Identification of RNAs that co-purify with eukaryotic Sm proteins
As mentioned above, the Sm and Sm-like proteins comprise a family of ancient evolutionary origin that functions to modulate the stability and translation of several classes of RNA, including mRNAs [1, 35]. Based on these ancestral roles, the involvement of eukaryotic Sm proteins in splicing is generally thought to be a derived function, and additional RNA targets of Sm proteins remain to be discovered.
To characterize the repertoire of RNA targets that are associated with Sm proteins in Drosophila ovarian lysates, we performed RIP-seq analysis of individual subunits of the canonical Sm ring. We also performed RIP-seq on Trailer Hitch (Tral), a protein that contains an Sm domain (Figure 1c). Tral is not incorporated into the canonical Sm ring; therefore, we expected it to associate with a distinct subset of transcripts [36]. An outline of the experimental strategy and data analysis pipeline is shown in Figure 1a. Immunoprecipitations (IPs) were carried out using either anti-SmB (monoclonal antibody Y12) or anti-green fluorescent protein (anti-GFP) antibodies (for the GFP- and Venus fluorescent protein (VFP)-tagged proteins). Normal goat serum was used as control for the IP. Immunoprecipitated RNA was reverse transcribed to cDNA, fragmented, ligated with adapters, PCR-amplified and sequenced on an Illumina Genome Analyzer II.
To reduce potential non-specific interactions and artifacts, we carried out RIP-seq on several Sm proteins expressed from three different genomic contexts: (i) native endogenous genes, (ii) VFP-tagged transgenes, or (iii) a gene-trapped (GFP-tagged) endogenous gene (Figure 1c). Comparisons among this wide variety of experimental conditions helps to minimize problems associated with genetic background, transgene overexpression, and antibody specificity. Four different transgenic lines were employed, including VFP-tagged SmD3, SmB, SmD1 and SmE [21]. Transgenes were expressed using the UAS/Gal4 system, crossed to a nanos-Gal4 driver for germline-specific expression or, in the case of VFP-SmD1, to a daughterless-Gal4 driver for ubiquitous expression [37]. SmB and SmD3 form an obligate dimer (Figure 1b), whereas SmD1 and SmE are present in distinct sub-complexes within the heteroheptameric ring structure [9]. Thus, IPs targeting different components of the Sm ring further reduced potential artifacts resulting from epitope tagging, as these proteins form a complex that is expected to bind a similar set of RNAs. RIP-seq experiments were performed on SmB, SmD3 and SmE, whereas RIP-qRT-PCR was performed on VFP-SmD1 for identified targets. To broaden the scope of our study, we also performed RIP-seq analysis in cultured human HeLa cells, using the Y12 antibody mentioned above (Figure 1d; see details in Table S1 in Additional file 1).
Enrichment analysis of Sm RIP-seq experiments
We obtained between 8 and 28 million 35-nucleotide single-end reads per Drosophila ovary RIP-seq library, and roughly 20 million 48-nucleotide paired-end reads per human HeLa cell RIP-seq library. All of the fly and human sequencing data are of high quality (Figure S1 in Additional file 1). Despite differences in total read numbers, the IPs consistently yielded many more mappable reads than did the controls (Table S2 in Additional file 1, ‘mapped’ and ‘%mappable’ columns). This was to be expected; due to the low amount of input cDNA, most of the reads in the control IPs are not mappable (for example, rRNAs, primer/adapter dimers or even random sequences; Table S3 in Additional file 1) and those that do map to the genome typically correspond to abundant RNAs that stick to the beads non-specifically Library statistics show that random hexamer priming yielded more mappable reads than did oligo(dT)20 priming (Table S4 in Additional file 1). Thus, we used the random hexamer-primed libraries for the subsequent enrichment analyses.
We built a data analysis pipeline (Figure 1a) by integrating previously published programs (see Materials and methods for details). Sequence reads for the Drosophila RIP-seq experiments were mapped to the Drosophila expanded genome and quantified using ERANGE [38]. Then, for each experiment, we filtered out transcripts with read coverage less than 10. Assuming that the majority of RNA species are not associated with Sm proteins, we normalized the remaining transcripts against the median of all enrichment ratios: (raw_IP + 2)/(raw_Ctrl + 2). After normalization, we defined the enrichment ratio as (norm_IP + 2)/(norm_Ctrl + 2). The use of median-normalized raw read numbers is similar to the upper-quartile normalization method used by others [39]. In this way, we made a conservative estimate of the enrichment of RNAs in IPs versus controls.
To visualize the enrichment data, scatter plots were constructed using the log-transformed and normalized read numbers. Data for the native SmB-associated RNAs (Oregon R, Y12 IPs) are shown in Figure 2a; data for the other Sm protein constructs are presented in Figure S1 in Additional file 1. In any co-IP experiment, there are two populations of molecules: those that interact specifically with the antibody and those that stick non-specifically to the beads. Non-specific interaction was observed for many transcripts, as depicted by the main cluster along the diagonal line (Figure 2a). The dots located above the main cluster represent the enriched RNAs. In order to objectively identify Sm-associated RNAs, we employed Gaussian mixture modeling [40], which has been used to analyze RIP-chip experiments [41]. Distributions of the enrichment ratios were first plotted as histograms. Next, we used mixtools to fit a combination of two Gaussian functions to the enrichment ratio distribution [42].
As shown in Figure 2b, the distribution of the log-transformed enrichment ratios (red line) can best be explained by two different Gaussian functions, one that corresponds to the background RNAs (black dotted line) and one that represents the Sm-associated RNAs (blue dotted line). The cutoff between Sm-associated and background mRNAs was defined by the log of the odds (LOD) ratio between the two Gaussian functions. The transcripts with a LOD > 1 (that is, those that had a greater likelihood of being in the Sm distribution) were considered to be Sm-associated RNAs. Using this threshold, we then mapped these assignments back onto the scatter plots. As shown in Figure 2a (blue dots), the enriched RNAs are clearly seen to be above the diagonal (black dots represent the background distribution). This same analysis was performed on the other Sm protein datasets, with strikingly similar results (Figure S2 in Additional file 1). Thus, the Gaussian mixture modeling procedure provides an unbiased and less arbitrary method for identifying enriched RNAs [41]. Using the aforementioned analysis pipeline, we identified roughly 200 Sm-associated RNAs in any given RIP-seq experiment, representing 0.7% of the Drosophila transcriptome, or 4% of the significantly expressed transcripts.
A multi-targeting RIP strategy identifies highly reproducible Sm-associated RNAs
To assess the robustness and reproducibility of the Drosophila RIP-seq experiments and analysis pipeline, we visualized the log-transformed enrichment ratios for the transcripts with a read coverage greater than 10. Out of the >15,000 annotated genes in the fruitfly genome, 5,296 of them showed sufficient read depth (d > 10). To determine the relationship between the profiles of the seven RIP-seq experiments without prior assumptions, we performed an unsupervised hierarchichal clustering analysis. The top of the map represents RNAs that are significantly enriched (Figure 2c). As shown by the dendrogram (Figure 2c) and consistent with expectation, the six canonical Sm protein RIP-seq experiments clustered together, whereas the data from the Tral IP formed an outgroup. The most-highly enriched transcripts among the random hexamer-primed libraries from six Sm IP experiments (including one VFP-SmD3 biological replicate) revealed extensive overlap. Detailed analysis showed that 25 RNAs (9 snRNAs, 16 mRNAs) were common among all 6 Sm protein IPs, and 52 transcripts (12 snRNAs, 40 mRNAs) were shared in 5 of the 6 (see Table S5 in Additional file 1 for detailed enrichment ratios). The top 86 transcripts (13 snRNAs, 1 small nucleolar RNA (snoRNA), and 72 mRNAs) were shared by at least 4 of the experiments. Since four Drosophila snRNAs (U1, U2, U4, and U5) have multiple variant paralogs, we reassigned uniquely mappable reads to them and we found that all of the snRNAs with significant coverage are enriched in all Sm IPs (Table S6 in Additional file 1). In addition, we analyzed the consensus set of 86 Sm-associated RNAs in the oligo(dT)20 primed libraries, and we found that they are also highly enriched, despite the lower number of mappable reads (Figure S4 in Additional file 1). Thus, our multi-targeting RIP-seq approach is robust despite the differences in library statistics (Table S2 in Additional file 1). We operationally defined the Sm-associated RNAs as being those that were enriched in at least four of the six experiments.
Next, we carried out pair-wise comparisons among the seven RIP-seq experiments and performed Fisher’s exact test to assess the significance of any overlapping subsets (Figure 2d). Interestingly, among the top 200 RNAs in the Tral IP experiment, very few of them overlapped with any of the RNAs that associated with canonical Sm proteins. As seen in the heat map (Figure 2c), the enrichment ratios for the VFP-SmE IP were typically lower than those of the other Sm proteins. However, the pairwise comparisons show that SmE associates with a similar group of RNAs (see also Figure S4 in Additional file 1). The overlaps between the different Sm protein IPs were highly significant, as shown by their extremely small P-values (10-32 to 10-135, plotted as negative logarithms; Figure 2d). Even when all of the snRNAs were taken out of the pair-wise comparisons, the P-values remained extremely small (Figure 2d; Figure S3 in Additional file 1). Despite the different experimental parameters (tagged versus untagged, native versus ectopic, and so on), the lists of enriched RNAs are essentially the same. This high degree of reproducibility suggests that the multi-subunit targeting approach is superior to the conventional biological replication of experiments for RNP analysis. Indeed, the variability between biological replicates was greater in the case of VFP-SmD3 than it was between some of the other RIPs (Figure 2c). Collectively, these data demonstrate a high degree of specificity in the Sm protein IPs, showing that canonical Sm proteins co-precipitate with essentially the same set of mRNAs.
Sm proteins associate with three major classes of RNAs
The RIP-seq experiments in both Drosophila and human cells confirmed the well-studied snRNAs as major targets of Sm proteins, and in addition indicate novel classes of Sm targets. A detailed analysis of the known and newly discovered RNAs from our study suggests that Sm proteins associate with three major classes of RNAs (Figures 3 and 4; Figures S4 and S6 in Additional file 1).
RIP-seq identifies Sm class snRNAs
The Sm-associated transcripts and their enrichment ratios are listed in Figure 3. As expected, all spliceosomal snRNAs were among the top-scoring transcripts in terms of their enrichment ratios. The only missing Sm class snRNA from the list of Sm-associated RNAs is U7 snRNA, because it is too short (71 nucleotides in Drosophila, and 63 nucleotides in human) to be included in the size-selected cDNA libraries (Figure 3a; Table S5 in Additional file 1) [43, 44]. Other highly abundant non-coding RNAs (ncRNAs; for example, 7SK snRNA, SRP RNA, 5.8S ribosomal RNA and so on, data not shown) were not enriched in the IPs, demonstrating the specificity of the approach. Multiple distinct paralogs exist for four of the Drosophila snRNAs, U1, U2, U4 and U5, and they share long stretches of identical regions (Figure S5 in Additional file 1). In order to accurately analyze each paralog without the confounding repetitive reads, we reassigned uniquely mappable reads to U1, U4 and U5 paralogs (Table S6 in Additional file 1). We used the variant nucleotides in U2 to calculate the fractions of each isoform and redistribute the total number of U2 reads among the gene paralogs. Not surprisingly, all snRNAs with significant read coverage are enriched in the IPs (Table S6 in Additional file 1). With regard to the HeLa cell analysis, there are hundreds of snRNA genes in the human genome, and only a small fraction of them are properly annotated. Not surprisingly, most of the annotated human spliceosomal snRNAs were identified in our IPs, all of which have very high enrichment ratios (Figure 3b).
ERANGE analysis and manual inspection of the Drosophila RIP-seq data revealed several clusters of reads that could not be mapped to gene models. Four of them are new genes that had not been previously annotated. During preparation of this manuscript, two transcriptomic studies have since identified these putative new transcripts [45, 46]: CR43708, CR43600, snoRNA:2R:9445410 (CR43574) and snoRNA:2R:9445205 (CR43587). Two of the four novel transcripts, CR43708 and CR43600, showed significant enrichment in the IPs.
We characterized the two Sm-associated ncRNAs and found that one, CR43708, has features typical of an snRNA. CR43708 is located in the second intron of fas2 (CG3524, fatty acid synthase 2), a homolog of the human fatty acid synthase gene (Figure 5a). We defined the accurate 5′ and 3′ ends of CR43708, and found that this transcript is 116 nucleotides long (ZL and AGM, unpublished). Detailed analysis of sequences upstream of CR43708 revealed conserved proximal sequence elements PSEA and PSEB, highly similar to Sm-class snRNA promoters (Figure 5a; Figure S7a in Additional file 1) [47, 48]. To examine the subcellular localization of CR43708, we carried out in situ hybridization in Drosophila S2 cells and found that this RNA accumulates in the nucleus (Figure 5c). Using the transcribed region and the promoter sequences, we searched genome and transcriptome databases for homologs. We recovered matches in nine species, all of which are in the melanogaster group of the Drosophila genus, and all are located within the same intron of the fas2 gene (Figure 5e,f). Among the sequenced Drosophila species in the melanogaster group, the Drosophila erecta genome does not appear to contain CR43708, suggesting that it may have been lost. Interestingly, we found a truncated version of this gene within an intron of the Ac3 gene in D. melanogaster (Figure S7c in Additional file 1). The homology extends through the first 70 bp of CR43708, and lacks the promoter and the 3′ end, suggesting that this paralog is a pseudogene. The predicted secondary structure of CR43708 closely resembles that of a canonical snRNA, including the presence of 5′ and 3′ end stem loops that flank a putative Sm binding site (Figure 5c). Structured sequence alignments clearly show that the putative Sm binding site (except in Drosophila kikkawai) and the terminal stem loops are well conserved. In addition, we identified many covariant base pairs within the two stem loops, supporting the predicted secondary structure (Figure 5f). Uridine-rich, Sm-class snRNAs such as U1 and U2 are known to contain a trimethyl-guanosine (TMG) 5′ cap structure that is generated upon formation of the Sm core RNP [9]. As expected, CR43708 was efficiently immunoprecipitated by anti-TMG antibodies (Figure 6a). Taken together, these features led us to conclude that this transcript is a novel Sm-class snRNA, which we termed snRNA:LU (Like U).
Interestingly, the U5:23D snRNA gene is located near LU, within a neighboring intron of the fas2 protein coding gene (Figure 5a). We were unable to deduce the precise origin of LU; however, its juxtaposition with U5:23D suggests that it could have evolved from a U5 gene duplication, followed by rapid divergence. Supporting this notion, the 3′ end stem-loops of the LU snRNA homologs are quite similar to those of U5 snRNAs (Figure S7 in Additional file 1), although there is a lack of overall sequence similarity between the two genes.
To study the function of LU snRNA, we first considered the possibility that it might base pair with other snRNAs, as we found a nearly invariant single-stranded region located in the middle of LU snRNA (Figure 5d,f). Notably, we identified extensive base complementarity between this region of LU and the 5′ end of U6 (Figure S7d in Additional file 1). This putative base-pairing suggests that LU may be involved in splicing regulation. We identified four independent transposon insertions in and around the LU gene locus (see Materials and methods), and we confirmed that one of these insertion lines, fas2k05816, disrupts expression of both the fas2 host gene and the LU snRNA gene (Figure 5a; Figure S7e in Additional file 1). Although homozygotes die around eclosion; complementation analysis between fas2k05816 and two other deletion lines uncovering this region suggests that neither the fas2 host gene nor the LU snRNA gene are required for organismal viability (Figure 5b). We conclude that, although it may well contribute to organismal fitness, LU is not an essential gene. This conclusion is supported by the independent loss of LU snRNA in D. erecta. Taken together, our RIP-seq analysis of Sm proteins reveals that a total of 11 distinct species of Sm-class snRNAs are present in Drosophila: U1, U2, U4, U5, U6, U7, U4atac, U6atac, U11, U12 and LU.
Sm proteins associate with evolutionarily conserved and rapidly evolving scaRNAs
scaRNAs are ncRNAs that guide methylation and pseudouridylation of snRNAs, the specificity of which is determined by base-pairing with targets [49]. A previous study showed that in human cells, several scaRNAs specifically associate with SmB and SmD3, including U85, U87, U89 and human telomerase RNA (hTR) [50]. Co-precipitation of SmB/D3 with these scaRNAs was shown to require the conserved CAB box [50], which is essential for scaRNA localization to Cajal bodies [51]. To determine whether other ncRNAs co-purify with Sm proteins in Drosophila and human cells, we systematically analyzed the enrichment values of snoRNAs and scaRNAs in our RIP-seq datasets. Consistent with the findings of Fu and Collins [50], we found that two previously identified Drosophila scaRNAs, U85 (CR32863 or snoRNA:MeU5-C46) and CR33716 (snoRNA:MeU5:U42), were enriched in the Sm protein IPs (Figure 4a; Table S5 in Additional file 1). Interestingly, the new Sm-associated ncRNA identified in this study (CR43600 or snoRNA:Prp8) also appears to have features of box H/ACA scaRNAs. Indeed, evolutionary comparisons identify conserved H/ACA and CAB box elements present within the detected orthologs (Figure S6b,c in Additional file 1). snoRNA:Prp8 folds into a predicted secondary structure similar to that of other box H/ACA scaRNAs, which is further supported by the presence of multiple covariant base pairs. In support of the notion that snoRNA:Prp8 is an H/ACA box scaRNA, we searched snRNAs for sequence complementarity to the pseudouridylation pocket sequences, and found potential target sites in U1, U5, U7 and U11 (Figure S6d in Additional file 1). Therefore, we have renamed this transcript scaRNA:Prp8. We detected homologs of scaRNA:Prp8 in both Diptera (Drosophilids, Anopheles gambiae) and Hymenoptera (Apis mellifera), but not in Coleoptera (Tribolium castaneum) (Figure S6b in Additional file 1). The orthologous scaRNA:Prp8 RNAs are highly conserved, suggesting their functional importance. Many scaRNA and snoRNA genes reside within introns of splicing and translation-related genes, respectively [52]. The nested gene structures are thought to facilitate transcriptional co-regulation. Thus, it is not surprising that the Prp8 host gene encodes a splicing factor (Figure S6a in Additional file 1) [53, 54]. Although Fu and Collins [50] reported that only SmB and SmD3 co-purified with scaRNAs such as hTR, we found that IP targeting VFP-SmD1 also pulled down snoRNA:Prp8 (Figure 7a). It has been shown that many H/ACA box scaRNAs are TMG-capped [55–58]; consistent with these studies, we also found that scaRNA:Prp8 co-immunoprecipitates with anti-TMG antibodies (Figure 6a).
To identify additional Sm-associated ncRNAs in HeLa cells, we examined known human sno/scaRNA loci. Several of the previously reported scaRNAs, including U85, U87 and U89, showed moderate but significant enrichment in Y12 IPs (Figure 4b; Table S7 in Additional file 1). In addition, we found several other scaRNAs that are highly enriched (Figure 4b; Table S7 in Additional file 1). However, we did not detect any significant enrichment of hTR as previously reported [50] (data not shown). We identified a novel, unannotated Sm-associated ncRNA, which we named SHAN (Sm-associated Hybrid tRNAAsp-containing NcRNA); its predicted secondary structure is shown in Figure S8c in Additional file 1. This new transcript appears to be a chimera between a tRNA gene and an H/ACA type scaRNA gene. Supporting this hypothesis, we detected H box, ACA box and CAB box motifs in the orthologous sequences from other primates (Figure S8b,c in Additional file 1). In summary, our RIP-seq analysis revealed both evolutionarily conserved and newly evolved interactions between Sm proteins and scaRNAs, suggesting that Sm proteins play roles in the biogenesis/function of a subset of scaRNAs. However, we did not identify sequence/structural features that distinguish Sm-associated scaRNAs from other scaRNAs.
Sm proteins associate with mRNAs encoding mitochondrial and translation-related proteins
Due to a relative lack of comprehensive annotation of Drosophila gene ontology, we manually annotated the Sm-associated mRNAs by homolog searching, protein domain analysis, and literature mining. This analysis surprisingly revealed two major categories of mRNAs: those encoding ribosome/translation-related proteins (13/86), and mitochondrial proteins (including mitochondrial ribosomal proteins, 19/86). As discussed above, the enrichment of ribosomal protein mRNAs is not simply due to high levels of expression. Only a subset of ribosomal protein mRNAs is enriched in the Sm protein IPs. For example, mRNAs encoding RpS11 (CG8857) and RpL39 (CG3997) are highly enriched in Sm protein IPs (Figure 3a; Table S5 in Additional file 1), whereas RpL19 (CG2746) and RpL4 (CG5502) are not enriched at all (Figure 4a and data not shown). Anecdotally, the mRNA encoded by CG3776, which is highly enriched, is located immediately adjacent to RpL19 in the Drosophila genome, demonstrating the high degree of specificity of our approach.
Two other Drosophila Sm-associated mRNAs merit special interest. CG4692 encodes a predicted mitochondrial F1-FO ATP synthase subunit that was consistently enriched in our IPs. We found that this mRNA localizes to the actin-rich oocyte cortex of late-stage Drosophila egg chambers (Figure S4 in Additional file 1), in a pattern that is very similar to that of VFP-tagged Sm proteins, as described previously [21]. Analysis of several other high-scoring mRNAs from Figure 3a and Figure S4 in Additional file 1 did not display this pattern (data not shown), so it is not a general feature of Sm-associated mRNAs, but was nonetheless interesting. CG1349 (dj-1beta) encodes a Drosophila homolog of the human DJ-1/PARK7 (Parkinson autosomal recessive, early onset 7) gene. DJ-1/PARK7 is one of 10 genes identified to date that cause familial Parkinson disease [59]. A subpopulation of DJ-1 protein is localized to mitochondria in a regulated manner, and is required for proper mitochondrial function [60]. Thus, it is possible that Sm proteins play a role in regulating the localization and/or translation of associated mRNAs.
In contrast to the more than 70 Sm-associated mRNAs in the fruitfly (Figure 3a), we identified roughly 30 high-scoring mRNAs in human cells (Figure 3b). The lower number in the human dataset is potentially due to a reduced coverage of the transcriptome. Nevertheless, we found that one of the replication-dependent histone mRNAs, HIST2H2AB, is highly enriched in the IPs (Figures 3b and 4b). In contrast, two adjacent histone genes, HIST2H2BE and HIST2H2AC, were not enriched (Figure 4b). Another histone mRNA (HIST1H2AM), was also significantly enriched (Figure 3b). Interestingly, Steitz and colleagues [34] previously showed that the U2 snRNP binds to (intronless) histone pre-mRNAs and stimulates 3′ end processing. Our identification of histone mRNAs in Sm protein co-IPs may reflect a snRNP-mediated interaction between Sm proteins and mRNAs. However, none of the Drosophila replication-dependent histone mRNAs were enriched in the Sm protein IPs (Figure S10 in Additional file 1). Taken together, our data suggest that the mode of interaction between Sm proteins, snRNPs and mRNAs is conserved between vertebrates and invertebrates.
Validation and tissue-specificity of RNA-Sm protein interactions in Drosophila
We have shown that the B/D3 and E/F/G subcomplexes bind essentially the same set of target RNAs. To determine whether SmD1 (which forms heterodimers with SmD2; Figure 1b) also associates with the RNAs listed in Figure 3a, we immunopurified ovarian RNA from daGal4, VFP-SmD1 flies (using anti-GFP) and carried out qRT-PCR. Furthermore, to assay the observed interactions in another cell type, we also performed qRT-PCR on RNAs immunopurified from S2 cells using anti-Sm antibody Y12. We chose six of the top-ranking mRNAs that were identified in the RIP-seq experiments (targeting SmB, SmD3 and SmE), and found that they were all highly enriched in the VFP-SmD1 IPs (Figure 7a). Two snRNAs (U1 and LU) were used as positive controls, whereas three RNAs not expected to interact with Sm proteins (Act5C and Smt3 mRNAs and 5S rRNA) were used as negative controls (Figure 7a). In contrast to the results in ovaries, only four out of the six mRNAs we tested were significantly enriched in the S2 cell IPs (Figure 7a). Given that the Sm proteins and the six mRNAs we tested all have comparable expression levels in both ovaries and S2 cells (Figure 7b and data not shown), these findings suggest that the interactions between mRNAs and Sm proteins can be tissue-specific. A potential concern in all RIP experiments is that the co-purification of the components might be due to reassortment of complexes following cell lysis [61, 62]. However, the fact that CG3997 and CG13410 fail to associate with Sm proteins despite the fact that they are well expressed in S2 cells argues strongly against this artifact.
Sm proteins associate with fully spliced and polyadenylated mRNAs
The identification of significantly enriched mRNAs in the co-IP fractions led us to ask whether the association between Sm proteins and mRNAs was due to the splicing reaction itself. In other words, do Sm proteins interact with partially spliced or fully mature mRNAs? A quick glance at Figure 3 shows that the read depth over intronic sequences is very low. Meta-gene analysis of both Drosophila and human Sm-associated intron-containing mRNAs showed that the vast majority of reads map to exons, and the IPs did not pull down more pre-mRNAs than the controls did (Figure 8a). Among the few transcripts that showed significant numbers of intronic reads, most of those were actually candidates for either new exons or new genes (for example, scaRNA:Prp8 and snRNA:LU; Figure 4a). Thus, this analysis demonstrates that the mRNAs that associate with canonical Sm proteins are fully spliced. Importantly, 6 of the 72 Drosophila Sm-associated mRNAs (CG6008, CG13151, CG13951, CG17531, CG11076 and CG7137), and 2 of the 30 human Sm-associated mRNAs (HIST2H2AB and HIST2H2AM) are intronless, suggesting that splicing is not a prerequisite for Sm protein interaction.
The highly conserved eukaryotic Lsm1-7 complex is known to bind to mRNA degradation intermediates, preferentially those with oligoadenylated tails [14, 63]. We therefore asked whether the canonical Sm ring shares this same recognition specificity. Taking advantage of the oligo(dT)20 and random hexamer primed RIP-seq cDNA libraries, we compared the read coverage patterns for the various mRNAs. As shown in Figure 8b,c, there is a dramatic 3′ end bias in the oligo(dT)20 primed libraries compared to the randomly primed ones. We also confirmed the presence of adenylated tails of Sm-associated and non-associated mRNAs by examining the unmappable reads in the oligo(dT)20 primed RIP-seq files (Figure S11 in Additional file 1). In order to measure polyA tail lengths, we performed RACE-PAT (rapid amplification of cDNA ends-poly(A) tail assay) on immunopurified RNAs from S2 cells [64]. This analysis demonstrates that the poly(A) tails of the Sm-associated mRNAs are roughly the same length as the input mRNAs (Figure 8d). Taken together, these data show that Sm and Lsm proteins have distinct specificities and modes of mRNA interaction.
Sm protein interaction with mRNAs is mediated by snRNPs
The association of snRNAs and scaRNAs with Sm proteins is thought to be mediated by direct binding to Sm sites and CAB boxes, respectively [50, 65, 66]. We therefore wanted to determine whether Sm proteins associate with mRNAs directly or indirectly. Toward that end, we carried out PAR-CLIP (photoactivatable ribonucleoside-enhanced crosslinking and immunoprecipitation) on native and VFP-tagged Sm complexes [67]; however, we were unable to detect any significant crosslinking events in the precipitated RNA (data not shown). We note that canonical Sm proteins are notoriously poor at crosslinking. Even on extremely abundant targets such as U1 snRNA, the UV crosslinking efficiency was rather low, with SmG being the predominant crosslinked member of the heptameric ring [68]. More recently, Castello et al. [69] carried out UV- and PAR-CLIP in parallel to generate a comprehensive mRNA interactome in HeLa cells. As part of their studies, they identified the Lsm1-7 proteins as mRNA binding proteins, but the canonical Sm proteins were not detected, again supporting the idea that Sm proteins are not efficiently crosslinked to mRNAs.
However, the fact that we found all three Sm sub-complexes in association with the same set of mRNAs (Figures 2 and 3) suggested interaction with a complex that contains an intact Sm ring. Furthermore, the previously reported binding between histone mRNAs and U2 snRNPs [34], coupled with our identification of H2A mRNAs in our RIP-seq data (Figure 4) led us to ask whether the mRNA-Sm interaction might be indirect, mediated by snRNPs. Sm-class spliceosomal snRNAs are transcribed by a specialized form of RNA polymerase II and contain a 5′ TMG cap structure [9]. Using anti-TMG antibodies, we immunopurified RNPs from S2 cell lysate and used qRT-PCR to assess the enrichment of mRNAs. As expected, the U1 and LU snRNAs (positive controls) were highly enriched in the anti-TMG IPs, whereas CG7939 (RpL32) mRNA was not (Figure 6a). Notably, the scaRNA:Prp8 transcript and all three of the Sm-associated mRNAs we tested (CG1349, CG3776 and CG4692) were significantly enriched in the anti-TMG pulldowns (Figure 6a). In parallel, we performed anti-TMG IPs using purified S2 cell RNA (that is, the IP was not performed in lysates). We detected significant enrichment of U1 snRNA but not the mRNAs (Figure S12 in Additional file 1). Therefore, the Sm-associated mRNP complex contains a TMG cap component that is structurally distinct from the mRNAs themselves, suggesting the presence of snRNPs.
In order to test whether the interactions with mRNAs are indirectly mediated by snRNPs, we took advantage of a database from a large-scale Drosophila S2 cell RIP-seq analysis of 29 RNA binding proteins, including U1-70 K [70]. The U1-70 K protein binds to U1 snRNA directly and specifically, thus allowing it to be used as an additional, independent epitope for pulldown experiments [68]. We mined the database for RNAs that associate with U1-70 K by analyzing RNAs that were enriched in IPs from U1-70 K transfected versus non-transfected cells. The RIP-seq data were displayed on a volcano plot to identify transcripts that are highly enriched in the IPs. As shown in Figure 6b, U1 snRNA, but not the other spliceosomal snRNAs, was dramatically enriched in the IP fractions, along with a number of other ncRNAs and mRNAs. Among this latter category, three mRNAs were particularly noteworthy: CG3776, CG8108 and U1-70 K (CG8749) itself. Although U1-70 K protein may well bind to its own mRNA for some type of autologous feedback, one must view this result with caution because the cells were transiently transfected with U1-70 K cDNAs, artificially inflating expression of this transcript. However, CG3776 and CG8108 remain good candidates. Interestingly, CG3776 was one of the top-ranking candidates in our ovarian RIP-seq experiments (Figures 3 and 4), but CG8108 was not identified as being enriched, even though it is expressed at similar levels in S2 cells (Figure 6d,e). Because the U1-70 K data were generated from S2 cells, we performed anti-TMG and anti-SmB (Y12) IPs in S2 cells, followed by qRT-PCR. As shown in Figure 6c, we detected significant enrichment of CG8108 in both the TMG and Sm protein IPs. These data provide additional support for the idea that the Sm-mRNA interactions are cell-type specific and not due to reassortment, as CG8108 is expressed in Drosophila ovaries (Figure 6d) but not significantly enriched in Sm protein IPs (Figure 6e).
In addition to CG3776, we also found other U1-70 K associated RNAs that overlapped with our Sm protein dataset, including CG5972 and CR32863. Although it is likely that U1-70 K binds to certain RNAs in a manner that is independent of the U1 snRNP, the overlap between our anti-Sm and anti-TMG data suggests that a cadre of mature mRNAs interacts with intact snRNPs outside of the spliceosome. Thus, we checked for sequence complementarity in CG3776 mRNA and found a 12 bp perfect duplex with the 5′ end of U1 snRNA (Figure 9a). The complementary region is in the middle of the second exon of CG3776, far from any intron-exon boundaries and the base-pairing potential is much greater than is typical for a 5′ splice site. Similarly, we found stretches of complementarity between U1 snRNA and exonic regions of CG8108, CG5972 and many other transcripts (Figure S13 in Additional file 1). Those mRNAs within our dataset that are missing from the U1-70 K pulldowns (for example, CG1349 and CG4692) are plausibly bound by other Sm snRNPs such as U2, U4/U6, U5, U11 and U12. A list of such potential base pairing interactions was compiled by taking known single-stranded regions from snRNAs, and using them to find putative binding sites on the list of Sm- and U1-70 K-associated mature mRNAs (Figure S13 in Additional file 1). We found many potential sites with a duplex length and minimum free energy profile similar to the ones shown in Figure 6f. Taken together with the Sm and TMG IPs, these data suggest that snRNPs associate with subsets of mature Drosophila mRNAs, in a mode that is distinct from their interactions within the spliceosome.
To test whether base pairing between U1 snRNP and CG3776 mRNA is responsible for their interaction, we introduced three synonymous point mutations within the twelve-nucleotide complementary region in CG3776 mRNA that should completely block putative pairing with U1 snRNA (Figure 9a). We then transfected both wild-type and mutant CG3776 mRNA expression constructs into S2 cells (Figure 9b). The constructs are transcribed by an Act5C promoter and are terminated using the SV40 polyA signal and a heterologous 3′ UTR. We confirmed that both transfections produced similar levels of chimeric CG3776 mRNAs (Figure 9c) and then performed Y12 IPs on S2 cell lysates, using normal goat serum as a control. As expected, 5S rRNA was not enriched in the IP fractions, whereas CG1349 mRNA and U1 snRNA were both significantly enriched in the transfections. Both endogenous and transfected CG3776wt mRNAs were pulled down by the Y12 antibody, whereas transfected CG3776mut mRNA was not (Figure 9d). These results support two conclusions. First, splicing is not required for U1 snRNP binding, and the binding site for U1 snRNP is located within the CG3776 mRNA coding sequence, since it can be efficiently pulled down by Y12 antibody. Second, the predicted U1 binding site is indeed necessary for U1 snRNP binding. Taken together, our results suggest that snRNPs bind mature mRNAs, and that at least one mechanism requires U1 snRNP base pairing with target mRNAs.