digestion using T4 DNA polymerase (NEB), A-tailing using an exo-Klenow fragment (Epicentre), and ligation of the Barcoded Y-linker (Illumina) using T4 DNA ligase (NEB). Resulting cDNAs were separated on an agarose gel (E-gel; Invitrogen), and products in the 250-450 bp size range were excised and purified using a gel purification kit (ZYMO Research). Products were amplified for 16 cycles of PCR (initial denaturation of 30 s at 98uC; then 16610 s at 98uC, 10 s at 67uC, and 30 s at 72uC; and a final extension of 10 min at 72uC) using Phusion DNA polymerase (Finnzymes), and HPLC purified primers from IDT (PE_PCR_AF: 59-AATGATACGGCGACCACCGA-GATCTACACTCTTTCCCTACACGACGCTCTTCCGATC-T-39; PE_PCR_BR: 59-CAAGCAGAAGACGGCATACGA-GATCGGTCTCGGCATTCCTGCTGAACCGCTCTTCCG-ATCT-39). Products for sequencing were size selected (,250-400 bp) from an agarose gel.
Paired-end sequencing was completed on an Illumina Hi-Seq2000 instrument. Raw sequencing reads from the Illumina platform were sorted based upon perfect match to a sample specific barcode (at the first 5 bases of read 1), trimmed, and finally split into separate fastq files for the 39 and 59 paired-end reads. Project data have been deposited to the NIH short read archive (SRP041394).

Mapping of Paired-End Reads
Illumina sequencing data was sorted relative to barcode, and trimmed accordingly. The 39 end of the sequence was evident based upon the presence of the paired end read that initiated with three T's. Paired end mapping to the mouse genome (mm9) was completed largely as described [56] -briefly: using the Burrows-Wheeler Aligner (BWA) mapping algorithm [57] to identify positions within the genome that differ by no more than 2 bp for the 39 read, include a corresponding match to the 59 read in the correct orientation (k = 2 & n = 2). Each paired end read that was successfully mapped in this manner was reported as a 'hit' for adenylation being initiated at the specific nucleotide in the genome that corresponded with the nucleotide adjacent to the initiating TTTs of the 39 read. Sequence reads that corresponded with highly repetitive genomic elements were ignored. Paired end reads that did not map to a unique genomic position, but matched fewer than ten genomic positions, were each randomly assigned to one of those genomic positions. In general, adenylation sites with fewer than five mapped sequencing reads were ignored.

Identification of Adenlyation Peaks
Clusters (a.k.a. peaks of adenylation) were identified largely as previously described [56] using the F-Seq program [58] with the exception that Narrow Peaks were defined as those whose 95% interval was less than 10 bp wide. PA clusters that correspond to a string of adenosines (either six continuous, or 15 out of 20 nt) encoded in the genomic DNA were removed from future analysis. Although their identification within our data suggests that these genomic regions are transcribed, they cannot be definitively attributed to 39 adenylation versus template adenosines potentially internal to the relevant transcript. A series of Python scripts and the Galaxy portal (main.g2.bx.psu.edu) were used for additional comparative analyses among the sequencing replicates [59][60][61].

Data Normalization
All data were normalized to account for the sometimes significant difference in sequencing depth between sequencing runs for each of the knockdowns. The total number of PA 'hits' at a given genomic location were normalized to reads per million for that knockdown sequencing run. Rather than pooling the sequencing results from biological replicates, most data are reported from one sequencing run (samples designated -1 in Table 1) where the total number of reads were comparable among the knockdowns, but the additional sequencing run always exhibited consistent trends.

Molecular validation of differential polyadenylated transcripts
Abundance of adenylated transcripts in the mMtr4KD vs. mControlKD was independently validated using qPCR. RNA was reverse transcribed using oligo-dT 18 and either M_MLV (Promega) or SuperScript II (Invitrogen) reverse transcriptase according to manufacturer's recommendations. Equivalent aliquots of each RT reaction were used in triplicate qPCR reactions assaying the levels of adenylated U3 snRNAs, and of the 59 leader vs. the pri-miRNAs of several miRNAs. In addition to melt-curve analysis to confirm single amplicon yield, real-time amplification data were also analyzed using the DART (Data analysis for Real-Time) platform [62], which uses the kinetics of the real-time reaction to determine the efficiency and consistency of amplification for each primer set. After validating each set of primers for reasonable consistency and efficiency, Ct values determined by the MyIQ software were used to calculate the relative DDCT statistics for each amplicon relative to CycB as a housekeeping control gene. Oligonucleoltide primers used for PA-Seq data validation are also included in Table S4.

TaqMan assay of mature miRNA abundance
Abundance of the mature mir322 miRNA was quantified relative to U6 snRNA using their respective TaqMan assays (Applied Biosystems) performed according to manufacturer's recommendations.

Results
siRNA Knockdown of mMtr4 in mouse N2A cells N2A cells transfected with siRNA specific to mMtr4 were assayed relative to cells transfected with a control siRNA. RNAs targeted for exosome degradation by fully functional TRAMP and NEXT complexes should be rare and largely undetectable in control knockdown cells, whereas these same RNAs should accumulate in cells depleted for mMtr4. In contrast, we hypothesized that canonical polyadenylation of messenger RNA would be unaffected by the mMtr4KD. Thus, adenylated RNAs that are substantially more abundant in the mMtr4KD are presumed to be substrates of mMTR4, and possible targets of the mammalian NEXT or TRAMP complexes. The ideal concentration and corresponding knockdown efficiency for individual siRNAs were empirically determined using qRT-PCR assays of the target mRNA (Fig. 1A). Immunoblots confirmed successful knockdown of the mMtr4 protein (Fig. 1B). RNA samples from two replicate knockdowns were used for high-throughput paired-end sequencing designed to specifically capture the 39 ends of adenylated RNAs. The paired-end reads were mapped to the mouse genome assembly using BWA (See Materials and Methods). Roughly 75% of raw sequence reads were successfully mapped to unique sites in the genome ( Table 1). Comparison of the two mMtr4KD replicates, sequenced to different depths, demonstrates that these sequencing data approach saturation. The ratio of mapped reads between mMtr4-1 and mMtr4-2 is 3.15, while the ratio of total genomic positions mapped between the same samples is only 1.73. All data were normalized to account for differences in sequencing depth. As described in the next section, comparisons between replicate KDs demonstrated reproducibility of the PA-seq data. Within each replicate sequencing experiment, the mMtr4KD and mControlKD, were compared to identify similarities and differences in the respective populations of adenylated RNAs. The replicate sequencing experiments exhibit a strong positive correlation for all target RNAs reported, but unless indicated otherwise, all data shown are from one dataset (those samples designated by 21).

Polyadenylation of messenger RNAs are largely unaffected by depletion of mMTR4
To test the hypothesis that mMtr4KD should not affect the abundance of protein coding mRNAs, normalized polyadenylated sequencing reads terminating within +50 bases of the annotated 39 end of RefSeq protein coding genes were identified for each KD. For each of the KDs, roughly 8500 RefSeq genes are represented in the PA-seq data with a minimum of five sequencing reads at the modal position. Data from the modal position were used to test the experimental reproducibility between sequencing experiments. A dot plot of the two mControlKDs (mControl-1 vs. mControl-2), with RNA isolation completed in parallel but library construction completed independently, demonstrated reproducibility of the PA-Seq data (R 2 = 0.87). The mMtr4KD and mControlKD sequencing reads of 8,935 RefSeq modal PA-sites are shown as a dot plot ( Fig. 2A), and the ratio of mMtr4KD versus mControlKD reads are presented as a histogram (Fig. 2B). The dot plot of mMtr4KD vs mControlKD indicated little variability in polyadenylation of the RefSeq protein coding mRNAs under conditions where mMtr4 has been depleted ( Fig. 2A, R 2 = 0.95). The histogram provides an alternative view of the data for mMtr4KD vs mControlKD, revealing a normal distribution, with ,94% of the RefSeq genes used in the analysis exhibiting less than a two-fold difference (|log 2 |,1) between mMtr4KD and mControlKD. The ,6% of RefSeq genes falling outside the two-fold difference are largely observed at the lowest expression levels ( Fig. 2A), consistent with higher variability at lower detection thresholds. Moreover, when examined in the replicate PA-Seq data set, the relative expression of these same RefSeqs in the mMtr4KD and mControlKD were equally split among three categories: equivalent expression levels (,2-fold difference), comparable difference (.2-fold difference), or an inverse correlation (.2-fold difference in the opposite direction). The data were also examined for evidence of differential polyadenylation position usage between the mMtr4KD and mControlKD. Although RNAs corresponding to alternative positions exist within the data, no significant differences in the relative distribution of site usage were apparent.
The 59 ETS of rRNA is a conserved target of yeast and mammalian RNA surveillance Convinced that changes in amplitude of PA-seq reads between mMtr4KD and mControlKD at a given locus would be the result of differences in RNA abundance, we focused on one of the known targets of the yeast TRAMP complex; the 59 ETS of pre-rRNA. The 59 ETS has also been observed to accumulate in human tissue culture cells depleted of hMtr4 [49]. The PA-seq data demonstrate that the mouse 59 ETS accumulates in the mMtr4KD relative to mControlKD. Processing intermediates of the mouse pre-rRNA 59 ETS include the region from the transcription start to the A9 processing site, from A9 to A 0 , and from A 0 to the 18S cleavage junction ( [63,64] ; Fig. 3A). Accumulation of A 0 terminating fragments (+2 bp) is 8-fold higher in the mMtr4KD than the mControlKD (Fig. 3B). In contrast, adenylated fragments terminating at A9 or the 18S junction were below the PA-seq detection threshold in any of our PA-seq libraries ( Fig. S1; also see discussion). Thus, accumulation of 59 ETS adenylated RNAs in the mMtr4KD is specific to the A 0 processing site. We conclude from this observation that degradation of the 59 ETS at the A 0 site is affected by depletion of mMtr4.

U3 snoRNA adenylated transcripts accumulate in cells depleted of mMtr4
Many of the snoRNAs function as guides directing accurate processing or modification of precursor rRNA transcripts to release 18S, 5.8S and 28S rRNAs, and facilitate their maturation and function through nucleotide modification [65][66][67][68]. Specifically, U3 snoRNPs form duplexes at several sites along the 59 ETS to direct endonucleolytic cleavage [64]. There are four identical copies of the U3B snoDNA (U3B.1-U3B.4) gene located in tandem within the 2 nd intron of Tex14 mRNA (NM_031386) on mouse chromosome 11 [69][70][71]. We observed accumulation of adenylated U3 snoRNA transcripts within our PA-seq data corresponding to all four copies of U3B, and sought to characterize these data further. First, we determined that all of the 59 paired-end reads map within the 214 bp of the mature snoRNA. This suggests that the reads derive from bona fide U3B processing rather than read through transcription. We report the mapped PA-seq data pooled from all four U3B loci (Fig. 4). The level of U3 snoRNAs adenylated precisely at their annotated 39 ends was nearly equivalent in mMtr4KD vs. mControlKD (Fig. 4). These data may indicate that routine turnover of aberrant or perhaps even mature copies of U3 snoRNA occurs via an adenylation dependent pathway. In stark contrast, shorter U3 snoRNAs adenylated at nucleotide positions immediately 59 of the mature snoRNA 39 end are 9-fold more abundant in the mMtr4KD relative to the mControlKD (Fig. 4, Fig. S2). Slightly longer U3 snoRNAs are also observed, and are generally more abundant in the mMtr4KD. A comparable distribution of adenylated transcripts from the U3A locus on chromosome 10 are observed (data not shown). Given that normal U3 processing involves 39R59 exosome-mediated trimming [72], these data may indicate that U3 snoRNAs are occasionally 'over-trimmed', and that such errors are eliminated by MTR4-facilitated degradation.
Quantitative RT-PCR was used to confirm differential accumulation of adenylated U3 snoRNA in mMtr4KD relative to mControlKD. Total RNA from two replicate knockdowns was reverse transcribed using oligo-dT, and subjected to qPCR to measure the level of poly(A)+U3 snoRNA in the mMtr4KD and mControlKD using U3 specific primers. Given that oligo-dT primed reverse transcription (RT) would not distinguish between fulllength and shortened adenylated U3 snoRNAs, our qPCR results reflect the sum of all adenylation sites reported in Fig. 4. Thus, the observed ,5-fold difference between mMtr4KD and mControlKD by qPCR is consistent with the PA-seq data ( Table S1).

Adenylated transcripts accumulate in the mMtr4KD
To identify additional targets of mMTR4 mediated degradation, the PA-seq data was analyzed using the F-Seq feature density estimator [58] to identify clusters, or peaks, of adenylation. Peak data were filtered to eliminate sequencing reads corresponding to templated adenosines (see Methods). All remaining peaks were categorized as corresponding to annotated mRNA polyadenylation sites, repetitive genomic elements (RepeatMask track), or neither. Peaks in the neither category were compared to the mControlKD to identify RNAs that exhibit differential accumulation. Those exhibiting the largest ratios are shown in Table 2, along with information regarding their relative genomic context. Strikingly, four of the top five RNAs exhibiting increased accumulation in the mMtr4KD map to a microRNA or microRNA host gene. The above analyses identified two significant peaks of adenylation on chromosome X corresponding to a full length non-coding cDNA (RIKEN-C43009B03RIK) containing a miRNA cluster (mir322-mir503-mir351). To gain a better appreciation for the distribution of adenylation across this cDNA, we examined all F-Seq peaks across the 4 kb span of C43009B03RIK. Distribution of the mMtr4KD peaks reveal three prominent areas of adenylation in this region of chromosome X: 1) at the 39 annotated end of cDNA C43009B03RIK, 2) a loose clustering near the 39 annotated end of a shorter RIKEN cDNA AK021262 that encodes a hypothetical 92aa protein but does not include the miRNA cluster, and 3) at mir322 (Fig. 5). The presence of PA-seq peaks in our data corresponding to the annotated 39 ends of both RIKEN cDNAs indicates that our technique is effectively interrogating transcriptional activity at this genomic locus. Consistent with the existence of two RIKEN cDNAs at this locus, our data also indicate there are two outcomes of transcription starting at the 59 end of this genomic locus; one transcription unit which includes the miRNA322, 503 and 351 cluster and a shorter one that does not include the miRNAs. Strikingly, the substantial cluster of adenylated RNAs around the 59 end of mir322 is notably absent in the mControlKD (Fig. 5). The mode of this highly active adenylation cluster in the mMtr4KD precisely corresponds with the Drosha cleavage site (Fig. 6A). Directed interrogation of the mControlKD data identified only three PA-seq reads at this position, indicating that there is a 67-fold increase for these adenylated transcripts in the mMtr4KD. Drosha cleavage of the primary miRNA transcript would release three distinct products; the pre-miRNA, a 59 leader fragment (hereafter referred to as the 59 leader) and a 39 fragment (Fig. 6A). In the case of mir322, the 39 fragment will contain two additional microRNAs, mir503 and mir351, which may subsequently influence the fate of that fragment (discussed below). The notable PA-RNAs that accumulate in the mMtr4KD (Table 3), represent the 59leader, suggesting that mMTR4 may play a role in targeting the 59leaders of microRNAs to the exosome for degradation, in an adenylation dependent manner.
The vast increase in abundance of adenylated mir322 59 leader sequences in mMtr4KD relative to the mControlKD, along with a modest 6-fold increase in transcripts adenylated at the annotated 39 end of the full length RIKEN cDNA, might reflect increased expression of mir322. To determine whether the mMtr4KD causes differential expression or accumulation of mir322, we employed a TaqMan miRNA assay to quantify the relative levels of the mature mir322-5p microRNA in mMtr4KD and mControlKD RNAs. The mir322 TaqMan assays were normalized to U6 snRNA levels that were assayed in parallel. Analyses in three biological replicates suggest that mature mir322 levels may fluctuate in either direction by as much as two fold in the mMtr4KD and mControlKD, but that no consistent trend exists (Table S2). Although we cannot completely rule out that the mir322-mir503-mir351 cluster is more actively transcribed upon mMTR4 depletion, even a 2-fold increase would seem too small to explain the large, 67-fold increase in adenylated mir322 59 leader by a mechanism of increased transcription alone. In a related experiment, PA-seq detected a modest increase in the mir322 59leader and annotated 39 end of RIKEN-C43009B03RIK cDNA in N2A cells depleted for the exonuclease mRrp6 (data not shown/unpublished). Taken together, these data specifically support the idea that differences between mMtr4KD and mControlKD are from impaired degradation of those transcripts, rather than increased transcription.

MTR4 facilitates degradation of miRNA 59 leaders
Two additional microRNAs, let7b and mir138-1, display clusters of adenylation for mMtr4KD ( Table 2). Comparable F-Seq analysis using the deeper of the two mMtr4KD sequencing data sets (mMtr4-2) identified a total of six miRNA peaks, expanding the list to include let7c-2, mir106b and mir26b ( Table 3). In all cases, the modal position of adenylation maps precisely to the predicted Drosha cleavage site, such that the adenylated RNA corresponds to the 59 leader. In sharp contrast, no corresponding peaks were identified in the mControlKD. In an effort to determine whether the mControlKD may contain corresponding 59 leader sequences below the level of detection by F-Seq, or whether mMtr4KD may contain 59 leader sequences corresponding with additional microRNAs, we asked whether reads in the complete PA-seq data sets map in or around any annotated miRNAs. This analysis identified a small number of PAseq reads in the mControlKD (Table 3), and PA-seq reads in mMtr4KD corresponding with over 100 additional miRNAs, eighty-five percent of which (105/124) include reads mapping consistent with the predicted 59 leader terminating at the Drosha cleavage site. Although a majority of the 105 additional miRNAs are represented by fewer than 5 total PA-seq reads, it is difficult to dismiss the correlation with the position of the 59 leader as  coincidental. All miRNA 59 leader PA-seq data are available in Table S3.
Oligo-dT primed qRT-PCR was used to independently test miRNA 59 leader sequence accumulation in the mMtr4KD relative to mControlKD. We used the same RNA samples used to generate the PA-seq data, and focused on mir322, mir503, let7b, let7c2 and mir138-1. qRT-PCR analysis of miRNA 59 leader sequences was analyzed relative to qRT-PCR that spans the junction between the 59 leader and the pre-miRNA (as depicted by representative amplicons 1 and 2 in Fig. 6A). Amplicon 2 serves as a critical control for the level of detectable adenylated pri-miRNA that would be amplified by the primers located within the 59 leader. qRT-PCR results support accumulation of the miRNA 59 leader relative to the pri-miRNA for mir322, let7b and mir138-1 ( Figure 6B). In contrast, qRT-PCR results for amplicon 2 primers spanning the Drosha cleavage site were uniformly low. For comparison, Figure 6C presents normalized PA-seq reads for each of these microRNA leaders from both mMtr4KD and mControlKD, as well as their relative ratio. The qRT-PCR data for let7c2 and mir503 are less dramatic than the other microRNA 59 leaders, suggesting that the respective pri-miRNAs are more abundant than the 59 leaders produced by Drosha, either due to less frequent Drosha processing for these miRNAs, or instability of the 59 leaders of these miRNAs despite knockdown of mMtr4.

Discussion
We have analyzed the effect of depleting mMtr4 on RNA adenylation in mouse N2A cells using small interfering RNAs (siRNA). We used a PA-seq strategy to selectively amplify and sequence adenylated RNAs. Depletion of mMtr4 does not alter the abundance of adenylated protein coding mRNAs. In contrast, depletion of mMtr4 results in the accumulation of a number of adenylated non-coding RNA. Both the rRNA 59 ETS and U3 snRNA have been previously reported as targets of TRAMP. The accumulation of adenylated 59 leader transcripts from pri-miRNAs is novel, and suggests a role for mMtr4 in targeting these transcripts for degradation by the exosome. Together, these results are consistent with observations linking MTR4 homologs to RNA surveillance [47,49,[73][74][75][76]. Isolation of these target RNA sequences based upon presence of a poly(A) tail further suggests that, similar to extensive data in yeast [39,40,42,77] and more recent observations in humans [49], at least some mMTR4mediated RNA surveillance in mouse also involves a Poly(A) polymerase. In contrast, others have observed significant accumulation of adenylated PROMPT RNAs in hMtr4KD human cell lines [49,78], whereas significant PROMPT accumulation was not detected in our PA-Seq data. This may reflect differences in sensitivity of the respective gene specific RT-PCR vs. PA-Seq methods, or issues related to oligo-dT RT reactions or adenylate tail length.
The 59 ETS of pre-rRNA is a conserved target of MTR4mediated decay in mouse RNA surveillance has been most extensively characterized in yeast, where the TRAMP complex functions to identify and target specific RNAs for degradation. The yeast TRAMP complex consists of the DExD/DExH box RNA helicase Mtr4p, as well as a poly(A) polymerase (either Trf4p or Trf5p), and a RNA-binding protein containing five Zn-knuckles (either Air1p or Air2p). Although Mtr4 is an essential gene, repressible or reduced function alleles of Mtr4 have been used to identify targets of TRAMP and characterize the role(s) played by Mtr4p as part of the TRAMP complex [33,73,79]. Allmang and colleagues [33] determined that depletion of mtr4 resulted in time-course dependent accumulation of the 59 ETS of yeast rRNA.
We observe significant accumulation of 59ETS sequences adenylated at the A 0 processing site in our mMtr4KD, consistent with this rRNA byproduct being a conserved target of the mammalian TRAMP complex. Nevertheless, given the relative abundance of rRNA, the frequency with which we observe these transcripts (,28 reads per million in mMtr4KD, Fig. 3B) seems surprisingly low. We also explored whether rRNA transcripts 59 of the A9 processing site, or immediately adjacent to the 18S rRNA (see diagram in Fig. 3A, and Fig. S1), accumulate in a similar fashion to those at A 0 . Among ,31 million raw 39 reads queried, including both the mMtr4KD and mControlKD, only 3 adenylated transcripts corresponding with the region 59 of the A9 processing site and no reads corresponding to the 18S proximal segment were observed. Abundant data suggest that pre-rRNA processing should occur frequently at the A9 site [63,64,80], although failure to accumulate adenylated fragments terminating at this position in the mMtr4KD is consistent with the fact that the analogous portion of the yeast 59ETS is not degraded by the exosome, but rather by 59R39 exonucleases Xrn1p and Rat1p [67,81]. Overall, substantive accumulation of 59ETS adenylated RNAs in the mMtr4KD is specific to the A 0 processing site.
PA-seq reveals a heterogeneous population of adenylated U3 snoRNAs in mMtr4KD snoRNAs have also been observed to be targets of Mtr4p in yeast [33,82]. Whether Mtr4p is targeting adenylated snoRNAs for degradation, for 39 end processing by an exonuclease, or some combination of the two has not been explored, but our data would suggest that both may be true in mouse. Detection of U3 Figure 5. Abundant PA-seq reads map near mir322. The top diagram shows the mir322 -mir503 -mir351 cluster and associated RIKEN transcripts on chromosome X. Note that the chromosomal orientation has been reversed such that the mm9 chromosomal coordinates are in decreasing order from left to right because the transcripts are on the minus strand. The rightmost portion of this region has been expanded below to highlight both the location and relative abundance of adenylation in the mMtr4KD (black) vs. mControlKD (grey). doi:10.1371/journal.pone.0099430.g005 transcripts adenylated at or just beyond the annotated 39 end in the mControlKD is consistent with the idea that adenylation is part of the normal U3 life-cycle, from 39 end maturation to normal turnover of snoRNAs. Derti and colleagues [83] observe adenylated U3A in an assortment of wild-type tissues. The fact that we observe significantly higher levels of adenylated U3 in the mMtr4KD, with a substantial increase in transcripts adenylated just 59 of the annotated 39 end, is consistent with the concept that mMTR4 functions not only in the normal life-cycle of U3 as observed in normal cells, but also in turnover of defective U3 snoRNAs, such as those that may have been over-trimmed by an exonuclease.

Nuclear RNA surveillance of miRNA 59 leaders
It is commonly accepted that extraneous portions of miRNA primary transcripts released during miRNA biogenesis are 'simply degraded', but the cellular components responsible for targeting those RNAs for degradation have not been previously identified. Our PA-seq data suggest that mMTR4, possibly as part of the mammalian TRAMP complex, plays a role in directing some of these miRNA byproducts, particularly the 59 leader, to the nuclear exosome for degradation.
While we were able to detect adenylated 59 leader sequences corresponding to over 100 miRNAs in the mMtr4 knockdown, only a handful of these accumulate to highly significant levels ( Table 3, Fig. 6, and Table S3). Interestingly those that accumulate to the highest levels correspond to abundant miRNAs. Both mir322 and let7 are relatively ubiquitous across a wide range of tissues, and among the most abundant microRNAs detected [84][85][86]. Several of the other microRNAs play critical functional roles in neural or brain development, or in cases where their function is as yet unknown, they are abundantly represented in neuronal cells, or during brain development [87][88][89][90][91][92][93][94][95].
Much of the small RNA deep sequencing data in mirbase also suggests that the 59 most member of putative co-transcribed miRNAs is often more abundant than those progressively 39 within the cluster [84,[96][97][98], consistent with the idea that Drosha may preferentially liberate specific pre-microRNAs from polycistronic pri-microRNAs, and that in some cases, the remaining transcript may be degraded rather than processed further. Consistent with this observation, miR322 is the 59 most member of its microRNA cluster.
We explored the genomic context for several more of the most abundant microRNA 59 leaders (let-7b, let-7c2, mir138-1, mir106b and mir26b) in an effort to determine whether context might influence accumulation of the 59 leader in the mMtr4KD. Indeed, many of the most abundant 59 leaders correspond to microRNAs presumed to be co-transcribed with other micro-RNAs. miR138-1 is one exception, residing in an intergenic region of chromosome 9 with no other annotated miRNAs or mRNAs nearby. The other is mir26b, which is found within the fourth intron of a protein coding transcript (NM_153088, Ctdsp1). Similar to mir322, mir106b is the 59 most microRNA within the mir106b-mir93-mir25 cluster, which is located within intron 13 of the Mcm7 gene. Finally, let-7c2 and let-7b are less than 1.0kb apart and presumed to be part of the same primary transcript. While the 59 leader sequences for both of these microRNAs accumulate, we observed a greater amplitude of PA-seq accumulating in mMtr4KD for the more distal let-7b than for let-7c2, both being appreciably above that seen in the mControlKD. Overall, the reason we see preferential (sometimes exclusive) accumulation of the 59 leader for one member of a miRNA cluster is unknown, but could reflect differences in processing order or efficiency, the possibility that only some of the processing byproducts are Figure 6. Adenylated 59 leader sequences from pri-miRNAs accumulate in the mMtr4KD. A. Schematic depicting relevant aspects of miRNA processing. Primary miRNA transcripts are produced by RNA polII and include both a 59 cap and poly(A) tail. The miRNA and complementary mir* contribute to formation of a hairpin, and Drosha cleaves near the base of the hairpin to liberate the pre-miRNA [103]. mMtr4KD PA-seq identified adenylated 59 leader transcripts whose 39 end specifically maps to the Drosha cleavage site. qRT-PCR validation of these data compared amplification results for primers within the 59leader (eg. Amplicon 1, depicted on uppermost diagram) to primers that span the Drosha cleavage site (Amplicon 2) as a control for pri-miRNA abundance. B. Relative expression of miRNA 59 leader sequences was calculated using the DDCt method from triplicate reaction realtime PCR data. All reactions were performed in triplicate, further averaged over 3-4 independent qPCR runs for each miRNA 59 leader, and shown relative to parallel qPCR reactions detecting Cyclophilin B mRNA. All Yaxis values represent relative arbitrary units. C. Combined PA-seq data for miRNA 59 leader sequences reported as reads per million, and ratio of mMtr4KD over mControlKD. doi:10.1371/journal.pone.0099430.g006 adenylated, or RNA structural determinants of the accumulating PA-seq.
Although it remains unclear why specific miRNA 59 leaders accumulate in the mMtr4KD, it is worth noting that we do not observe notable accumulation of a corresponding 39 fragment that would simultaneously be liberated by Drosha cleavage of the pre-miRNA from the pri-miRNA. One factor that may influence miRNA 59leader accumulation is the 59 cap on these RNApolII derived transcripts preventing 59R39 exonuclease degradation, whereas the 39 fragment would be susceptible to both 59R39 and 39R59 exonuclease degradation. Consistent with this distinction, 59 proximal versus 39 proximal RNAs resulting from cotranscriptional cleavage (CoTC) of the human beta-globin gene have alternate fates. The 39 proximal RNAs are degraded by the Xrn2 59R39 exonuclease [99], whereas the 59 proximal RNAs are subject to exosome mediated degradation [100]. Presence of a 59 cap could also help explain the correlation that several of the 59 leaders that accumulate are from the 59 most members of miRNA clusters thought to be co-transcribed. Overall, no single aspect appears to readily predict which microRNA 59leaders will accumulate in mMtr4KD cells, or the relative extent to which they will accumulate, but a select few are by far the most abundant adenylated RNAs we have thus far identified in the mMtr4KD.

Implications for nuclear RNA surveillance in mammals
Although extensive research has uncovered a wealth of TRAMP substrates in yeast, the absence of microRNAs in yeast and thus our inability to predict this novel finding underscores the importance of expanding research into a wider array of organisms. This finding also reveals the extent to which nuclear RNA surveillance complexes may have evolved to target an ever increasing diversity of substrates as genomes evolve and increase in complexity. Finally, our observation suggests that depleting MTR4 proteins in a wide array of tissues or organisms may be a useful mechanism for characterizing the 59 ends of pri-miRNAs, such as determining transcription start sites and identifying promoter elements.
Well conserved homologs of Mtr4, Trf4 and Air2 form a TRAMP-like complex in humans, but immunofluorescence localization suggests that the Air2 homolog ZCCHC7 exclusively localizes to the nucleolus [49]. Given that the Air2 protein of yeast is critical for assembly of the yeast TRAMP complex [101], and that ZCCHC7 is the only human protein with significant similarity to the Air2 and Air1 paralogs of yeast, it is unclear how mammalian Trf4 (also known as Papd5) and Mtr4 would collaboratively function outside the nucleolus. Although the NEXT complex exhibits a complementary localization within the mammalian nucleus, and is excluded from the nucleolus, neither Trf4 nor any other poly(A) polymerase co-immunoprecipitated with this complex [49]. Nevertheless, surveillance may involve one or more canonical poly(A) polymerases [78,102], or the mammalian Trf4 may function autonomously outside the nucleolus [50]. Definitive identification of the poly(A) polymerase(s) responsible for adenylating will further illuminate RNA surveillance in mammalian systems. Overall, our PA-Seq analyses confirm the role of mMtr4 in adenylation mediated degradation of the 59 ETS of rRNA and U3 snoRNAs, and suggest that some miRNA 59 leader sequences utilize mMtr4 in a similar degradation pathway.   * Data are reported as total raw reads combined from both biological replicates, and summed over positions +2 bp relative to the predicted Drosha cleavage site. ** Raw reads from mControlKD are not directly comparable to mMtr4KD due to ,3-fold higher coverage in one of the sequencing experiments. Figure 6C