Unconventional splicing of XBP1 mRNA occurs in the cytoplasm during the mammalian unfolded protein response

The endoplasmic reticulum (ER) is an organelle in which both secretory and membrane proteins are synthesized. Proteins correctly folded with assistance from ER chaperones are selectively exported to the Golgi complex (Gething, 1997; Helenius and Aebi, 2004), whereas unfolded or malfolded proteins are degraded by ER-associated protein degradation (ERAD) (Brodsky, 2007; Ruddock and Molinari, 2006). The amounts of ER chaperones and ERAD components are tightly regulated by a mechanism called the unfolded protein response (UPR) or ER-stress response, and increase through transcriptional induction of genes encoding ER chaperones and ERAD components when unfolded proteins are accumulated in the ER (ER stress) (Kohno, 2007; Malhotra and Kaufman, 2007; Mori, 2003; Ron and Walter, 2007; Yoshida, 2007a). In yeast, the ER-stress response is regulated solely by the IRE1 pathway. Ire1p is a transmembrane protein located in the ER membrane, which contains an RNase domain in its cytoplasmic portion (Cox et al., 1993; Mori et al., 1993). In response to ER stress, unspliced (U) HAC1 mRNA is cleaved by Ire1p, and then ligated by tRNA ligase Rlg1p (Cox and Walter, 1996; Gonzalez et al., 1999; Kawahara et al., 1997; Kawahara et al., 1998; Mori et al., 1996; Shamu and Walter, 1996; Sidrauski et al., 1996; Sidrauski and Walter, 1997). Hac1pⁱ translated from a spliced HAC1 mRNA binds to the cis-acting element UPRE in the promoters of ER chaperone genes as well as ERAD genes, and induces their transcription (Kohno et al., 1993; Mori et al., 1998; Mori et al., 2000; Mori et al., 1992; Ng et al., 2000; Travers et al., 2000). The IRE1 pathway is conserved from yeast to mammals (Iwawaki et al., 2001; Tirasophon et al., 1998; Wang et al., 1998), but IRE1α and IRE1β, which are mammalian homologs of Ire1p, cleave unspliced XBP1 mRNA instead of unspliced HAC1 mRNA in response to ER stress (Calfon et al., 2002; Yoshida et al., 2001a). Spliced XBP1 mRNA encodes an active transcription factor, pXBP1(S), which activates transcription of its target genes, including genes encoding ER chaperones and ERAD components, to protect cells from ER-stress-induced apoptosis (Lee et al., 2003; Yoshida et al., 2003; Yoshida et al., 2006a). Interestingly, unspliced XBP1 mRNA is efficiently translated to produce pXBP1(U), whereas translation of unspliced HAC1 mRNA is blocked by its intron. pXBP1(U) binds pXBP1(S) and enhances its degradation; pXBP1(U) is thought to be a negative regulator of the IRE1 pathway (Tirosh et al., 2006; Yoshida et al., 2006b).

Mammalian cells have developed two additional regulatory pathways, the ATF6 and PERK pathways. ATF6 is a transmembrane protein located in the ER. Upon ER stress, it is transported to the Golgi and sequentially cleaved by the proteases S1P and S2P, resulting in release of its cytoplasmic part, pATF6(N) (Haze et al., 1999; Okada et al., 2003; Ye et al., 2000). pATF6(N) containing both DNA-binding and transcriptional activation domains translocates into the nucleus, and activates transcription of its target genes, such as ER chaperones and ERAD components as well as XBP1 (Adachi et al., 2008; Haze et al., 2001; Okada et al., 2002; Yamamoto et al., 2007; Yoshida et al., 1998; Yoshida et al., 2000; Yoshida et al., 2001b). PERK is another transmembrane protein located in the ER: it is a protein kinase that attenuates translation in response to ER stress by phosphorylating the α-subunit of eukaryotic translational initiation factor 2 (eIF2α) (Harding et al., 2000a; Harding et al., 2001; Harding et al., 2000b; Harding et al., 1999; Harding et al., 2003). The PERK pathway is also responsible for translational induction of ATF4, a transcription factor whose target genes include translational components as well as antioxidant proteins.

Importantly, the splicing of HAC1 and XBP1 mRNAs is unconventional (Calfon et al., 2002; Gonzalez et al., 1999; Kawahara et al., 1998; Sidrauski et al., 1996; Sidrauski and Walter, 1997; Yoshida, 2007b; Yoshida et al., 2001a). Conventional splicing is catalyzed by the spliceosome, and involves a consensus sequence at the border between an exon and an intron, such as GU-AG or AU-AC, according to the so-called Chambon's rule (Tarn and Steitz, 1997). During the conventional splicing reaction, the 5′ splice site is cleaved first, and then the 3′ splice site is cut after formation of a lariat structure. By contrast, the unconventional splicing reaction is catalyzed by IRE1 and a specific RNA ligase, and is completely independent of the spliceosome. There is a pair of characteristic stem-loop structures at the exon-intron border instead of a Chambon's consensus sequence, and the cleavage of the 5′ and 3′ sites occurs randomly.

Conventional splicing takes place exclusively in the nucleus, but it has been controversial where the splicing reaction of unconventional splicing occurs (see Discussion). Unconventional splicing of yeast HAC1 mRNA is thought to occur in the cytoplasm (Ruegsegger et al., 2001), although Gething and colleagues argued that it occurs in the nucleus as well as cytoplasm (Goffin et al., 2006). Regarding mammalian unconventional splicing, it was reported that IRE1 is localized in the inner nuclear membrane, suggesting that it takes place in the nucleus (Lee et al., 2002). However, it was also reported that the unconventional splicing reaction could occur in the cytoplasm without nuclear processing when the catalytic domain of IRE1 is ectopically expressed in the cytoplasm (Back et al., 2006; Iwawaki and Akai, 2006).

In this study, we aimed to determine whether mammalian unconventional splicing occurs in the nucleus, cytoplasm or both by analyzing the subcellular localization of the splicing machinery, including IRE1 and RNA ligase, in mammalian cells. We also examined whether the major source of substrate for unconventional splicing is the cytoplasmic or nuclear pool of XBP1(U) mRNA.

Although IRE1α is assumed to be located in the ER membrane (Niwa et al., 1999; Tirasophon et al., 1998), it is not clear whether the catalytic domain of some fraction of IRE1 faces the nucleus or cytosol, because the ER membrane is connected with the nuclear membrane and IRE1 is a type I transmembrane protein. Preferential localization to the inner nuclear membrane was also reported in a subcellular fractionation study (Lee et al., 2002). To clarify this, a plasmid expressing the C-terminal region of IRE1α (IRE1α-CTR, corresponding to residues 469-977 of human IRE1α) tagged with HA epitopes was transfected into cells (Fig. 1A). IRE1α-CTR was clearly localized to the cytoplasm and excluded from the nucleus (Fig. 2Aa-c). Although large molecules are known to be excluded from the nucleus because the nuclear pore cannot accommodate them, exclusion of IRE1α-CTR from the nucleus is not due to its molecular size, because IRE1α-CTR fused with a nuclear localization signal of the transcription factor ATF6α (NLS-IRE1α-CTR) was localized in the nucleus (Fig. 1A and Fig. 2Ad-f). This strongly suggests that the catalytic domain of IRE1 is equipped with a nuclear exclusion signal (NES) to avoid mislocalization to the nucleus, and that its cytoplasmic localization is relevant to its function. The exogenous expression of each IRE1α construct was confirmed by immunoblotting (Fig. 1B, lanes 3 and 4). Two bands of IRE1α deletions were detected in lanes 3 and 4: the upper band might represent an autophosphorylated form (activated form), because only one band was detected when a kinase-defective mutant, IRE1α-K599A, was expressed (Fig. 1C, lanes 10 and 14).

To determine the location of the NES contained in the IRE1α-CTR, the subcellular localization of deletion mutants lacking distinct domain(s) was analyzed (Fig. 2B). Expression of these mutants was confirmed by immunoblotting (Fig. 1B). A deletion mutant retaining the RNase domain (IRE1α [571-977]) was excluded from the nucleus (Fig. 2Bd-f), whereas other deletions lacking the RNase domain (IRE1α [469-834], IRE1α [469-570] and IRE1α [571-834]) were detected in both the nucleus and the cytoplasm (Fig. 2Ba-c,g-l). These findings suggest that the NES of IRE1α is located in the RNase domain, although a deletion mutant containing only the RNase domain (IRE1α [835-977]) formed aggregates in the cytoplasm and therefore its localization could not be determined (data not shown). Next, we searched the RNase domain for an NES-like sequence in silico, and found an NES-like sequence, as expected, that was highly similar to the NES consensus [L-x(2,3)-(LIVFM)-x(2,3)-L-x-(LI)] (la Cour et al., 2003) and well conserved from yeast to humans (Fig. 1E). An IRE1α-CTR-NES mutant in which two critical hydrophobic residues (Val918 and Leu922) were mutated to glycine was found in the cytoplasm as well as nucleus (Fig. 2C), suggesting that the NES-like sequence found in the RNase domain actually functions as an NES. Since IRE1α-CTR does not seem to have an NLS consensus sequence, the IRE1α-CTR-NES mutant might stray into the nucleus through the nuclear pore by diffusion (the nuclear pore complex can passively transport macromolecules whose nucleocytoplasmic domain is smaller than 67 kDa, and IRE1α-CTR is 57.7 kDa). From these results, we conclude that IRE1α contains an NES sequence so that its catalytic domain faces the cytosol.

The above findings prompted us to investigate the subcellular localization of the specific RNA ligase that joins XBP1 mRNA cleaved by IRE1. Since the gene for RNA ligase involved in mammalian unconventional splicing has not yet been cloned, we adopted an indirect approach, investigating the subcellular location of RNA ligase activity. Our strategy was based on the notion that ectopic expression of IRE1α-CTR in the subcellular compartment where the RNA ligase resides would result in the production of XBP1(S) mRNA and pXBP1(S), since XBP1(U) mRNA cleaved by IRE1α-CTR could be ligated by the endogenous RNA ligase there. By contrast, when IRE1α-CTR was expressed at a location lacking RNA ligase activity, the cleaved XBP1(U) mRNA would not be ligated, leading to accumulation of cleaved mRNA. Since IRE1α-CTR is known to form a dimer through its residual dimerization potential in the cytoplasmic domain and can cleave substrate mRNA (Gonzalez et al., 1999; Lee et al., 2008; Sidrauski and Walter, 1997), we did not fuse an additional dimerization motif to IRE1α-CTR. Upon immunoblotting, two bands of IRE1α-CTR (with or without NLS) were detected (Fig. 3B, lanes 13-18), whereas only one band was detected in the case of kinase-defective IRE1α-CTR-K599A mutant (Fig. 1C, lanes 10 and 14), suggesting that the upper band represents an autophosphorylated form (activated form), and that IRE1α-CTR forms a dimer that is indeed active.

When HeLa cells were transfected with a plasmid expressing XBP1(U) mRNA, 2.0 kb XBP1(U) mRNA was detected (Fig. 3B, upper panel, lane 1), from which a pXBP1(U) protein of 29 kDa was translated (Fig. 3B, middle panel, lane 7). Simultaneous transfection with a plasmid expressing IRE1α-CTR resulted in splicing of XBP1(U) mRNA, which led to the production of 50 kDa pXBP1(S), as expected (Fig. 3B, middle panel, lane 8). It should be noted that the size of XBP1 mRNA did not change appreciably (Fig. 3B, middle panel, lane 2), because the intron removed by unconventional splicing was very small (26 nucleotides). Evidently, cytoplasmic expression of IRE1α-CTR resulted in both cleavage and ligation of XBP1(U) mRNA, leading to the production of XBP1(S) mRNA. By contrast, co-expression of NLS-IRE1α-CTR with XBP1(U) mRNA produced a 0.7 kb truncated XBP1 mRNA, whereas 2.0 kb XBP1(U) mRNA was reduced (Fig. 3B, middle panel, lane 3). The amount of pXBP1(U) was markedly decreased and pXBP1(S) was produced at a very low concentration (Fig. 3B, middle panel, lane 9). Since the 0.7 kb band was detected using a cDNA probe for the 5′ portion of XBP1(U) mRNA, and since the splicing site is located at residue 164 (see Fig. 3A), this mRNA band must have come from the 5′ portion of XBP1(U) mRNA, which was cleaved by NLS-IRE1α-CTR, remained unligated and accumulated in the nucleus.

To confirm this, 3′ deletion mutants of XBP1(U) mRNA were expressed together with NLS-IRE1α-CTR in HeLa cells (Fig. 3C). Deletion mutants containing the unconventional splicing site yielded the 0.7 kb band (Fig. 3C, lanes 1-3), whereas a mutant lacking the splicing site (1-133) did not (Fig. 3C, lane 4). This suggested that the 3′ end of the 0.7 kb protein resides between 133 and 185, where the unconventional splicing site of XBP1 exists. The 0.7 kb band is smaller than a band of XBP1 [1-133] containing a poly(A) tail (lanes 1-4), suggesting that the 0.7 kb band is not polyadenylated, and that it is not produced by premature termination or altered initiation of transcription. Although bisected mRNAs are thought to be rapidly degraded by an mRNA quality control mechanism such as nonstop mRNA decay (Frischmeyer et al., 2002; van Hoof et al., 2002), Peter Walter and colleagues reported that truncated HAC1 mRNA was indeed observed in the mutant of RNA ligase rlg1-100 (Sidrauski et al., 1996). The small amount of bisected XBP1 mRNA observed in Fig. 3A seemed to survive the mRNA quality control process, although most of the bisected XBP1 mRNA was degraded. These results indicate that the activity of the RNA ligase required for unconventional splicing resides predominantly in the cytoplasm, not in the nucleus. We found that the expression level of IRE1α-CTR was somewhat greater than that of NLS-IRE1α-CTR (lanes 13-18), suggesting that it is not likely that excessive expression of NLS-IRE1α-CTR caused a shortage of RNA ligase and the consequent accumulation of truncated XBP1 mRNA. It is also unlikely that low expression of NLS-IRE1-CTR resulted in reduced cleavage of XBP1 mRNA, leading to little induction of pXBP1(S), because cleaved XBP1 mRNA accumulated (Fig. 3B, lane 3) and the amount of pXBP1(U) was also decreased (lane 9). Moreover, expression of cytoplasmic IRE1α-CTR resulted in cleavage of about one-third of XBP1(U) mRNA (judging from reduction of pXBP1(U) protein in lanes 7 and 8 in Fig. 3B), whereas that of NLS-IRE1α-CTR resulted in cleavage of more XBP1(U) mRNA (Fig. 3B, lanes 7 and 9). The expression of cytoplasmic IRE1α-CTR was higher than that of NLS-IRE1α-CTR (Fig. 3B, lanes 14 and 15), suggesting that the RNase activity of NLS-IRE1α-CTR is as effective as that of cytoplasmic IRE1α-CTR. Thus, it is not likely that NLS-IRE1α-CTR is an ineffective RNase or that overaccumulation of NLS-IRE1α-CTR in the nucleus disrupts IRE1α-RNA ligase complexes (which might need to form in a defined stoichiometric ratio), leading to cleavage but not ligation in this situation. Since the amount of XBP1(U) mRNA located in the nucleus is very limited (Fig. 4A), nascent XBP1(U) mRNA might be cleaved by NLS-IRE1α-CTR upon transcription in the nucleus, leading to accumulation of truncated XBP1 mRNA.

Next, we analyzed the subcellular localization of XBP1(U) mRNA, a substrate of unconventional splicing. First, the distribution of XBP1(U) mRNA in the nucleus and cytoplasm was examined. HeLa cells were incubated in PBS or HEPES containing 0.5% NP40, which solubilizes the plasma membrane without disrupting the nuclear membrane. The nucleus was separated from the cytoplasm by centrifugation, and RNA and protein were extracted from the nuclear and cytoplasmic fractions, and subjected to northern and western blotting, respectively (Fig. 4). The purity of each fraction was verified by detecting a nuclear marker (lamin B) as well as a cytoplasmic marker (GAPDH) (Fig. 4D,E). When cells were disrupted in HEPES-NP40 solution, the nuclear fraction was contaminated with cytoplasmic material (Fig. 4, lanes 3 and 4), but purity was improved when cells were disrupted in PBS-NP40 solution (Fig. 4, lanes 1 and 2). Most rRNA was found in the cytoplasmic fraction, although a small amount of rRNA was recovered in the nuclear fraction (Fig. 4C). It is possible that rRNA newly synthesized in the nucleolus might have been responsible for such a nuclear localization. Most XBP1(U) mRNA as well as GAPDH mRNA was localized in the cytoplasmic fraction (Fig. 4A,B), suggesting that XBP1(U) mRNA is rapidly exported from the nucleus after transcription and accumulates in the cytoplasm, and that there is only a small pool of XBP1(U) mRNA in the nucleus.

We next analyzed more precisely the subcellular location of XBP1 mRNA in the cytoplasm (Fig. 5). HeLa cells were disrupted by three cycles of freeze-thawing, and then centrifuged to separate the soluble materials, including cytosol and luminal materials of organelles, from the insoluble materials, including the cellular membrane. RNA and protein were extracted from each fraction, and subjected to northern and western blotting. The purity of each fraction was verified by checking for the presence of an ER membrane marker (calnexin, CNX) and a cytosolic marker (GAPDH) (Fig. 5C,D). CNX was found only in the insoluble fraction, whereas most GAPDH protein was found in the soluble fraction (although a small amount was detected in the insoluble fraction) (Fig. 5C,D, lanes 1-4). Most GAPDH mRNA was found in the soluble fraction, as expected (Fig. 5B, lanes 1-4). By contrast, most mRNA encoding BiP (human gene HSPA5) was localized in the insoluble fraction, reflecting the fact that it docks at the ER membrane through the signal peptide contained in the BiP protein. Interestingly, most XBP1(U) mRNA was found in the insoluble fraction rather than in the soluble fraction in the absence of an ER-stress-inducing agent, thapsigargin (Fig. 5A, lanes 1 and 2), suggesting that a large amount of XBP1(U) mRNA is associated with some cellular membrane other than the nuclear membrane. One possibility is that XBP1 mRNA is associated with this membrane (possibly the ER membrane) so that it can be rapidly processed by the unconventional splicing machinery in response to ER stress. This idea is consistent with published results (Stephens et al., 2005; Yanagitani et al., 2009) (see Discussion). When cells were treated with thapsigargin for 8 hours, the amount of XBP1 mRNA was increased by transcriptional induction in response to ER stress (Yoshida et al., 2000), and XBP1 mRNA fractionated into the soluble fraction was increased, suggesting that XBP1(S) mRNA was released from the membrane.

To confirm this, we performed RT-PCR analysis using fractionated RNAs (Fig. 5E). In the absence of ER stress, only XBP1(U) mRNA was found in both soluble and insoluble fractions (Fig. 5E, lanes 1 and 5). After 8 hours of thapsigargin treatment, the ratio between XBP1(S) mRNA and XBP1(U) mRNA in the insoluble fraction was almost 1:1, whereas the ratio was approximately 2:1 in the soluble fraction (Fig. 5E, lanes 4 and 8), suggesting that XBP1(S) mRNA tends to be released from the membrane. The purity of each fraction was checked by immunoblotting using anti-CNX and anti-GAPDH antisera (Fig. 5F,G).

The above observations indicate that most XBP1(U) mRNA is localized in the cytoplasm, suggesting that unconventional splicing machinery uses the cytoplasmic pool of XBP1(U) mRNA. However, it remained possible that the reaction of unconventional splicing was coupled with transcription, and that nuclear XBP1 mRNA was spliced in the nucleus and rapidly transported to the cytoplasm. To exclude this possibility, de novo transcription was halted by adding α-amanitin, an inhibitor of RNA polymerase II (Stirpe and Fiume, 1967) (Fig. 6A-C). In the absence of ER stress, XBP1 mRNA was not spliced, and only small amounts of XBP1 and BiP (HSPA5) mRNAs were observed (Fig. 6A-C, lane 1). When cells were treated with thapsigargin for 8 hours to induce ER stress, most XBP1(U) mRNA was spliced and the levels of XBP1 and BiP (HSPA5) mRNAs were increased (Fig. 6A-C, lane 2). When cells were treated with both α-amanitin and thapsigargin, transcriptional induction of BiP and XBP1 was completely inhibited, indicating that de novo transcription was completely abolished (Fig. 6A,B, lane 4). In this situation, most XBP1(U) mRNA was converted to mature mRNA (Fig. 6C, lane 4), suggesting that unconventional splicing of XBP1 can occur without de novo transcription, and that the unconventional splicing machinery splices the pre-existing pool of XBP1(U) mRNA, which is mostly cytoplasmic (see Fig. 4A, lanes 1-4). In other words, the cytoplasmic pool of XBP1(U) mRNA is not a `dead-end' product, but is rather a cryptic substrate for unconventional splicing. We obtained essentially the same results using other inhibitors of RNA polymerase II, such as actinomycin D (Goldberg et al., 1962) and DRB (Sehgal et al., 1976) (Fig. 6D,E). Thus, it is highly likely that the major source of substrate for unconventional splicing is cytoplasmic XBP1(U) mRNA. Considering the above evidence concerning IRE1, RNA ligase and XBP1(U) mRNA together, we conclude that the unconventional splicing of XBP1(U) mRNA occurs predominantly in the cytoplasm, where both the machinery and substrate for the unconventional splicing exist.

In the past, it has been controversial whether unconventional mRNA splicing occurs in the nucleus, cytoplasm or both. This question is very important, not only for research on RNA splicing, but also for research on the ER-stress response, because the location of the splicing reaction is closely linked with the biological significance of the unconventional splicing (see below). Since XBP1 splicing occurs very fast (15-30 minutes after ER stress) (see Fig. 6D,E) compared with the cell cycle (approximately 24 hours in HeLa cells), and ER stress arrests cells at G1 phase (Brewer et al., 1999), the nucleus is separated from the cytoplasm during ER stress. Walter and colleagues clearly demonstrated that unspliced HAC1 mRNA is located in the cytoplasm and associated with stalled polysomes, and that splicing of HAC1 can occur even if de novo transcription is inhibited, strongly suggesting that unconventional splicing occurs in the cytoplasm in yeast cells (Ruegsegger et al., 2001). However, Gething and colleagues recently reported that yeast Ire1p contains a nuclear localization sequence, and that HAC1 splicing requires both nuclear localization of Ire1p and components of the nuclear import machinery such as RanGAP and RanGEF (Goffin et al., 2006). From these observations, they argued that newly synthesized unspliced HAC1 mRNA is spliced in the nucleus, whereas the pre-existing pool of unspliced HAC1 mRNA is spliced in the cytoplasm. However, if this is the case, yeast cells have unconventional splicing machinery in both the nucleus and cytoplasm, and point mutations in the NLS of Ire1p should not abolish splicing of HAC1 mRNA upon ER stress, contradicting their observations. Moreover, the NLS sequences that they found in yeast Ire1p (KKKRKR and KKGR) are not conserved in species. Thus, it seems plausible that yeast unconventional splicing takes place in the cytoplasm.

As for mammalian unconventional splicing, the location of the splicing reaction has remained elusive. Kaufman and colleagues showed that IRE1 is localized to the inner nuclear envelope, supporting the notion that XBP1 splicing occurs in the nucleus (Lee et al., 2002), although they reported in a previous paper that IRE1 is located in the ER and the nuclear membrane (Holmer and Worman, 2001; Tirasophon et al., 1998). They also reported that XBP1(U) mRNA either transcribed in the cytosol by T7 RNA polymerase or delivered to the cytosol by RNA transfection can be spliced in response to ER stress, and that ectopic expression of the catalytic domain of IRE1 in the cytoplasm induced XBP1 splicing (Back et al., 2006). Iwawaki and colleagues reported similar results (Iwawaki and Akai, 2006). These observations suggest that upon artificial expression of XBP1 mRNA or IRE1 in the cytoplasm, unconventional splicing can occur in the cytoplasm of mammalian cells in the absence of nuclear processing. However, Walter and colleagues reported that IRE1 residing in the ER is translocated to the nucleus upon ER stress to cleave a substrate mRNA (Niwa et al., 1999; Tirasophon et al., 1998). Thus, it is unclear whether XBP1 splicing occurs in the nucleus, or whether it actually occurs in the cytoplasm in vivo.

Here, we tried to clearly determine whether mammalian unconventional splicing occurs in the cytoplasm, nucleus or both. We first examined whether IRE1 is located in the nucleus, especially in the inner nuclear membrane (INM). Our finding that the catalytic domain of IRE1α has a nuclear exclusion signal strongly suggests that the NES of IRE1α has been evolved to prevent mislocalization of the IRE1α catalytic domain to the nucleus (Figs 1 and 2). Mislocalization of IRE1 would be detrimental to cells because the nucleus is not the place where the unconventional splicing occurs under normal physiological conditions, and mislocalization might cause accidental cleavage of XBP1(U) mRNA. It is possible that the NES is important to prevent mislocalization of IRE1α to the inner nuclear envelope after reassembly of the nuclear envelope after nuclear division.

The targeting of INM proteins such as lamin B receptor to the INM has been well studied (Holmer and Worman, 2001). INM proteins are synthesized in the ER and freely diffuse in the interconnected membranes of the ER and nucleus; they move along the nuclear pore complex and arrive at the INM. As a result of binding to nuclear ligands such as lamins or chromatin, INM proteins are retained there. Movement through the nuclear pore complex is limited by protein size. Membrane proteins whose nucleoplasmic domain is larger than 67 kDa cannot diffuse through the nuclear pore complexes. INM proteins contain targeting signals to the inner nuclear membrane called the inner nuclear membrane sorting motif (INM-SM) (Braunagel et al., 2004). The INM-SM contains two major features, namely, a hydrophobic stretch of 18-20 amino acid residues constituting a transmembrane sequence and a cluster of charged residues exposed within the cytoplasm or nucleoplasm, which is positioned within 4-8 residues of the end of the transmembrane sequence. Importin-α-16 is located adjacent to the translocon protein Sec61α and recognizes the sorting motif of INM proteins, and facilitates their sorting to the INM (Saksena et al., 2006). Human IRE1 has neither an INM-SM nor a classical NLS motif, strongly suggesting that human IRE1 is not an INM protein. Since the cytoplasmic domain of human IRE1 is 57.7 kDa, it can stray into the INM by diffusion through the nuclear pore complex, and NES is crucial for preventing such mislocalization.

Our second finding, that ectopic expression of the catalytic domain of IRE1α in the nucleus resulted in impairment of unconventional splicing (Fig. 3), is consistent with the above results, and strongly suggests that very little of the RNA ligase involved in unconventional splicing resides in the nucleus. It is possible that our findings that overexpression of cytoplasmic IRE1α-CTR enhanced unconventional XBP1 splicing (Fig. 3B, lane 8) might merely be an artifact caused by colocalization of the enzymes with an otherwise quiescent pool of XBP1 mRNA. However, we propose that the RNA ligase responsible for unconventional splicing actually exists in the cytoplasm, because we overexpressed only cytoplasmic IRE1, and not RNA ligase. Moreover, overexpression of NLS-IRE1 resulted in the truncation of XBP1 mRNA (Fig. 3B, lane 3), indicating that the nuclear function required for cleavage of XBP1 mRNA is not impaired. The expression of cytoplasmic IRE1α-CTR was higher than that of NLS-IRE1α-CTR (Fig. 3B, lanes 14 and 15), meaning that it is unlikely that overaccumulation of NLS-IRE1α-CTR in the nucleus disrupts IRE1-RNA ligase complexes. Kaufman and colleagues reported that both NLS-IRE1α-CTR and cytoplasmic IRE1α-CTR showed a low level of XBP1 to GFP mRNA splicing that did not increase upon AP20187 treatment (AP20187 was used to dimerize membrane bound-IRE1), whereas membrane-bound IRE1-CTR produced mostly spliced XBP1-GFP mRNA in response to AP20187 treatment (Back et al., 2006). The discrepancy between their results and ours might be due to some difference in experimental methods. They used reverse transcription PCR to measure exogenously transfected XBP1-GFP mRNA, whereas we examined XBP1 mRNA by northern blot analysis.

The mammalian RNA ligase responsible for unconventional splicing has not yet been identified, and the mammalian genome seems to contain no ortholog of yeast tRNA ligase RLG1, which ligates unspliced HAC1 mRNA cleaved by Ire1p. Although it is thought that Rlg1p localizes to the nucleus (Clark and Abelson, 1987) and tRNA splicing occurs in the nucleus, Yoshihisa and colleagues reported that Sen2p and Sen54p, components of tRNA splicing endonuclease, are located not in the nucleus, but on the mitochondrial surface (Yoshihisa et al., 2003), suggesting that Rlg1p could be distributed in the cytoplasm. Indeed, Weissman and colleagues showed that Rlg1p localizes to the cytoplasm (Huh et al., 2003). In plants, Arabidopsis and Oryza tRNA ligases, as well as tRNA splicing endonuclease and 2′-phosphotransferase, are reported to be preferentially located in the nucleus, although they also seem to be targeted to other cellular compartments, including the chloroplasts and the mitochondria (Englert and Beier, 2005). Thus, it is possible that mammalian tRNA ligase is located in the cytoplasm, although it remains unclear whether the RNA ligase involved in mammalian unconventional splicing is a tRNA ligase. There is accumulating evidence supporting the notion that the mechanism of tRNA ligation is highly diverged between yeast and mammals (Abelson et al., 1998; Hopper and Phizicky, 2003; Laski et al., 1983; Sawaya et al., 2003), although HeLa cells have a minor RNA ligase activity with the same biochemical hallmarks as yeast tRNA ligase (Zillmann et al., 1991). Actually, XBP1 splicing is intact in mice deficient in Trpt1, which encodes murine tRNA splicing 2′-phosphotransferase (Harding et al., 2008). Inactivation of Trpt1 eliminates all detectable 2′-phosphotransferase activity from cultured mouse cells, suggesting that tRNA-splicing 2′-phosphotransferase is not involved in XBP1 splicing.

Finally, we clearly demonstrated that the pool of XBP1(U) mRNA in the cytoplasm is very large compared with that in the nucleus (Fig. 4), and that the cytoplasmic XBP1(U) mRNAs are spliced in response to ER stress even if de novo transcription is blocked (Fig. 6). This suggests that unconventional splicing uses the cytoplasmic pool of XBP1(U) mRNA as the source of its substrate, although it is still possible that nuclear XBP1(U) mRNA is also used as a source if an unidentified mRNA transport mechanism shuttles the cytoplasmic pool of XBP1(U) mRNA into the nucleus for splicing. Kaufman and colleagues previously reported that XBP1(U) mRNA was equally detected in both the nuclear and cytoplasmic fractions (Back et al., 2006), which is inconsistent with our observation that most XBP1(U) mRNA is localized in the cytoplasm. It is possible that this discrepancy is again due to differences in the experimental procedures. Kaufman and colleagues measured the levels of exogenously transfected XBP1-GFP transcripts by reverse-transcription PCR, which might be at saturation and consequently could grossly underestimate the pool of cytoplasmic XBP1 mRNA.

Based on experiments using a temperature-sensitive mutant of RNA polymerase II, Ruegsegger and colleagues reported that yeast HAC1 splicing does not require de novo transcription (Ruegsegger et al., 2001), suggesting that the mechanism of unconventional splicing is conserved between yeast and mammals. Back and co-workers reported that de novo transcription is essential for XBP1 splicing (Back et al., 2006), contradicting our results. The discrepancy between these and our data might be derived from the differences of the methods used. We measured endogenous XBP1 mRNA, whereas they measured exogenously transfected XBP1-GFP mRNA.

The discovery that most XBP1(U) mRNA is associated with membranes in the cytoplasm (Fig. 5) is consistent with a previous report by Nicchitta and colleagues, in which they separated cytosolic RNA from ER-associated RNA by centrifugation through a sucrose gradient, and found that XBP1 mRNA was rich in the ER-associated fraction (Stephens et al., 2005). Kohno and colleagues also independently found that XBP1(U) mRNA associates with the ER membrane (Yanagitani et al., 2009). It is reasonable to suggest that XBP1(U) mRNA is concentrated on the ER membrane, where IRE1 senses the situation of the ER. Since XBP1(U) mRNA showed more affinity for membranes than its spliced product, it is possible that IRE1 retains XBP1(U) mRNA by binding its splicing site. From these three lines of evidence, i.e. that the catalytic domain of IRE1 and RNA ligase activity were localized in the cytoplasm and cytoplasmic XBP1(U) mRNA was the main source of splicing substrate, together with the previous reports, we concluded that mammalian unconventional splicing occurs predominantly in the cytoplasm, as in the case for yeast unspliced HAC1 mRNA, and that the mechanism of the unconventional splicing is well conserved from yeast to mammals.

This conclusion raises another important question: why do eukaryotic cells use cytoplasmic splicing to regulate ER stress instead of nuclear splicing? One of the possible answers to the above question is that cytoplasmic splicing is simpler than nuclear splicing. In the case of nuclear splicing, signals from the ER must be transmitted to the nucleus to activate the splicing machinery. After the splicing reaction, spliced mRNA should be transported to the cytosol for translation. After translation, XBP1 protein has to go back to the nucleus to activate the transcription of ER chaperone genes. Thus, signaling molecules need to cross the nuclear membrane three times. In the case of cytoplasmic splicing, XBP1 protein is the only signal molecule that crosses the nuclear membrane, because splicing and translation can occur in the cytosol. Another possible answer to the question is that cytoplasmic splicing is more energy efficient. There is a large pool of XBP1(U) mRNA in the cytoplasm, from which a negative regulator, pXBP1(U), is translated (Yoshida et al., 2006b). However, nuclear splicing cannot use this pool of substrate because cytoplasmic XBP1(U) mRNA might not be transported to the nucleus (shuttling of cytoplasmic mRNA to the nucleus has not been discovered thus far), and requires de novo transcription and translation as well as transport of mRNA to the cytosol, thus probably consuming more ATP. Induction of transcription and translation are usually time-consuming processes (XBP1 mRNA and protein require 2 hours and 4 hours to reach their maxima in response to ER stress, respectively), compared with nuclear transport of transcription factors and it takes 3-9 minutes for NF-AT to translocate into the nucleus upon Ca²⁺ signaling (Kwon et al., 2008). These additional processes could still be a good point for introducing a layer of regulation, but the ER-stress response does not seem to require it. Moreover, unintended translation of pXBP1(U) from unspliced mRNA still continues during ER stress. By contrast, cytoplasmic splicing can convert cytoplasmic XBP1(U) mRNA to XBP1(S) mRNA in response to ER stress, and can rapidly produce an active transcription factor, pXBP1(S), simultaneously shutting off pXBP1(U) production, while consuming less time and energy. We speculate that eukaryotic cells have evolved such a mechanism of cytoplasmic splicing to optimally regulate the ER-stress response during evolution.

HeLa cells were cultured in Dulbecco's modified Eagle's medium (DMEM: glucose at 4.5 g/l) supplemented with 10% fetal calf serum, 2 mM glutamine and antibiotics, 100 U/ml penicillin-100 μg/ml streptomycin. Cells were maintained at 37°C in a humidified 5% CO₂ 95% air atmosphere.

Transient transfection of HeLa cells was carried out by the standard calcium phosphate method (Sambrook et al., 1989; Yoshida et al., 2006a). HeLa cells cultured in 24-well or 60-mm dishes were incubated with precipitates of calcium phosphate containing plasmids for 6 hours at 37°C. After washing with phosphate-buffered saline (PBS) to remove CaPO₄-DNA precipitates, cells were incubated in fresh medium for 24 hours and harvested for analysis.

pcDNA-human IRE1α was a kind gift from Randal J. Kaufman (University of Michigan, Ann Arbor, MI). A plasmid expressing IRE1α-CTR (pCMV-HA-IRE1α-CTR) was constructed by inserting a cDNA encoding residues 468-997 of human IRE1α into an XhoI site of pCMV-HA vector, whereas pCMV-HA-NLS-IRE1α-CTR was made by inserting residues 308-330 of ATF6α (containing a nuclear localization signal) into a BglII site of pCMV-HA-IRE1α-CTR. Expression plasmids for a series of IRE1α deletions fused with a HA tag were made by ligating the PCR product of the corresponding region with pCMV-HA. pCMV-HA-IRE1α-CTR-NES mutant was constructed by PCR-based site-directed mutagenesis, using a commercially available kit (QuikChange Site-Directed Mutagenesis Kit: Stratagene) and a pair of primers (AGCTGCCTGCAGAGGGGCGGGAGACGGGGGGGTCCCTCCCCGA and TCGGGGAGGGACCCCCCCGTCTCCCGCCCCTCTGCAGGCAGCT). To construct a plasmid expressing XBP1(U) mRNA [pcDNA-pXBP1(U)], a 1787 bp fragment of XBP1 cDNA encoding pXBP1(U), including the 3′-untranslated region, was cloned into an XhoI site of pcDNA3.1 vector (Invitrogen). Expression plasmids for a series of pXBP1(U) deletion mutants were made by ligating the PCR product of the corresponding region with pcDNA3.1.

HeLa cells grown on coverslips were transiently transfected with appropriate expression plasmids by the calcium phosphate method described above. Cells were fixed with 2% paraformaldehyde in PBS containing 2 mg/ml NaIO₄ and 10 mg/ml lysine for 10 minutes, permeabilized with 0.2% Triton X-100 in PBS for 10 minutes, and stained with appropriate antisera. Coverslips were mounted with 90% glycerol, 10% PBS containing 100 ng/ml DAPI. Fluorescent microscopy was carried out with an E800 microscope (Nikon) and ORCA-ER digital camera (Hamamatsu photonics).

Cells grown in a 60 mm culture dish were harvested with a cell scraper and pelleted by centrifugation. The pellet was suspended in 20 μl of ice-cold PBS containing protease inhibitors (100 μM AEBSF, 80 μM aprotinin, 1.5 μM E-64, 2 μM leupeptin, 5 μM bestatin and 1 μM pepstatin A, 10 μM MG132), mixed with 20 μl of 4× SDS-sample buffer (200 mM Tris-HCl (pH 6.8), 400 mM DTT, 8% SDS and 40% glycerol), and immediately boiled at 100°C for 10 minutes. Portions of samples (10 μl) were subjected to SDS-polyacrylamide gel electrophoresis using 4-20% gradient gels, transferred onto a Hybond-P membrane (GE), and incubated with various antisera, according to standard protocols (Sambrook et al., 1989). Anti-XBP1-A (produced previously) detects both pXBP1(U) and pXBP1(S) (Yoshida et al., 2001a). Anti-IRE1α antiserum was raised by immunizing rabbits with the 218-366 region of human IRE1α fused with glutathione S-transferase. An ECL western blotting detection kit (GE) and a LAS-3000 lumino image analyzer (Fuji Film) were used to detect antigens.

Total RNA extracted from cells with guanidine-phenol was separated by electrophoresis on a 2% or 3% agarose gel containing 2.2 M formaldehyde, blotted onto a Hybond-N⁺ membrane (GE), hybridized with alkaline phosphatase-conjugated cDNA probes, and detected with a LAS-3000 lumino image analyzer using the Gene Images AlkPhos Direct Labeling and Detection System (GE).

RT-PCR of XBP1 mRNA was performed essentially as described previously (Yoshida et al., 2001a). 10 μg total RNA was reverse-transcribed with MMLV reverse transcriptase (Invitrogen) and amplified with Ex-Taq polymerase (Takara) using a pair of primers that correspond to nucleotides 493-512 (CGCGGATCCGAATGAAGTGAGGCCAGTGG) and 834-853 (GGGGCTTGG TATATATGTGG) of XBP1 mRNA, respectively. Amplified fragments covering a 26 nucleotide intron (nucleotides 531-556) and flanking exon fragments were separated on 4-20% polyacrylamide gels, visualized by staining with Gel Red (Biotium) and detected using an LAS-3000 (Fuji Film).

To prepare the nuclear and cytoplasmic fractions of cells, HeLa cells grown on 60-mm dishes were collected using a cell scraper, and incubated in 0.5% NP40-PBS solution or 0.5% NP40-HEPES solution (20 mM HEPES-KOH, pH 7.9, 100 mM KCl, 10% glycerol, 1 mM MgCl₂, 1 mM 2-mercaptoethanol) at 4°C for 10 minutes to solubilize the plasma membrane. After centrifugation at 500 g for 5 minutes at 4°C, the precipitate containing the nuclei was separated from the supernatant containing the cytoplasm, and both fractions were subjected to western and northern blotting. To separate the soluble and insoluble fractions, HeLa cells collected from 60 mm dishes were subjected to three cycles of freeze-thawing in PBS, and centrifuged at 17,800 g for 10 minutes at 4°C. Precipitates and supernatants containing the insoluble and soluble fractions, respectively, were subjected to western, northern and RT-PCR analyses.