RBM39 modulates UPR signaling through alternative splicing of IRE1α/ERN1

A genome-wide phenotypic screen identifies spliceosome related genes as regulators of noncanonical XBP1 cytoplasmic splicing

To systematically identify regulators of noncanonical cytoplasmic splicing of XBP1, we applied our recently developed CRASP-seq platform (Behera et al. 2026). CRASP-seq couples high-throughput paired-end sequencing of splicing reporter transcripts—in this case, an XBP1 open reading frame containing the 26 nt intron—with direct identification of the corresponding knockout genotype through physically linked hgRNA expression cassettes (Fig. 1A,B). The XBP1 reporter is expressed under the control of a doxycycline-inducible tetracycline response element (TRE), driven by the reverse tetracycline-controlled transactivator (rtTA) fused to a puromycin resistance cassette via a self-cleaving 2A peptide (Fig. 1A). Each reporter transcript harbors the hgRNA cassette in reverse orientation within its 3′ untranslated region, ensuring that the guide sequence is captured during sequencing. This design enables the direct coupling of specific genetic perturbations to their corresponding splicing outcomes, with read 1 identifying the hgRNA and read 2 quantifying XBP1 intron retention levels (Fig. 1B). As a result, CRASP-seq provides a robust, quantitative, and cost-effective framework for dissecting alternative splicing regulation at genome scale—without the need for fluorescent reporters, flow cytometry, or specialized instrumentation (Behera et al. 2026).

To validate the system, we cloned the XBP1 reporter (Fig. 1A) into CRASP-seq lentiviral vectors expressing hgRNAs targeting ERN1 (encoding IRE1α, the endoribonuclease required for XBP1 splicing) or a nontargeting control. Cas9- and Cas12a-expressing HAP1 cells were transduced with these vectors, and reporter expression was induced with doxycycline. Upon tunicamycin treatment—an ER stressor that inhibits N-glycosylation—we observed robust XBP1 reporter splicing in control cells, whereas ERN1-targeted cells display almost complete splicing inhibition (Fig. 1C). These results demonstrate that the reporter faithfully responds to ER stress and that the CHyMErA system enables efficient gene perturbation in UPR signaling pathways.

We next cloned the XBP1 reporter into a newly designed genome-wide CRASP-seq lentiviral library (Supplemental Table S1), which targets 18,302 human genes using dual Cas9 and Cas12a gRNAs, with four independent hgRNAs designed per gene to control for guide-to-guide variability. This dual-targeting approach enhances knockout efficiency and enables perturbation of paralogs and selected noncoding RNAs by targeting their promoters. The library comprises 91,078 hybrid gRNAs (hgRNAs), including 504 intergenic and nontargeting controls to account for background effects and establish a baseline for phenotypic comparisons. The XBP1 lentiviral genome-wide plasmid library exhibited a tight distribution (Supplemental Fig. S1A) and was packaged into lentivirus to transduce HAP1, RPE1, and HepG2 cells expressing Streptococcus pyogenes Cas9 and an optimized Acidaminococcus sp. Cas12a variant (opCas12a) (Gier et al. 2020; Xiao et al. 2024). Transduction was performed at a low multiplicity of infection (MOI), maintaining >200-fold coverage. Following puromycin selection and gene editing, we induced XBP1 reporter expression with doxycycline for 24 h. Four hours before harvesting, cells were treated with either tunicamycin or DMSO to assess XBP1 splicing under both ER stress and steady-state conditions.

XBP1 reporter transcripts were analyzed following our optimized CRASP-seq pipeline. Total RNA was extracted from pooled cells and poly(A)⁺ mRNA was isolated using oligo(dT) beads. cDNA synthesis was performed using a reporter-specific oligo containing unique molecular identifiers (UMIs), enabling PCR duplicate removal during downstream analysis. The resulting Illumina libraries were subjected to paired-end sequencing (Fig. 1B). This RNA-coupled chemogenetic screen directly links genetic perturbations to changes in XBP1 intron splicing, quantified as percent intron retention (PIR) for each guide RNA in the library (Fig. 1D; Supplemental Fig. S1B; Supplemental Table S2).

As expected, ERN1 emerged as the strongest positive regulator of XBP1 splicing in the presence of tunicamycin, consistent with its role in noncanonical intron cleavage (Fig. 1D; Supplemental Fig. S1B). Among additional hits, CALR, which encodes a calcium-binding ER chaperone frequently mutated in myeloproliferative neoplasms and implicated in UPR regulation (Ibarra et al. 2022; Ying et al. 2026), as well as genes involved in cell cycle control—including cyclin A1, cyclin A2 (identified as a paralog hit) (Supplemental Table S2), and PLK1—were identified as suppressors of XBP1 splicing upon knockout. These findings are consistent with the established link between cell cycle arrest and alleviation of ER stress (Chowdhury et al. 2024).

Conversely, HSPA5, encoding the ER chaperone BiP/GRP78 that maintains IRE1α in an inactive state, emerged as the strongest negative regulator under basal conditions. Knockout of HSPA5 alone is sufficient to induce reporter splicing, reinforcing its established role in UPR regulation. Similarly, loss of DNAJB9, a cochaperone that recruits HSPA5 to IRE1α, also promotes XBP1 splicing, likely by facilitating IRE1α activation through dimerization (Amin-Wetzel et al. 2017). Additional prominent hits include core components of the nonsense-mediated decay (NMD) pathway, such as UPF1, UPF2, SMG1, SMG5, and SMG7 (Karam et al. 2015; Sieber et al. 2016; Goetz and Wilkinson 2017; Li et al. 2017), as well as members of the Sec61/Sec63 translocon complex (including SEC61A1, SEC61G, and SEC63), which has been shown to associate with IRE1α (Plumb et al. 2015; Li et al. 2020). Both the NMD machinery and the Sec61/Sec63 complex have been previously implicated in the modulation of ER stress responses. Notably, ALG11, a glycosyltransferase involved in N-linked glycosylation—the process targeted by tunicamycin—also scores as a strong negative regulator, particularly under untreated conditions. Importantly, hits are highly consistent across HAP1, RPE1, and HepG2 cells, with significant overlap in top-ranking genes (Fig. 1D; Supplemental Fig. S1C) and strong correlation in PIR values (Supplemental Fig. S1D), underscoring the robustness and reproducibility of our screening approach.

Gene ontology (GO) analysis of high-confidence hits revealed strong enrichment for expected terms such as “response to ER stress,” “ER unfolded protein response,” and “protein catabolic processes”—all consistent with established XBP1 functions. However, the most significantly enriched GO terms were unexpectedly associated with “RNA splicing, via transesterification reactions” (Fig. 1E; Supplemental Fig. S2A). Among the top hits was RBM39, a known alternative splicing regulator that enhances U2 snRNP recruitment. In addition, core U2 spliceosomal components such as SF3B1 and SF3B3, among others, were also strongly enriched. Given that previous studies in yeast showed that canonical spliceosome components do not influence HAC1 (the yeast XBP1 homolog) splicing (Sidrauski et al. 1996; Sidrauski and Walter 1997; Gonzalez et al. 1999), these findings were unexpected and intriguing. We therefore pursued this connection between the nuclear spliceosome and cytoplasmic noncanonical XBP1 splicing in greater detail.

RBM39 and U2 snRNP components regulate alternative splicing of ERN1 exon-18

To validate the role of RBM39 and U2 snRNP components in XBP1 regulation, we first assessed endogenous XBP1 splicing following tunicamycin-induced ER stress. We used both genetic and pharmacological approaches to perturb spliceosome function. Specifically, siRNA-mediated knockdown of RBM39, SF3B1, and SF3B3 was performed (Supplemental Fig. S3A). In parallel, we treated cells with indisulam, a sulfonamide that induces proteasomal degradation of RBM39 by recruiting it to the CUL4–DCAF15 E3 ubiquitin ligase complex (Han et al. 2017), and pladienolide B (PladB), a potent inhibitor of SF3B1, a core U2 snRNP subunit (Kotake et al. 2007).

RT-PCR analysis in HAP1 cells revealed that depletion of RBM39, SF3B1, or SF3B3 markedly reduces splicing of XBP1 intron-4 following ER stress induction (Fig. 2A). Similar inhibition of XBP1 splicing was observed upon treatment with indisulam or PladB, which leads to near-complete suppression of the spliced isoform (Fig. 2B). To further corroborate these effects, we performed RNA-seq in cells depleted of the same splicing regulators (Fig. 2C; Supplemental Fig. S3B; Supplemental Tables S3–S6). Transcriptome-wide analysis demonstrated that disruption of RBM39 or U2 snRNP components dampens the transcriptional activation of UPR target genes in response to ER stress (Fig. 2D; Supplemental Fig. S3C). These transcriptomic results were independently validated by qRT-PCR analysis of selected UPR-responsive genes, including HSPA5 and DNAJB9, confirming that perturbation of U2 snRNP components suppresses their expression (Fig. 2E). Together, these findings establish RBM39 and the U2 snRNP as essential mediators of efficient UPR signaling and functionally validate the novel regulators of XBP1 splicing identified through our CRASP-seq screen.

Given that XBP1 splicing occurs in the cytoplasm, we hypothesized that RBM39 regulates XBP1 indirectly by modulating alternative splicing of upstream regulators. To investigate this hypothesis, we analyzed our RNA-seq data sets to identify alternative splicing changes induced by genetic and pharmacological perturbation of RBM39 and U2 snRNP components (Fig. 3A). We focused on genes identified as top hits in our CRASP-seq screen. This analysis revealed that knockdown of RBM39, SF3B1, or SF3B3, as well as treatment with indisulam and PladB, consistently led to increased skipping of exon-18 in ERN1—the strongest positive regulator of XBP1 splicing identified in our screen (Fig. 3A,B; Supplemental Tables S4, S6).

Exon-18 skipping is predicted to generate a previously uncharacterized IRE1α isoform lacking the C-terminal RNase domain but retaining the dimerization and kinase domains (Fig. 3C). We validated the role of RBM39 in regulating ERN1 exon-18 splicing through RT-PCR assays, which showed that knockdown of RBM39 or SF3B components, as well as treatment with indisulam or PladB, all lead to robust exon-18 skipping (Fig. 3D). These findings support the hypothesis that alternative splicing of ERN1 represents a regulatory mechanism through which the spliceosome modulates noncanonical cytoplasmic XBP1 splicing and UPR signaling.

ERN1 exon-18 skipping produces an unstable transcript and a dominant negative protein that impairs XBP1 splicing

To evaluate the functional significance of the IRE1α Δexon-18 isoform, we engineered HEK293 Flp-In cell lines allowing doxycycline-inducible expression of either the full-length (FL) or Δexon-18 (ΔEx18) variant of IRE1α, each resistant to siRNAs targeting endogenous ERN1 (Supplemental Fig. S4A). Endogenous ERN1 was knocked down using siRNA, while exogenous expression of the two isoforms was induced with doxycycline in cells treated with either tunicamycin or DMSO. Knockdown conditions were selected to retain residual endogenous IRE1α, enabling assessment of the two protein isoforms in the presence of endogenous IRE1α (Fig. 4A; Supplemental Fig. S4B). RT-PCR assays showed that tunicamycin robustly induces XBP1 intron removal, and this effect is further enhanced by IRE1α overexpression. In contrast, expression of the Δexon-18 IRE1α isoform markedly reduces XBP1 splicing (Fig. 4A,B, cf. lanes 10 and 12). These findings are consistent with a dominant negative mechanism, where the Δexon-18 isoform (lacking the RNase domain but retaining dimerization and kinase domains) may heterodimerize with IRE1α, thereby blocking its RNase function.

RNA-seq analyses further confirmed that expression of the Δexon-18 isoform attenuates XBP1 splicing and dampens downstream UPR gene expression (Supplemental Fig. S4C; Supplemental Tables S7, S8). This transcriptional repression was validated by qRT-PCR analysis of the XBP1 target gene DNAJB9 (Fig. 4C, cf. lanes 10 and 12). Taken together, these results suggest that ERN1 exon-18 skipping produces a nonfunctional IRE1α variant that acts in a dominant negative fashion to suppress XBP1 splicing and the broader UPR.

Next, to test the physiological relevance of endogenous exon-18 skipping, we employed CRISPR base editing to mutate the 5′ splice site of ERN1 exon-18 and induce its exclusion from the transcript at the endogenous level (Fig. 4D). We transfected HEK293T cells with an all-in-one vector expressing the ABE8e base editor and a guide RNA targeting the exon-18 5′ splice site. Control clones were generated using a guide targeting an intergenic region. From the screen, we identified three clones with nearly complete exon-18 skipping and two clones showing partial skipping, suggesting heterozygosity (Fig. 4E). Upon tunicamycin treatment, clones lacking exon-18 inclusion fail to splice XBP1 intron-4 (Fig. 4E). Notably, even heterozygous clones exhibit minimal XBP1 splicing, consistent with a dominant negative effect of the Δexon-18 isoform. This defect results in reduced induction of DNAJB9, confirming impaired UPR output (Fig. 4F).

Because skipping of exon-18 alters the ERN1 reading frame, we investigated its impact on mRNA stability and protein expression. We observed that Δexon-18 transcripts are significantly less stable, consistent with their degradation by NMD triggered by the premature stop codon (Fig. 4G,H). Treatment with SMG1i, a selective NMD inhibitor, restores the stability of Δexon-18 transcripts and also increases levels of wild-type ERN1 mRNA, in line with previous reports identifying ERN1 as a natural NMD target (Karam et al. 2015). Consistent with these findings, exon-18 skipping leads to reduced levels of IRE1α protein. Importantly, SMG1i treatment rescues protein expression by stabilizing Δexon-18 transcripts, resulting in accumulation of the corresponding truncated IRE1α isoform (Fig. 4I).

These findings suggest that exon-18 inclusion ensures transcript stability and functional protein production, while its skipping promotes transcript degradation and expression of a dominant negative isoform. Interestingly, this mechanism mirrors the regulation of XBP1 isoforms, where the unspliced isoform (XBP1u) is less stable and can inhibit XBP1s activity playing a dominant negative role (Calfon et al. 2002; Lee et al. 2003a). Together, these results highlight a convergent strategy within the UPR pathway, where both spliceosome-mediated and noncanonical alternative splicing shape protein stability and function, ultimately tuning the cellular response to stress.

Heat shock induces ERN1 exon-18 skipping to restrain maladaptive IRE1α–XBP1 signaling

To investigate the physiological contexts that induce ERN1 exon-18 skipping, we hypothesized that this splicing event might act as a regulatory mechanism to attenuate the IRE1α–XBP1 arm of the UPR in response to severe cellular stress. Such regulation could allow cells to fine-tune their transcriptional responses, potentially dampening prosurvival signaling when stress becomes overwhelming. To test this, we exposed HEK293T cells to a range of cellular stressors—including genotoxic, heat shock, ER, osmotic, hypoxic, hypotonic, and oxidative stress—and assessed ERN1 exon-18 splicing. We observed a clear increase in exon-18 skipping following heat shock, with milder effects under hypertonic stress. In contrast, ER, hypoxic and osmotic stresses have minimal or no impact on exon-18 splicing (Fig. 5A,B). These findings suggest that ERN1 exon-18 skipping is specifically induced by certain stressors beyond ER stress, potentially serving to dampen IRE1α–XBP1 activity under conditions where prolonged or inappropriate UPR signaling could be detrimental.

Next, to assess the functional relevance of XBP1 splicing under heat shock conditions, we generated HEK293 Flp-In cell lines expressing doxycycline-inducible forms of XBP1s, XBP1u, or a splicing-deficient mutant (XBP1u_mut), the latter containing silent mutations that prevent IRE1α-mediated splicing without altering its protein sequence (Fig. 5C). As expected, XBP1s accumulates at higher levels than XBP1u, and tunicamycin treatment converts XBP1u—but not XBP1u_mut—into XBP1s (Fig. 5D; Supplemental Fig. S5A).

Constitutive expression of XBP1s markedly impacts cellular fitness under heat shock conditions. Specifically, XBP1s expression reduces cell fitness and proliferation, as assessed by direct quantification of viable cells (Fig. 5E) and by Alamar Blue assays (Fig. 5F) following heat shock. These effects are stress-specific, as XBP1s expression has only a modest impact on cell fitness under basal growth conditions (Supplemental Fig. S5B,C). In contrast, expression of XBP1u or the splicing-incompetent XBP1u_mut modestly enhances cell fitness following heat shock (Fig. 5F), suggesting that active IRE1α–XBP1 signaling is maladaptive under these conditions. Consistent with this interpretation, overexpression of full-length IRE1α, which robustly promotes XBP1 splicing, decreases cell numbers after heat shock, whereas expression of the Δexon-18 IRE1α isoform results in the opposite effect (Fig. 5E,F). Importantly, these fitness effects are not attributable to differential induction of apoptosis, as neither heat shock nor manipulation of XBP1 isoform expression substantially activates apoptotic signaling (Supplemental Fig. S5D).

Together, these results indicate that hyperactivation of the IRE1α–XBP1 pathway compromises cellular fitness during heat shock and that exon-18 skipping mitigates this cytotoxic signaling by limiting XBP1 splicing.

Heat shock attenuates RBM39 activity to promote ERN1 exon-18 skipping

We next investigated how ERN1 exon-18 splicing is regulated during cellular stress by examining whether RBM39 itself responds to stress signals. RBM39 is known to undergo autoregulation through the inclusion of a poison exon, which provides a sensitive readout of its splicing regulatory activity (Campagne et al. 2023). Under heat shock, we observed significantly reduced inclusion of the RBM39 poison exon, consistent with a reduction in RBM39 functional activity under these conditions (Fig. 6A). Indeed, heat shock leads to modest but reproducible decrease in RBM39 protein abundance (Fig. 6B,C). In parallel, we detected reduced levels of full-length IRE1α and a corresponding increase in the Δexon-18 isoform, which becomes more apparent upon inhibition of NMD (Fig. 6B,C). These findings support a model in which RBM39 activity is attenuated under stress conditions, at least in part through reduced protein levels, leading to altered ERN1 splicing.

To directly investigate the role of RBM39 in ERN1 exon-18 regulation, we constructed wild-type and mutant ERN1 minigenes containing exon-18 and flanking intronic regions (Fig. 6D). The mutant version harbors point mutations in a T-rich motif overlapping the polypyrimidine tract, a sequence resembling known RBM39 binding motifs (Wang et al. 2019; Campagne et al. 2023). RBM39 knockdown leads to reduced exon-18 inclusion in the wild-type but not the mutant minigene, whose splicing is already impaired (Fig. 6D). Furthermore, heat shock enhances exon-18 skipping in the wild-type minigene, mimicking effects observed at the endogenous locus (Supplemental Fig. S5E). These results suggest that RBM39 promotes ERN1 exon-18 inclusion under basal conditions, while its functional attenuation during heat shock stress facilitates exon skipping and IRE1α inactivation.

To assess whether RBM39 regulates ERN1 splicing through direct interaction, we first performed RNA immunoprecipitation experiments and observed significant enrichment of ERN1 transcripts in RBM39-associated complexes compared to controls (Supplemental Fig. S5F). To further probe this interaction, we carried out RNA pull-down assays using biotinylated probes corresponding to the region upstream of ERN1 exon-18, either in its wild-type form or with mutations disrupting a putative RBM39 binding motif (Fig. 6E,F). RBM39 is robustly recovered with the wild-type probe, whereas its association is substantially diminished upon motif disruption. These findings support a model in which RBM39 directly engages ERN1 pre-mRNA to regulate exon-18 splicing. However, we cannot fully exclude the possibility that indirect mechanisms also contribute to this regulation.

Together, these data indicate that RBM39 is required for efficient UPR signaling by promoting ERN1 exon-18 inclusion, whereas proteotoxic stressors such as heat shock attenuate RBM39 activity, increase exon-18 skipping, and consequently dampen UPR output to support cellular adaptation.

Cell culture

HEK293T, HEK293 Flp-In, HAP1, and RPE1 cells were cultured in Dulbecco's modified Eagle medium (DMEM; Gibco 11995073) supplemented with 10% heat-inactivated fetal bovine serum (HI-FBS; Gibco 16140071) and 1% penicillin–streptomycin (Gibco 15140122). HepG2 cells were maintained in Eagle's Minimum Essential Medium (EMEM; ATCC 30-2003) containing 10% fetal bovine serum (FBS; Gibco A5670701). Cells were routinely passaged with 0.25% trypsin–EDTA (Gibco 25200056) and seeded at experiment-appropriate densities in tissue culture-treated plates. Culture medium was replaced every 2–3 days, confluency was monitored using an EVOS M5000 Imaging System (Invitrogen), and viability was assessed with a Vi-Cell BLU Cell Viability Analyzer (Beckman Coulter).

Generation of a genome-wide CRASP-seq lentiviral KO library expressing the XBP1 reporter

A second-generation genome-wide CRASP-seq lentiviral library (v2; Addgene 250494) was constructed by cloning an updated hgRNA pool into the CRASP-seq CBC Backbone vector (Addgene 232068) as previously described (Behera et al. 2026). The revised hgRNA library was designed following the original framework with several modifications, including the use of a comprehensive transcript annotation set to identify overlapping transcript models and preferentially select guides targeting multiple isoforms. Guides containing homopolymer stretches of greater than or equal to four identical nucleotides (e.g., AAAA and CCCC) were excluded. The updated library also contains an increased number of intergenic negative-control guides (412 vs. 331 in v1). Additionally, to minimize template switching during pooled PCR amplification, oligos were synthesized with randomized spacer sequences between the Cas9 and Cas12a guide elements.

To generate the XBP1 reporter cassette, the full XBP1 open reading frame (ORF), including the 26 nt retained intron (Supplemental Table S9). was amplified from a cDNA library using NEBNext Ultra II Q5 Master Mix (New England Biolabs M0544X). PCR conditions were 30 sec at 98°C; 14 cycles of 10 sec at 98°C, 20 sec at 60°C, and 1 min at 72°C; and followed by a 1 min final extension at 72°C. The resulting amplicon was cloned into pCR-Blunt II-TOPO (Thermo Fisher Scientific K280002) according to the manufacturer's protocol. The XBP1 insert was sequence-verified and reamplified with flanking homology arms compatible with NEBuilder HiFi DNA assembly (New England Biolabs E2621L).

For reporter integration into the CRASP-seq library backbone, 1.5 µg of the lentiviral library plasmid pool (Supplemental Table S1) was digested with Eco32I (Thermo Fisher Scientific FD0304) and assembled with 500 ng of the PCR-amplified XBP1 reporter using NEBuilder HiFi. Assembly reactions were purified by sodium acetate/ethanol precipitation as described previously (Aregger et al. 2021) and transformed into Endura electrocompetent E. coli (LGC Biosearch Technologies 60242-2) using electroporation (1 mm cuvette; 25 µF, 200 Ω, 1600 V). Transformants were plated across multiple 15 cm LB agar plates containing 100 µg/mL ampicillin to achieve >500-fold representation of the library. After overnight growth at 30°C, colonies were scraped, pooled, and plasmid DNA was purified using a Qiagen Maxi Kit (Qiagen 12362).

CRASP-seq screening protocol

The generated CRASP-seq XBP1 minigene reporter lentiviral library was used to generate lentivirus as described previously (Behera et al. 2026), and then infect HAP1, RPE1, or HepG2 cells each expressing spCas9 and AsCas12a nucleases (Xiao et al. 2024). Infections were performed at a multiplicity of infection (MOI) of 0.2 to ensure a minimum coverage of 250 cells per hgRNA. After 24 h, selection pressure was applied using 2 µg/mL puromycin. Forty-eight hours after selection, surviving cells were pooled, replated, and doxycycline was added to induce reporter expression. To induce endoplasmic reticulum (ER) stress, cells were treated with either DMSO or 10 µg/mL tunicamycin (Sigma-Aldrich SML1287-1ML) 4 h prior to harvesting. Twenty-four hours after doxycycline induction, cells were collected and divided into aliquots of 10 million cells for downstream analyses.

Total RNA was extracted using the RNeasy Plus mini kit (Qiagen 74136) in accordance with the manufacturer's protocol. Polyadenylated mRNA was isolated using oligo(dT) Dynabeads (Thermo Fisher Scientific 61002), followed by first-strand cDNA synthesis with the Maxima H Minus first strand cDNA synthesis kit (Thermo Fisher Scientific K1651) and a custom primer (listed in Supplemental Table S9). The resulting cDNA was purified using the QIAquick PCR purification kit (Qiagen 28104). PCR amplification with the Illumina barcoded primers (Supplemental Table S9) was optimized using the NEBNext Ultra II Q5 master mix (New England Biolabs M0544X) with the following conditions: an initial denaturation for 30 sec at 98°C; followed by 14 cycles of 10 sec at 98°C, 20 sec at 60°C, and 30 sec at 72°C; and concluding with a final extension for 1 min at 72°C. PCR products were resolved on 1% agarose gel stained with SYBR Safe (Thermo Fisher Scientific S33102), and the target bands were excised and purified using a gel extraction kit (Thermo Fisher Scientific K0691).

Purified libraries were quantified using the Qubit dsDNA HS Assay (Thermo Fisher Scientific Q32851) and the Agilent TapeStation system (Agilent 5067-5582 and 5067-5583). Validated libraries were then pooled and sequenced on the NovaSeq 6000 platform with a 20%–25% PhiX spike-in using 300-cycle kits. The sequencing strategy was as follows: read 1: 210 cycles, index read 1: eight cycles, index read 2: eight cycles, and read 2: 85 cycles.

Analysis of CRASP-seq data

The CRASP-seq FASTQ reads were analyzed to identify and annotate different functional features using a custom bioinformatics pipeline built in Python programming language. The R1 reads were used to extract Cas9 and Cas12a guides, cell barcodes (CBC), and unique molecular identifiers (UMI) by mapping adjacent anchor sequences. STAR aligner tool (Dobin et al. 2013) was used to annotate Cas9 and Cas12a guide pairs by aligning them to the reference guide library. The R2 reads were analyzed to assess XBP1 intron-4 reporter splicing phenotypes and calculate percent intron retention (PIR) values using the following formula:

PIR = [RI/(RI + JSE)] × 100, where PIR is percent intron retention, RI is reads with retention of intron, and JSE is reads with joined or spliced exons.

Then we combined information from R1 and R2 reads to link genomic perturbations and splicing phenotypes and kept only fully annotated reads for downstream analysis. We calculated ΔPIR values with the following formula:

ΔPIR = PIR (knockout) − PIR (intergenic control), where PIR (knockout) represents the PIR value of any knockout guide or gene, and PIR (intergenic control) represents the cumulative PIR value of all the intergenic guides in the library.

We assessed ΔPIR false discovery rate (FDR) values and determined ΔPIR thresholds at 0.05 FDR for hit calling, considering both positive and negative regulation of splicing separately. We applied this ΔPIR thresholds for each screen at the guide level as well as gene level. For a gene to be a hit, it was required to exceed this threshold at the gene level, and more than half of the guides should exceed this threshold for better consistency. Additionally, a gene was also required to be a hit in at least two screens from a total of six screens (three cell lines × two treatment conditions) for greater confidence in the splicing regulation.

For data representations and various downstream assessments and comparisons for splicing regulations, these gene hits were further filtered. For example, the heat map in Figure 1D contains genes that are hits in at least two cell lines in either untreated or tunicamycin-treated conditions. In the ranked scatter plots for HAP1, RPE1, and HEPG2 cells (Supplemental Fig. S1B), the highlighted genes are the hits that were identified in all of the three cell lines in either untreated or tunicamycin-treated conditions. To identify regulated splicing events in strong XBP1 regulators (Fig. 3A) we considered the hits in all the three cell lines either in untreated or tunicamycin treated conditions.

Cloning ERN1 and XBP1

ERN1 (ENST00000433197) and XBP1 (ENST00000344347) coding sequences were PCR-amplified, purified, and cloned into the pDONR223 entry vector using BP Clonase II Enzyme Mix (Thermo Fisher Scientific 11789020). BP reaction products were transformed into NEB stable competent E. coli cells (New England Biolabs C3040H), and transformants were selected on LB agar containing spectinomycin. The resulting pDONR223-ERN1 entry clone was modified by site-directed mutagenesis using the Q5 site-directed mutagenesis kit (NEB E0552S) to introduce silent mutations conferring resistance to the most effective siRNA (Horizon Discovery ON-TARGETplus J-004951-21; sequence: AUAAUGAAGGCCUGACGAA). Next, the modified ERN1 and XBP1 sequences were transferred from the pDONR223 entry vector into pcDNA5-3xFLAG destination vectors using Gateway LR Clonase II enzyme mix (Thermo Fisher Scientific 11791020). These pcDNA5-3xFLAG vectors encode an N-terminal 3xFLAG tag. LR reaction mixes were transformed into NEB stable competent E. coli cells, and transformants were selected on LB agar plates containing 100 µg/mL ampicillin. Additional site-directed mutagenesis was performed on the siRNA-resistant pDONR-ERN1 and pDONR-XBP1 entry clones to generate truncated and mutant constructs, including ERN1 exon-18 skipping (Δex18), XBP1 spliced, and XBP1 mutant. All plasmids used in this study have been deposited with Addgene (Supplemental Table S9).

Generation of HEK293 Flp-In cell lines expressing ERN1 (and XBP1) splice isoforms

To generate HEK293 Flp-In cell lines expressing control plasmids (EGFP or empty vector), IRE1α isoforms, or XBP1 isoforms, one million cells were transfected with the corresponding pcDNA5 constructs using Lipofectamine 2000 (Thermo Fisher Scientific 11668027) according to the manufacturer's instructions. After 4.5 h, the transfection medium was replaced with fresh growth medium to reduce Lipofectamine-associated toxicity. Two days after transfection, cells were selected in DMEM containing 200 µg/mL hygromycin B (Thermo Fisher Scientific 10687010), and the hygromycin-containing medium was refreshed every 4–5 days to enrich for stably transfected cells. Once cultures reached confluency, doxycycline titrations were performed to optimize expression of FLAG-tagged proteins. Final induction conditions were 0.001 µg/mL doxycycline for FLAG-EGFP and 0.5 µg/mL doxycycline for FLAG-Empty, FLAG-ERN1 and FLAG-XBP1 constructs, which yielded consistent protein expression.

ERN1 rescue experiments

HEK293 Flp-In cells were plated at 700,000 cells per condition and transfected with either a nontargeting control siRNA (siControl) or siERN1 (sequence: AUAAUGAAGGCCUGACGAA; Horizon Discovery ON-TARGETplus J-004951-21) with or without tunicamycin (10 µg/mL final; Sigma SML1287) for 4 h. Cells were collected and frozen 72 h after transfection for downstream applications.

Establishment of monoclonal cell lines with endogenous ERN1 Δexon-18 isoform expression

Adenine base editor (ABE8e) and a gRNA targeting the 5′ splice site of ERN1 exon-18 (Supplemental Table S9) were expressed in HEK293T cells by transfecting the PX461 ABE8e–SpCas9 ERN1 ex18 5′ss plasmid using Lipofectamine 2000 (Thermo Fisher Scientific 11668027). After 2 days, cells were counted with a Vi-Cell BLU cell viability analyzer (Beckman Coulter), diluted, and ∼50 cells were seeded per 96 well plate (∼0.5 cells/well) for clonal isolation. Wells containing single colonies (monoclones) were selected and expanded. ERN1 splicing patterns were assessed by RT-PCR, and clones were classified as wild type, heterozygous, or homozygous.

RT-PCR and qRT-PCR assays

RNA was extracted from frozen cell pellets using the RNeasy Plus universal mini kit (Qiagen 74136) according to the manufacturer's instructions. For most targets, 40 ng of RNA was amplified by one-step RT-PCR using the Qiagen OneStep RT-PCR kit (Qiagen 210215). For the XBP1 reporter, first-strand cDNA was synthesized from 1 µg of RNA with the Maxima H Minus first strand cDNA synthesis kit (Thermo Fisher Scientific K1651) using a U6 primer; the cDNA was then diluted 1:4, and 2 µL of the diluted cDNA was used in a 20 µL PCR with Taq polymerase (Thermo Fisher Scientific 10342020). Primer sequences are provided in Supplemental Table S9. PCR cycling conditions were 28 cycles for the XBP1 reporter, 30 for RBM39, 32 for endogenous XBP1, and 32 or 35 for ERN1.

For real-time quantitative RT-PCR validation, 40 ng of RNA was processed using the SensiFAST SYBR No-ROX one-step kit (Thomas Scientific C755H72) following the manufacturer's protocol. Each sample was run in two to three technical replicates, and outliers were excluded prior to analysis. Primer sequences are listed in Supplemental Table S9.

Immunoblotting

For experiments with HEK293 Flp-In cells and HEK293T base-edited monoclones, pellets were lysed for 10 min on ice in F buffer (10 mM Tris at pH 7.05, 50 mM NaCl, 30 mM Na₄ pyrophosphate, 50 mM NaF, 5 µM ZnCl₂, 10% glycerol, 0.5% Triton X-100) supplemented with a protease inhibitor cocktail (Roche 11836170001). Lysates were centrifuged at 14,000 rcf for 5 min at 4°C, and supernatants were collected. Samples were mixed with NuPAGE LDS sample buffer (Thermo Fisher Scientific NP0007) containing 100 mM DTT (Thermo Fisher Scientific R0861) and heated for 5 min at 70°C. Protein concentrations were measured using Bradford reagent (Bio-Rad 5000006), and equal amounts (10–30 µg) were separated on 4%–12% Bis-Tris gels (Thermo Fisher Scientific NP0323BOX). Proteins were transferred to PVDF membranes, blocked with 5% milk for 1 h at room temperature, and probed with the following primary antibodies: IRE1α (ERN1; 1:1000; Proteintech 27528-1-AP for Flp-In cells and Thermo Fisher MA5-51210 for base-edited monoclones), FLAG M2 (1:2000; Sigma F3165), GAPDH (1:10,000; Proteintech 60004-1-Ig), XBP1 (1:1000; Abcam ab220783), RBM39 (1:1000; Proteintech 67420-1-Ig), PARP1 (1:1000, Proteintech 13371-1-AP), and Caspase3/p17/p19 (1:1000; Proteintech 19677-1-AP). Primary antibodies were diluted in 5% milk. After washing, membranes were incubated with HRP-conjugated secondary antibodies (antirabbit, Cell Signaling Technology 7074; antimouse, Cell Signaling Technology 7076) at 1:10,000 for 1 h at room temperature. Signals were developed using SuperSignal West Pico PLUS (Thermo Fisher Scientific 34580) or SuperSignal West Femto (Thermo Fisher Scientific A45916), and images were acquired on an Invitrogen iBright CL1500 imaging system (Thermo Fisher Scientific A44114).

Splicing factors inhibition

To inhibit the splicing factors RBM39, SF3B1, and SF3B3, we used siRNA-mediated knockdown and small-molecule inhibitors. HAP1 cells (1 × 10⁶) were seeded, and siRNAs targeting RBM39 (Horizon Discovery ON-TARGETplus LQ-011965-00-0005), SF3B1 (LQ-020061-01-0005), and SF3B3 (LQ-020085-00-0005) were transfected at the time of plating using Lipofectamine RNAiMAX (Thermo Fisher Scientific 13778150). Where indicated, cells were treated with indisulam (10 µM final; MedChemExpress HY-13650) to target RBM39 and pladienolide B (PladB; 10 nM final; MedChemExpress HY-16399) to inhibit SF3B1/SF3B3, with or without tunicamycin (10 µg/mL final; Sigma SML1287), for 4 h. After 48 h, cells were harvested for RT-PCR analysis.

In vitro RNA–protein pull-down assay

Twenty nucleotide T7 promoter sequences followed by a 100 nt fragment spanning ERN1 intron-17 and exon-18, either wild-type or mutant (RBM39 recognition motif mutated from TCTCTTTT to TCCCCCCC), were synthesized as single-stranded oligonucleotides. Complementary oligonucleotides were annealed to generate double-stranded DNA templates, and in vitro transcription was performed overnight according to the manufacturer's protocol (Thermo Fisher Scientific A57620). RNA concentration was determined using the Agilent TapeStation system (Agilent 5067-5576 and 5067-5577). A total of 100 pmol of RNA was desthiobiotinylated using the Pierce RNA 3′ end desthiobiotinylation kit (Thermo Fisher Scientific 20163) overnight at 16°C, and labeling efficiency was assessed by dot blot according to the manufacturer's protocol.

For the RNA–protein binding assay, 100 pmol of desthiobiotinylated RNA bait was bound to 50 µL of streptavidin magnetic beads provided in the Pierce magnetic RNA–protein pull-down kit (Thermo Fisher Scientific 20164). For cell lysate preparation, HEK293T cells were lysed in IGEPAL buffer (50 mM Tris-HCl at pH 7.5, 150 mM NaCl, 1% IGEPAL CA-630, 5% glycerol, supplemented with protease inhibitors). Briefly, 1 × 10⁷ cells were lysed in 150 µL of lysis buffer to obtain a protein concentration of ∼7 µg/µL. After incubation for 10 min on ice, lysates were centrifuged at 4°C, and the supernatant was collected for the RNA–protein binding assay. Proteins bound to the RNA bait were eluted in 30 µL of biotin elution buffer and analyzed by immunoblotting. For this immunoblot, the RBM39 antibody was used at 1:1000 (Proteintech 67420-1-Ig), and signal was detected using a highly sensitive ECL substrate (SuperSignal West Atto ultimate sensitivity substrate, Thermo Fisher Scientific A38555).

RBM39 RNA immunoprecipitation

HEK293T cells were washed with room-temperature PBS, trypsinized, collected in DMEM, and centrifuged at 120g for 5 min at room temperature. Cells were resuspended in 1 mL of ice-cold PBS and UV-cross-linked on ice at 150 mJ/cm². Cell number was then counted, and 4 million cells were collected and stored at −80°C. Frozen cell pellets were lysed in 600 µL of Igepal buffer supplemented with 1 µL of murine RNase inhibitor (NEB M0314L). After incubation on ice for 10 min, lysates were centrifuged for 5 min at 4°C, and the supernatant was collected for immunoprecipitation.

For immunoprecipitation, 30 µL of each cleared lysate was reserved as input, and the remaining lysate was incubated for 2 h at 4°C with 3.5 µL of Dynabeads Protein G (Invitrogen 10003D) prebound to 1.5 µg of either control mouse IgG (Diagenode C15400001) or RBM39 antibody (Proteintech 67420-1-Ig). After incubation, beads were washed three times with IGEPAL buffer. Beads were then resuspended in 100 µL of Igepal/SDS buffer (99 µL of Igepal buffer plus 1 µL of 10% SDS; final SDS concentration: 0.1%) and incubated for 20 min at 55°C with agitation in the presence of proteinase K (0.2 mg/mLf inal concentration; Thermo Fisher Scientific 25530049). RNA was purified using TRIzol (Thermo Fisher Scientific 15596018), and the RNA pellet was resuspended in 100 µL of TURBO DNase (Thermo Fisher Scientific AM2238) reaction mix (88 µL of water, 10 µL of 10× TURBO DNase buffer, 1 µL of Turbo DNase, 1 µL of murine RNase inhibitor) and incubated for 20 min at 37°C. RNA was then purified again by TRIzol extraction and eluted in 20 µL of water. Eluted RNA was diluted 100-fold and used for qRT-PCR with an ERN1-targeting primer set. Enrichment was calculated as 2^{(Cqinput − CqIP)}, representing the relative amount of immunoprecipitated ERN1 RNA compared with the input sample.

Cloning of ERN1 exon-18 minigene vectors

A DNA fragment encompassing the exon-18 of ERN1 together with its native flanking intronic sequences was PCR-amplified using primers listed in Supplemental Table S9. The amplified product was cloned into the ApaI and NotI sites of the Exontrap cloning vector pET01 (MoBiTec GmbH; Addgene 250491) using the NEBuilder HiFi DNA Assembly Master Mix (New England Biolabs E2621L). To mutate RBM39 recognition motif (TCTCTTTT to TCCCCCCC), site-directed mutagenesis was performed using the Q5 Site-Directed Mutagenesis Kit (New England Biolabs E0552S; Addgene 250492).

Splicing of ERN1 exon-18 minigene constructs

To assess the role of RBM39 in regulating ERN1 exon-18 splicing, HEK293T cells were initially transfected with 25 nM RBM39-targeting siRNA using Lipofectamine RNAiMax (Thermo Fisher Scientific 13778150) for 48 h, followed by transfection with minigene constructs harboring either the wild-type or mutated RBM39 recognition motif using X-tremeGENE 9 DNA transfection reagent (Sigma-Aldrich 6365809001). Twenty-four hours later, total RNA was extracted, and splicing patterns were evaluated with primers annealing to the flanking 5′ and 3′ exons of the pET01 vector.

The effect of stress on ERN1 and XBP1 splicing

To systematically examine the regulation of endogenous ERN1 and XBP1 splicing under diverse cellular stress conditions, HEK293T cells were exposed to a comprehensive panel of perturbations. These included heat shock (4 h at 42°C), osmotic stress (600 mM D-sorbitol for 2 h; Sigma-Aldrich S1876), ER stress (10 µg/mL tunicamycin for 4 h; Sigma-Aldrich SML1287), oxidative stress (250 µM hydrogen peroxide for 4 h, Sigma-Aldrich H1009), genotoxic stress (5 µM mitoxantrone for 4 h; MedChemExpress HY-13502 or 250 µM Cisplatin for 24 h; Sigma-Aldrich 232120), hypoxia (1% O₂ for 24 h), and hypertonic stress (100 mM NaCl for 4 h).

Heat shock and viability assays

HEK293 Flp-In cells stably expressing control, ERN1, or XBP1 constructs were plated at 2 × 10⁵ cells per well in 12 well plates in the presence of doxycycline (0.5 µg/mL). After 24 h, cells were transferred to 42°C for a 16 h heat shock treatment and then returned to 37°C for 48 h of recovery. Cell viability was assessed with either the Vi-Cell BLU cell viability analyzer (Beckman Coulter) or alamarBlue (Thermo Fisher Scientific DAL1025). For alamarBlue, 1 h after reagent addition, fluorescence was measured on a plate reader using excitation at 570 nm and emission at 585 nm. Relative viability was normalized to time-matched, non-heat-shocked controls.

Gene expression and splicing regulation analyses

We used FastQC and MultiQC tools (Ewels et al. 2016) for sequencing read quality assessments. Cutadapt tool (Martin 2011) was used to trim adapters from the raw FASTQ reads. The trimmed reads were mapped to hg38 human reference genome with gencode v44 primary assembly FASTA sequence and gencode v44 primary assembly comprehensive annotation GTF files using STAR aligner tool (Dobin et al. 2013). SAMtools (Li et al. 2009) was used to merge, sort, and index Binary Alignment (BAM) files. The gene read count results obtained from STAR alignment were further assessed for differential expression using DESeq2 (Love et al. 2014) based statistical analysis on R. FPKM (fragments per kilobase of gene per million reads) values were calculated as follows:

FPKM = [(gene read counts/total sequencing read counts) × 1000000)/gene length in base pairs] × 1000.

Genes with expression |log₂FoldChange| ≥ 1, Padj ≤ 0.05, and FPKM ≥ 1 were determined to be significantly regulated for the RNA-seq screens with spliceosome knockdown and inhibition in HAP1 cells. Genes with expression |log₂FoldChange| ≥ 1, Padj ≤ 0.05, and FPKM ≥ 0.1 were determined to be significantly regulated for the RNA-seq screens with ERN1/ERN1-ΔEx18 rescue in HEK293 Flp-In cells. DESeq2 calculates log₂FoldChange values using the following formula based on normalized read counts:

log₂FoldChange (L2FC) = log₂ [normalized read counts (test)/normalized read counts (control)].

Differential splicing analysis was performed using rMATS-Turbo tool (Wang et al. 2024). Alternative splicing events (ASEs) with ΔPSI ≥ 20, FDR ≤ 0.05, and gene FPKM ≥ 0.1 were determined as significantly regulated in all the RNA-seq screens. These differentially spliced significant events were assessed for overlap across RNA-seq and CRASP-seq screens.

UCSC utility tool wigToBigWig (Kent et al. 2010) was used for conversion of wig files to bigwig binary expression signal files and the genome browser tracks were visualized using UCSC web genome browser as well as IGVTools (desktop version) (Robinson et al. 2011). Sashimi plots were generated with ggsashimi (Garrido-Martín et al. 2018).

Patient sample RNA-seq data analysis

Myelodysplastic syndrome (MDS) patient sample RNA-seq data (GSE128805; Zhang et al. 2019) were downloaded from SRA (Sequence Read Archive) using fasterq-dump module of SRAtoolkit (Sequence Read Archive toolkit). Mapping and gene expression analysis were performed as described above. The splice junction read count files generated from STAR aligner were used to estimate percent spliced in (PSI) values to assess relative distributions of ERN1 exon-18 inclusion across different sample groups; for example, SF3B1 mutant (MUT) grouped together versus SF3B1 wild type (WT). An independent t-test was applied to assess statistical significance between groups.

GSEA/gene ontology analysis and gene network visualization

g:Profiler (Kolberg et al. 2023) was used to perform gene set enrichment analysis (GSEA) with XBP1 CRASP-seq hits with CRASP-seq library protein coding genes serving as background gene set. clusterProfiler (Yu et al. 2012) tool was used to perform gene ontology (GO) analysis with differentially expressed genes (DEGs) obtained from RNA-seq data analysis. The GO biological process (GO: BP) pathways were further analyzed for visualization as bar plots or heat maps using standard python libraries.

Cytoscape software (Shannon et al. 2003) was used to generate network for the GO: BP pathways identified for XBP1 splicing regulators identified from CRASP-seq. The network node size represents gene subset size. The boundaries of the nodes are color coded in three shades of gray to depict levels of statistical significance, with darkest color representing highest significance ≤0.001. The edges of the nodes represent average percent overlap with XBP1 CRASP-seq hits and are color coded with a continuous scale of gray.

Use of generative AI in the writing process

Following generation of the original draft, the authors used ChatGPT to edit the manuscript, correct grammar, and enhance readability. After using this tool, the authors reviewed and further edited the content as needed and take full responsibility for the content of the published article.

Data and materials availability

Generated key plasmids and the CRASP-seq plasmid library have been deposited to Addgene (https://www.addgene.org/browse/article/28263732; see Supplemental Table S9): . CRASP-seq screening data generated by this study have been deposited at GEO and are publicly available (series GSE312839). RNA-seq data sets generated by this study have been deposited at GEO and are publicly available (series GSE312841, series GSE312843, and series GSE312847). Raw data underlying the figures have been deposited in Mendeley Data under doi:10.17632/nc7ffvyfdp.1 (https://data.mendeley.com/datasets/nc7ffvyfdp/1).