Pharmacological inhibition of the spliceosome subunit SF3b triggers exon junction complex-independent nonsense-mediated decay

The pathway from transcription of a DNA template in the nucleus to protein translation in the cytoplasm involves a multitude of steps that are highly interconnected and interdependent. In human cells, most newly synthesized precursor messenger RNA (pre-mRNA) molecules are first capped at the 5′ end, spliced, and cleaved and polyadenylated at the 3′ end before they are released from the site of transcription to the nucleoplasm. Mature mRNAs then diffuse throughout the nucleus until they encounter a nuclear pore complex (NPC) and are translocated to the cytoplasm. Each stage of mRNA biogenesis involves the dynamic assembly of a complex repertoire of proteins that are essential for processing, transport, translation and stability (for reviews, see Moore and Proudfoot, 2009; Mitchell and Parker, 2014). Coordination and integration among the sequential steps in gene expression occurs through the deposition of specific RNA-binding proteins that are read as signals of successful completion of the previous step in the pathway. This is thought to ensure a quality control process that discriminates between correctly processed and faulty RNA products. RNAs that fail proper maturation tend to be selectively eliminated in the cell by surveillance mechanisms that are not yet fully understood.

Nonsense-mediated mRNA decay (NMD) is a translation-coupled quality control pathway that degrades cytoplasmic transcripts with restricted coding capacities, such as mRNAs containing a premature translation termination codon (PTC) or short upstream open reading frame (for a review, see Kervestin and Jacobson, 2012). PTCs can arise by random nonsense and frameshift mutations, DNA rearrangements or inaccurate transcription, but most often they result from errors in splicing (i.e. when introns are retained or inappropriately spliced). How NMD distinguishes PTCs from normal stop codons is still unclear (for reviews, see Chang et al., 2007; Popp and Maquat, 2013; Brogna et al., 2016). Initially, the presence of an exon junction complex (EJC) downstream of a termination codon was thought to be essential to identify that termination codon as premature (Maquat, 2004). However, it subsequently became clear that NMD can occur in the absence of an EJC (Bühler et al., 2006; Matsuda et al., 2007; Eberle et al., 2008; Singh et al., 2008; Metze et al., 2013). In comparison to EJC-independent NMD, the presence of an EJC downstream of the PTC increases degradation efficiency, suggesting that the EJC enhances NMD (Metze et al., 2013).

The EJC is composed of a set of proteins that are stably loaded onto mRNAs upon splicing ∼24 nucleotides upstream of exon–exon junctions (Le Hir et al., 2000). EJC formation is tightly coupled to the splicing process, and constitutes a molecular signature of splicing events. During spliceosome assembly, the splicing factor CWC22 mediates recruitment of the EJC core protein eukaryotic initiation factor 4A-III (eIF4A-III; also known as EIF4A3) (Barbosa et al., 2012; Steckelberg et al., 2012; Alexandrov et al., 2012), which then stably binds to spliced mRNA (Shibuya et al., 2004). The protein eIF4AIII is locked into a stable RNA-binding configuration by its binding partners Y14 (also known as RBM8A) and Magoh. Another EJC core protein, MLN51 (metastatic lymph node 51, also known as Barentsz or CASC3), increases the affinity of eIF4AIII for RNA (Ballut et al., 2005). Once locked onto the mRNA, the EJC escorts messengers to the cytoplasm until they are disassembled by translating ribosomes (Gehring et al., 2009a). A multitude of factors have been found to be associated with the EJC, namely RNA-binding protein S1 (RNPS1), Acinus (also known as ACIN1) and Pinin (Bono and Gehring, 2011). EJC additionally interacts with UAP56 (also known as DDX39B) and ALYREF proteins, which are components of the transcription-export complex TREX (Strässer et al., 2002) and act together with serine/arginine-rich (SR) proteins (Singh et al., 2012), which are also involved in mRNA export (Müller-McNicoll et al., 2016). These observations, together with several other lines of evidence, strongly suggest that the EJC plays a role in export of spliced mRNA to the cytoplasm (for a review, see Le Hir et al., 2003).

In order to be exported from the nucleus to the cytoplasm mRNA molecules must bind a transport receptor that interacts with NPC components to activate translocation through the pore (for reviews, see Natalizio and Wente, 2013; Björk and Wieslander, 2014; Wickramasinghe and Laskey, 2015; Elbarbary and Maquat, 2016). Most mRNAs use the export receptor NXF1 (nuclear export factor 1, also known as Tap in humans and Mex67p in yeast) and its cofactor NXT1 (also known as p15). In its free form, the RNA-binding domain (RBD) of NXF1 is hidden, thus preventing association of the receptor with mRNA. The interaction of NXF1 with adaptor proteins assembled on the mRNA triggers a conformational switch that exposes the RBD and enables binding of the receptor to mRNA (Viphakone et al., 2012). One of the best-characterized NXF1 adaptors is ALYREF, a protein that is co-transcriptionally recruited to the nascent mRNA (Cheng et al., 2006). When NXF1 interacts with an ALYREF–RNA complex, the RNA is handed over from ALYREF to NXF1 (Hautbergue et al., 2008). ALYREF is essential for export of many yeast mRNAs (Hieronymus and Silver, 2003), but its requirement for mRNA export in metazoans appears to be redundantly shared with other adaptor proteins (Gatfield and Izaurralde, 2002; MacMorris et al., 2003; Hautbergue et al., 2009). ALYREF interacts with additional mRNA-binding proteins forming the transcription-export complex TREX (Strässer et al., 2002). TREX subunits associate with multiple components of the transcription and mRNA processing machineries, and its recruitment to nascent transcripts is mediated by splicing (Masuda et al., 2005) and the cap-binding proteins (Cheng et al., 2006). However, although splicing promotes RNA export of intron-containing genes, several studies show that export factors can interact with mRNAs that do not naturally contain introns or that have been synthesized by using cDNA as a template (Rodrigues et al., 2001; Izaurralde, 2002; Reed and Hurt, 2002).

Recently, natural microbial products were discovered that inhibit splicing through binding the SF3b subunit of the U2 small nuclear ribonucleoprotein (snRNP). These include pladienolide B, the primary analogs of a family of macrocyclic lactones isolated from Streptomyces platensis (Kotake et al., 2007; Yokoi et al., 2011), and two analogs of natural product FR901464 isolated from the culture broth of the bacterium Pseudomonas sp.2663, namely, spliceostatin A (Kaida et al., 2007) and meayamycin (Albert et al., 2009). These compounds are attracting much attention because they have potent anti-proliferative and anti-tumor activities (Kaida et al., 2007; Kotake et al., 2007; Furumai et al., 2010; Yokoi et al., 2011) and also because they provide a tool for arresting the spliceosome at an early stage of its assembly.

In this study, we investigated the effects of these pharmacological inhibitors of the SF3b complex on mRNA transport and stability in human cells. We observed a massive accumulation of unspliced pre-mRNAs in the nucleus, indicating that the vast majority of unprocessed transcripts are not immediately recognized and degraded by a surveillance mechanism. We also found that the export adaptor ALYREF associates with intron-containing pre-mRNAs in the nucleus and mediates their transport to the cytoplasm. In the cytoplasm, we demonstrate that intron-containing pre-mRNAs are targeted for NMD. As the EJC core protein eIF4AIII was not detected in a stable complex with unspliced transcripts, we conclude that inhibition of spliceosome assembly triggers an EJC-independent NMD pathway.

As a model system to study the effect of pharmacological inhibitors of the spliceosome on mRNA biogenesis, we used human osteosarcoma-derived cells (U2OS) that have been engineered to stably express doxycycline-inducible full-length wild-type (WT) β-globin transgenes integrated in tandem at a single locus in the genome (Martins et al., 2011). Cells were incubated with either spliceostatin A (SSA) at 100 ng ml⁻¹, meayamycin (MEA) at 5 nM, or pladienolide B (PlaB) at 1 µM. RNA was isolated after 2, 4 and 8 h of treatment and analyzed by RT-PCR using primers that detect spliced and unspliced transcripts (Fig. 1A–C; Fig. S1A,B). We analyzed RNAs transcribed from the WT β-globin transgene as well as transcripts from endogenous PCNA- and β-actin-encoding genes. The results show a time-dependent reduction in the level of spliced RNAs and accumulation of intron-containing transcripts with similar kinetics for the three inhibitors (Fig. 1A–C; Fig. S1A,B), which were therefore used interchangeably throughout this work. Next, we analyzed the subcellular distribution of transgenic β-globin transcripts by RNA fluorescence in situ hybridization (FISH) using probes that hybridize either to introns or to the full-length pre-mRNA (Fig. 1D,E). In untreated cells, staining for unspliced β-globin RNA revealed a bright focus in the nucleus upon induction of transgene expression with doxycycline (Fig. 1D, −SSA; Fig. S1C). This spot corresponds to nascent RNAs transcribed from the integrated transgene (Martins et al., 2011). After exposure to the inhibitor, unspliced β-globin transcripts localized throughout the nucleoplasm with higher concentration in compartments that are reminiscent of nuclear speckles (see Discussion for details on nuclear speckles) (Fig. 1D, +SSA). To further characterize these compartments, we combined RNA FISH with immunofluorescence using antibodies directed against the SR-related protein SRm160 (also known as SRRM1), which is known to be highly concentrated in nuclear speckles (Wagner et al., 2003). As shown in Fig. 1E, β-globin RNA and SRm160 colocalize in speckles after SSA treatment. Remarkably, nuclear speckles became larger and acquired a more homogeneous round shape as soon as 2 h after exposure to the drug (Fig. S1D). This type of morphological re-organization of nuclear speckles was previously observed in cells upon disruption of pre-mRNA splicing (O'Keefe et al., 1994).

We also analyzed the subcellular distribution of endogenous β-actin transcripts (Fig. 1F). Unspliced β-actin RNA in untreated cells was detected in discrete foci (Fig. 1F, −SSA) that correspond to the endogenous sites of transcription. The number of foci ranged between one and four per nucleus, depending on the cell cycle stage and whether or not both β-actin alleles were simultaneously active, as previously reported (Lionnet et al., 2011). In SSA-treated cells, unspliced β-actin transcripts accumulated in enlarged nuclear speckles (Fig. 1F, +SSA).

The observed accumulation of unspliced pre-mRNAs in cells treated with inhibitor indicates that intron-containing transcripts are not targeted for rapid degradation in the nucleus, implying that they are not immediately recognized by surveillance mechanisms. These observations further raise the question of whether unspliced transcripts accumulate in the nucleus because they are not competent to be exported to the cytoplasm or because they are retained in nuclear speckles.

To determine whether SF3b inhibition interferes with the binding of proteins that normally associate with nascent transcripts, we made use of a microscopy assay that combines DNA FISH to identify the site of β-globin transcription and immunofluorescence to track RNA-binding proteins. Antibodies were used against the U1 snRNP U1A protein, the spliceosomal protein CWC22, the EJC core proteins eIF4AIII, BTZ and Magoh, the EJC-associated protein Pinin, and the TREX components ALYREF and UAP56. Radial profiles of fluorescence intensity were calculated by defining multiple concentric annuli around the transcription site (Fig. 2A). The average value of fluorescence intensity in each annulus was subtracted from background (defined as the level of fluorescence detected in the most external region) and plotted as a function of distance from the center of the transcription site. The results show that protein components of the spliceosome, EJC and TREX are detected in close proximity to the β-globin transgenic locus (Fig. 2B–I, green traces; Fig. S2A). Accumulation of these proteins around the β-globin transcription site is even more striking after treatment with the spliceosome inhibitor (Fig. 2B–I, magenta traces; Fig. S2B). For comparison, we analyzed cells containing a β-globin deletion mutant transgene (βIVS1) that gives rise to unspliced transcripts restricted to the site of transcription (Custódio et al., 2004; Martins et al., 2011). In clear contrast with the results observed in inhibitor-treated cells, none of the analyzed proteins was enriched close to the mutant transcription sites (Fig. 2B–I, blue traces; Fig. S2C).

Spliceosome inhibitors of the SSA family do not prevent formation of complexes containing U1 and U2 snRNPs (Roybal and Jurica, 2010; Corrionero et al., 2011). This is consistent with our observation that U1A is similarly recruited to β-globin transcription sites in untreated and inhibitor-treated cells. In addition to transcription sites, U1A is localized in nuclear speckles (Fig. S3). In agreement with previous reports showing that nuclear speckles tend to localize in the close vicinity of actively transcribed genes (Carmo-Fonseca and Carvalho, 2007), we observed that the vast majority of cells had a speckle juxtaposed to the β-globin integration locus (Fig. S3). Because speckles enlarge after treatment with the spliceosome inhibitor, the region intensely labeled for U1A extends further away from the transcription site in treated cells compared to untreated cells (Fig. 2B). Similarly to U1A, ALYREF (Fig. 2C), UAP56 (Fig. 2D), eIF4AIII (Fig. 2E), Magoh (Fig. 2F), BTZ (Fig. 2G), Pinin (Fig. 2H) and CWC22 (Fig. 2I) were all detected at the sites of transcription and adjacent speckles, suggesting that these protein components of the TREX and EJC complexes can be recruited to intron-containing nascent β-globin transcripts. Because β-globin pre-mRNAs contain two introns, one possibility would be that in the presence of inhibitors some transcripts are still partially spliced, so that TREX and EJC complex assembly could occur on RNAs containing at least one excised intron. To address this possibility, we carried out RT-PCR using primers that amplify the full-length transcript (Fig. S4). The results show a band corresponding in size to spliced mRNA and a band corresponding in size to β-globin pre-mRNA with both introns included (Fig. S4), arguing against the presence of partially spliced β-globin RNAs. However, based on the detection of fully spliced β-globin mRNA after 2 and 8 h of drug treatment, we cannot exclude the possibility that some nascent transcripts are still normally spliced and therefore assemble TREX and EJC complexes.

To validate the immunofluorescence assay, we performed quantitative radial analysis of the distribution of Magoh after downregulation of eIF4AIII by RNAi (Fig. 3A). According to several lines of evidence (Ballut et al., 2005; Bono et al., 2006; Andersen et al., 2006; Gehring et al., 2009a), depletion of eIF4AIII should prevent EJC core assembly. Accordingly, we detected significantly reduced enrichment of Magoh at transcription sites (Fig. 3B). Enrichment of ALYREF remained unaltered (Fig. 3C), as expected since binding of ALYREF to nascent transcripts is known to be independent of EJC. We therefore conclude that the colocalization assay has sufficient sensitivity to detect changes in protein recruitment to nascent transcripts.

To further investigate how spliceosome inhibition affects the co-transcriptional recruitment of proteins to β-globin RNA, we carried out chromatin RNA immunoprecipitation (RIP). This technique involves in vivo crosslinking of proteins to RNA with formaldehyde, cell lysis, isolation of soluble chromatin and immunoprecipitation. The immunoprecipitated fraction is then treated with DNase. Finally, RNA is extracted, reverse-transcribed and analyzed by quantitative real-time RT-PCR (qRT-PCR) (Fig. 4). The results show that antibodies to eIF4AIII precipitated significantly less pre-mRNA in treated cells compared to untreated cells (Fig. 4A,B). In contrast, anti-ALYREF antibodies precipitated similar levels of pre-mRNA in treated and untreated cells (Fig. 4A,B). Overall these results are consistent with the prevailing model that ALYREF can bind to RNA independently from splicing, whereas deposition of EJC is strictly dependent on splicing. However, anti-eIF4AIII antibodies precipitated a significant amount of unspliced RNA in untreated cells (Fig. 4A,B), suggesting that EJC proteins interact with pre-mRNA before the splicing reaction is completed. In agreement with this idea, the spliceosomal protein CWC22, which directly binds eIF4AIII and initiates EJC assembly (Steckelberg et al., 2012; Barbosa et al., 2012; Alexandrov et al., 2012), was found to already be associated with the spliceosome at the precatalytic stage (Will and Lührmann, 2011; Yeh et al., 2011). Possibly, in the presence of spliceosome inhibitor, CWC22 and EJC proteins can still form loose transient interactions with nascent pre-mRNA resulting in the recruitment detected by immunofluorescence (Fig. 2E–I); however, without splicing, a stable EJC complex does not assemble, and therefore anti-eIF4AIII antibodies fail to immunoprecipitate pre-mRNA accumulated in treated cells (Fig. 4).

Our RIP results indicate that ALYREF associates with β-globin pre-mRNA in cells treated with spliceosome inhibitor (Fig. 4). As previous studies have shown that ALYREF is recruited to a region near the 5′ cap of mRNA (Cheng et al., 2006), we asked whether binding of ALYREF to unspliced β-globin pre-mRNA is also restricted to the 5′ end. For this, we used an oligonucleotide-targeted RNase H cleavage assay. RNase H-mediated cleavage was directed by hybridization of three DNA oligonucleotides that were complementary to the first, second and third exons of the β-globin transcript (Fig. 5A). The resulting fragments were then immunoprecipitated and analyzed by performing gel electrophoresis of reverse transcription PCR products and qRT-PCR (Fig. 5B–D). In untreated cells, anti-ALYREF antibodies precipitated a significant amount of RNA from the promoter-proximal region of the transcript but low levels from the body of pre-mRNA (Fig. 5B), consistent with the model that binding of ALYREF is restricted to the 5′ end. However, after SSA treatment ALYREF, but not eIF4AIII, was found to be associated with fragments spanning the first and the second introns (Fig. 5C,D). Most likely, SSA has no effect on ALYREF binding other than to increase the levels of introns present in the nucleus.

If unspliced, β-globin pre-mRNAs bind the export adaptor ALYREF and they should be competent for export to the cytoplasm. Indeed, FISH experiments revealed that although the bulk of RNA is detected in the nucleus in cells treated with splicing inhibitor, staining for pre-mRNA is also observed in the cytoplasm (Fig. 1E,F). Moreover, evidence that SSA does not block transport of unspliced pre-mRNA to the cytoplasm was previously reported for p27 (CDKN1B) transcripts (Kaida et al., 2007). To study the intracellular transport of unspliced transcripts in our model system, we analyzed RNA species present in subcellular fractions. We adapted previously established protocols for the isolation of cytoplasmic, nucleoplasmic and chromatin-associated RNA (Wuarin and Schibler, 1994; Dye et al., 2006; Pandya-Jones and Black, 2009). Immunoblot analysis confirms that tubulin is present predominantly in the cytoplasmic fraction, whereas histone H3 is mainly associated with chromatin (Fig. 6A). Lamin A/C and U2 snRNP specific protein B″ (U2B″) were detected both in the chromatin and nucleoplasmic fractions (Fig. 6A), in agreement with previous studies (Pandya-Jones and Black, 2009). Next, we analyzed the subcellular localization of unspliced transgenic β-globin and endogenous β-actin and p27 transcripts by gel electrophoresis (Fig. 6B). As expected, in untreated cells, unspliced RNAs are predominantly associated with chromatin. After treatment with spliceosome inhibitor, unspliced transcripts accumulate in the nucleoplasm and are also detected in the cytoplasm. Then, we measured by qRT-PCR the relative levels of unspliced transcripts in treated versus untreated cells in each fraction. Unspliced pre-mRNA is more abundant in all fractions isolated from cells treated with the inhibitor. Although in treated cells, the vast majority of pre-mRNA is detected in association with the chromatin and nucleoplasmic fractions, a small amount is exported to the cytoplasm (Fig. 6B).

To show that export of unspliced RNAs to the cytoplasm is dependent on ALYREF, cells were transfected with specific small interfering (si)RNAs. Immunostaining and immunoblotting results show that at 72 h after transfection, the levels of ALYREF protein are significantly reduced when compared with cells transfected with a negative control GL2 siRNA, which targets firefly luciferase (Elbashir et al., 2001) (Fig. 6C). Quantitative analysis by qRT-PCR revealed that downregulation of ALYREF induced a significant reduction in the levels of unspliced β-actin and β-globin transcripts detected in the cytoplasm (Fig. 6D), as would be expected if ALYREF is required for export of these RNAs. If RNAs are no longer transported to the cytoplasm, they should accumulate in the nucleus. However, we found similar levels of RNA in the nucleus, with a tendency for reduction in the absence of ALYREF (Fig. 6E). This could be due to less-efficient transcription upon ALYREF depletion, as recently reported (Stubbs and Conrad, 2015).

Intron retention is often associated with disruption of the reading frame through an appearance of a premature termination codon (PTC) that can target the mRNA for degradation by NMD (Nagy and Maquat, 1998). To determine whether the unspliced transcripts that leak into the cytoplasm of cells treated with spliceosome inhibitors are degraded by NMD, we used RNAi to reduce the level of UPF1 (Fig. 7), the essential core protein of the NMD machinery (Chang et al., 2007). A synthetic siRNA duplex was designed with a sequence unique to UPF1 (Pacheco et al., 2004). As control, we used the GL2 duplex (Elbashir et al., 2001). Immunoblot analysis of total cellular proteins collected 4 days after transfection with siRNAs shows that UPF1 protein levels are specifically and efficiently downregulated (Fig. 7A). qRT-PCR analysis of RNA extracted from cells transfected with control GL2 siRNAs shows that treatment with spliceosome inhibitor leads to a significant increase in the cytoplasmic level of unspliced β-globin (Fig. 7B) and β-actin (Fig. 7C) transcripts. After blocking NMD by downregulation of UPF1, the level of unspliced β-globin and β-actin transcripts detected in the cytoplasm was even higher (Fig. 7B,C), as would be expected if intron-containing RNAs were recognized and degraded by NMD. We further analyzed p27 transcripts because it was previously reported that unspliced p27 RNA is translated into a functional protein in cells treated with the spliceosome inhibitor SSA (Kaida et al., 2007; Satoh and Kaida, 2016). We observed similar levels of unspliced p27 RNA in the cytoplasm of inhibitor-treated cells irrespectively of UPF1 knockdown (Fig. 7D), suggesting that this transcript escapes NMD. Indeed, unspliced p27 transcripts contain an in-frame STOP codon in the first intron. The p27 pre-mRNA consists of three exons and two introns, and the mature mRNA is translated from the start codon in exon 1 to the stop codon in exon 2. When splicing is inhibited translation occurs from the start codon in exon 1 to the first in-frame stop codon in intron 1. Thus, a shorter p27 protein is synthesized, which inhibits Cdk2 activity and blocks cell cycle progression (Kaida et al., 2007; Satoh and Kaida, 2016).

To date, spliceostatin A, meayamycin, and pladienolide B are part of a well-characterized group of small-molecule splicing inhibitors (for a recent review, see Effenberger et al., 2016). These compounds target a specific spliceosome component, the U2 snRNP SF3b complex, and were originally discovered as substances that inhibited the growth of human tumors (Nakajima et al., 1996). Currently, these molecules hold great promise both as tools to manipulate splicing and as potential anti-cancer drugs (Bonnal et al., 2012; Balaian et al., 2013).

Because pre-mRNA splicing is intimately connected with transport, translation and turnover of mRNA, one would expect that SF3b inhibitors impact on multiple stages of mRNA biogenesis. Kaida and colleagues first reported that poly(A) RNA accumulated in the nucleus of human cells treated with SSA, and inferred that the poly(A) signal detected by FISH corresponded to unspliced pre-mRNA (Kaida et al., 2007). Here, using specific FISH probes in SSA- and MEA-treated cells, we show nuclear accumulation of intron-containing β-globin and β-actin transcripts that appear particularly enriched in enlarged nuclear speckles (Fig. 1). Nuclear speckles are compartments enriched in spliceosomal components that are located adjacent to transcription sites (Lamond and Spector, 2003). Because the spliceosome is a multi-megadalton complex that assembles at splice sites on nascent pre-mRNAs and disassembles immediately after exon–exon ligation, spliceosomal components in the nucleus are either actively engaged in splicing or waiting for the next ‘call’ to assemble a spliceosome (Rino et al., 2007). While not engaged in active splicing, spliceosome components are predominantly detected in nuclear speckles. Nuclear speckles are very dynamic structures that reflect the transcriptional and splicing status of a cell. A massive activation of gene transcription and pre-mRNA splicing causes nuclear speckles to practically disappear, whereas general gene inactivation increases the pool of ‘reserve’ splicing factors that accumulate in enlarged speckles (Rino et al., 2007). It is therefore expected that in cells treated with splicing inhibitors nuclear speckles become larger and filled with splicing factors. But why are unspliced RNAs also enriched in speckles? We propose that this localization pattern is mediated by the establishment of loose nonproductive interactions between the splicing factors concentrated in the speckles and the vast amount of splice site sequences in unspliced transcripts available for protein binding. Consistent with this view, it was recently shown that transcripts synthesized from DNA constructs microinjected in the nucleus require functional splice sites in order to associate with nuclear speckles (Dias et al., 2010).

For the vast majority of mRNAs, splicing is completed in the nucleus before export to the cytoplasm, but whether cells have mechanisms to actively recognize unspliced transcripts and block their transport to the cytoplasm is uncertain. In yeast, previous studies have shown that mutations in the spliceosomal protein Prp2, an RNA-dependent ATPase that activates the spliceosome before the first catalytic step of splicing, lead to the accumulation of unspliced RNAs from most intron-containing genes (Rosbash et al., 1981; Plumpton et al., 1994). Human cells expressing a dominant-negative DHX16 mutant, the human ortholog of Saccharomyces cerevisiae Prp2, also show retention of unspliced RNAs in the nucleus, suggesting a spliceosome-mediated mechanism of nuclear retention of unspliced pre-mRNAs (Gencheva et al., 2010). These observations are in very good agreement with our view that in cells treated with SF3b inhibitors, pre-mRNAs are retained through interactions with spliceosomal components localized in the speckles. The establishment of interactions between RNAs and spliceosomal components at speckles is further supported by recent evidence suggesting the involvement of nuclear speckles in post-transcriptional splicing (Girard et al., 2012; Mor et al., 2016).

Nuclear retention of unspliced transcripts does not appear to be a tight scrutiny mechanism that prevents export of incompletely processed mRNAs to the cytoplasm. Indeed, in their pioneer study, Kaida and colleagues identified a protein product translated from an unspliced pre-mRNA and concluded that despite nuclear retention of poly(A) RNA, SSA did not block RNA export to the cytoplasm (Kaida et al., 2007). More recently, an RNA-seq analysis on nuclear and cytoplasmic fractions of SSA-treated cells identified additional intron-containing pre-mRNAs in the cytoplasm (Yoshimoto et al., 2017). Our qRT-PCR analysis of subcellular fractions demonstrates the presence of unspliced β-globin, β-actin and p27 transcripts in the cytoplasm of inhibitor-treated cells (Fig. 6B). Moreover, previous work performed in yeast showed that mutations in several early-acting splicing factors, including U2 snRNP components, cause cytoplasmic leakage of pre-mRNA (Casolari and Silver, 2004).

Our results reveal that the export adaptor ALYREF interacts with unspliced β-globin transcripts (Figs 4 and 5) and mediates their export to the cytoplasm in cells treated with SF3b inhibitors (Fig. 6C). As the RIP method used in our experiments involves immunoprecipitation of chromatin-associated RNA, which presumably corresponds to nascent RNA emanating from the elongating RNA polymerase II (Percipalle and Obrdlik, 2009), we conclude that ALYREF is recruited to intron-containing newly synthesized RNA. This finding expands previous work that revealed the interaction of export factors with cellular and viral mRNAs that do not naturally contain introns, or with RNAs that were synthesized by using cDNA as template (Rodrigues et al., 2001; Taniguchi and Ohno, 2008; Tunnicliffe et al., 2011).

Moreover, in cells treated with SF3b inhibitor, we found that ALYREF was associated with β-globin transcripts not only at the 5′ end as previously reported (Cheng et al., 2006) but also throughout the full length of the transcript (Fig. 5). In the case of naturally intron-less mammalian mRNAs, specific regions were identified that recruit the TREX export machinery (Guang and Mertz, 2005; Lei et al., 2011, 2013). This suggests that splicing-dependent export can be bypassed by sequence elements that recruit the mRNA export machinery. Thus, one possibility is that similar elements occur in introns; normally such elements would not be used because introns are rapidly excised, but in the presence of splicing inhibitors introns are retained and the sequences available to recruit export factors. Finally, it is noteworthy that binding of ALYREF was shown to stabilize nuclear RNAs independently of their export (Stubbs et al., 2012). This raises the possibility that the observed interaction between ALYREF and unspliced β-globin pre-mRNAs may further contribute to their stabilization in the nucleus of cells treated with splicing inhibitors.

Previous studies have shown that by binding SF3b, SSA interferes with the normal recruitment of U2 snRNP to the pre-mRNA and prevents the subsequent assembly of a catalytically active spliceosome (Roybal and Jurica, 2010; Corrionero et al., 2011). However, all five spliceosomal snRNPs were found to be associated with pre-mRNA during SSA inhibition, suggesting that the inhibitor may promote nonproductive interactions between spliceosome components (Roybal and Jurica, 2010; Corrionero et al., 2011). Biochemical analysis of splicing complexes assembled in vitro has revealed that the EJC core proteins eIF4AIII, Y14 and Magoh are already present before spliceosome activation (Jurica et al., 2002; Makarov et al., 2002; Reichert et al., 2002; Merz et al., 2007; Zhang and Krainer, 2007; Bessonov et al., 2008), but formation of a stable tetrameric EJC core occurs only upon recruitment of MLN51 after the second splicing transesterification reaction (Degot et al., 2004; Tange et al., 2005; Gehring et al., 2009b). The ability of EJC proteins to establish loose interactions with spliceosome components could explain the observation that EJC proteins are detected in close proximity to transcription sites in SSA-treated cells (Fig. 2).

In our RIP experiments, antibodies against eIF4AIII7 did not precipitate β-globin pre-mRNA in SSA-treated cells (Fig. 4), consistent with the prevailing model that assembly of a stable EJC is strictly dependent on splicing (Chang et al., 2007; Popp and Maquat, 2013). However, we found that in cells treated with splicing inhibitors unspliced β-globin RNAs are degraded in the cytoplasm by a mechanism that requires UPF1 (Fig. 7). This suggests that unspliced β-globin transcripts in the cytoplasm are targeted for NMD by an EJC-independent pathway. Accordingly, several studies have reported that NMD can occur in the absence of an EJC downstream from the PTC (Bühler et al., 2006; Matsuda et al., 2007; Eberle et al., 2008; Singh et al., 2008; Metze et al., 2013). It should be noted, however, that not all 3′-splice sites are equally sensitive to inhibitors (Corrionero et al., 2011), raising the possibility that in a multi-exon transcript, retained introns may co-exist with EJCs deposited on normally spliced exon–exon junctions. In that scenario, NMD could follow the most prevalent EJC-enhanced pathway.

In conclusion, our results indicate that pharmacological inhibitors targeting SF3b cause a massive accumulation of unspliced pre-mRNAs in the nucleus. However, intron-containing RNAs can still bind the export factor ALYREF and be transported to the cytoplasm, where they trigger an alternative NMD pathway. We envisage a kinetic competition scenario where splicing inhibition leads to the stabilization of pre-mRNA molecules that are normally very short lived. These RNA species can then engage in interactions with protein factors and give rise to RNA–protein complexes that would not normally form, highlighting the importance of timely kinetics of RNA biogenesis for gene expression fidelity.

U2OS-βWT and U2OS-βIVS1 cells were described previously (Martins et al., 2011). Cells were cultured at 37°C in Dulbecco's modified Eagle's medium (DMEM) containing 10% fetal calf serum (FCS, GIBCO-BRL). β-globin transgene transcription was induced by incubation with 6 µg/ml doxycycline for 24 h. Cells were then exposed to each splicing inhibitor, spliceostatin A (SSA), meayamycin (MEA) and pladienolide B (PlaB), for the indicated period of time, at a final concentration of 100 ng/ml, 5 nM and 1 µM, respectively.

One day prior to siRNA transfection, 20×10⁴ cells were platted in 35 mm Petri dishes. Cells were then transfected with 150 nM chemically synthesized, annealed siRNAs (Eurogentec, Belgium) by using Lipofectamin RNAi Max reagent (Invitrogen). After 2 days, cells were re-transfected and cultured for 1 or 2 more days. All siRNAs used are described in Table S1.

FISH and immunofluorescence were carried out as described previously (Martins et al., 2011). To visualize pre-mRNA, intronic probes were made by in vitro transcription using, as a template, PCR products with a T7 promoter sequence. The full genomic sequence of the human β-globin gene was additionally labeled by nick translation and used as a probe to detect total β-globin RNA. The genomic site of integration of the β-globin transgene was detected by DNA FISH with a probe that corresponds to the cloning vector.

Images were acquired on a Zeiss LSM 510 confocal microscope (Carl Zeiss, Jena, Germany) using a PlanApochromat 63×1.4 NA objective, as described previously (Martins et al., 2011). To assess recruitment of proteins to nascent β-globin transcripts, we adapted a previously described protocol (Spiluttini et al., 2010). Briefly, the site of transcription was identified by DNA FISH. The fluorescence intensity resulting from immunostaining was automatically quantified at and around the transcription site by using ImageJ (http://imagej.nih.gov/ij/). Fluorescence intensity was measured in 45 concentric regions delineated by circumferences centered at the transcription site and separated from each other by ∼1.3 pixels. The average value of fluorescence intensity in each annulus was subtracted from background (defined as the level of fluorescence detected in the most external region located at 60 pixels from the center of the transcription site).

Nuclear and cytoplasmic RNA fractions were isolated as described previously (Wang et al., 2006). Nuclei were further separated into chromatin-associated and nucleoplasmic fractions according to previously described protocols (Wuarin and Schibler, 1994; Dye et al., 2006; West et al., 2008; Pandya-Jones and Black, 2009). Fractions were frozen at −80°C in TRIzol^® (Invitrogen) or PureZOL^® (Bio-Rad) reagent.

RNA was extracted by using TRIzol RNA isolation reagent and cDNA was made using random primers and Superscript II Reverse Transcriptase (Invitrogen). RT-PCR products were separated by gel electrophoresis, detected by GelRed, scanned with Typhoon (GE Healthcare) and quantified with the ImageQuant software (Molecular Dynamics, Sunnyvale, CA). qRT-PCR analysis was performed in the 7500 Real-Time PCR System (Applied Biosystems, Foster City, CA), using the SYBR Green PCR master mix (Applied Biosystems). The primers used are described in Table S2.

RIP was performed according to a described protocol (Percipalle and Obrdlik, 2009). Briefly, cell extracts were sonicated on ice with a Sanyo Soniprep 150 at amplitude of 10 μm with 6×20 s bursts. The chromatin solution was pre-cleared with protein-A–Sepharose beads (Sigma) for 1 h prior to overnight incubation (4°C) with 2.5 µg of each immunoprecipitating antibody. Pre-immune serum (sc-2025 or sc-2027, Santa Cruz Biotechnology) was used as negative control. Complexes were immunoprecipitated with protein-A–Sepharose beads (Sigma) for 3 h at 4°C. RNA was extracted with TRIzol followed by DNase I treatment. cDNA synthesis was performed using a past poly(A) β-globin reverse primer and samples analyzed by gel electrophoresis of RT-PCR products and qRT-PCR. The primers used are described in Table S2.

DNA oligonucleotide-mediated RNase H cleavage of β-globin transcripts was carried out as described previously (Kurosaki and Maquat, 2013) using three different DNA oligonucleotides (DNA oligoExI, DNA oligoExII and DNA oligoExIII; Table S3).

Western blotting was carried out as described elsewhere (de Almeida et al., 2010).

The following primary antibodies were used: mouse monoclonal antibodies against ALYREF [immunofluorescence (IF) 1:100, clone 11G5; Abcam], eIF4AIII (IF, 1:5, provided by Dr Adrian Krainer, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY), Magoh (IF, 1:50, clone 21B12, ab10686, Abcam), U1A (IF 1:50, ab55751, Abcam), U2 snRNP specific protein B″ [western blotting (WB) 1:250, clone 4G3, PROGEN Biotechnik], U2AF65 (IF, 1:10; Gama-Carvalho et al., 1997) and tubulin (WB, 1:5000, clone B-5-1-2, Sigma); and rabbit polyclonal antibodies against SRm160 (IF, 1:200; Blencowe et al., 1998), BTZ/MLN51 (IF, 1:10, ab90651, Abcam), Pinin (IF, 1:10, ab108485, Abcam), UAP56 (IF, 1:100, ab47955, Abcam), CWC22 (IF: 1:10, HPA036748, Sigma), UPF1 (WB, 1:1000, #9435, Cell Signaling Technology), lamin A/C (WB, 1:1000, H-110, Santa Cruz Biotechnology.) and total H3 (WB, 1:1000, ab1791, Abcam).

We thank Dr Adrian Krainer from Cold Spring Harbor Laboratory and Dr Benjamin Blencowe from University of Toronto for kindly providing eIF4AIII and SRm160 antibodies, respectively. We are also grateful to Dr Lynne Maquat from University of Rochester Medical Center for insightful discussion and advice.