Stalled Spliceosomes Are a Signal for RNAi-Mediated Genome Defense

SUMMARY

Using the yeast Cryptococcus neoformans, we describe a mechanism by which transposons are initially targeted for RNAi-mediated genome defense. We show that intron-containing mRNA precursors template siRNA synthesis. We identify a Spliceosome-Coupled And Nuclear RNAi (SCANR) complex required for siRNA synthesis and demonstrate that it physically associates with the spliceosome. We find that RNAi target transcripts are distinguished by suboptimal introns and abnormally high occupancy on spliceosomes. Functional investigations demonstrate that the stalling of mRNA precursors on spliceosomes is required for siRNA accumulation. Lariat debranching enzyme is also necessary for siRNA production, suggesting a requirement for processing of stalled splicing intermediates. We propose that recognition of mRNA precursors by the SCANR complex is in kinetic competition with splicing, thereby promoting siRNA production from transposon transcripts stalled on spliceosomes. Disparity in the strength of expression signals encoded by transposons versus host genes offers an avenue for the evolution of genome defense.

INTRODUCTION

RNA interference (RNAi)-related RNA silencing pathways constitute a group of small-RNA-based silencing mechanisms that antedate expansion of the eukaryotic lineage and function throughout this domain of life (Shabalina and Koonin, 2008). Enzymes required for RNA silencing are numerous and can differ between species but universally include Argonaute or PIWI clade proteins, which bind small RNAs. Some RNA silencing pathways also utilize Dicer ribonucleases, which produce small interfering RNA (siRNA) from double-stranded RNA (dsRNA) precursors, and RNA-dependent RNA polymerases, which produce dsRNA. Although RNAi-related systems perform disparate roles in different organisms—from histone modification to translational regulation—a deeply conserved and biologically critical function for these systems, observed from protists to man, is to defend genome integrity by silencing transposable elements (Behm-Ansmant et al., 2006; Cam et al., 2009; Ghildiyal and Zamore, 2009). Yet transposons occur in many families that bear little or no resemblance to each other (Malone and Hannon, 2009), raising the question of how they are recognized as nonself DNA.

One RNAi-related system that suppresses transposon mobilization is the Piwi-interacting small RNA (piRNA) pathway, best understood in Drosophila. piRNAs derive from specific genomic clusters of transposon-related sequences and act with Argonaute proteins of the PIWI clade to silence homologous sequences throughout the genome (Brennecke et al., 2007; Malone and Hannon, 2009). Such a mechanism constitutes an adaptive immunity to transposons, as it silences only transposon families that had previously been incorporated into a piRNA cluster (Khurana et al., 2011). These constraints raise the question of whether eukaryotes also demonstrate innate immunity to transposons, in which prior exposure to a transposon is not required for its recognition.

The processing of long dsRNA into siRNA can be viewed as an innate immune mechanism for transposon defense, capable of recognizing even novel transposons by virtue of their tendency to generate dsRNA. For instance, transposons can produce dsRNA by mobilizing into an existing transcriptional unit or by virtue of transposon-encoded inverted repeats and internal antisense promoters; such dsRNAs template the production of repressive endogenous siRNA in a manner that requires Dicer (Conley et al., 2008; Drinnenberg et al., 2009; Ghildiyal et al., 2008; Sijen and Plasterk, 2003; Yang and Kazazian, 2006). Mutations that block exogenous RNAi, which is triggered by long dsRNA, concomitantly increase endogenous transposon mobilization, providing additional evidence for the role of dsRNA processing in transposon recognition (Ketting et al., 1999; Tabara et al., 1999). Another class of RNAi-related system potentially involved in innate transposon immunity is thought to have evolved to recognize unusual DNA arrangements. The quelling pathway of Neurospora crassa targets repetitive transgene arrays (Lee et al., 2010; Nolan et al., 2008), whereas meiotic silencing of unpaired DNA (MSUD) mechanisms silence transgenes that lack a partner during homolog pairing in meiosis I (Kelly and Aramayo, 2007). Unlike mechanisms of innate transposon immunity that rely on dsRNA recognition, quelling and MSUD can silence loci that do not naturally produce dsRNA, and both require RNA-dependent RNA polymerases (Cogoni and Macino, 1999; She et al., 2009; Shiu et al., 2001). Because transposons are particularly likely to form tandem arrays or occur asymmetrically on homologs during meiosis, quelling and MSUD may suppress transposon mobilization, but their underlying mechanisms remain poorly understood.

The human pathogenic yeast Cryptococcus neoformans offers a genetically tractable model system for approaching the mechanisms of small-RNA-mediated transposon suppression. This organism displays an active RNAi pathway comprising one Argonaute clade member (Ago1), two redundant Dicer orthologs (Dcr1/2), and an RNA-dependent RNA polymerase ortholog (Rdp1). These factors play a key role in defending the C. neoformans genome: null mutations in their corresponding genes result in increased transposon expression, transposon mobilization, and transposon-induced drug resistance mutations (Janbon et al., 2010; Wang et al., 2010). In this paper, we investigate the mechanism of siRNA biogenesis in C. neoformans. Six observations indicate a key role for introns and the spliceosome in this process. First, small RNA analysis reveals that unspliced messenger RNA (mRNA) precursors are preferred substrates for siRNA production. Second, we describe a nuclear RNA-dependent RNA polymerase complex required for siRNA production, termed SCANR (Spliceosome-Coupled And Nuclear RNAi), and find that it physically associates with the spliceosome. Third, we observe that RNAi target transcripts encode suboptimal splicing signals and exhibit unusually high accumulation on spliceosomes. Fourth, we find that deletion of introns from a strong RNAi target blocks the accumulation of siRNA corresponding to this transcript. Fifth, we find that experimental stalling of a pre-mRNA’s splicing dramatically increases its siRNA production in a manner that requires entry into the splicing pathway. Sixth, we demonstrate that the lariat debranching enzyme is required for siRNA synthesis. These results indicate that stalled spliceosomes are a signal for RNAi and that splicing intermediates may be a favored substrate. We propose that a competition between the recognition of spliceosome-associated mRNA precursors by SCANR and the completion of their ongoing splicing plays an important role in specifying sequences from which repressive siRNA is produced.

RESULTS

Endogenous siRNA of C. neoformans Targets Transposon mRNA Precursors

To investigate the mechanism of small RNA biogenesis in C. neoformans genome defense, we used high-throughput sequencing to identify the siRNAs in haploid, vegetatively growing cells. Wild-type cells produced 21–23 nt siRNAs with a strong preference for a uridine residue at their 5′ termini, a characteristic feature of Argonaute-bound siRNAs in other fungal systems (Figure 1A) (Drinnenberg et al., 2009). These siRNAs were absent from cells lacking Ago1 or Rdp1 (Figures 1B and 1C). siRNAs mapped most prominently to transposons: 22% of reads mapped to centromeres, which are composed primarily of transposons and transposon remnants (Loftus et al., 2005), and 32% of reads mapped to intergenic regions that have sequence similarity to centromeres (Figure 1D). An additional 39% of siRNA reads mapped to mRNA-encoding genes, many of which had sequence similarity to centromeres. Genic siRNAs displayed a strand bias, as 95% of reads had a polarity opposite that of the predicted transcript at their corresponding loci. These findings indicate that the previously described broad targeting of transposon transcripts by siRNA that occurs during mating is also a feature of vegetative growth in C. neoformans (Wang et al., 2010).

Figure 1. Endogenous siRNA of C. neoformans.

(A–C) Read counts for siRNAs based on length and 5′ nucleotide identity.

(D) Genomic mapping of siRNA sequences.

(E) Density plot of siRNAs mapping to an RNAi target locus, CNAG_7721, which comprises two exons and one intron.

See also Tables S1 and S2.

Transcripts targeted by RNAi corresponded to loci on each of the 14 C. neoformans chromosomes (e.g., Figure 1E). These RNAi targets were not enriched for convergently transcribed genes, suggesting that dsRNA generated by bidirectional transcription is not a major driver of siRNA generation (Table S1 available online). Strikingly, siRNA reads in genes mapped not only to exons but also to introns; more than 15% of these siRNAs contained intronic sequence, whereas introns constitute 18% of all genic sequences. Moreover, the 50 genes targeted by the greatest number of siRNA reads exhibited not only intronic siRNAs (in 45 cases) but also siRNAs that spanned intron-exon junctions (in 42 cases) (Table S2). siRNA reads spanning exon-exon junctions were also observed within this gene set at a frequency only 2.45-fold that of intron-exon junction reads, despite the fact that the vast majority of cellular RNAs are expected to be fully spliced (Table S2). The sequence features of siRNAs, together with their dependence on Rdp1, are consistent with a model in which Rdp1 acts preferentially upon intron-containing mRNA precursors to generate dsRNA, which is processed into siRNA to effect silencing of transposons and other RNAi targets. This model is further developed below.

Identification and Characterization of Proteins Associated with RNA-Dependent RNA Polymerase

To gain further insights into how the RNAi machinery might be targeted to mRNA precursors in the nucleus, we used tandem affinity purification and mass spectrometry to identify proteins associated with Rdp1, the only component of the C. neoformans RNAi machinery that has been reported to be nuclear (Wang et al., 2010). Purification of Rdp1-CBP-2xFLAG, expressed from its endogenous locus, yielded Rdp1 itself as well as four additional proteins: Ago1 and three proteins we named Qip1, Gwc1, and Srr1 (Figure 2A). Qip1 is an ortholog of QIP, an N. crassa exonuclease that binds Argonaute and degrades the passenger strand of siRNA duplexes (Maiti et al., 2007). Gwc1 (GW-containing) contains five GW/WG dipeptides—a motif commonly found in Argonaute-binding proteins (El-Shami et al., 2007)—but no other obvious domain or homolog. Srr1 (Serine/Arginine-rich) has no clear ortholog in other organisms, but, intriguingly, its domain structure resembles those of several mammalian splicing factors. In particular, Srr1 contains 30 RS/SR dipeptides and three RNA recognition motifs.

Figure 2. Functional Characterization of Rdp1-Associated Proteins.

(A) Proteins associated with Rdp1-CBP-2xFLAG by tandem affinity purification. Likely contaminants and proteins with <10% sequence coverage have been excluded.

(B) Transcript levels of three RNAi target genes, assessed by RT-qPCR and normalized to 18S rRNA levels. Error bars represent SEM.

(C) sRNA Northern blot. Loading control: U18 snoRNA.

(D) Localization of Rdp1-associated proteins. Proteins were fused to fluorescent protein tags and detected by epifluorescence microscopy. Nuclei were labeled using Hoechst dye. Percentage of cells exhibiting cytoplasmic localization versus dual nuclear and cytoplasmic localization is indicated. White arrow: cytoplasmic localization; white arrowhead: nuclear and cytoplasmic localization.

(E) Proteins that interact with the Rdp1-associated proteins Gwc1, Qip1, and Ago1.

(F) Protein interaction network of Rdp1-associated proteins. The purification results of (A) and (E) are represented graphically.

(G) Predicted protein domains of SCANR subunits.

(H) Protein-protein interactions among SCANR components, as assessed by yeast two-hybrid assay.

See also Figure S1 and Table S3.

To assess the potential function of each protein that copurified with Rdp1, we attempted to create deletion mutations in their corresponding genes. We were successful for Ago1, Qip1, and Gwc1 but were unsuccessful for Srr1, suggesting that it may be essential for viability. Strains lacking Ago1, Rdp1, Gwc1, or Qip1 exhibited increased levels of three transcripts highly targeted by siRNA in our sequencing experiments: CNAG_6757, an unannotated gene with homology to a transposon; CNAG_6844, a RecQ helicase; and CNAG_6705, an unannotated open reading frame (ORF) (Figure 2B). The same strains also displayed a loss of siRNA corresponding to these transcripts, indicating a functional requirement for Ago1, Rdp1, Gwc1, and Qip1 in RNAi (Figure 2C).

The isolation of Ago1 as an Rdp1-associated protein was surprising because it had been reported to localize exclusively to cytoplasmic P bodies (Wang et al., 2010). However, we found that, although an mCherry-Ago1 fusion protein was cytoplasmic in most cells, a substantial fraction of cells (26%) displayed localization in both the nucleus and cytoplasm (Figure 2D). The cytoplasmic signal of mCherry-Ago1 localized both diffusely and in foci corresponding to P bodies (Figure S1A). GFP-Gwc1 and GFP-Qip1 fusion proteins displayed a nuclear and cytoplasmic pattern in all cells, which is consistent with a fraction of these proteins associating with Rdp1 in the nucleus (Figure 2D).

To characterize the physical relationships among Rdp1-associated proteins, we performed tandem affinity purifications of Gwc1, Qip1, and Ago1. Protein identification by mass spectrometry revealed a dense network of interactions among these proteins, which is consistent with the existence of a protein complex (Figures 2E and 2F). Gwc1 and Qip1 purifications yielded only proteins that were also identified in an Rdp1 purification, whereas an Ago1 purification yielded all of these proteins as well as six additional proteins, which we named Skp1, Gwo1, Aga1, Aga2, Bre1, and Aga3. These results can be most simply explained by proposing that Rdp1, Gwc1, Ago1, and Qip1 form a protein complex required for siRNA production, whereas Ago1 participates additionally in at least one other protein complex. For reasons described below, we refer to the Rdp1 complex as SCANR (Figure 2G). We note that the dual nuclear and cytoplasmic localization of Ago1, Gwc1, and Qip1 suggests the existence of a cytoplasmic subcomplex that lacks Rdp1. Yeast two-hybrid experiments using each SCANR component as bait and prey revealed interactions involving every subunit, further supporting the associations identified by tandem affinity purification (Figure 2H and Table S3).

We tested whether any of the proteins that copurified with Ago1, but not with any other SCANR protein (Skp1, Gwo1, Aga1, Aga2, Bre1, and Aga3), are required for silencing of mRNAs targeted by RNAi. We were able to delete each corresponding gene except SKP1, which may be essential for viability, as it is in S. cerevisiae (Connelly and Hieter, 1996). Of the five mutants, only the gwo1Δ strain exhibited derepression of RNAi target transcript levels, albeit to a lesser extent than that of strains lacking SCANR components (Figures 3A and 3B). Loss of Gwo1 did not affect siRNA accumulation, however, indicating that Gwo1 is not required for siRNA biogenesis (Figure 3C). Tandem affinity purification of Gwo1, followed by mass spectrometry, identified only Gwo1 itself and Ago1, suggesting that these two factors form a protein complex distinct from SCANR (Figures 3D and 3E). Indeed, yeast two-hybrid analysis detected an interaction between Gwo1 and Ago1, but not between Gwo1 and the SCANR components Rdp1, Qip1, or Gwc1 (Figure 3F and Table S3). Furthermore, analytical tandem affinity purification of Ago1 yielded Gwo1, whereas purification of the SCANR component Rdp1 did not yield Gwo1, as assessed by immunoblotting (Figure 3G). Finally, localization experiments indicated that a GFP-Gwo1 fusion protein resided in cytoplasmic foci that colocalized with Ago1 foci and with the P body marker Dcp1 (Figures 3H and S1B). We therefore refer to the Ago1-Gwo1 complex as PRSC (P-body-associated RNA Silencing Complex) and conclude that it is distinct from the nuclear SCANR complex. Investigation of the function of PRSC is ongoing.

Figure 3. PRSC and SCANR Subunits Are Physically and Functionally Distinct.

(A and B) Transcript levels of three RNAi target genes, assessed by RT-qPCR and normalized to 18S rRNA levels. Error bars represent SEM. *p < 0.05 by Student’s t test.

(C) sRNA Northern blot. Loading control: U18 snoRNA.

(D) Proteins associated with Gwo1-CBP-2xFLAG. Assayed as in Figure 2A.

(E) Predicted protein domains of PRSC subunits.

(F) Protein-protein interactions among PRSC components, as assessed by yeast two-hybrid assay. No interactions were detected between Gwo1 and any exclusive member of SCANR.

(G) Coimmunoprecipitation of Gwo1 with Ago1, but not with Rdp1. Strains expressing Gwo1-13xMyc and CBP-2xFLAG-Ago1 or Rdp1-CBP-2xFLAG were subjected to FLAG-CBP tandem purification. Input and purified material were analyzed by immunoblot using anti-FLAG, anti-Myc, or anti-PSTAIRE, which detects the negative control protein p31.

(H) Colocalization of PRSC component Gwo1 with the P body marker Dcp1. Unfixed cells were examined after incubation in yeast nitrogen base (YNB) media.

See also Figure S1 and Table S3.

SCANR Associates with the Spliceosome

Given that unspliced mRNA precursors appear to be preferred substrates for dsRNA synthesis, we were intrigued by the fact that Srr1, which was identified by mass spectrometry in purifications of two SCANR components, resembles mammalian splicing regulators (Figure 4A). To confirm a physical interaction between SCANR and Srr1, we performed analytical tandem affinity purifications of SCANR components and used immunoblotting to detect associated proteins. Purifications of each SCANR subunit yielded Srr1 (Figures 4B–4D). In contrast, purification of the RNAi factor Dcr1, which has been reported to localize to P bodies (Wang et al., 2010), did not yield Srr1, nor was an unrelated control protein, p31, detected in any purification. Consistent with its SCANR association, an Srr1-mCherry fusion protein displayed a nuclear localization (Figure 4E).

Figure 4. SCANR Physically Associates with Srr1 and the Spliceosome.

(A) Predicted domains of Srr1.

(B–D) Coimmunoprecipitation of Srr1 with SCANR components Ago1 (B), Gwc1 (C), Qip1 (C), and Rdp1 (D), but not with Dcr1 (B). Assayed as in Figure 3G.

(E) Localization of Srr1. An Srr1-mCherry fusion protein was detected in fixed cells. Nuclei were stained using Hoechst dye.

(F) Interaction of Srr1 and spliceosomal snRNAs. Levels of individual RNAs coimmunoprecipitated with Srr1 were assessed by RT-qPCR and normalized to their abundance in whole-cell extract; transcript level is relative to that obtained in purifications from wild-type (untagged) lysates. Error bars represent SD.

(G) Proteins associated with CBP-2xFLAG-Srr1. Genes were named based on S. cerevisiae orthologs or, in the absence of one, based on metazoan orthologs. Known spliceosome components are colored in red.

(H) Coimmunoprecipitation of the spliceosome component Syf1 with Ago1, but not with Dcr1. Assayed as in Figure 3G.

See also Figure S2.

Given its domain structure, we tested whether Srr1 physically associates with spliceosomal snRNAs by immunoprecipitating Srr1 and detecting bound RNAs by RT-qPCR. All four annotated C. neoformans spliceosomal snRNAs (U2, U4, U5, and U6), but not a control small nucleolar RNA (snoRNA) (U18), associated with Srr1 (Figure 4F). We next identified proteins associated with Srr1 using tandem affinity purification and mass spectrometry. Remarkably, of the 23 proteins that copurified with Srr1, 17 were orthologs of known spliceosomal proteins (indicated in red, Figure 4G). These included components of spliceosomal snRNPs, such as Smb1, Prp4, and Msl1, and other spliceosome-associated complexes, such as Prp19, Syf1, Cef1, Cwc2, and Isy1 of the NineTeen complex (NTC). Detection of SCANR components among Srr1-associated proteins was below our 10% coverage threshold, which is consistent with the lower sensitivity of this assay relative to immunoblotting. These data establish Srr1 as a nuclear protein that interacts with spliceosomal complexes.

The observed interaction between Srr1 and RNAi factors of SCANR, together with the associations between Srr1 and the spliceosome, suggested that SCANR might physically associate with the spliceosome but too weakly to be detected by mass spectrometry methods. To test this hypothesis, we determined whether Ago1 and Rdp1 physically interact with the NTC component Syf1, which is part of the spliceosome at multiple stages of its assembly and during the catalytic steps of splicing (Hogg et al., 2010). Using immunoblotting, we detected Syf1 in tandem affinity purifications of Ago1 or Rdp1, but not upon purification of Dcr1, which is not a SCANR component (Figures 4H and S2). These findings indicate that SCANR physically associates with at least a subset of spliceosomal complexes marked by Syf1.

RNAi Target Transcripts Display High Spliceosome Occupancy

Our finding that mRNA precursors are a substrate not only for splicing but also for siRNA production (Figure 1) led us to hypothesize that the spliceosome and RNAi may compete for their substrates in some way (Figure 5A). In this scenario, the slow splicing of a transcript would, by virtue of increasing its time spent as an mRNA precursor, increase its availability as a substrate for the RNAi pathway, thereby biasing RNAi targeting toward poorly spliced genes.

Figure 5. Transcripts Targeted by RNAi Display Suboptimal Splicing Features and Are Stalled on Spliceosomes In Vivo.

(A) Model of competing pathways for mRNA precursor maturation.

(B) Analysis of longest intron lengths of siRNA targets (>150 siRNA reads) versus all genes. Cumulative distribution functions of longest intron length for these two gene classes are plotted.

(C) Analysis of weakest 5′ splice sites of siRNA targets (>150 siRNA reads) versus all genes. Cumulative distribution functions of weakest 5′ splice site strengths for these two gene classes are plotted. Inset: sequence logo generated from all C. neoformans 5′ splice sites.

(D) Effect of intron deletion on the production of siRNA from the CNAG_6705 transcript. Above: RNA was isolated from cells that expressed, from a GAL7 promoter, either wild-type CNAG_6705 or a mutated form of the gene in which one or both introns were deleted. Below: the transcript level of each CNAG_6705 mutant was measured by RT-qPCR and normalized to levels of actin RNA. Error bars represent SD.

(E) Association of siRNA target transcripts and non-siRNA target transcripts with the spliceosome. Spliceosomes tagged with Prp19-CBP-2xFLAG were immunoprecipitated, and copurified RNAs were detected by RT-qPCR. Levels of individual RNAs were normalized to their abundance in whole-cell extract; IP/WCE values are relative to those of purifications from wild-type (untagged) lysates. Error bars represent SD. Bold: transposon-related transcripts.

See also Figure S3.

An important prediction of such a model is that genes targeted by RNAi should encode inefficiently spliced transcripts. As an initial test of this idea, we examined intron features likely to affect splicing efficiency. Unlike S. cerevisiae, in which most genes have no introns and genes with introns typically have only one, C. neoformans genes average five introns per gene, and it is thought that nearly all, if not all, genes harbor introns (Loftus et al., 2005). Furthermore, whereas S. cerevisiae intronic splicing signals conform closely to consensus, C. neoformans splice sites are considerably more degenerate, and in this respect they resemble those of metazoans (Irimia et al., 2007; Loftus et al., 2005). Although C. neoformans intron splice sites possess generally low information content, intron size is highly constrained: intron lengths are tightly distributed around a mode of 52 nt, and evolutionary studies of the Cryptococcus species complex suggest that an optimal intron size is under selection (Hughes et al., 2008).

We examined the aforementioned splicing features of RNAi target and nontarget transcripts to determine whether the RNAi pathway acts preferentially on transcripts with suboptimal introns. The vast majority of both gene classes contained introns—RNAi target transcripts encode 3.9 introns per gene on average, as compared to 5.1 for all C. neoformans genes. Because a single poorly spliced intron in a multi-intron transcript would, in principle, be sufficient to stall maturation of the entire transcript, we classified each gene based on the properties of its weakest intron with respect to a given feature. We observed that the longest annotated introns of RNAi target genes were, on average, considerably longer than those of all genes (Figure 5B; p < 1 × 10⁻⁸ by two-tailed Kolmogorov-Smirnov [K-S] test). In contrast, the strengths of the weakest 5′ splice sites of RNAi target genes were, on average, only modestly weaker than those of all C. neoformans genes (Figure 5C; p = 0.013 by two-tailed K-S test). Because the 5′ splice site is critical for initiation of spliceosome assembly, whereas experimentally increasing intron length can, depending on the site, block later steps of the splicing pathway (Cellini et al., 1986; Chua and Reed, 2001), these findings raise the possibility that siRNA target transcripts enter the splicing pathway but become stalled at a later stage. Stalled splicing may thereby provide an opportunity for the spliceosome-associated SCANR complex to target mRNA precursors for RNAi.

As an initial test of this hypothesis, we examined whether causing an RNAi target transcript to bypass splicing would reduce its ability to template the production of siRNA. To do so, we generated mutants of an RNAi target transcript, CNAG_6705, in which its two introns were deleted, singly or together. Each of these mutations dramatically reduced the accumulation of siRNA corresponding to CNAG_6705 (Figure 5D). Intron deletion also decreased steady-state levels of the CNAG_6705 transcript, but this effect appeared insufficient to explain the loss of CNAG_6705 siRNA caused by the mutations (Figure 5D). These results are consistent with a model—tested further below—in which spliceosome engagement is important for siRNA production.

To examine more directly the interactions between endogenous RNAi target transcripts and the splicing machinery, we sought to measure the in vivo occupancy of mRNA precursors on the spliceosome. We immunoprecipitated NTC-containing spliceosomal complexes using an epitope-tagged Prp19 and examined levels of associated transcripts by RT-qPCR, normalizing the data for total transcript levels. Strikingly, the 12 transcripts most highly targeted by siRNA exhibited dramatically greater spliceosome occupancy than did six non-RNAi target genes that spanned a broad range of expression level, intron number, and gene size (Figure 5E). These 12 RNAi targets include seven transposon-related genes (indicated in bold, Figure 5E), whereas the non-RNAi targets include genes encoding actin (CNAG_0483) and GAPDH (CNAG_6699). We found that the accumulation of RNAi target transcripts on spliceosomes was maintained in the context of an rdp1Δ strain, demonstrating that this property is not caused by siRNAs themselves (Figure S3). Thus, transcripts targeted by RNAi exhibit intronic sequence features predictive of poor splicing and accumulate abnormally on spliceosomes in vivo.

Mutations that Stall Splicing Trigger RNAi

Our finding that RNAi target transcripts accumulate on spliceosomes led us to hypothesize that these transcripts proceed inefficiently through the splicing pathway, during which some fraction is redirected to the RNAi pathway and converted into siRNA. This hypothesis predicts that the entry of a transcript into the RNAi pathway would be improved by intronic mutations that stall its splicing at an intermediate stage. To test this prediction, we introduced intronic mutations into a single-copy, endogenous gene that is very weakly targeted by RNAi: CNAG_7888 (Figure S4A). CNAG_7888 contains two predicted introns whose locations we verified by complementary DNA sequencing. To facilitate detection of siRNA generated from the CNAG_7888 locus by Northern hybridization, its promoter was replaced with the strong GAL7 promoter. We assessed two classes of intronic mutations: 5′ splice site mutations, which we predicted to block splicing prior to spliceosome assembly, and 3′ splice site mutations, which we predicted to allow spliceosome assembly but block the second catalytic step of splicing.

Cells expressing wild-type CNAG_7888 exhibited a very low level of siRNA, comparable to the background observed in cells in which the CNAG_7888 locus was deleted (Figure 6A). No change in siRNA levels was detected in the context of CNAG_7888 alleles containing 5′ splice site mutations of either intron. In contrast, a 3′ splice site mutation—specifically, that of intron 2—resulted in dramatically increased siRNA production from CNAG_7888. This siRNA production required Rdp1, confirming that it represented entry of CNAG_7888 into the RNAi pathway (Figure S4B). The inability of other CNAG_7888 mutant alleles to promote siRNA generation could not be explained by defects in their expression, as all mutants showed similar transcript levels (Figure S4C). Additionally, it was not the case that the sequence of intron 2 is uniquely capable of promoting RNAi, as a CNAG_7888 construct in which the sequences of introns 1 and 2 had been exchanged similarly triggered siRNA synthesis only when the 3′ splice site of the downstream intron was mutated (Figure S4D). Evidently, the position of the mutated intron in CNAG_7888 (or perhaps its contiguous exonic sequence) influences its ability to stimulate siRNA production. The nature of the 3′ splice site mutation is unimportant: multiple, distinct 3′ splice site mutations in the second intron stalled CNAG_7888 splicing at the second catalytic step, as assessed by primer extension detection of lariat intermediate, and each strongly triggered siRNA production (Figure 6B).

Figure 6. Stalled Spliceosomes Promote RNAi in a Manner that Depends on Entry into the Splicing Pathway.

(A) Effect of perturbed splicing on the production of siRNA from the CNAG_7888 transcript. RNA was isolated from cells that expressed, from a GAL7 promoter, either wild-type CNAG_7888 or a mutated form of the gene in which a single splice site was mutated.

(B) Effect of 3′ splice site mutations on siRNA production from and splicing of the CNAG_7888 transcript. RNA was isolated from cells that expressed, from a GAL7 promoter, either wild-type CNAG_7888 or a mutated form of the gene in which the 3′ splice site of intron 2 was mutated. siRNA production was assessed by riboprobe hybridization, and splicing was assessed by primer extension using a labeled primer complementary to CNAG_7888 exon 3. Loading control: U6 snRNA primer extension product.

(C) Spliceosome dependence of siRNA accumulation triggered by 3′ splice site mutations. RNA was isolated from cells that expressed, from a GAL7 promoter, either wild-type CNAG_7888 or a mutated form of the gene in which the 5′ and 3′ splice sites of intron 2 were mutated, singly or together, and assayed as in (B).

(D) Association of the spliceosome with CNAG_7888 transcripts containing intron 2 splice site mutations, assayed as in Figure 5E. Error bars represent SD.

See also Figure S4.

Based on our finding that RNAi target transcripts are enriched for introns predicted to be longer than optimal (Figure 5B), we hypothesized that introns containing an increased distance from the branchpoint adenosine to the 3′ splice site would promote RNAi because such introns stall splicing in other systems (Cellini et al., 1986; Chua and Reed, 2001). We generated alleles of CNAG_7888 that contained 75 or 100 nt insertions of adenosine-free sequence between the mapped branchpoint and 3′ splice site of intron 2 and then examined their splicing efficiency and siRNA production. As predicted, both of these insertion mutations reduced intron 2 splicing efficiency, although not as severely as did mutation of the 3′ splice site, as assessed by level of lariat intermediate (Figure S4E). Accordingly, both insertion mutations triggered siRNA production from CNAG_7888 but to a lesser extent than that triggered by a 3′ splice site mutation. This inverse correspondence between splicing efficiency and siRNA accumulation across a range of splicing efficiencies is consistent with a kinetic competition between splicing and siRNA production and indicates that multiple types of suboptimal introns can promote RNAi.

Entry into the Splicing Pathway Is Required for RNAi

Our finding that the suboptimal introns that promote RNAi tend to cause stalling of the spliceosome, together with our observation that RNAi target transcripts accumulate on spliceosomes, led us to hypothesize that spliceosome assembly at an intron is required for that intron to promote RNAi. This hypothesis predicts that a 5′ splice site mutation, which disrupts the initial recognition of an intron by the spliceosome, should suppress the RNAi-promoting effects of 3′ splice site mutations in CNAG_7888. Strikingly, a 5′ splice site mutation of intron 2, which caused no effect on RNAi by itself, fully suppressed the siRNA production triggered by a 3′ splice site mutation (Figure 6C). As expected, the 5′ splice site mutation also suppressed the accumulation of intron 2 lariat intermediate observed in a 3′ splice site mutant.

To confirm that 3′ splice site mutations stall spliceosomes and that this effect is eliminated by a 5′ splice site mutation, we employed the spliceosome occupancy assay described above. As expected based on its exceedingly weak production of siRNA, wild-type CNAG_7888 exhibited low spliceosome occupancy, similar to that of non-RNAi target transcripts (Figure 6D). A 5′ splice site mutation of intron 2 caused no change in spliceosome occupancy, whereas a 3′ splice site mutation dramatically increased spliceosome occupancy, which is consistent with our finding that this mutation blocks the second catalytic step of splicing. In accord with the well-established role of the 5′ splice site sequence in mediating entry into the splicing pathway, a 5′ + 3′ double splice site mutant displayed low spliceosome occupancy. The correlation between spliceosome occupancy and siRNA levels in this series of CNAG_7888 mutant transcripts mirrors our observation that RNAi target transcripts generally display greater spliceosome occupancy than do non-RNAi target transcripts, supporting the view that stalled spliceosomes are a signal for RNAi.

Splicing Intermediates May Be a Preferred Substrate for siRNA Biogenesis

Our finding that inefficient progression of CNAG_7888 through the splicing pathway increases its targeting by RNAi raised the possibility that splicing intermediates of this transcript, such as the intron 2 lariat intermediate, represent preferred substrates for siRNA biogenesis. We therefore tested whether siRNA production from CNAG_7888 requires the lariat debranching enzyme (Dbr1), which debranches discarded lariat intermediates and excised lariats in order to allow processing of these RNA species (Chapman and Boeke, 1991; Mayas et al., 2010). We found that Dbr1 was absolutely required for the siRNA production stimulated by a 3′ splice site mutation of CNAG_7888 intron 2 (Figure 7A). Strikingly, we also observed that debranchase was essential for accumulation of siRNA corresponding to the strong RNAi target transcripts CNAG_6757, CNAG_6844, and CNAG_6705 (Figure 7B). These results suggest that debranching of splicing intermediates may enable their use as preferred templates for siRNA production. Such a model predicts that the first exon of either a single- or multi-intron pre-mRNA would be an underrepresented substrate for siRNA biogenesis because this region cannot be encoded by any lariat intermediate. Consistent with this prediction, we found that, among the top 50 RNAi target transcripts, the siRNA read density in annotated first exons—relative to the overall exonic density in their corresponding genes—was significantly less than the same metric calculated for annotated last exons (Figure 7C). Together, these observations suggest that debranching of splicing intermediates stalled on spliceosomes may be a generally important step in siRNA production.

Figure 7. Lariat Debranchase Is Required for siRNA Production.

(A) siRNA accumulation triggered by a 3′ splice site mutation of CNAG_7888 requires the lariat debranching enzyme. RNA was isolated from wild-type or dbr1Δ cells that expressed, under the control of a GAL7 promoter, wild-type CNAG_7888 or a mutated form of the gene.

(B) sRNA Northern blot. Loading control: U6 snRNA.

(C) Relative siRNA read density in the first or last exon of RNAi target genes. For each of the top 50 RNAi target genes, the following ratio was calculated: siRNA read density in the first (or last) exon over siRNA read density across all exons of the corresponding ORF. These values were log₁₀ transformed and are presented as box plots; blue line indicates median. Values for first versus last exons were compared by the two-tailed Mann-Whitney U test.

(D) Kinetic competition model. The utilization of particular mRNA precursors by SCANR is influenced by their splicing efficiency due to a kinetic competition between splicing and dsRNA synthesis: transcripts that are inefficiently spliced, such as foreign genetic elements, exhibit increased spliceosome association, thereby facilitating their conversion to dsRNA, and, ultimately, siRNA. In this hypothetical example, a lariat intermediate is produced by stalled splicing of a transcript’s first intron; downstream introns remain incompletely spliced. This intermediate is acted upon by SCANR and Dbr1 to produce a dsRNA substrate for Dcr1/2. See text for details.

DISCUSSION

RNAi-related systems that target transposons are important guardians of genome integrity, but our understanding of how self and nonself DNA are distinguished is incomplete. Whereas piRNA systems offer an adaptive mechanism for transposon recognition, the existence of innate mechanisms can also be inferred, particularly for cases in which siRNA is produced against transposons that form dsRNA. Our studies of the human pathogenic yeast Cryptococcus neoformans provide multiple lines of evidence that the process of pre-mRNA splicing by the spliceosome plays an important role in defining targets for the RNAi machinery.

Specifically, we find that transcripts targeted by RNAi are stalled in spliceosomes and that stalling promotes siRNA synthesis through a spliceosome-associated RNAi complex, SCANR. Our results are compatible with a competition between the SCANR-mediated recognition of unspliced mRNA precursors and the completion of their splicing, in which inefficiently spliced transcripts are channeled toward dsRNA synthesis. Although previous studies in S. pombe, C. elegans, and A. thaliana have demonstrated that mutation or knockdown of RNA processing factors, including essential splicing factors, can affect siRNA levels, in no case has it been clarified whether these effects are direct, in part because the underlying mechanisms have not been elucidated (Bayne et al., 2008; Herr et al., 2006; Kim et al., 2005; Robert et al., 2005; Tabach et al., 2012). Below, we summarize evidence that intron-containing mRNA precursors are preferred substrates for siRNA synthesis. We discuss how this finding, together with the observation that transposon transcripts display unusually high spliceosome occupancy, suggests a kinetic competition model for RNAi targeting, and we describe potential advantages of such a mechanism. Finally, we speculate on the evolutionary forces that may have led to the high spliceosome occupancy observed for transposon RNAs.

Intron-Containing mRNA Precursors Are a Template for siRNA in C. neoformans

Our siRNA sequencing data demonstrate that siRNAs map to the introns and intron-exon junctions of most RNAi target genes, suggesting that either DNA or intron-containing mRNA precursors are a template for siRNA synthesis. Several lines of evidence point to mRNA precursors as relevant templates. First, our studies of SCANR show that this RNA-dependent RNA polymerase complex is essential for siRNA production and associates with the spliceosome, which itself assembles on pre-mRNA. More importantly, we show that deletion of introns from a strong RNAi target gene reduces the level of its corresponding siRNA, whereas introduction of a 3′ splice site mutation into a weak RNAi target gene, which stalls its splicing, increases the level of its corresponding siRNA. Crucially, the latter effect is blocked by a 5′ splice site mutation that prevents spliceosome association with the transcript. Finally, we find that the lariat debranching enzyme, which acts on branched RNA species generated by the action of the spliceosome (both lariat intermediate and excised lariat), is required for siRNA accumulation. Taken together, these data argue that intron-containing mRNA precursors are templates for dsRNA synthesis and siRNA production.

Spliceosomal Stalling of Transposon Transcripts Suggests a Kinetic Competition

A second key finding of our studies is that RNAi target transcripts display much higher occupancy on spliceosomal complexes than do transcripts that do not enter the RNAi pathway. Importantly, RNAi itself is not responsible for this difference, as cells lacking siRNA display the same behavior. These findings suggest a kinetic competition model for the specification of transposon transcripts as substrates for siRNA synthesis (Figure 7D). In this model, mRNA precursors stalled on the spliceosome remain available for recognition by SCANR, whereas transcripts that complete splicing undergo spliceosome disassembly, intron degradation, and export out of the nucleus. The nuclear localization of SCANR and its physical association with the spliceosome would facilitate the proposed competition. Such a spliceosome-based mechanism is advantageous in that it does not respond only to particular sequences but rather to splicing efficiency itself and, therefore, could in principle recognize multiple types of suboptimal introns, including those of foreign genetic elements not previously encountered.

Although our experiments addressing transcript occupancy on Prp19-containing spliceosomal complexes do not speak to what specific kinetic steps are delayed during the splicing of transposon transcripts, other results suggest that spliceosome stalling events subsequent to the first catalytic step may be particularly capable of engaging SCANR. In fact, we find that a 3′ splice site mutation of CNAG_7888, which stalls splicing at the second catalytic step, promotes siRNA production from this locus. Furthermore, this siRNA production requires the lariat debranching enzyme, as does the production of siRNA corresponding to several other endogenous RNAi targets. Because loss of Dbr1 in other systems causes an accumulation of lariat RNAs but does not perturb splicing or spliceosome disassembly (Mayas et al., 2010; Nam et al., 1997), this finding suggests that lariat RNAs generated by the first step of splicing, once de-branched, may be preferred substrates for Rdp1 and/or Dicer in C. neoformans.

The Inefficient Splicing of Pre-mRNAs that Contain Transposons

The genome of C. neoformans is rich in introns, as more than 97% of annotated genes contain at least one intron. In other organisms with intron-rich genomes, such as mammals, splicing enhances gene expression by promoting 3′ end formation, nuclear export of mRNA, and mRNA translation (Lu and Cullen, 2003; Valencia et al., 2008). Intron-dependent gene expression is also a feature of basidiomycetous yeast, raising the possibility that successful transposons in C. neoformans require introns (Burns et al., 2005; Lugones et al., 1999). In fact, all of the 12 genes most targeted by siRNA in C. neoformans, the majority of which are related to transposons, contain introns. Yet all of these transcripts exhibit abnormally high spliceosome occupancy, suggesting that their introns may be poorly spliced. Furthermore, RNAi target transcripts generally exhibit sequence features predictive of inefficient splicing, including relatively low 5′ splice site strength and increased intron length. But why would these foreign genetic elements tend to be poorly spliced as compared to endogenous genes? We suggest two nonmutually exclusive possibilities.

First, transposons that have entered the C. neoformans genome recently by horizontal transfer have had limited time to adapt to its specific splicing preferences. Efficient splicing in C. neoformans likely requires appropriate intron size, consensus-matched splice site sequences, and proper exonic splicing enhancer sequences (Warnecke et al., 2008); these particular splicing preferences differ among organisms, creating a barrier to the efficient expression of horizontally transferred genes. Thus, the suboptimal 5′ splice sites and intron sizes of C. neoformans RNAi targets could be due in part to their limited coevolution with endogenous genes.

A second possibility is based on an extensive literature describing cryptic introns in transposons that appear to minimize the impact of transposon insertion into host genes (Purugganan, 2002). Transposon disruptions of essential host genes can result in the host cell’s death, which is deleterious for both the transposon and the host. These same transposon insertions, however, can be viable if transposon sequences are spliced out of the essential gene’s transcript. Therefore, splicing allows transposons to circumvent some negative fitness consequences associated with their propagation (Gierl, 1990). Consistent with this idea, multiple transposons—including Ds in maize, the 412 retrotransposon and P elements in Drosophila, and Tc1 in C. elegans—contain splice sites near their termini such that the whole transposon can be spliced out of a larger transcription unit, thereby limiting disruption of endogenous loci (Belancio et al., 2008; Purugganan and Wessler, 1992; Purugganan, 2002). Many other transposons, including Harbinger family DNA transposons and LINE elements, also contain introns, although these introns do not span the entire transposon (Belancio et al., 2006; Kapitonov and Jurka, 2004). If a transposon, after its insertion into a host gene, is spliced in such a way that mitigates its negative effects on host gene expression but also removes sequences important for mobilization, then weak splicing would be favored as a compromise between transposon expression and host organism health (Menssen et al., 1990). Such inefficiently spliced transcripts may become stalled in the spliceosome and recognized by SCANR, thereby contributing to the high spliceosome occupancy of transposon transcripts and to the production of transposon-specific siRNA for genome defense. Whether these principles apply to other organisms remains to be elucidated, but we are intrigued by the recent report of a class of C. elegans endo-siRNA that appears to be derived from intron-containing mRNA precursors (Warf et al., 2012).

EXPERIMENTAL PROCEDURES

Yeast Strains

Yeast strains used in this study are listed in Table S4. All C. neoformans strains were derived from strain H99 using standard procedures (Chun and Madhani, 2010).

Tandem Affinity Protein Purification

C. neoformans cultures grown in YPAD media were harvested, snap frozen, and then lysed using a coffee grinder and mortar and pestle. Proteins tagged with CBP-2xFLAG were purified using anti-FLAG M2 resin (Sigma) and calmodulin resin (Stratagene) according to manufacturers’ instructions. Purified proteins were analyzed by immunoblot or mass spectrometry as described in Extended Experimental Procedures and Table S6.

RNA Immunoprecipitation

Epitope-tagged proteins were purified using anti-FLAG resin as described above, and associated RNA was isolated by phenol-chloroform extraction, as described in Extended Experimental Procedures. Table S5 lists primers used for RT-qPCR.

RNA Isolation and sRNA Northern Blot

Total RNA was isolated using TRIzol (Invitrogen), whereas small RNA was isolated using a modified mirVana (Ambion) protocol, as described in Extended Experimental Procedures. Small RNA Northern blots were performed as described previously (Pall and Hamilton, 2008).