SPLiCR-seq: A CRISPR-Based Screening Platform for RNA splicing Identifies Novel Regulators of IRE1α-XBP1 Signaling Under ER Stress

Abstract

RNA splicing is fundamental to cellular function, yet systematic investigation of its complex regulation has been limited by existing methods. Here, we present SPLiCR-seq (SPLicing regulator identification through CRISPR screening), a high-throughput CRISPR screening platform that enables direct measurement of RNA splicing outcomes for pooled genetic perturbations, overcoming limitations of traditional fluorescence-based approaches. Applying SPLiCR-seq to investigate XBP1 splicing during the unfolded protein response (UPR), we conduct targeted and genome-wide screens across diverse cellular contexts, revealing both common and cell-type specific regulators. Notably, we identify GADD34 (PPP1R15A) as a novel modulator of IRE1α-XBP1 signaling, demonstrating that it directly interacts with IRE1α and functions independently of its canonical role in eIF2α dephosphorylation. Pharmacological inhibition of GADD34 using Sephin1 effectively suppressed XBP1 splicing and alleviated CAR-T cell exhaustion in an ex vivo model, leading to enhanced tumor-killing capacity across multiple cancer models. This work not only establishes a powerful new tool for systematically studying RNA splicing regulation but also uncovers a promising therapeutic strategy for improving CAR-T cell immunotherapy through modulation of the IRE1α-XBP1 pathway.

Systematic investigation of RNA splicing regulation remains challenging. Here, authors present SPLiCR-seq, a CRISPR platform enabling direct measurement of splicing outcomes for pooled perturbations. They identify GADD34 as a novel IRE1α-XBP1 regulator, with inhibition enhancing CAR-T cell activity.

Introduction

RNA splicing is an essential process in eukaryotic cells that ensures the precise removal of non-coding introns from precursor mRNA transcripts and the joining of protein-coding exons to produce functional mature RNA¹. This process enables the generation of diverse protein isoforms from a single gene and contributes to complex cellular functions and responses. Dysregulation in RNA splicing is linked to numerous diseases, including cancer, neurological disorders, and immune dysfunction^2–5.

Conventional RNA splicing occurs in the nucleus and is mediated by the spliceosome, a highly dynamic complex of proteins and small nuclear RNAs that recognize specific splicing signals. The spliceosome precisely modulates splicing outcomes based on cell type and environmental cues, ensuring proper gene regulation. In addition to canonical nuclear splicing, RNA molecules can also undergo unconventional splicing in response to specific cellular stress signals. A notable example is the unconventional splicing of X-box binding protein 1 (XBP1) mRNA, which occurs in the cytoplasm during the unfolded protein response (UPR)⁶. The UPR is triggered when misfolded or unfolded proteins accumulate in the endoplasmic reticulum (ER), creating stress on the cell’s protein-folding machinery^7,8. Under these stress conditions, the transmembrane protein IRE1α (inositol-requiring enzyme 1 alpha, encoded by the ERN1 gene), a key UPR sensor, becomes activated and catalyzes the cytoplasmic excision of a 26-nucleotide segment from unspliced XBP1 (XBP1u) mRNA^6,9. This excision induces a frameshift, producing the mature, spliced XBP1 (XBP1s) mRNA, which encodes the transcriptionally active XBP1 protein. This protein functions as a potent transcription factor, upregulating the expression of genes involved in protein folding, ER-associated protein degradation (ERAD), and lipid biosynthesis, thereby enabling cells to restore ER homeostasis¹⁰. In cancer, the IRE1α-XBP1 axis emerges as a critical regulator of tumor progression and immune evasion¹¹. In tumor cells, it supports adaptive survival under stress and metabolic reprogramming, creating an immunosuppressive microenvironment^12–14, while in immune cells, it contributes to T cell dysfunction and exhaustion^15–17. Therapeutic strategies targeting this axis hold significant potential for improving cancer immunotherapy outcomes.

Recent advances in CRISPR-based genetic perturbation technologies have enabled large-scale screens to identify regulators of specific splicing events. However, existing genetic screening approaches for RNA splicing often rely on fluorescent or luminescent reporter systems^18–30, where splicing outcomes are converted into fluorescent or luminescent protein expression. These methods have several limitations. First, they provide an indirect measurement of splicing, which lacks precision and temporal resolution, as the fluorescent signal lags behind actual RNA splicing events. Second, they rely on fluorescence-activated cell sorting (FACS), restricting application in primary cells, organoids, or other contexts where sorting is impractical. These challenges hinder the study of dynamic and context-specific RNA splicing regulation.

To overcome these limitations, we developed SPLiCR-seq (SPLicing regulator identification through CRISPR screening), a CRISPR-based screening platform that directly assesses RNA splicing phenotypes by next-generation sequencing (NGS) for pooled genetic perturbations. By eliminating the need for fluorescent reporters and FACS, SPLiCR-seq provides a more direct, efficient, and precise approach for identifying splicing regulators across diverse biological settings. The simplicity of SPLiCR-seq enables us to conduct multiple large-scale screens to investigate XBP1 splicing during UPR across various cellular contexts, including diverse cell types, ER stressors, and treatment timepoints. These screens reveal numerous known and novel regulators of XBP1 splicing under different conditions. Among our key findings, we discovered that GADD34 directly interacts with IRE1α to modulate XBP1 splicing under ER stress. Furthermore, we demonstrate that pharmacological inhibition of GADD34 reduces CAR-T cell exhaustion and enhances their tumor-killing capacity, highlighting the therapeutic potential of targeting GADD34 in cancer immunotherapy.

Results

Development and validation of the SPLiCR-seq vector

The SPLiCR-seq platform was designed to link sgRNA-induced genetic perturbations with direct RNA splicing readouts via NGS. To achieve this, we engineered the SPLiCR-seq vector based on a previously published CROP-seq vector (pMK1334)^31,32. The SPLiCR-seq vector includes a U6-driven sgRNA expression cassette for CRISPR-based gene perturbation and a splicing reporter designed to monitor the splicing level of a specific event of interest. The splicing reporter is co-transcribed with the U6-sgRNA sequence under the control of an EF1α promoter. Using paired-end NGS, the sgRNA identity is captured from one read and the splicing phenotype of the reporter is captured from the other, enabling precise mapping of genetic perturbations to their impact on RNA splicing (Fig. 1A).

Fig. 1. Development and Validation of the SPLiCR-seq Vector.

A Schematic of the SPLiCR-seq platform. The SPLiCR-seq vector enables the co-transcription of a splicing reporter and sgRNA driven by the EF1α promoter, while sgRNA expression for gene perturbation is driven by the U6 promoter. Paired-end sequencing links the splicing phenotype of the reporter to the corresponding sgRNA identity. B Diagram of the IRE1α/XBP1 pathway. Under ER stress, IRE1α’s RNase domain cleaves unspliced XBP1 mRNA (XBP1u), removing a 26-nucleotide intron. The resulting spliced XBP1 mRNA (XBP1s) is translated into a potent transcription factor that induces the expression of ER quality control genes to mitigate stress. C Design of the SPLiCR-seq vector for XBP1 splicing. The reporter contains the 26-nucleotide intron and 106-nucleotide 5’ and 32-nucleotide 3’ flanking sequences from XBP1u. D, E Dose-dependent splicing of endogenous XBP1 and the XBP1 reporter in the SPLiCR-seq vector in HEK293T cells treated with increasing concentrations of either Tg for 3 h (D) or Tm for 6 h (E), as assessed by RT-PCR. Spliced and unspliced products are indicated. F Time-course analysis of XBP1 splicing dynamics for endogenous XBP1 and the XBP1 reporter in the SPLiCR-seq vector in HEK293T cells treated with Tg (100 nM or 500 nM) for the indicated durations, as assessed by RT-PCR. Spliced and unspliced products are indicated. G, H Quantification of time-course splicing as the ratio of spliced to unspliced XBP1 (s/u) for endogenous XBP1 (G) and the XBP1 reporter (H), determined by grayscale analysis of RT-PCR bands. I Representative RT-PCR analysis showing splicing of endogenous XBP1 and XBP1 reporter in SPLiCR-seq vectors expressing a control, ERN1 or EIF2AK3 sgRNA in CRISPRi-HEK293T cells. J, K Quantification of endogenous XBP1 splicing (J) and XBP1 reporter splicing (K) from (I) (n = 2 biological replicates). Source data are provided as a Source Data file.

To demonstrate the utility of SPLiCR-seq, we applied this platform to study the unconventional splicing of XBP1 during UPR (Fig. 1B). For this purpose, we incorporated an XBP1 splicing reporter into the SPLiCR-seq vector. This reporter contains the 26-nucleotide intron and flanking sequences from unspliced XBP1 (XBP1u) mRNA, which undergo IRE1α-mediated cleavage under ER stress conditions (Fig. 1B, C).

We first validated whether the XBP1 splicing reporter in the SPLiCR-seq vector could accurately capture the splicing dynamics of endogenous XBP1 under ER stress. Cells expressing the reporter were treated with increasing concentrations of the ER stress inducers thapsigargin (Tg) or tunicamycin (Tm)^33–35. We found that the splicing of the reporter closely mirrored the dose-dependent splicing of endogenous XBP1 mRNA, with both showing increased splicing at higher concentrations of Tg (Fig. 1D) and Tm (Fig. 1E). Similarly, time-course experiments demonstrated that the reporter splicing followed the same temporal dynamics as endogenous XBP1, with splicing levels increasing with treatment time from 0 to 8 h and subsequently declining by 24 h, consistent with previously reported XBP1 splicing patterns⁹ (Fig. 1F–H).

Next, we evaluated whether the SPLiCR-seq vector could reliably detect the effects of genetic perturbations on XBP1 splicing. To this end, we cloned sgRNAs targeting two well-characterized regulators of the IRE1α-XBP1 pathway into the SPLiCR-seq vector: ERN1, which encodes the enzyme IRE1α responsible for XBP1 splicing, and EIF2AK3, which encodes PERK, a kinase that mediates a parallel UPR pathway whose inhibition is known to enhance IRE1α-XBP1 signaling^36,37. The vectors were transduced into a HEK293T cell line (CRISPRi-HEK293T) engineered to stably express the CRISPR interference machinery (dCas9-BFP-KRAB) for gene knockdown. As expected, ERN1 knockdown significantly reduced the splicing of both endogenous XBP1 and the reporter, while EIF2AK3 knockdown showed the opposite effect (Fig. 1I–K). These results demonstrate that the SPLiCR-seq vector accurately reflects endogenous splicing dynamics and can reliably monitor the effects of genetic perturbations on splicing.

RBP-focused SPLiCR-seq screens identify cell type-specific regulators of XBP1 splicing

Next, we performed large-scale SPLiCR-seq screens to identify regulators of XBP1 splicing in two distinct cellular contexts: HEK293T cells, representing transformed cells, and induced pluripotent stem cells (iPSCs), representing normal cells with a normal karyotype. We constructed an SPLiCR-seq library that contains sgRNAs targeting all human RNA-binding proteins (RBPs) and transduced the library via lentivirus into HEK293T cells (CRISPRi-HEK293T) and iPSCs (CRISPRi-iPSC) that stably express CRISPRi machinery (dCas9-BFP-KRAB) from the CLYBL safe harbor locus. Following puromycin selection and cell expansion, we induced ER stress in these cells with Tg treatment for 3 h (Fig. 2A). Subsequently, RNA was extracted and reverse transcribed, followed by PCR amplification of regions containing the splicing reporter and sgRNA sequences to generate NGS libraries for paired-end sequencing. From each read pair, Read2 identified the sgRNA while Read1 revealed the splicing outcome of the XBP1 reporter. Splicing phenotypes for each sgRNA were calculated as a ratio of spliced to unspliced read counts, normalized to non-targeting controls (Methods, Fig. 2A, Supplementary Fig. 1A–F).

Fig. 2. RBP-focused SPLiCR screens identify common and cell-type specific regulators of IRE1α-XBP1 Signaling.

A Schematic illustration of the SPLiCR screen workflow. HEK293T cells and iPSCs expressing the CRISPRi machinery were transduced with an sgRNA library targeting 1,350 human RNA-binding proteins (MOI < 0.3). After puromycin selection and cell expansion, cells were treated with Tg for 3 h to induce ER stress. RNA was extracted, reverse-transcribed, and regions containing the XBP1 splicing reporter and sgRNA were PCR-amplified to generate paired-end NGS libraries. For each sgRNA, splicing phenotypes were calculated as the ratio of spliced reads (S_i) to unspliced reads (U_i), and normalized to the median control sgRNA levels. The screen was performed in two biological replicates. B, C Volcano plots showing results for SPLiCR-seq screens in HEK293T cells (B) and iPSC (C). log₂ fold change (LFC) and P values were calculated using MAGeCK⁷⁵. Representative hits are labeled. D Comparison of gene knockdown phenotypes from HEK293T and iPSC SPLiCR-seq screens, revealing both common and cell-type-specific regulators of XBP1 splicing. E, F Distributions of phenotype scores for non-targeting control sgRNAs (gray) and sgRNAs targeting selected negative (blue) and positive (red) hits in HEK293T (E) and iPSC (F) SPLiCR-seq screens. The shaded area represents the 10th–90th percentile range of the LFC distribution for control sgRNAs. G Representative RT-PCR analysis showing XBP1 splicing in HEK293T cells following CALR or UPF3B knockdown with two independent sgRNAs. XBP1 splicing levels are indicated at the bottom (mean ± SD, n = 3 biological replicates). Student’s t-test: *p < 0.05, **p < 0.01. H Western blot showing protein levels of XBP1s, IRE1α and phosphorylated IRE1α (p-IRE1α) in HEK293T cells under ER stress conditions following knockdown of CALR or UPF3B. GAPDH was used as a loading control. I Co-immunoprecipitation showing a direct physical interaction between CALR and IRE1α. J, K Representative RT-PCR analysis showing XBP1 splicing under ER stress following DBR1 knockdown in HEK293T cells (J) and iPSCs (K), with splicing levels indicated at the bottom (mean ± SD, n = 3 biological replicates). Student’s t-test: **p < 0.01; ns, not significant. Source data are provided as a Source Data file.

These screens identified genes whose knockdown either enhanced (positive hits) or suppressed (negative hits) XBP1 splicing in both cell types (Fig. 2B, C, Supplementary Data 1). As anticipated, ERN1 emerged as the top negative hit in both screens (Fig. 2B, C, Supplementary Fig. 1C, F), validating our screening approach.

The screens revealed both shared and cell type-specific hits (Fig. 2D). For instance, CALR, which encodes calreticulin-an ER chaperone involved in protein folding and calcium homeostasis that was recently identified as an ATF6 repressor³⁰-showed a similar negative knockdown phenotype in both cell types. Likewise, UPF3B, a component of the nonsense-mediated mRNA decay (NMD) pathway recently shown to interact with IRE1α³⁸, showed a similar positive knockdown phenotypes in both cell types. In contrast, DBR1, which encodes an RNA debranching enzyme essential for intronic lariat degradation during splicing^39,40, exhibited a strong negative phenotype in iPSCs but not in HEK293T cells (Fig. 2D–F).

We next performed validation experiments for these three genes to confirm the screen results. In HEK293T cells, knockdown of UPF3B significantly enhanced endogenous XBP1 splicing (Fig. 2G, Supplementary Fig. 1G), as well as XBP1s protein levels (Fig. 2H) under ER stress, while knockdown of CALR showed the opposite effect (Fig. 2G, H). Further investigation revealed that these effects were mediated through modulation of IRE1α activation, with UPF3B knockdown increasing and CALR knockdown decreasing IRE1α phosphorylation (Fig. 2H). In addition, CALR directly interacts with IRE1α, as shown by co-immunoprecipitation (co-IP) (Fig. 2I), suggesting a direct regulatory mechanism.

Consistent with the screen results, DBR1 knockdown exhibited cell-type-specific effects, reducing XBP1 splicing in iPSCs but not in HEK293T cells (Fig. 2J, K), highlighting a cell-type-specific role for DBR1 in modulating IRE1α-XBP1 signaling. These results demonstrate the power of SPLiCR-seq to identify both common and cell-type-specific regulators of RNA splicing across diverse cellular contexts.

Genome-wide SPLiCR-seq screens identify regulators of IRE1α-XBP1 signaling under different ER stress conditions

Building on the success of the RBP-focused SPLiCR-seq screens, we expanded the platform to genome-wide screens to comprehensively identify regulators of IRE1α-XBP1 signaling under different ER stress conditions. To this end, we constructed a genome-wide CRISPRi library containing more than 22,000 sgRNAs targeting over 11,000 genes into the XBP1 SPLiCR-seq vector. The screens were conducted following the same workflow as the RBP screen: CRISPRi-HEK293T cells were transduced with the library, treated with either Tg or Tm for 24 h to induce ER stress, and RNA was extracted. SPLiCR libraries were then prepared for paired-end NGS to determine sgRNA-associated splicing phenotypes (Fig. 3A, Supplementary Fig. 2A–I; Supplementary Data 1).

Fig. 3. Genome-Wide SPLiCR-Seq Screens Identify Regulators of IRE1α-XBP1 Signaling Under Different ER Stress Conditions.

A Schematic of the genome-wide SPLiCR-seq screen workflow. CRISPRi-HEK293T cells were transduced with a genome-wide sgRNA library targeting 11,120 genes that are expressed in HEK293T cells (MOI < 0.3). Following puromycin selection and cell expansion, cells were treated with either Tg or Tm for 24 h to induce ER stress. RNA was extracted and SPLiCR libraries were prepared for NGS. B, C Volcano plots showing gene knockdown phenotypes from genome-wide screens in HEK293T cells treated with Tg (B) or Tm (C). LFC and P values were calculated using MAGeCK⁷⁵. Key known regulators of the IRE1α-XBP1 pathway are labeled, with PPP1R15A, a novel regulator investigated in this study, highlighted in bold. D Comparison of gene knockdown phenotypes between Tg and Tm screens, showing a strong overlap in common regulators of XBP1 splicing. E, F Functional categorization of hits from the genome-wide screens under Tg (E) and Tm (F) treatment. Red circles, positive hits; blue circles, negative hits. Genes involved in protein homeostasis and ER function were identified in both screens, whereas ER–Golgi trafficking genes were unique to the Tg screen. G GO Biological Process enrichment analysis of the top 100 positive hits (red) and top 100 negative hits (blue) for the Tg (left) and Tm (right) screens. P values were calculated using the Fisher’s Exact test. Source data are provided as a Source Data file.

As expected, ERN1 emerged as the top negative hit in both the Tg and Tm screens (Fig. 3B, C, Supplementary Data 1). We also recovered many known regulators of the IRE1α-XBP1 pathway and broader UPR, including negative hits such as VMP1⁴¹, RTCB⁴², and PPP1R15A⁴³, and positive hits such as UPF1^38,44, UPF2^38,44, DERL1⁴⁵, DERL2⁴⁵, ATF6⁶, FAF2⁴⁶, and EIF2AK3^36,37 (Fig. 3B, C, Supplementary Data 1). Furthermore, most strong hits from the previous RBP screen recapitulated their phenotypes in the genome-wide screen (5 out of 7 with an FDR < 0.1 and |lfc | > 0.5; Supplementary Fig. 2J).

A comparison of the Tg and Tm screens revealed significant overlap (Fig. 3D, Supplementary Fig. 2K): among the top 100 positive hits, 36 were shared between screens (Fisher’s exact P = 4.20e − 50), while 14 of the top 100 negative hits overlapped (Fisher’s exact P = 2.58e − 13). Both screens identified regulators enriched in pathways related to protein homeostasis and ER function, including protein translation and degradation, UPR, ERAD, ER chaperones, and protein translocation into the ER (Fig. 3E–G). Notably, a substantial fraction of shared hits overlapped with findings from a previous FACS-based genome-wide screen of IRE1α-XBP1 signaling⁴⁷ (12 of 36 positive hits and 5 of 14 negative hits; Supplementary Fig. 2K). Together, these findings demonstrate the reliability and robustness of our SPLiCR-seq screens.

Importantly, our screens also identified stressor-specific regulators, indicating distinct mechanisms modulating IRE1α-XBP1 signaling based on the type of ER stress. For instance, genes involved in COPII vesicle-mediated ER-to-Golgi trafficking, including SEC24A, PREB, SAR1A, and SEC31A, were significantly enriched among the top negative hits in the Tg screen but not the Tm screen (Fig. 3E–G). In contrast, pathways related to post-translational modifications including ubiquitination and phosphorylation were uniquely enriched in the Tm screen (Fig. 3G), underscoring the distinct cellular responses to these two stressors.

These genome-wide SPLiCR-seq screens provide a comprehensive catalog of regulators involved in IRE1α-XBP1 signaling and UPR, uncovering both conserved and stressor-specific regulatory mechanisms. Together, these results establish SPLiCR-seq as a robust and scalable platform for probing RNA splicing.

Batch validation screens identify PPP1R15A (GADD34) as a novel regulator of IRE1α-XBP1 signaling

To validate hits from our genome-wide screens, we conducted a secondary batch validation screen using a targeted library containing sgRNAs against 70 selected hits, following the same workflow as the primary screen (Fig. 4A). The smaller library size enabled us to conduct multiple parallel screens; considering the dynamic regulation of XBP1 splicing during the UPR (Fig. 1F)⁹, three treatment durations, 1, 8, and 24 h, were tested in the validation screen.

Fig. 4. Validation Screens Identify PPP1R15A (GADD34) as a Novel Regulator of IRE1α-XBP1 Signaling.

A Schematic of the batch validation SPLiCR-seq screen workflow. B Heatmap showing gene knockdown phenotypes from the validation screens (top), compared to the primary screens (bottom). Genes are hierarchical clustered based on their validation screen phenotypes. Genes selected for individual validation are highlighted in blue (negative hits) and red (positive hits). C RT-PCR analysis of XBP1 splicing in HEK293T cells under Tg-induced ER stress following knockdown of PPP1R15A. Splicing levels are indicated below each lane. Three independent sgRNAs were tested. D RT-PCR analysis of XBP1 splicing in control and PPP1R15A knockdown HEK293T cells expressing either empty vector or GADD34 vector. Splicing levels are indicated below each lane. E Western blot analysis of p-IRE1α, total IRE1α, and XBP1s protein levels in control and PPP1R15A knockdown cells. GAPDH was used as a loading control. Three independent sgRNAs were tested for PPP1R15A. F RT-PCR analysis of XBP1 splicing in HEK293T cells treated with Sephin1. Splicing levels are indicated below each lane (mean ± SD, n = 3 experimental replicates). Student’s t-test: **p < 0.01, ***p < 0.001. Source data are provided as a Source Data file.

Most genes in the validation screen exhibited comparable knockdown phenotypes across different stress durations and were consistent with their effects observed in the primary screens, confirming the reliability of our findings from the genome-wide screens (Fig. 4B, Supplementary Data 1).

From these high-confidence hits, we selected five genes for individual validation of their effects on endogenous XBP1 splicing in UPR. These included the positive hit SLC35B1, which encodes a nucleotide sugar transporter required for protein glycosylation^48,49, and negative hits MAN2A2, SEC24A, PREB, and PPP1R15A. Among these, MAN2A2 regulates N-glycan processing⁵⁰, SEC24A and PREB are involved in the COPII vesicle-mediated ER-Golgi trafficking pathway^51,52, and PPP1R15A encodes GADD34, known to promote ER stress recovery through eIF2α dephosphorylation^43,53,54.

Consistent with the screen results, knockdown of SLC35B1 dramatically enhanced endogenous XBP1 splicing at all Tg treatmemt time-1, 8 and 24 h, suggesting enhanced IRE1α-XBP1 signaling (Supplementary Fig. 3G–I). Conversely, knockdown of MAN2A2, SEC24A, PREB, PPP1R15A significantly suppressed XBP1 splicing (Fig. 4C, Supplementary Fig. 3A–F, J–L, Supplementary Fig. 4A–C).

Given its strong phenotypes in both Tm and Tg screens and previously uncharacterized role in IRE1α-XBP1 signaling, we further investigated PPP1R15A (GADD34). Overexpression of GADD34 in PPP1R15A (GADD34) knockdown cells rescued the XBP1 splicing phenotype, while overexpression in wild-type cells further enhanced XBP1 splicing (Fig. 4D, Supplementary Fig. 4D). At the protein level, western blot analysis confirmed that both knockdown and knockout of PPP1R15A significantly reduced XBP1s protein levels under ER stress (Fig. 4E, Supplementary Fig. 4E–K). Additionally, both knockdown and knockout of PPP1R15A suppressed IRE1α activation, as evidenced by reduced phosphorylated IRE1α (p-IRE1α) levels (Fig. 4E, Supplementary Fig. 4H, K). These results suggest that GADD34 regulates XBP1 splicing by modulating IRE1α activation.

To complement our genetic approaches, we examined the effects of pharmacological GADD34 inhibition using Sephin1, a small-molecule inhibitor of GADD34⁵⁵. Similar to genetic knockdown, Sephin1 treatment potently suppressed XBP1 splicing in a dose-dependent manner and across different ER stress treatment durations (Fig. 4F, Supplementary Fig. 4L).

Together, these results confirm the discovery of novel IRE1α-XBP1 signaling regulators from the genome-wide screens and establish GADD34 as a previously unrecognized modulator of this pathway.

GADD34 regulates IRE1α-XBP1 signaling independently of eIF2α dephosphorylation

GADD34 has previously been proposed to function as a negative feedback regulator of the UPR by recruiting Protein Phosphatase 1 (PP1) to dephosphorylate eIF2α^56–58, which is phosphorylated by PERK during ER stress. However, our findings reveal that GADD34 regulates IRE1α-XBP1 signaling through a mechanism independent of eIF2α dephosphorylation, supported by several lines of evidence:

First, we found no increase in eIF2α phosphorylation levels in HEK293T cells subjected to either genetic or pharmacological inhibition of GADD34 under Tg- or Tm-induced ER stress conditions (Fig. 5A, B, Supplementary Fig. 4H, Supplementary Fig. 4K). Moreover, the ability of GADD34 to regulate eIF2α phosphorylation was dependent on its expression levels: notable dephosphorylation of eIF2α was only observed with massive GADD34 overexpression achieved through transient transfection, while moderate overexpression via lentiviral infection produced no notable changes in eIF2α phosphorylation (Fig. 5C).

Fig. 5. GADD34 Regulates IRE1α-XBP1 Signaling Through an eIF2α-Independent Mechanism.

A Western blot analysis of PERK, phosphorylated eIF2α (p-eIF2α) and total eIF2α levels in control and PPP1R15A knockdown HEK293T cells treated with Tg (500 nM) for the indicated durations (0, 6, or 24 h). GAPDH was used as a loading control. Three independent sgRNAs were tested for PPP1R15A. B Western blot analysis of PERK, p-eIF2α, total eIF2α, p-IRE1α, total IRE1α, and XBP1s in HEK293T cells treated with increasing concentrations of Sephin1 (0, 1, or 5 μM) in the presence or absence of Tm (2 μM, 24 h). GAPDH was used as a loading control. C Western blot analysis of p-eIF2α levels in cells with GADD34 overexpression via transient transfection (left) or lentiviral infection (right). GAPDH was used as a loading control. D, E RT-PCR analysis of XBP1 splicing in control and EIF2S1 (eIF2α) knockdown cells (D), and EIF2S1 partial knockout cells (E) treated with Tm (2 μM, 24 h) and Sephin1 (5 μM, 24 h) as indicated. Splicing ratios are indicated below each lane (mean ± SD, n = 3 biological replicates). Student’s t-test: *P < 0.05, **P < 0.01, ***P < 0.001. Two independent sgRNAs were tested for EIF2S1 knockdown, and three monoclonal lines were analyzed for EIF2S1 knockout. F Representative western blot analysis of p-eIF2α, total eIF2α, p-IRE1α, total IRE1α, and XBP1s in control and EIF2S1 knockout HEK293T cells treated with Tm (2 μM, 24 h) and Sephin1 (5 μM, 24 h) as indicated. GAPDH was used as a loading control. Three independent monoclonal lines were tested. G RT-PCR analysis of XBP1 splicing in HEK293T cells treated with Tm (2 μM, 3 h), Sephin1 (5 μM, 3 h), and ISRIB (1 μM, 3 h) as indicated. Splicing levels are indicated below each lane (mean ± SD, n = 3 biological replicates). Student’s t-test: **P < 0.01, ****P < 0.0001. H Representative western blot analysis of p-eIF2α, total eIF2α, p-IRE1α, total IRE1α and XBP1s in HEK293T cells treated with Tm (2 μM, 3 h), Sephin1 (5 μM, 3 h), and ISRIB (1 μM, 3 h) as indicated. GAPDH was used as a loading control. Quantifications of p-eIF2α and XBP1s levels are shown (mean ± SD, n = 3 biological replicates). Student’s t-test: *P < 0.05, **P < 0.01. Source data are provided as a Source Data file.

Second, we generated eIF2α knockdown and partial knockout HEK293T cell lines (complete knockout of eIF2α was found to be lethal) (Supplementary Fig. 5A–C). Notably, GADD34 inhibition by Sephin1 in these cells remained effective in suppressing IRE1α-XBP1 signaling in these eIF2α-depleted cells, as evidenced by reduced XBP1 splicing, XBP1s expression and IRE1α phosphorylation (Fig. 5D–F, Supplementary Fig. 5D).

Finally, treatment with ISRIB, a small-molecule activator of eIF2B that reverses the translational inhibition caused by eIF2α phosphorylation⁵⁹, failed to prevent Sephin1’s inhibitory effect on XBP1 splicing during ER stress (Fig. 5G, H).

These results collectively demonstrate that GADD34 regulates IRE1α-XBP1 signaling independently of its known function in dephosphorylating eIF2α.

GADD34 regulates IRE1α-XBP1 signaling through direct interaction with IRE1

To investigate how GADD34 regulates IRE1α-XBP1 signaling, we examined whether GADD34 directly interacts with IRE1α. Co-IP experiments revealed that full-length GADD34 (GADD34 FL) physically interacts with IRE1α under both basal and ER stress conditions (Fig. 6A, B).

Fig. 6. GADD34 Directly Interacts with IRE1α.

A Schematic representation of full-length GADD34 (GADD34 FL) and its truncation mutants analyzed in the study. The ER-targeting domain (residues 1-60), PEST motifs (residues 323-513), and PP1-binding domain (residues 513-674) are indicated. GFP was fused to the N-terminus of each construct. B, C Co-IP analysis of the interaction between IRE1α and full-length GADD34 (B) or truncated GADD34 variants (C). D Immunofluorescence showing subcellular localization of GADD34 variants. COS-7 cells expressing GFP-vector or indicated GFP-GADD34 constructs (green) were treated with or without Tg (2 μM, 3 h) and immunostained for the ER marker calnexin (CANX, red). Nuclei were counterstained with DAPI (blue). Scale bar = 10 μm. E RT-PCR analysis of XBP1 splicing in control and PPP1R15A knockdown cells expressing full-length or truncated GADD34 constructs. Splicing levels are indicated below each lane (mean ± SD, n = 3 biological replicates). Student’s t-test: *p < 0.05; ns, not significant. Source data are provided as a Source Data file.

GADD34 contains a putative ER-targeting domain at its N-terminus, four PEST repeats, and a PP1-binding domain at the C-terminus⁶⁰(Fig. 6A). To determine which domains are crucial for IRE1α binding, we generated several GADD34 truncation mutants (Fig. 6A): a large N-terminal deletion (Δ1–323), a smaller N-terminal deletion that removes the putative ER-targeting domain (Δ1–60), and a C-terminal deletion that removes the PP1-binding domain (Δ513-674). Co-IP experiments revealed that deletion of the ER-targeting domain (GADD34 Δ1-60) or the PP1-binding domain (GADD34 Δ513-674) greatly reduced the GADD34-IRE1α interaction, while the large N-terminal deletion (GADD34 Δ1-323) completely abolished the interaction (Fig. 6C). These results suggest that the N-terminal region, specifically the ER-targeting domain, and the PP1-binding domain play a critical role in mediating the interaction between GADD34 and IRE1α. Further work is needed to delineate the precise interaction interfaces and the key residues involved.

Confocal microscopy analysis showed that full-length GADD34 predominantly localizes to the ER, as evidenced by co-localization with the ER marker calnexin (CANX), under both basal and Tg-induced ER stress conditions (Fig. 6D). In contrast, both N-terminal truncations (GADD34 Δ1-323 and GADD34 Δ1-60) disrupted this ER localization and predominantly accumulated in the nucleus. Notably, cells expressing these N-terminal truncations displayed abnormal ER morphology. Deletion of the PP1-binding domain (GADD34 Δ513-674) did not affect GADD34’s ER localization (Fig. 6D).

We next assessed whether these structural domains affect GADD34’s regulatory function on IRE1α-mediated XBP1 splicing (Fig. 6E). Overexpression of full-length GADD34 rescued the XBP1 splicing defect observed in GADD34-knockdown cells. In contrast, the PP1-binding domain deletion mutant (GADD34 Δ513-674) failed to restore normal XBP1 splicing, indicating that the PP1-binding domain is essential for GADD34’s regulation of IRE1α-XBP1 signaling.

Interestingly, overexpression of either N-terminal truncation mutant (GADD34 Δ1-323 or Δ1-60) resulted in enhanced XBP1 splicing. However, since these truncations also caused abnormal ER morphology (Fig. 6D), their effect on XBP1 splicing may be indirect and mediated through perturbations in ER homeostasis, rather than through direct regulation of IRE1α activity. This possibility warrants further investigation.

Collectively, these results demonstrate that GADD34 directly interacts with IRE1α and that both its N-terminal ER-targeting domain and C-terminal PP1-binding domain contribute to this interaction, with the PP1-binding domain being particularly crucial for GADD34’s regulatory effect on IRE1α-XBP1 signaling.

GADD34 inhibition by sephin1 mitigates CAR-T cell exhaustion and enhances tumor killing

Previous studies have implicated ER stress and XBP1 splicing in driving T cell functional exhaustion and tumor immunoevasion^15,16,61. Hence, we speculated that GADD34 inhibition may delay or reverse T cell exhaustion, leading to enhanced anti-tumor activity. To model T cell exhaustion in vitro, we leveraged a serial co-culture assay where human chimeric antigen receptor T cells (CAR-T cells) were repetitively stimulated with target tumor cells. CAR-T cells exhibit gradual loss of effector functions and reduction in anti-tumor capacity upon each round of co-culture, indicative of functional exhaustion⁶².

To test our hypothesis, B7-H3-specific CAR-T cells were co-cultured with pancreatic cell line BXPC3 in the presence of DMSO or GADD34 inhibitor Sephin1 (Fig. 7A). We observed a significant increase in XBP1 splicing in CAR-T cells after 3 rounds of stimulation (D9), which was effectively suppressed upon Sephin1 addition (Fig. 7B, C). Importantly, CAR-T cells treated with Sephin1 maintained superior effector functions, as indicated by sustained IFN-γ and TNF-α secretion, especially after the second and third rounds of co-culture (Fig. 7D–G).

Fig. 7. Sephin1 reduces XBP1 splicing in ex-vivo exhaustion model thereby alleviating CAR-T cell exhaustion.

A Schematic diagram of the ex-vivo CAR-T cell exhaustion model. B Representative agarose gel showing RT-PCR products of endogenous XBP1 splicing in CAR-T cells repetitively co-cultured with tumor cells for 3, 6 and 9 days in the presence of DMSO or Sephin1. C XBP1 splicing level in CAR-T cells repetitively co-cultured with tumor cells for 3, 6, and 9 days in the presence of DMSO or Sephin1 and normalized to that of the DMSO treatment on day 0 and presented as fold change. Data are presented as mean ± SD, paired Student’s t-test; ***P < 0.0001, n = 4 technical replicates). D Representative flow cytometry plots showing intracellular cytokine staining of IFN-γ and TNF-α in CAR-T cells repetitively co-cultured with tumor cells for 3, 6 and 9 days in the presence of DMSO or Sephin1. E Quantification of the percentage of IFN-γ + /TNF-α + (top panel) and IFN-γ-/TNF-α- (bottom panel) double positive/negative populations in CAR-T cells at day 9. *P < 0.05, **P < 0.01, n = 5 experiments; paired Student’s t-test. F Representative histograms showing expression of IFN-γ (top panel) and TNF-α (bottom panel) in CAR-T cells at day 9, with mean fluorescent intensity (MFI) values indicated. G Quantification of mean fluorescent intensity (MFI) of IFN-γ (top panel) and TNF-α (bottom panel) in cells stained positive for respective cytokines at day 9. *P < 0.05, **P < 0.01, n = 5 experiments; paired Student’s t-test. H Cell counts of CAR-T cells and BXPC3 cells at day 3, 6, 9 upon treatment of DMSO or Sephin1 (Data are presented as mean ± SEM, paired Student’s t-test; *P < 0.05, **P < 0.01, n = 5 experiments). Source data are provided as a Source Data file.

Furthermore, Sephin1 treatment enhanced both the expansion and anti-tumor activity of CAR-T cells, as evidenced by increased numbers of CAR-T cells and reduced residual tumor cells after each round of co-culture (Fig. 7H). To rule out potential direct effects of Sephin1 on cell viability, we confirmed that addition of Sephin1 to tumor or CAR-T cells alone does not alter their cell numbers (Supplementary Fig. 6A, B).

Notably, the protective effect of Sephin1 was not limited to the B7-H3 CAR-T/BXPC3 model. We observed similar improvements when testing CD19-specific CAR-T cells co-cultured with breast cancer cell line HCC1806 expressing the CD19 antigen (Supplementary Fig. 6C–G), suggesting that the beneficial effects of GADD34 inhibition on CAR-T cell function are not limited to specific antigens or tumor models.

Collectively, these findings demonstrate that GADD34 inhibition by Sephin1 suppresses XBP1 splicing and alleviates CAR-T cell exhaustion, leading to enhanced CAR-T cell expansion, sustained effector function, and improved tumor killing. These results highlight the therapeutic potential of GADD34 inhibition for improving CAR-T tumor immunotherapy.

Discussion

In this study, we developed SPLiCR-seq, a novel CRISPR-based screening platform for studying RNA splicing regulation (Fig. 1A). Unlike existing screening methods that rely on indirect fluorescence-based readouts and FACS-based sorting^18–30, SPLiCR-seq links genetic perturbations to direct readout of RNA splicing phenotypes via NGS. This approach provides several key advantages: it enables high-resolution analysis of splicing dynamics with improved temporal precision, eliminates the lag between splicing events and readout signals inherent to fluorescent reporters, and extends applicability to diverse biological contexts where FACS-based sorting is impractical. Beyond RNA splicing, similar RNA-linked CRISPR screening strategies have recently been employed to explore other aspects of RNA biology, including RNA dynamics^63,64, modification^65,66, translation⁶⁴ and alternative polyadenylation⁶⁷.

The power of SPLiCR-seq was demonstrated through our targeted and genome-wide screens, which revealed both conserved and context-specific regulators of XBP1 splicing under ER stress. Our RBP-focused screens across different cell types uncovered cell type-specific regulation of splicing, exemplified by DBR1, which showed strong phenotypes specifically in iPSCs but not in HEK293T cells (Fig. 2J, K). This finding highlights the importance of cellular context in splicing regulation and demonstrates SPLiCR-seq’s utility in uncovering context-dependent mechanisms. The genome-wide screens further expanded our understanding of IRE1α-XBP1 regulation, identifying both conserved regulators across different ER stressors and stressor-specific modulators. For instance, genes involved in COPII vesicle-mediated ER-to-Golgi trafficking were specifically enriched among negative regulators under Tg treatment but not Tm (Fig. 3G), revealing distinct regulatory mechanisms engaged by different types of ER stress.

Our screens yielded comprehensive catalogs of XBP1 splicing regulators across diverse contexts, providing valuable insights into both IRE1α-XBP1 signaling and broader UPR regulation. Among the identified regulators, PPP1R15A (GADD34) emerged as a particularly intriguing hit due to its strong and consistent phenotypes across multiple conditions. GADD34 has been previously established as a negative feedback regulator of the unfolded protein response (UPR)⁶⁸, functioning to dephosphorylate eIF2α by recruiting protein phosphatase 1 (PP1). The small molecule inhibitor Sephin1 (also known as Icerguastat) was first reported by Das et al.⁵⁵. as a selective inhibitor of GADD34 that disrupts the GADD34-PP1 interaction. Sephin1 has demonstrated therapeutic potential in several diseases, including Charcot-Marie-Tooth disease type 1B⁵⁵, amyotrophic lateral sclerosis (ALS)⁵⁵, multiple sclerosis⁶⁹, and spinocerebellar ataxia⁷⁰. It is currently undergoing Phase 2 clinical trials for ALS. However, the proposed mechanism of action of Sephin1 has been contested. Crespillo-Casado et al.⁷¹. found that Sephin1 did not disrupt the formation or stability of the GADD34-PP1 complex in vitro biochemical assays and had no measurable effect on eIF2α dephosphorylation in cells.

In our study, we observed that neither genetic inhibition of GADD34 via CRISPR nor pharmacological inhibition using Sephin1 resulted in increased phosphorylation levels of eIF2α in HEK293T cells (Fig. 5A, B, Supplementary Fig. 4H, Supplementary Fig. 4K). Interestingly, significant eIF2α dephosphorylation was only observed when GADD34 was massively overexpressed in cells using transient transfection, but not when expressed at more physiological levels via lentiviral infection (Fig. 5C). This raises the possibility that GADD34’s canonical role in eIF2α dephosphorylation may be context-dependent or secondary to other functions, particularly under conditions of supraphysiological expression.

Instead of affecting eIF2α phosphorylation, we found that GADD34 inhibition strongly reduced XBP1 splicing under ER stress. Further analysis revealed that GADD34 modulates IRE1α activation and directly interacts with IRE1α (Fig. 6B, C). This interaction likely involves multiple regions of GADD34, including its ER-targeting region and PP1-binding domain. These findings suggest that GADD34 plays a previously unrecognized role in regulating IRE1α-XBP1 signaling, independent of its canonical function in eIF2α dephosphorylation. The precise molecular mechanism by which GADD34 modulates IRE1α activation remains to be elucidated.

Building on these findings, we explored the therapeutic potential of targeting GADD34 in cancer immunotherapy. ER stress and XBP1 splicing have been implicated in T cell exhaustion, a major barrier to the efficacy of CAR-T cell immunotherapy^12,15,16,61. By inhibiting GADD34 with Sephin1, we were able to suppress XBP1 splicing (Fig. 7B, C) and alleviate CAR-T cell exhaustion in an ex vivo serial co-culture model. Sephin1-treated CAR-T cells exhibited enhanced expansion, sustained effector function, and improved tumor-killing capacity across multiple cancer models, including both B7-H3 and CD19 CAR-T cells (Fig. 7E–H, Supplementary Fig. 6D–G), suggesting that GADD34 inhibition may represent a broadly applicable strategy for improving CAR-T cell performance. However, further studies are needed to confirm whether Sephin1’s beneficial effects are mediated specifically through IRE1α and to elucidate the downstream mechanisms. Future studies should also evaluate the in vivo efficacy and safety profile of GADD34 inhibition to fully assess its potential as a therapeutic strategy for improving CAR-T cell performance.

We anticipate that our SPLiCR-seq platform will be broadly applicable to diverse splicing events. While we demonstrated its utility in XBP1 splicing in this work, the system can be readily adapted by swapping the splicing reporter to interrogate other splicing events-including those implicated in human diseases-such as SMN2 splicing in spinal muscular atrophy⁷², LMNA splicing in progeria⁷², cryptic splicing in TDP-43 proteinopathies⁴ and MAPT splicing in Alzheimer’s disease and frontotemporal dementia⁴. By systematically identifying the key regulators of these splicing events, SPLiCR-seq holds the potential to advance our understanding of RNA splicing regulation and facilitate the development of novel therapeutic strategies against splicing-related diseases.

Methods

Cell culture

HEK293T (ATCC, CRL-3216) and COS-7 (ATCC, CRL-1651) cells were purchased from ATCC. 293FT (Procell, CL-0313) cells, the human PDAC cell line BXPC-3 (Procell, GCL-0042), and the TNBC cell line HCC1806 (Procell, GCL-0726) were obtained from Procell (Wuhan, China). iPSCs (WTC11 background) were a gift from Bruce R. Conklin (Gladstone Institutes). Frozen peripheral blood mononuclear cells (PBMCs; STEMCELL, 70025.3) from healthy donors were purchased from STEMCELL Technologies.

HEK293T, 293FT and COS-7 cells were cultured in Dulbecco’s Modified Eagle’s Medium (Gibco, C11995500BT) supplemented with 10% fetal bovine serum (TransGen Biotech, FS301-02) and 1% penicillin-streptomycin (Aladdin, P301861). iPSCs were cultured in StemFlex medium (Gibco, A3349401) and plated on Matrigel (Corning, 356231) coated plates for at least 30 min before seeding the cells. 10 nM ROCK inhibitor (Selleck, S1049) was added on the first day of plating iPSCs. BXPC3 and HCC1806 cells were maintained in RPMI1640 (Gibco, C11875500BT) supplemented with 10% FBS (TransGen, FS301-02), 2 mM GlutaMax (Thermo, 35050061), and 100 U/mL penicillin and 100 μg/mL streptomycin. Human T cells were maintained in complete human T cell medium (1/2 RPMI-1640, 1/2 Click’s Media (Mesgen, MCM4125), supplemented with 10% FBS (Hyclone, SV30208.02), 2 mM GlutaMAX, 100 U/mL of Penicillin, and 100 μg/mL of streptomycin). Expression of surface markers and functional readouts were validated by FACS. All cells were maintained at 37 °C and 5% CO₂, and were tested negative for mycoplasma contamination using the MycAway™ Plus-Color One-Step Mycoplasma Detection Kit (Yeasen, 40612ES25).

Generation of the XBP1 SPLiCR-seq vector

Sequences flanking the splicing site of XBP1 was amplified from HEK293T cDNA by PCR. The amplified product was inserted into the pMK1334 vector by replacing the original WPRE cassette between EcoRI and SalI sites using the ClonExpress® Ultra One Step Cloning Kit (Vazyme, C115). pMK1334 was a gift from Martin Kampmann (Addgene plasmid # 127965; http://n2t.net/addgene:127965; RRID:Addgene_127965).

sgRNA cloning

sgRNAs for CRISPRi were cloned into the SPLiCR-seq vector or pLG15 via BstXI and Bpu1102I sites as previously described³¹. sgRNAs for CRISPR knockout were cloned into the pX459 vector via BsmBI sites. The pX459 vector was a gift from Feng Zhang (Addgene plasmid # 62988; http://n2t.net/addgene:62988; RRID:Addgene_62988). A complete list of sgRNA sequences used in this study is listed in Supplementary Data 2.

Lentivirus production

For lentivirus production of the sgRNA library, 15 × 10⁶ HEK293T cells were seeded in a 15 cm dish for 24 h before transfection. Fifteen micrograms of sgRNA library plasmid and 15 µg of third-generation packaging mix (1:1:1 mix of the three plasmids: pMDLg/pRRE (Addgene, 12251), pRSV-Rev (Addgene, 12253), and pMD2.G (Addgene, 12259)) were diluted in 3 mL of Opti-MEM (Gibco, 31986-07). Subsequently, 45 µL of 2 mg/mL Polyethylenimine Linear (PEI) MW40000 (Yeasen, 40816ES03) was added to the 3 mL DNA dilution, vortexed for 10 seconds, and thoroughly mixed. After incubating at room temperature for 15–20 min, the mixture was added to the 15 cm dish containing HEK293T cells. 48 h later, the viral supernatants were collected and filtered through a 0.45 μm filter (Millipore, SLHV033RB).

For small-scale lentivirus production, 0.9 × 10⁶ HEK293T cells were seeded on 6-well plates. After 24 h, 100 µL of Opti-MEM was used to dilute 1 µg of transfer plasmid and 1 µg of third-generation packaging mix, thoroughly mixed to form the DNA dilution. Subsequently, 3 µL of PEI was added, and the mixture was vortexed for homogeneity. The remaining procedures were carried out as described above.

Quantitative real-time polymerase chain reaction (qPCR)

Total RNA was extracted using the MolPure® Cell RNA Kit (Yeasen, 19231ES50) according to the manufacturer’s instructions. RNA was reverse transcribed to cDNA with the TransScript® One-Step gDNA Removal and cDNA Synthesis SuperMix (TransGen, AT311-03). Quantitative real-time PCR was performed using AceQ qPCR SYBR Green Master Mix (Vazyme, CQ111-02) according to the manufacturer’s protocol and run on a Fluorescence Quantitative PCR detection system (FDQ-96A). GAPDH was used as an endogenous control.

Detection of XBP1 splicing in SPLiCR vector by RT-PCR and grayscale analysis

To measure XBP1 splicing levels, total RNA was extracted using the MolPure® Cell RNA Kit (Yeasen, 19231ES50) according to the manufacturer’s instructions. One microgram of total RNA was reverse-transcribed into cDNA using the TransScript® One-Step gDNA Removal and cDNA Synthesis SuperMix (TransGen, AT311-03).

For PCR amplification, 1.5 µL of cDNA was used as the template in a 20 µL reaction mixture prepared with 2×Hieff®PCR Master Mix (With Dye) (Yeasen, 10102ES03), following the manufacturer’s protocol.

For endogenous XBP1 splicing, the PCR cycling conditions were: 98 °C for 3 min; 25 cycles of 98 °C for 10 s, 62 °C for 20 s, and 72 °C for 15 s; followed by a final extension at 72 °C for 5 min. The primers used were endo-XBP1-For and endo-XBP1-Rev (Supplementary Data 2).

For XBP1 reporter splicing, the PCR cycling conditions were: 98 °C for 3 min; 25 cycles of 98 °C for 10 s, 65 °C for 20 s, and 72 °C for 15 s; followed by a final extension at 72 °C for 5 min. The primers used were exo-XBP1-For and exo-XBP1-Rev (Supplementary Data 2).

PCR products were separated on a 4% agarose gel and imaged using an automated gel imaging system (OI1000). Gel images were analyzed using Fiji (ImageJ). Background subtraction was performed using the Subtract Background function in the Process menu. Regions of interest (ROIs) corresponding to the spliced and unspliced bands were manually selected, and their mean gray values were obtained using the Measure function. The XBP1 splicing level under each condition was calculated as the ratio of the mean gray value of the spliced band to that of the unspliced band (spliced/unspliced, s/u).

Drug treatment

Drug treatments were performed using thapsigargin (Tg; Santa Cruz, sc-24017), tunicamycin (Tm; Beyotime, SC0393), dithiothreitol (DTT; BBI, B645939), Sephin1 (MCE, HY-111022), and trans-ISRIB (Tocris, 5284). Compounds were prepared as stock solutions in appropriate solvents and stored according to the manufacturers’ recommendations. Cells were seeded at least 24 h before treatment and treated at 50–70% confluence. Concentrations and exposure durations were selected based on prior reports as follows: thapsigargin, 10 nM–2 µM for 0–48 h^9,12,33; tunicamycin, 1–20 µM for 0–24 h^33,36,43; DTT, 0.4 mM for 1 h^54,73,74; Sephin1, 1–5 µM for 3 or 24 h^55,75; and trans-ISRIB, 1 µM for 3 h^36,41.

Generation of the RBP and genome-wide SPLiCR-seq sgRNA libraries

For the RBP library, we used a previously reported, manually curated RBP gene set⁷³ and selected the top five sgRNAs for each gene from the CRISPRi-v2 library⁷⁴. This resulted in a library comprising 6,602 unique sgRNA sequences targeting 1,350 RBPs, along with 250 non-targeting control sgRNAs (Supplementary Data 3).

For the genome-wide library, we selected genes expressed in HEK293T cells (CPM > 5). For each gene, two sgRNAs were selected from the CRISPRi-v2 library, prioritized based on empirical activity determined from our in-house screening data. For genes lacking empirical data, the original activity scores from the CRISPRi-v2 library were used to rank and select sgRNAs. This yielded a library containing 22,240 unique sgRNA sequences targeting 11,120 genes, along with 1760 non-targeting control sgRNAs (Supplementary Data 3).

All sgRNA oligonucleotides were synthesized by GENEWIZ and cloned into the XBP1 SPLiCR-seq vector using the BstXI and BlpI restriction sites. To evaluate library quality, the sgRNA-containing fragment was amplified using Phanta Flash Master Mix (Vazyme, P520) according to the manufacturer’s instructions, and the resulting PCR products were analyzed by next-generation sequencing.

SPLiCR-seq screening and data analysis

Lentivirus production of the SPLiCR-seq library was carried out as described above. The lentivirus was transduced into CRISPRi-HEK293T and CRISPRi-iPSC cells at a multiplicity of infection (MOI) less than 0.3. 48 h later, the transduced cells were selected with 2 µg/mL puromycin for 48 h to eliminate uninfected cells and generate a genome-edited cell pool. After 7 days of passage in medium without puromycin and treated with Thapsigargin (Tg, Santa Cruz, sc-24017) or Tunicamycin (Tm, Beyotime, SC0393), 10 million cells were collected for total RNA extraction using the MolPure® Cell RNA Kit. The mRNA with a poly-A tail was purified from the total extracted RNA using the BeyoMag™ mRNA Purification Kit with Magnetic Beads (Beyotime, R0071M) and reverse transcribed into cDNA using Oligo-dT primers and TransScript® II One-Step gDNA Removal and cDNA Synthesis SuperMix (TransGen, AH311). The resulting cDNA was used as template to amplify the region encompassing the XBP1 splicing reporter and the sgRNA cassette with primers SPLICR-F and SPLICR-R (Supplementary Data 2). PCR was performed with 2×Phanta Flash Master Mix polymerase (Vazyme, P510) under the following cycling conditions: initial denaturation at 95 °C for 3 min; 22 cycles of 95 °C for 15 s, annealing at 58 °C for 15 s, and 72 °C for 30 s; final extension at 72 °C for 10 min. Approximately 200 ng cDNA was used per 50 µL reaction. Amplicons were purified using Hieff NGS® DNA Selection beads (Yeasen, 12601ES08) at a 0.45× bead-to-sample ratio and sequenced using paired-end 150 bp reads (PE150) on an Illumina NovaSeq 6000 (Berry Genomics).

Raw FASTQ files were processed separately for sgRNA identification (Read 2) and reporter splicing status (Read 1). Read 2 FASTQ files were trimmed to retain the sgRNA insert and aligned to the sgRNA reference library using Bowtie (v1.1.2). Read 1 FASTQ files were trimmed to the reporter region and aligned to a custom reference composed of the spliced and unspliced XBP1 reporter sequences to determine the splicing status of each read. Bowtie alignment outputs from Read 1 and Read 2 were joined by read ID to associate each reporter splicing call (spliced or unspliced) with its corresponding sgRNA. These associations were aggregated into count tables reporting, for each sgRNA in the library, the number of spliced and unspliced reads (Supplementary Data 4).

Count tables were analyzed with MAGeCK(v0.5.9.2)⁷⁵ using the ‘mageck test -k’ workflow with the ‘--paired’ option. Spliced counts were designated as the treatment group and unspliced counts as the control group. Normalization employed non-targeting control sgRNAs, specified via the --norm-method option. This analysis produced sgRNA-level and gene-level splicing phenotypes (log2 fold change, lfc) and statistical significances. A negative phenotype indicates that the perturbation decreased splicing, whereas a positive phenotype indicates increased splicing. Gene-level scores reported in Figs. 2D, 3E, and 3F were derived from the MAGeCK gene_summary output as signed −log10 (RRA score), where the sign reflects the direction of lfc. P values were calculated from the negative binomial model using a modified robust ranking aggregation algorithm⁷⁵. FDR was computed from the empirical permutation P values using the Benjamini–Hochberg procedure⁷⁵.

Immunoblotting

Cells were lysed in RIPA buffer (Beyotime, P0013B) supplemented with a phosphatase inhibitor cocktail (Selleck, B15001) and protease inhibitor cocktail (MCE, HY-K0010). The protein concentration of each sample was determined using the Easy II Protein Quantitative Kit (TransGen, DQ111). Proteins were denatured in loading buffer (Solarbio, P1041) by boiling at 95 °C for 10 min, and 20 µg of protein from each lysate was electrophoresed on SDS-PAGE gels and transferred to 0.45 µm PVDF membranes (Immobilon, IPVH00010). The membrane was blocked with 5% BSA (Sigma-Aldrich, V900933) in TBST (Sangon, C520009) and incubated with primary antibodies as indicated at 4 °C overnight. The next day, the membrane was incubated with horseradish peroxidase–conjugated secondary antibodies for 1 h at room temperature. Protein signals were detected using the Clarity Western ECL Substrate (EpiZyme, SQ202L, SQ201). The antibodies used in this study are summarized here: anti-PERK (Proteintech, 20582-1-AP, 1:1000 dilution), anti-Phospho-IRE1α (Abcam, ab124945, 1:1000 dilution), anti-IRE1α (CST, 3294, 1:1000 dilution), anti-GADD34 (Proteintech, 10449-1-AP, 1:1000 dilution), anti-XBP1s (CST, 12782, 1:1000 dilution), anti-Phospho-eIF2α (CST, 9721S, 1:1000 dilution), anti-eIF2α (CST, 5324, 1:1000 dilution), anti-GAPDH (Proteintech, HRP-60004, 1:5000 dilution), anti-Flag (Absin, Abs830014, 1:5000 dilution), anti-HA (Sigma, H3663, 1:2000 dilution), anti-GFP (Ray Antibody, RM1108, 1:5000 dilution), anti-rabbit IgG (CST, 7074S, 1:5000 dilution), anti-mouse IgG (CST, 7076S, 1:5000 dilution).

Co-immunoprecipitation (Co-IP)

A total of 5 × 10⁶ HEK293T cells were plated onto a 10 cm diameter dish for 24 h before transfection. The pcDNA3.1-GFP, pcDNA3.1-GFP-GADD34 FL, pcDNA3.1-GFP-GADD34 Δ1-60, pcDNA3.1-GFP-GADD34 Δ1-323, pcDNA3.1-GFP-GADD34 Δ513-674, PCDH-vector-3×Flag, and PCDH-FL GADD34-3×Flag vectors were transfected into the cells. Forty-eight hours later, the cells were collected and lysed in lysis buffer (Beyotime, P2181S-1) supplemented with Deoxy Big CHAP (BBI, A600681-0250). The cell lysates were centrifuged at 12,000 rpm at 4 °C for 10 min, and the supernatant was collected. Anti-Flag magnetic beads (Beyotime, P2181S-4), Anti-IgG magnetic beads (Beyotime, P2171), Anti-HA magnetic beads (Beyotime, P2121), and anti-GFP agarose beads (ABMagic, MA108-25T) were blocked overnight at 4 °C using 5% BSA. The supernatant was then incubated with anti-Flag magnetic beads and anti-GFP agarose beads at 4 °C overnight. After washing, the beads bound to proteins were denatured and subjected to immunoblotting analysis.

Immunofluorescence (IF)

Cells were seeded at 5 × 10⁴ cells per well on Matrigel-coated 12-mm glass coverslips (CITOTEST, #10210012CE) in 24-well plates and cultured overnight. Cells were then fixed with 4% paraformaldehyde (Beyotime, #P0099) for 30 min at room temperature (RT), permeabilized with 0.1% Triton X-100 (Coolaber, #CT11451) in PBS for 10 min, and blocked with 5% BSA (Sigma-Aldrich, #V900933) in PBS for 1 h at RT. Primary antibody against CANX (Proteintech, #10427-2-AP, 1:500 dilution) was applied overnight at 4 °C. After washing three times with PBS (5 min each), cells were incubated with Alexa Fluor-conjugated secondary antibodies (1:1000 dilution) for 1 h at RT. Nuclei were counterstained with DAPI (Beyotime, #C1006, 1:1000) for 10 min at RT. Coverslips were washed three times with PBS and mounted using Mounting Medium (SouthernBiotech, #0100) on glass slides. Images were captured using a confocal microscope (Zeiss, LSM 980) and analyzed using Fiji (v2.0.0).

Transduction of human T cells

Human CAR-T cells were generated as previously described⁷⁶. Briefly, PBMCs were activated by anti-CD3 (1 μg/mL, Biolegend, 317326) and anti-CD28 (1 μg/mL, BD Biosciences, 555725) mAbs pre-coated on a non-tissue culture plate. Activated T lymphocytes were then transduced with retroviral supernatants using retronectin-coated plates (Takara, T100B). Three days after transduction, CAR-T cells are harvested and cultured in complete human T cell media medium containing IL-7 (10 ng/mL; PeproTech, 200-07-500) and IL-15 (5 ng/mL; PeproTech, 200-15-500) for 7–9 days. 24 hours before functional assays, cytokines were washed away, and CAR-T cells were rested in T cell medium overnight.

Ex-vivo exhaustion model

Tumor cells (BXPC3 or HCC1806-CD19) were seeded at 0.2 × 10⁶ cells per well in a 12 well plate overnight and B7-H3 or CD19 specific CAR-T cells were added in triplicate at 1-to-1 effector-to-target ratio in the presence of DMSO or 30 µM Sephin1 (MCE, HY-111022). 3 days after co-culture, cells from all 3 wells were pooled, washed and resuspended in 3 mL fresh media. 1 mL of cells were used for cell counting and intracellular cytokine staining while 2 mL of cells were added to freshly seeded tumor cells for additional rounds of co-culture.

Intracellular cytokine staining

CAR-T cells were stimulated with 50 μg/mL PMA (Beyotime, S1819) and 1 mM Ionomycin (Beyotime, S1672) with 5 μg/mL brefeldin A (Biolegend, 420601) for 4 h to induce cytokine production. Cells were then harvested and stained for CD8 and Live/Dead (Zombie Aqua, Biolegend, 423102) and subsequently fixed and permeabilized using BD Cytofix/Cytoperm™ Fixation/Permeabilization Kit (BD Bioscience, 554714) at 4 °C overnight, followed by intracellular staining of IFN-γ and TNF-α.

Flow cytometry analysis

The following flow cytometry antibodies were purchased from Biolegend: AF700-conjugated anti-CD8 (clone SK-1), AF647-conjugated anti-IFN-γ (clone 4S.B3), and PE-conjugated anti-TNF-α (clone MAB11). Flow cytometry data was acquired on a NovoCyte Quanteon flow cytometer (Agilent) using NovoExpress software and the flow data were analyzed by FlowJo software (v10.7, Tree Star).

Statistics and reproducibility

The data in most figure panels reflect multiple experiments performed using independent samples. Statistical differences between different groups were analyzed using the paired Student’s t test when comparing two variables or one-way ANOVA with Dunnett multiple comparisons test when comparing multiple, with the statistical differences displayed in the Source Data file. GraphPad Prism were used for plotting, graphing, and statistical analysis. Statistical significance was determined as P < 0.05, with the exact P values displayed in the Source Data file. For experiments with replicates, the results were shown as mean ± s.d., unless stated otherwise. Most representative western blot images and RT-PCR analysis were from at least three independent repeats with similar results. Representative micrographs were from at least three independent samples with similar results. No data were excluded from the analyses.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

R.T. conceived the project. R.T. and Y.X. supervised the project. Q.Y., Y.C., and L.S. conducted experiments and analyzed data with guidance from Y.X. and R.T. Q.Y., Y.C., Y.X., and R.T. wrote the manuscript with input from all authors.

Peer review

Peer review information

Nature Communications thanks the anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The data that support the findings of this study are available within the paper, supplementary information, and supplementary data files. The raw sequence data reported in this paper have been deposited in the Genome Sequence Archive in the National Genomics Data Center under accession code PRJCA036971. Source data are provided with this paper.

Competing interests

Q.Y., L.S., and R.T. are inventors on a patent application related to methods for high-throughput screening of RNA splicing phenotypes that has been filed by Southern University of Science and Technology (patent application number: CN119899872A). Q.Y., Y.C., Y.X., and R.T. are inventors on a patent application related to the use of GADD34 inhibitors in CAR-T cell tumour therapy that has been filed by Southern University of Science and Technology (patent application number: CN120478637A). The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-67633-4.