Splicing-mediated control of hnRNPD isoform switching by SRSF2 drives PD-L1-dependent immune evasion in gallbladder cancer

Abstract

Gallbladder cancer (GBC), a lethal malignancy of the biliary tract, is associated with a poor clinical prognosis. Although chemo-immunotherapy combinations demonstrate preliminary efficacy, the molecular determinants of treatment response remain elusive. Emerging evidence implicates aberrant alternative splicing in modulating tumor immunity. Through an in vitro CRISPR/Cas9 screen, we identified SRSF2 as a key RNA-binding protein regulating PD-L1 expression. Intriguingly, SRSF2 does not directly bind PD-L1 mRNA. Multi-omics analyses (mRNA-seq, RIP-seq, and proteomics) revealed that SRSF2 induces exon skipping in hnRNPD, shifting isoform expression from full-length P45 to truncated P40. Functional studies established that P45—but not P40—binds to AU-rich elements in the PD-L1 3’-UTR to promote mRNA degradation. Leveraging this mechanism, we designed splice-switching antisense oligonucleotides (ASOs) that block SRSF2-mediated exon skipping, restoring P45 expression. This intervention effectively reduced PD-L1 levels and potentiated T-cell-mediated cytotoxicity in vitro and in vivo. These findings elucidate a splicing-centric mechanism of immune evasion and highlight the therapeutic potential of splicing modulation in cancer immunotherapy.

Proposed model of the SRSF2-hnRNPD-PD-L1 axis in gallbladder cancer (GBC) immune evasion and its therapeutic targeting. Overexpression of SRSF2 drives hnRNPD exon skipping, shifting the isoform balance from the PD-L1-degrading P45 to the truncated P40. This transition stabilizes PD-L1 mRNA and facilitates tumor immune evasion. Conversely, therapeutic intervention with splice-switching ASOs blocks SRSF2-mediated alternative splicing, restores P45 expression, and effectively reactivates T-cell-mediated cytotoxicity against GBC cells.

Plain Language Summary

Gallbladder cancer is a highly aggressive disease that often evades the body’s natural immune defense, making current treatments less effective. Our study aimed to understand the hidden mechanisms allowing these cancer cells to hide from immune attacks. Using laboratory tumor models and patient tissue samples, we investigated the internal genetic processes of these cancer cells. We discovered that a specific protein (SRSF2) acts like a rogue editor, altering the genetic instructions of another molecule (hnRNPD). This specific modification ultimately creates a “shield” (PD-L1) on the surface of the cancer cell, effectively blinding the immune system and allowing the tumor to grow unchecked. Understanding this specific genetic editing process is highly significant, as it provides a vital new target. In the future, blocking this pathway could strip away the cancer’s shield, potentially making existing immunotherapies much more effective for patients.

Highlights

• SRSF2 drives PD-L1 upregulation in GBC via hnRNPD P45-to-P40 isoform shift.

• hnRNPD P45, but not P40, binds the PD-L1 3’-UTR to promote mRNA degradation.

• ASOs blocking SRSF2-mediated hnRNPD exon skipping restore P45, lowering PD-L1.

• Splicing dysregulation unveils a targetable immune evasion pathway in GBC.

Introduction

Alternative splicing serves as a crucial post-transcriptional regulatory mechanism, enabling a single pre-mRNA transcript to undergo differential processing through combinatorial splice site selection [1, 2]. This sophisticated processing generates multiple mature mRNA variants (isoforms) through the selective inclusion or exclusion of specific exons or intronic regions [3]. Such transcript diversification dramatically enhances proteomic complexity through the production of structurally and functionally distinct protein products from a single genetic locus. Alternative splicing dysregulation has emerged as a hallmark of human diseases, particularly in malignancies [4]. Growing evidence indicates that cancer cells exploit alternative splicing networks to generate pro-survival isoforms, remodel signaling pathways [5].

Gallbladder cancer (GBC), the most common biliary tract malignancy, is characterized by persistently dismal prognosis and rising global incidence. Current therapeutic approaches remain inadequate, highlighting an urgent need for novel targets and agents [6, 7]. Programmed death-ligand 1 (PD-L1) is a key immune checkpoint molecule expressed on tumor cells that binds to PD-1 on T cells, thereby inhibiting T-cell activation and enabling immune escape [8]. The publication of phase III clinical trial results from KEYNOTE-966 and TOPAZ-1 has substantially reshaped the therapeutic landscape for biliary tract cancer (BTC) through the integration of PD-L1/PD-1 immune checkpoint inhibitors [9, 10]. While existing trials for immune checkpoint inhibitors (ICIs) have largely comprised pan-biliary tract cancer populations, it is crucial to recognize the therapeutic implications of BTC heterogeneity. Data focusing specifically on GBC are notably scarce, which can be attributed to the rarity of this particular subtype. More importantly, a significant proportion of patients with advanced GBC exhibit primary resistance to these therapies. This underscores an urgent need to elucidate the precise regulatory networks governing PD-L1 expression, and raises the intriguing question of whether post-transcriptional mechanisms, such as alternative splicing, orchestrate immune evasion in GBC.

Serine/arginine-rich splicing factor 2 (SRSF2), a core member of the SR protein family, serves as a critical regulator of both pre-mRNA splicing and transcriptional regulation in mammalian systems [11]. This splicing factor is ubiquitously expressed across diverse cell types and participates in multiple pathological processes [12]. Emerging evidence indicates that SRSF2 dynamically regulates the alternative splicing of ACSL3 [13], YTHDF1 [14] and ASPM [15] to modulate tumor progression through precise splicing-mediated oncogenic reprogramming.

Here, using an in vitro CRISPR/Cas9 screen, we identified SRSF2 as a pivotal RNA-binding protein regulating PD-L1 expression. Functional studies demonstrated that SRSF2-mediated PD-L1 upregulation facilitated immune evasion in GBC. Mechanistically, SRSF2 enhances PD-L1 mRNA stability indirectly through its 3’ untranslated region (3’-UTR). Integrated multi-omics revealed hnRNPD as the critical downstream effector, wherein SRSF2 promoted hnRNPD exon skipping to drive an isoform switch from P45 to P40. Functional characterization demonstrated that P45 bound PD-L1 3’-UTR to promote mRNA degradation. Therapeutically, antisense oligonucleotides (ASOs) targeting the SRSF2-hnRNPD splicing sequence demonstrated potent growth suppression of gallbladder cancer in vitro and in vivo. This work elucidates a novel post-transcriptional regulatory paradigm wherein RNA-binding proteins fine-tune PD-L1 expression through alternative splicing control, while providing a clinically actionable strategy to overcome immune evasion in gallbladder cancer.

Methods

CRISPR/Cas9 screening of human mRNA-binding protein library

A CRISPR/Cas9 knockout screen utilizing a lentiviral Human mRNA-Binding Protein Library (Pooled Library #168791 targeting 725 genes) was conducted in NOZ cells to identify PD-L1 regulators. Briefly, cells were transduced with the library at a low multiplicity of infection. Following a 24-h incubation, transduced cells underwent puromycin selection (0.5 μg/ml) for 7 days to establish a stable knockout pool. For the analysis of PD-L1 expression, the cell pool was harvested, resuspended in cold PBS, and stained with an anti-PD-L1 antibody (BD Biosciences #558017). Using fluorescence-activated cell sorting (FACS), distinct subpopulations exhibiting the top 5% and bottom 5% PD-L1 expression intensity relative to the total population were isolated. Genomic DNA was then extracted from these sorted fractions, followed by PCR amplification and sequencing of integrated sgRNA cassettes. The resulting sequencing data, reflecting sgRNA abundance profiles, were subsequently analyzed using the MAGeCK (Model-based Analysis of Genome-wide CRISPR-Cas9 Knockout) algorithm to statistically identify mRNA-binding proteins whose knockout significantly altered PD-L1 surface expression levels.

Patient data

Clinical specimens and associated clinicopathological data were obtained from the Department of General Surgery at Xinhua Hospital, Shanghai Jiao Tong University School of Medicine. All pathological materials underwent rigorous verification by board-certified pathologists prior to analysis. Eligibility criteria required histopathological confirmation of diagnosis and excluded patients who had undergone any form of adjuvant therapy (including chemotherapy or targeted interventions). Written informed consent was obtained from all participants prior to study enrollment.

Immunohistochemistry (IHC) and multiplex immunohistochemistry (mIHC)

For IHC, tissue sections underwent sequential processing including rehydration, endogenous peroxidase inactivation (3% H₂O₂, 5 min), and heat-mediated antigen retrieval using 0.01 M sodium citrate buffer. After blocking with 5% goat serum, primary antibody incubation (4 °C, overnight) was performed followed by horseradish peroxidase-conjugated secondary antibody (goat-anti-rabbit, 1 h at room temperature). Chromogenic development utilized 3,3’-diaminobenzidine with hematoxylin counterstaining.

For mIHC, tyramine signal amplification (TSA) technology was implemented. Following standard antibody staining cycles, fluorophore-conjugated TSA reagents (Beyotime) were applied (10 min at room temperature) with iterative antigen retrieval between sequential staining rounds. Nuclear visualization was achieved with DAPI before mounting. Three blinded investigators independently scored immunoreactivity using standardized criteria. Antibody specifications are detailed in Supplementary Table 1.

Cell culture and reagents

Human gallbladder carcinoma cell lines (NOZ and GBC-SD) were obtained from the Cell Resource Center of Shanghai Institutes for Biological Sciences, Chinese Academy of Sciences. These cell lines were maintained in high-glucose Dulbecco’s Modified Eagle Medium (DMEM; BasalMedia, China) supplemented with 10% fetal bovine serum (FBS; Vazyme, Nanjing, China), under standardized culture conditions of 37 °C with 5% humidified CO₂ atmosphere. The pharmacological agents employed in this investigation, including atezolizumab and proteasome inhibitor MG-132, were commercially sourced from MedChemExpress.

Cell transfection

Gene-specific small interfering RNAs (siRNAs) and antisense oligonucleotides with 2’-O-methoxyethyl modifications were synthesized by Genepharma Co., Ltd (Shanghai, China), with detailed sequences provided in Supplementary Table 2. For functional studies, transient gene silencing was achieved using Rfect transfection reagent (Baidai Biotech, China), while plasmid-mediated overexpression was performed with Lipofectamine 2000 (Invitrogen) according to standardized protocols. Full-length cDNA constructs cloned into mammalian expression vectors were obtained from Longqian Biotechnology (Shanghai, China).

Reverse transcription PCR (RT-PCR) and Quantitative Real-Time PCR (qPCR)

Total RNA was isolated from both clinical specimens and cultured cells using TRIzol reagent (Sangon Biotech, China), with purity verified by NanoDrop 2000 spectrophotometry. cDNA synthesis was performed using the HiScript II 1st Strand cDNA Synthesis Kit (Vazyme) following manufacturer-recommended thermal cycling parameters. For transcript detection, HiScript II One Step RT-PCR Kit (Vazyme) was used in RT-PCR and ChamQ SYBR qPCR Master Mix (Vazyme) was used for qPCR. Custom-designed primers (Supplementary Table 3) were synthesized by HuageneBio (Shanghai, China). Relative mRNA expression levels were calculated using the 2^−ΔΔCt method, with GAPDH serving as the endogenous normalization control.

Cell proliferation detection

The Cell Counting Kit-8 (Vazyme) assay was conducted following the manufacturer’s protocol to assess metabolic activity. The EdU-based DNA synthesis assay (Beyotime, China) was performed by incubating experimental cells with 10 μM EdU for 2 h at 37 °C. Subsequent processing included fixation with 4% paraformaldehyde, permeabilization with 0.3% Triton X-100, and incubation with click reaction cocktail containing Alexa Fluor 488-azide for 30 min protected from light. Nuclear counterstaining was achieved using DAPI (1 μg/mL) prior to fluorescence microscopy analysis.

Western blot analysis

Protein separation was achieved using SDS-PAGE gels (7.5%, 10%, 15%; Vazyme) selected based on the molecular weights of the target proteins. Following electrophoresis under constant voltage (100 V for 90 min), proteins were transferred to PVDF membranes (Millipore) using wet transfer methodology. Membranes were subsequently blocked with Protein-Free Rapid Blocking Buffer (EpiZyme) for 1 h at room temperature prior to overnight incubation at 4 °C with primary antibodies (information in Supplementary Table 1). After three 5-min TBST washes (Servicebio, Wuhan, China), membranes were exposed to HRP-conjugated secondary antibodies (1:5000; Beyotime) for 1 h at room temperature. Protein bands were visualized using an enhanced chemiluminescence detection system (Vazyme) with exposure times optimized between 10 s and 5 min based on signal intensity.

Luciferase assay

Firefly luciferase reporter constructs containing either the full-length PD-L1 promoter or its truncated variants were synthesized and cloned into pGL3-basic vectors by Longqian Biotechnology. Following experimental treatments, transfected cells were lysed and luciferase activity was quantified using a luciferase detection system (Yeasen) according to the manufacturer’s protocol.

RNA immunoprecipitation (RIP) and RIP Sequencing (RIP-seq)

For each RNA immunoprecipitation assay, approximately 5-20 × 10^6 cells were harvested. The final cell pellet was resuspended in an equal volume of polysome lysis buffer (100 mM KCl, 5 mM MgCl2, 10 mM HEPES (pH 7.2), 0.5% NP40, 1 mM DTT, RNase inhibitors, protease inhibitors), incubated on ice for 5 min and cellular debris was pelleted by 4 °C centrifugation at 15,000 x g for 15 min. After centrifugation 40 μL of supernatant was collected for RNA extraction as input sample. The remaining supernatant was used for immunoprecipitation. 100ul protein A/G magnetic beads were washed with NT2 buffer (50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 1 mM MgCl2 and 0.05% NP-40) three times. Magnetic beads resuspended in 850 μl of the immunoprecipitation reaction solution (200 units of an RNase inhibitor, 400 μM RVC, 10 μl of 100 mM DTT, 30 μl of 0.5 mM EDTA and 800 μl of cold NT2 buffer). Cleared cell lysates were incubated with immunoprecipitation reaction solution at 4 °C for 4 h on a tube rotator. The beads were then washed with 1 ml of ice-cold NT2 buffer five times. TRIzol was added to input and immunoprecipitation samples to isolate total RNA. RNA quality control was measured on a Qubit. Libraries were constructed using either poly-A enrichment-based library preparation with the SMART-Seq v4 Ultra Low Input RNA Kit (Takara Clontech). Libraries were sequenced on a NovaSeq 6000 (Illumina). Detailed antibody specifications and primer sequences are cataloged in Supplementary Table 1 and Supplementary Table 3, respectively.

mRNA sequencing (mRNA-seq)

Transcriptomic profiling was conducted through mRNA sequencing of TRIzol-extracted RNA (Invitrogen) from clinical specimens and cultured cells. Sequencing libraries were constructed using poly(A) selection and subjected to 2×150 bp sequencing on the Illumina NovaSeq 6000 system (LC Bio Technology). Raw sequencing data underwent quality control and alignment through the OmicStudio bioinformatics platform (https://www.omicstudio.cn/tool), with differential expression analysis performed using DESeq2.

Immunoprecipitation (IP)

Cellular extracts were incubated with antibody-coupled protein A/G magnetic beads (MedChemExpress) in binding buffer (20 mM Tris-HCl pH 7.4, 150 mM NaCl, 0.1% NP-40) under continuous rotation (16 h, 4 °C). The immunocomplexes were subjected to three stringent wash cycles with high-salt buffer (50 mM HEPES, 500 mM NaCl, 0.05% Tween-20). Protein elution was achieved through thermal denaturation (95 °C, 10 min) in Laemmli buffer (62.5 mM Tris-HCl pH 6.8, 2% SDS, 10% glycerol; Beyotime).

Enzyme-linked immunosorbent assay (ELISA)

Secreted proteins were analyzed using commercial ELISA kits. Following experimental treatments, cell culture supernatants were collected and centrifuged to remove debris. Granzyme B (Abcam, ab235635) and IFN-γ (Abcam, ab174443) levels were quantified according to manufacturer protocols, with absorbance measurements performed in technical duplicates.

Protein expression and purification

Recombinant hnRNPD-P45 and hnRNPD-P40, each fused at the N-terminus to a 6×His-SUMO tag removable by Ulp1 protease, were produced in E. coli BL21 (DE3). Cultures were grown at 37 °C in LB medium supplemented with 50 µg mL⁻¹ kanamycin until the OD₆₀₀ reached ~0.6, then induced with 0.2 mM IPTG and shifted to 18 °C for overnight expression. Cells were collected by centrifugation (4 000 rpm, 15 min, 4 °C) and disrupted at 4 °C using a French press (JNBIO).

Lysates were clarified and loaded onto a HisTrap HP column (Cytiva) equilibrated with binding buffer. Bound 6×His-SUMO-hnRNPD proteins were eluted with imidazole, the tag was removed by Ulp1 digestion, and the mixture was reapplied to the column to separate the cleaved proteins from tag and protease. Subsequent purification employed a HiTrap Q ion-exchange column (Cytiva) with a linear NaCl gradient, followed by size-exclusion chromatography on a Superdex G75 HiLoad 16/60. The final buffer contained 10 mM Tris-HCl (pH 8.0), 100 mM NaCl, and 1 mM DTT.

RNA Electrophoretic Mobility Shift Assay (EMSA) RNA-protein interactions were analyzed using the EMSA kit (Beyotime) following the manufacturer’s protocol. Briefly, biotin-labeled RNA probes were incubated with purified protein in binding buffer for 30 min at 25 °C. For competition assays, 200-fold excess unlabeled probes or mutant probes were pre-incubated with the purified protein for 15 min prior to labeled probe addition. RNA-protein complexes were resolved on 6% non-denaturing polyacrylamide gels at 100 V for 1.5 h in 0.5× Tris-borate-EDTA buffer and transferred to nylon membranes. Biotin signals were detected using streptavidin-horseradish peroxidase and chemiluminescent substrate, followed by imaging with a ChemiDoc system (Bio-Rad).

RNA pulldown

RNA-protein interactions were analyzed using biotinylated RNA probes incubated with NOZ cell lysates in hybridization buffer (10 mM HEPES pH 7.3, 3 mM MgCl₂, 150 mM KCl) for 60 min at 25 °C. Probe-bound complexes were captured with streptavidin magnetic beads (Vazyme) and subjected to three high-stringency washes. Eluted proteins were resolved by SDS-PAGE and immunoblotted with target-specific antibodies.

Animal experiments

Female NOD/Shi-scid (NOG) mice (5-week-old) from Shanghai Laboratory Animal Center of the Chinese Academy of Sciences were humanized via intravenous injection of healthy donor PBMCs. After 7 days, 2×10⁶ treated NOZ cells were subcutaneously implanted into the right axilla. Tumor xenografts were excised 28 days post-implantation for histopathological analysis. All experimental protocols were approved by the Xinhua Hospital Animal Ethics Committee (Animal experiment approval number: XHEC-NSFC-2024-058).

Flow cytometry

Apoptosis was quantified using Annexin V-FITC/PI staining (Yeasen) following manufacturer protocols. Human immune cell engraftment was assessed by PE-conjugated anti-CD45 antibody (BioLegend) staining of peripheral blood cells. Membrane-bound PD-L1 expression was evaluated using PE-labeled anti-PD-L1 antibody (Invitrogen). Samples were analyzed on a CytoFLEX flow cytometer (Beckman Coulter) with CytExpert software.

Statistical Analysis

Data analysis was performed using GraphPad Prism 8.0.1 (GraphPad Software, San Diego, CA, USA). Quantitative data are presented as the mean ± standard deviation (SD) from at least three independent biological replicates. Intergroup differences were analyzed using an unpaired two-tailed Student’s t-test for comparisons between two groups, or one-way analysis of variance (ANOVA) for multiple comparisons. Significance was denoted as *P < 0.05, **P < 0.01, and ***P < 0.001. Kaplan-Meier survival curves were compared via the log-rank test.

Results

SRSF2 regulates PD-L1 expression in gallbladder cancer cells

PD-L1 is a critical immune checkpoint molecule, and its expression level in tumor cells is a key determinant of the response to immune checkpoint inhibitor therapy in GBC. However, the post-transcriptional regulatory mechanisms controlling PD-L1 expression in GBC remain poorly understood. To systematically identify RNA-binding proteins (RBPs) that regulate PD-L1 expression in GBC, we performed a CRISPR/Cas9 screen using a human messenger RNA binding protein library (Fig. 1A). Through screening, we identified 219 positive genes (Fig. 1B and Supplementary Table 4). Then using transcriptomic data from our previously established cohort of 12 paired gallbladder cancer and adjacent normal tissues[16], we identified 3387 upregulated differentially expressed genes (DEGs) with |log2(fold change)| ≥ 2 and p ≤ 0.05 (Supplementary Fig. S1A). Subsequently, Pearson correlation analysis of PD-L1 expression levels across 12 tumor specimens revealed 2482 genes exhibiting strong positive correlations (correlation coefficient > 0.8) (Supplementary Fig. S1B). Through integrated analysis of these three cohorts, we identified SRSF2 potentially modulating PD-L1 expression (Fig. 1C). Subsequently, we knocked down SRSF2 in gallbladder cancer cells. qPCR results showed that PD-L1 mRNA levels were significantly reduced after SRSF2 knockdown (Fig. 1D and Supplementary Fig. S1C). At the protein level, western blot and flow cytometry experiments demonstrated that both total and membrane-localized PD-L1 were markedly decreased following SRSF2 knockdown (Fig. 1E and F). Corresponding results were also observed upon SRSF2 overexpression (Supplementary Fig. S1D and E). Immunohistochemical analysis of a gallbladder cancer tissue microarray revealed SRSF2 overexpression in tumor tissues, which was positively correlated with PD-L1 expression and negatively correlated with the abundance of CD3+ and CD8 + T cells (Fig. 1G and Supplementary Fig. S1F–I). Our single-cell sequencing data in gallbladder cancer [17] also revealed a negative correlation between SRSF2 and T cell activation (Fig. 1H). Furthermore, comprehensive bioinformatic analysis of integrated gallbladder cancer transcriptomes (GSE138109, GSE139682 and GSE202479) revealed a significant negative correlation between SRSF2 expression and CD8⁺ T-cell abundance (Supplementary Fig. S1J). These findings suggest that SRSF2 is closely associated with tumor immunity. To evaluate the clinical relevance of SRSF2 in gallbladder cancer, we quantified its mRNA expression levels in 44 paired tumor and adjacent normal tissues using qPCR. The cohort was stratified into two equal groups (low vs. high SRSF2 expression) (Fig. 1I and Supplementary Table 5). Statistical analysis revealed a significant association between elevated SRSF2 expression and advanced TNM staging (Fig. 1J). Notably, Kaplan-Meier survival curves demonstrated that patients with high SRSF2 expression exhibited markedly reduced overall survival compared to the low-expression group (Fig. 1K). The observed association between SRSF2-driven PD-L1 expression and deteriorated survival rates underscores its pivotal role in gallbladder cancer progression.

Fig. 1. SRSF2 regulates PD-L1 expression in gallbladder cancer cells.

A CRISPR/Cas9 screening workflow. In brief, NOZ cells transduced with an RBP-focused CRISPR/sgRNA library were cultured for 7 days. Subsequently, PD-L1-high and PD-L1-low populations were isolated by FACS and subjected to deep sequencing for sgRNA abundance analysis. B Identification of 219 candidate positive regulators of PD-L1 by CRISPR/Cas9 screening. The differential ranking plot visualizes the results of a functional genomics screen in NOZ cells. C Venn diagram of 219 candidate positive genes of CRISPR/Cas9 screening, tumor upregulated differentially expressed genes, and correlation DEGs in 12 tumor samples. D qPCR analysis of PD-L1 and SRSF2 mRNA expression following SRSF2 knockdown with siRNAs in NOZ gallbladder cancer cells. E Western blot analysis detecting PD-L1 protein expression in NOZ gallbladder cancer cells after SRSF2 knockdown. The result confirms that SRSF2 depletion downregulates PD-L1 expression. F Flow cytometry quantifying the cell surface PD-L1 expression in NOZ cells upon SRSF2 silencing. The data demonstrates that SRSF2 knockdown significantly reduces membrane-localized PD-L1. G Representative immunohistochemical staining for SRSF2, PD-L1, CD3, and CD8 on a gallbladder cancer tissue microarray. These staining reveal the co-expression patterns of SRSF2, PD-L1, and tumor-infiltrating lymphocytes in the tumor microenvironment. H Correlation analysis between SRSF2 expression and T cell activation signatures derived from single-cell RNA sequencing data of gallbladder cancer tissues. The plot shows a negative correlation, indicating that high SRSF2 expression is associated with suppressed T cell activity. I Slope chart comparing SRSF2 mRNA expression levels in 44 paired samples of gallbladder cancer tissues and their corresponding adjacent non-tumor tissues. The data indicates that SRSF2 is significantly upregulated in tumor tissues. J. Box plot illustrating the association between TNM stages and SRSF2 mRNA expression levels across the 44 gallbladder cancer patients. Higher SRSF2 expression is positively correlated with advanced TNM stage. K Kaplan-Meier overall survival curves for gallbladder cancer patients stratified by high and low SRSF2 mRNA expression levels. Patients with high SRSF2 expression exhibit significantly poorer survival outcomes.

SRSF2 promotes immune evasion in gallbladder cancer

To investigate SRSF2’s immunomodulatory function in GBC, siRNA-mediated knockdown was performed in GBC cell lines. Cell proliferation assays demonstrated that SRSF2 silencing did not alter tumor growth kinetics (Supplementary Fig. S2A–C). Subcutaneous implantation of SRSF2-silenced NOZ cells versus control cells in NOG immunodeficient mice demonstrated comparable tumor growth kinetics (Supplementary Fig. S2D–F). To evaluate the impact of SRSF2-modulated PD-L1 on T cell functionality in vitro, an assessment of T cell-mediated killing assay was performed. Co-cultivation experiments revealed elevated levels of cytotoxic markers (IFN-γ and granzyme B) in CD8 + T cells interacting with SRSF2-silenced GBC cells (Supplementary Fig. S2G and H). Concurrently, flow cytometry combined with EdU-488 DNA synthesis assay demonstrated amplified apoptotic rates in tumor cells following SRSF2 depletion during CD8 + T cell co-incubation (Fig. 2A and B). Pharmacological blockade of PD-L1 using atezolizumab (10 μg/mL) demonstrated no additive effect on SRSF2 knockdown-driven cytotoxic T cell enhancement, suggesting that SRSF2-mediated tumor immune evasion operates predominantly through a PD-L1-dependent mechanism (Fig. 2C–E and Supplementary Fig. S2I). The absence of murine gallbladder cancer cell lines precludes immune-related studies in immunocompetent murine models. To address this limitation, we established a humanized PBMC-reconstituted (huPBMC) mouse model[18]. Flow cytometry validation confirmed successful engraftment with >60% human CD45+ leukocytes in peripheral blood (Supplementary Fig. S2J). In subcutaneous xenograft huPBMC models, SRSF2 knockdown markedly suppressed tumor growth kinetics (Fig. 2F–I). Multiplex immunohistochemistry analysis of resected tumors revealed enhanced intratumoral CD8 + T cell infiltration in SRSF2-deficient cohorts (Fig. 2J and K). Collectively, our data demonstrate that SRSF2 orchestrates PD-L1-dependent immunosuppression in gallbladder cancer.

Fig. 2. SRSF2 promotes immune evasion in gallbladder cancer.

A Apoptosis of gallbladder cancer cells after co-culture with human CD8 + T cells (CD8 + T cells and tumor cells co-cultured at a ratio of 1:1 for 48 h). Cancer cell apoptosis was quantified by flow cytometry analysis of Annexin V staining. Data are presented as the percentage of Annexin V-positive cells (mean ± SD). B Proliferation of gallbladder cancer cells co-cultured with CD8 + T cells, assessed by the EdU-488 DNA synthesis assay. C Granzyme B production in CD8 + T cells after co-culture with SRSF2-knockdown or control cancer cells, in the presence or absence of atezolizumab (10 µg/mL). Granzyme B levels were measured by ELISA. D IFN-γ production by CD8 + T cells under the same co-culture conditions as in G. IFN-γ levels were measured and analyzed as in G. E Relative apoptosis rate of gallbladder cancer cells co-cultured with CD8 + T cells, with or without atezolizumab (10 µg/mL). F Schematic of the human peripheral blood mononuclear cell (huPBMC) humanized mouse model. Mice were engrafted with huPBMCs, followed by subcutaneous injection of SRSF2-knockdown or control NOZ cells to establish the gallbladder cancer model. G Representative gross images of resected tumors from the huPBMC model at the study endpoint. H Tumor growth kinetics of xenografts in the huPBMC model. Data are presented as mean tumor volume ± SEM (n = 6 per group). I Comparison of final tumor weights from the huPBMC model. J Representative multiplex immunohistochemistry images of xenograft tumor sections from the huPBMC model, stained for CD8 (T cells, green) and PD-L1 (tumor cells, red). Nuclei are counterstained with DAPI (blue). K Quantitative analysis of CD8 + T cell infiltration and PD-L1 expression from mIHC images in (N).

SRSF2 drives PD-L1 expression via 3’-UTR-dependent mRNA stabilization without direct RNA binding

To elucidate the mechanisms by which SRSF2 regulates PD-L1 expression, we first assessed protein stability using cycloheximide chase assays, which revealed that SRSF2 knockdown did not alter PD-L1 protein degradation kinetics (Supplementary Fig. S3A and B). Furthermore, luciferase reporter assays demonstrated that SRSF2 increased PD-L1 3’-UTR-driven luciferase activity, while showing no significant effect on promoter-driven reporter activity or 5’-UTR-driven reporter activities (Fig. 3A and Supplementary Fig. S3C). These findings collectively indicated that SRSF2 post-transcriptionally upregulated PD-L1 expression through a 3’-UTR-dependent mechanism, rather than by affecting its transcriptional initiation, translation mediated by the 5’-UTR, or protein stability. Given that the 3’-UTR serves as a critical regulatory domain in eukaryotic mRNA, exerting multifaceted control over post-transcriptional processes [19–21]. To determine whether SRSF2 regulates PD-L1 expression through mRNA stabilization, we performed RNA decay analysis using actinomycin D-mediated transcriptional arrest. Pharmacological inhibition of de novo RNA synthesis revealed that SRSF2 substantially prolonged PD-L1 mRNA half-life (Fig. 3B and Supplementary Fig. S3D). To investigate whether this regulatory mechanism is mediated through the PD-L1 3’-UTR, we engineered GFP-PD-L1 reporters that contain or do not contain PD-L1 3’-UTR (Fig. 3C)[22]. Flow cytometry and western blot analyses consistently demonstrated that the inclusion of the PD-L1 3’-UTR significantly diminished GFP-PD-L1 protein levels (Fig. 3D and E, and Supplementary Fig. S3E and F), supporting the role of this UTR in post-transcriptional regulation. These findings confirm that SRSF2 modulated PD-L1 expression via its 3’-UTR. Intriguingly, RIP assays revealed no direct binding between SRSF2 and PD-L1 mRNA (Fig. 3F), suggesting an indirect regulatory mechanism. Collectively, these data demonstrate that SRSF2 post-transcriptionally upregulates PD-L1 by enhancing the stability of its mRNA through an indirect mechanism that requires the PD-L1 3’-UTR.

Fig. 3. SRSF2 drives PD-L1 expression via 3’-UTR-dependent mRNA stabilization without direct RNA binding.

A Effect of SRSF2 knockdown on PD-L1 promoter, 5’-UTR and 3’-UTR reporter activity. Luciferase reporters fused with the PD-L1 promoter region, or 5’-UTR the PD-L1 3’-UTR were co-transfected into gallbladder cancer cells with either SRSF2-targeting or control siRNAs. Luciferase activity was measured 48 h post-transfection and normalized to the control group. B SRSF2 regulates PD-L1 mRNA stability. Control and SRSF2-knockdown GBC cells were treated with Actinomycin D (5 µg/mL) to halt transcription. PD-L1 mRNA levels at the indicated time points were quantified by qRT-PCR and normalized to time zero. The mRNA decay curves are shown (mean ± SEM, n = 3). The half-life of PD-L1 mRNA was significantly shortened upon SRSF2 knockdown. C Schematic of the GFP-PD-L1 reporter constructs. The “NON” reporter contains the coding sequence for a GFP-PD-L1 fusion protein, while the “UTR” reporter additionally includes the full-length PD-L1 3’-UTR. D Flow cytometric analysis of cell surface GFP-PD-L1 expression. GBC cells were transfected with the indicated NON or UTR reporter plasmids, along with SRSF2 or control siRNAs. E Western blot analysis of total GFP-PD-L1 protein expression. GBC cells were transfected as in D. Whole-cell lysates were immunoblotted with antibodies against GFP and GAPDH (loading control). SRSF2 knockdown specifically reduced the protein level of the UTR reporter, but not the NON reporter. F RNA immunoprecipitation was performed in GBC cells using an anti-SRSF2 antibody or control IgG. The co-precipitated PD-L1 mRNA was quantified by qPCR and is presented as the fold enrichment relative to the IgG control.

SRSF2 regulates hnRNPD exon skipping

Building on previous findings suggesting potential functional interplay between SRSF2 and PD-L1, we hypothesized the existence of an intermediate regulatory factor. Given that SRSF2 functions as an RNA-binding protein, we performed RIP-seq using an anti-SRSF2 antibody (Fig. 4A). The RIP-seq analysis revealed that SRSF2 predominantly binds to protein-coding transcripts and exonic regions (Supplementary Fig. S4A and B). Through stringent dual-filter criteria (log2 fold change ≥ 12 and p < 0.05), we identified 318 transcripts with statistically significant differential binding patterns (Supplementary Table 6). To delineate the global splicing regulatory role of SRSF2, we conducted mRNA-seq coupled with rMATS analysis following SRSF2 knockdown (Fig. 4B). Alternative splicing profiling revealed that SRSF2 depletion predominantly induced exon skipping events, with 301 and 302 differential exon skipping events identified in two independent knockdown replicates (Supplementary Fig. S4C, Supplementary Table 7 and Supplementary Table 8). To identify PD-L1 3’-UTR-interacting regulatory factors, we performed RNA pulldown using in vitro transcribed biotinylated PD-L1 3’-UTR probes followed by TMT-labeled quantitative mass spectrometry. Applying stringent selection criteria ( | log2 fold-change | ≥ 4, p < 0.05), we identified 543 proteins demonstrating significant differential binding to the PD-L1 3’-UTR. Through integrated analysis of three complementary approaches, we identified heterogeneous nuclear ribonucleoprotein D (hnRNPD) as a key downstream effector of SRSF2 (Fig. 4D). hnRNPD, a multifunctional RNA-binding protein, orchestrates gene expression through dynamic interactions with U/AU-rich elements (AREs) in both coding and noncoding RNAs [23, 24]. As demonstrated in Fig. 4E, the hnRNPD gene produces four functionally distinct protein isoforms (P37, P40, P42, P45) through differential splicing of exon 2 and exon 7 during mRNA maturation [25]. Bioinformatic analysis of RNA-seq data revealed that SRSF2 promoted exon 7 skipping. To validate this splicing alteration, we designed exon-spanning primers flanking exon 7 and performed RT-PCR amplification. Sanger sequencing of gel-purified amplicons confirmed both the exon 7 splicing pattern and primer specificity (Fig. 4F). RIP assays demonstrated direct binding of SRSF2 to hnRNPD pre-mRNA (Supplementary Fig. S4D). Next, to resolve the composition of hnRNPD splicing variants in gallbladder cancer, we designed exon 2-spanning primers for RT-PCR analysis. Electrophoretic separation revealed a single amplification product containing exon 2 (Fig. 4G). This suggests that the P37 and P42 isoforms, which lack exon 2, are likely not expressed in this malignancy. We then proceeded to validate the regulatory effect of SRSF2 on hnRNPD splicing. We performed RT-PCR and western blot analysis following SRSF2 knockdown and overexpression. The RT-PCR results demonstrated that SRSF2 significantly promoted exon 7 skipping of hnRNPD. Intriguingly, western blot revealed only two distinct protein isoforms in gallbladder cancer cells, suggesting selective expression of two predominant hnRNPD isoforms (P40 (exon 7-skipped) and P45 (full-length)) in this malignancy (Fig. 4H and Supplementary Fig. S4E). Besides we found that knockdown of hnRNPD also does not affect cell proliferation (Supplementary Fig. S4F), which is consistent with the phenotype observed upon SRSF2 knockdown. We systematically analyzed hnRNPD isoform expression patterns in 12 paired gallbladder carcinoma and adjacent non-tumor tissues. Notably, carcinoma tissues exhibited a significant shift towards P40 predominance, suggesting tumor-specific dysregulation of hnRNPD splicing favoring the exon 7-skipped isoform (Fig. 4I and Supplementary Fig. S4G). Consistent with this splicing shift, western blot analysis confirmed that SRSF2 protein expression was significantly upregulated in tumor tissues compared to matched non-tumor samples (Supplementary Fig. S4H). Furthermore, correlation analysis across these clinical specimens revealed a statistically significant negative relationship between SRSF2 protein levels and the P45 isoform (quantified as P45 PSI values) (Supplementary Fig. S4I), reinforcing the role of SRSF2 in promoting exon 7 skipping of hnRNPD in vivo. To delineate the precise binding sites of SRSF2 on hnRNPD pre-mRNA, we performed RNA pulldown assays which demonstrated SRSF2’s specific association with exon 6 of hnRNPD (Fig. 4J). Through RBPmap (http://rbpmap.technion.ac.il/index.html) bioinformatic analysis, four potential SRSF2-binding motifs (site1-site4) were identified within exon 6, showing partial sequence overlap (Supplementary Fig. S4J). To functionally validate these sites, we engineered five mutant minigene constructs (Mut1-Mut5) with site-specific mutation spanning the predicted motifs (Fig. 4K). In SRSF2-knockdown cells transfected with these mutants, only Mut1and Mut2 abolished SRSF2-mediated exon 7 skipping, confirming site1 as the critical binding locus (Fig. 4L).

Fig. 4. SRSF2 regulates hnRNPD exon skipping.

A RNA immunoprecipitation was performed in NOZ cells using an anti-SRSF2 antibody. The co-precipitated RNAs were subjected to high-throughput sequencing. The diagram illustrates the experimental procedure for identifying direct targets of SRSF2. B mRNA sequencing was conducted in NOZ cells following SRSF2 knockdown. C Biotin-labeled PD-L1 3’-UTR RNA was incubated with NOZ cell lysates. Proteins bound to the RNA were pulled down and identified by mass spectrometry. The schematic shows the process of isolating and identifying PD-L1 3’-UTR-binding proteins. D Venn diagram illustrates the overlap among exon skipping events from mRNA-seq B, SRSF2-bound transcripts from RIP-seq A, and proteins binding to PD-L1 3’-UTR from RNA pulldown C. This integrated approach identified hnRNPD as a critical target. E The diagram shows the domain structure and differences of the four hnRNPD isoforms (P37, P40, P42, P45), highlighting the inclusion or exclusion of exon 7 or exon 2. F (Left Panel) RT-PCR analysis of hnRNPD reveals two distinct isoforms in untreated NOZ cells, separated by agarose gel electrophoresis. (Right Panel) Sanger sequencing chromatograms of the purified upper and lower bands. The sequence alignment confirms that the difference corresponds to the exclusion/inclusion of exon 7. G RT-PCR assay using primers designed for exon 2 in GBC cells. H GBC cells were transfected with control or SRSF2-targeting siRNAs. Upper panel: RT-PCR analysis of hnRNPD shows a shift in splicing towards the exon 7-skipping upon SRSF2 knockdown. Lower panel: western blot analysis confirms the corresponding change in hnRNPD protein isoform levels. I RT-PCR and western blot analysis of hnRNPD isoforms in paired tumor (T) and adjacent normal (N) tissues from gallbladder cancer patients. J RNA pulldown assay was performed using in vitro transcribed biotin-labeled hnRNPD pre-mRNA fragments. Western blot with an anti-SRSF2 antibody confirms the direct binding of SRSF2 to specific regions of hnRNPD pre-mRNA. K Diagram of the minigene with mutations introduced at the predicted SRSF2 binding site within the hnRNPD pre-mRNA, designed to disrupt SRSF2 binding. L Wild-type (WT) or mutant (MUT) minigenes were transfected into NOZ cells along with control or SRSF2-targeting siRNAs. RT-PCR analysis shows that SRSF2 knockdown no longer alters hnRNPD splicing in the MUT minigene, confirming the functional importance of this binding site.

The P45 isoform of hnRNPD specifically binds to the PD-L1 mRNA 3’-UTR to promote its degradation

To investigate the impact of SRSF2-mediated hnRNPD exon skipping on PD-L1 expression, we constructed P40 and P45 overexpression plasmids. Western blot analysis and flow cytometry assays revealed distinct biological effects: while P40 overexpression showed no significant influence on PD-L1 protein levels, P45 overexpression resulted in a marked reduction of PD-L1 expression (Fig. 5A and Supplementary Fig. S5A). To further validate this observation, we designed exon 7-targeting siRNAs specifically against P45. Knockdown of P45 led to a substantial increase in PD-L1 protein content (Supplementary Fig. S5B and C). RNA stability experiments demonstrated that P45, but not P40, significantly accelerated PD-L1 mRNA degradation, suggesting a post-transcriptional regulatory mechanism underlying this differential effect (Fig. 5B). To assess the isoform-specific antitumor effects in an immunocompetent context, we overexpressed P45 or P40 in the gallbladder cancer huPBMC mouse model. We found that P45 overexpression significantly suppressed tumor growth, whereas P40 overexpression showed no notable effect on tumor volume (Supplementary Fig. S5D–G). Subsequently, RNA pulldown assays revealed an interaction between the PD-L1 3’-UTR and only an hnRNPD isoform (Fig. 5 C). Band pattern analysis suggested a predominant association with the P45 isoform (based on molecular weight-correlated electrophoretic mobility), though additional validation experiments are required to confirm this isoform-specific interaction. hnRNPD regulates mRNA stability by recognizing AREs in the 3’-UTR of target genes[26]. These regulatory sequences (typically spanning 50-150 nucleotides) derive their name from their characteristic adenosine- and uridine-rich composition. Structurally, AREs frequently contain clustered AUUUA pentamer motifs embedded within uracil-enriched regions of the 3’ untranslated region [27]. Using the RBPmap platform, we identified two potential candidate regions within the PD-L1 3’-UTR. Based on these findings, we designed mutant luciferase reporter plasmids and specific probes for functional validation (Supplementary Fig. S5H). Luciferase reporter assays revealed that overexpression of P45 isoform significantly reduced luciferase activity directed by the PD-L1 3’-UTR, whereas P40 isoform exhibited no regulatory effect. Notably, mutation of the predicted key binding sites partially reversed the regulatory effects of P45 overexpression (Fig. 5D). To further characterize the RNA-protein interaction, we purified P45 and P40 proteins and performed EMSA using probes spanning the predicted binding regions (Supplementary Fig. S5I). The EMSA results demonstrated specific binding capacity between P45 protein and ARE region of PD-L1 3’-UTR, while P40 showed no detectable binding activity (Fig. 5E and F, Supplementary Fig. S5J). Furthermore, RNA immunoprecipitation assays following individual overexpression of the P45 and P40 isoforms confirmed that only P45 specifically associated with PD-L1 mRNA (Supplementary Fig. S5K). These collective findings suggest that the hnRNPD P45 isoform specifically interacts with the ARE motif in PD-L1 3’-UTR to promote transcript degradation. To validate that SRSF2 regulates PD-L1 expression primarily through the hnRNPD P45 isoform, we performed a rescue experiment. Given that SRSF2 acts as an upstream negative regulator of P45, we reasoned that if this axis is the dominant pathway, then concurrent knockdown of P45 should attenuate the PD-L1 reduction induced by SRSF2 knockdown. As shown in Fig. 5G, the concurrent knockdown of P45 significantly rescued the suppression of PD-L1 expression caused by SRSF2 deficiency. Building on our previous findings, we designed antisense oligonucleotides (ASOs) targeting the SRSF2-regulated splicing site in hnRNPD pre-mRNA to inhibit exon skipping and promote P45 isoform production for PD-L1 degradation. Four ASOs (ASO1-4) were synthesized based on the predicted splicing regulatory elements (Fig. 5H). Functional analyses demonstrated that ASO3 and ASO4 significantly reduced hnRNPD exon skipping and downregulated PD-L1 expression in tumor cells (Fig. 5I and Supplementary Fig. S5L). In vitro experiments revealed that ASO3/4 treatment enhanced T cell-mediated tumor cell killing (Supplementary Fig. S5M). Using the gallbladder cancer huPBMC model, we further observed that ASO3/4 administration markedly suppressed tumor growth and promoted intratumoral T cell infiltration (Fig. 5J–N). In conclusion, targeted ASO therapy blocking SRSF2-driven hnRNPD exon skipping promotes P45-dependent PD-L1 mRNA decay and boosts T-cell cytotoxicity.

Fig. 5. The P45 isoform of hnRNPD specifically binds to the PD-L1 mRNA 3’-UTR to promote its degradation.

A Flow cytometry analysis of membrane PD-L1 expression in NOZ cells following overexpression of the P40 or P45 isoform. Data are presented as mean fluorescence intensity ± SD from three independent experiments. B PD-L1 mRNA stability assay. Gallbladder cancer cells overexpressing P40 or P45 were treated with actinomycin D (5 µg/mL) to block transcription. PD-L1 mRNA levels were measured by qRT-PCR at indicated time points and normalized to time zero. Data represent mean ± SEM (n = 3). C RNA pull-down assay using in vitro transcribed, biotinylated probes spanning the 5’-UTR, CDS, or 3’-UTR of PD-L1 incubated with whole-cell lysates. Bound proteins were analyzed by western blot with anti-hnRNPD antibody. The 3’-UTR specifically enriched hnRNPD isoforms. D Luciferase reporter assays with wild-type (WT) or ARE-mutated (Mut) PD-L1 3’-UTR constructs in cells overexpressing P40 or P45. Relative luciferase activity was measured 48 h post-transfection. Data show mean ± SD (n = 3). E, F Electrophoretic mobility shift assay showing direct binding of purified recombinant P45 protein to biotin-labeled RNA probes containing Site 1 E or Site 2 F from the PD-L1 3’-UTR. Specific binding was confirmed by cold probe competition. G Western blot analysis of PD-L1 protein expression in NOZ cells following siRNA-mediated knockdown of SRSF2, P45, or both. GAPDH served as a loading control. Concurrent P45 knockdown rescued the PD-L1 reduction induced by SRSF2 deficiency. H Schematic illustration of antisense oligonucleotides targeting the SRSF2-regulated splicing site in hnRNPD pre-mRNA. Four ASOs (ASO1-4) were designed to block exon skipping. I Flow cytometry analysis of membrane PD-L1 expression in NOZ cells treated with control or targeting ASOs for 48 h. Data are presented as mean fluorescence intensity ± SD (n = 3). J Tumor growth curves of huPBMC model xenografts treated with control or targeting ASOs. Data points represent mean tumor volume ± SEM (n = 6 mice per group). K Comparison of final tumor weights from the huPBMC model. Data are shown as mean ± SEM (n = 6). L Representative gross images of resected subcutaneous tumors from each treatment group at the study endpoint (day 28). M Representative multiplex immunohistochemistry images of xenograft tumor sections stained for PD-L1 (green) and CD8 (red). Nuclei were counterstained with DAPI (blue). N Quantitative analysis of PD-L1 expression (mean fluorescence intensity) and CD8 + T cell infiltration in tumor tissues from N Data represent mean ± SEM.

Discussion

Gallbladder cancer represents the most prevalent malignant tumor of the biliary tract system in humans [28]. Despite current therapeutic strategies including conventional chemotherapy, immunotherapy, and targeted therapies, the prognosis for GBC patients remains dismal [29, 30]. This underscores an urgent need for developing novel therapeutic agents and treatment modalities. Emerging evidence highlights the critical role of aberrant alternative splicing in tumorigenesis and cancer progression [31–33].

Recently, the efficacy of immune checkpoint inhibitors targeting PD-1/PD-L1 has been demonstrated in various cancers, including biliary tract cancers. However, the low incidence of GBC poses a significant obstacle to conducting dedicated clinical trials for novel therapeutic agents. While recent studies have revealed PD-L1 could be modulated through the circFOXP1-miR-4477a axis and OLFM4-mediated MAPK-AP1 signaling, the potential involvement of alternative splicing in PD-L1 dysregulation remains unexplored. In this study, we investigated the regulatory role of aberrant splicing on PD-L1 in gallbladder cancer. Through integrated analysis of gallbladder cancer transcriptomic sequencing data, we have identified the RNA-binding protein SRSF2 as a regulator of PD-L1. Through in vitro and in vivo experiments, we have validated that SRSF2 promotes immune escape in gallbladder cancer via PD-L1-mediated mechanisms.

Strikingly, rather than directly interacting with PD-L1 mRNA, SRSF2 exerts its immunomodulatory effects indirectly by orchestrating the alternative splicing of a downstream effector, hnRNPD. The hnRNPD gene undergoes alternative splicing to generate four distinct isoforms (P37/P40/P42/P45), characterized by differential inclusion/exclusion of exon 2 and exon 7 [25]. hnRNPD primarily modulates RNA stability and translational efficiency through binding to characteristic cis-acting motifs located within 3’-UTR of transcripts, constituting its fundamental mechanism in post-transcriptional regulation [34]. The structural heterogeneity among these four isoforms may confer differential transcript recognition capabilities. Notably, the P42 isoform exclusively regulates FGF9 mRNA stability through an ARE-mediated mRNA decay pathway, demonstrating isoform-specific regulatory selectivity [35]. In GBC, we demonstrated that SRSF2 promoted exon 7 skipping of hnRNPD. Subsequent RT-PCR analysis coupled with western blot validation revealed exclusive expression of two hnRNPD isoforms (P45 and P40) in both malignant and non-neoplastic gallbladder tissues.

3’-UTR of mRNA plays pivotal roles in post-transcriptional regulation by modulating mRNA stability, translation efficiency, and subcellular localization [36, 37]. Key regulatory elements within 3’-UTRs, such as AU-rich elements and microRNA binding sites, dynamically interact with RNA-binding proteins and non-coding RNAs to fine-tune gene expression [38–40]. hnRNPD dynamically modulates mRNA stability through selective interaction with AREs in 3’-UTR, exerting context-dependent stabilization or destabilization effects on target transcripts [41]. Our further studies demonstrated that P45 facilitated PD-L1 mRNA degradation, while P40 showed no significant effect. This finding is consistent with the SRSF2-mediated exon skipping mechanism, where SRSF2 reduced P45 production to suppress PD-L1 mRNA degradation. Further investigations revealed that P45 specifically bound to the AREs motif within the PD-L1 3’-UTR, a functional interaction absent in P40. The functional divergence among hnRNPD isoforms is well-documented in the literature [35, 42]. This discrepancy is likely attributable to the 49-amino-acid insertion encoded by exon 7, which is retained in P45 but excluded in P40. Although located outside the classical RNA recognition motifs (RRMs), this domain appears essential for optimal RNA-binding affinity and subsequent transcript decay in the context of PD-L1. This finding resonates with emerging evidence that intrinsically disordered regions or auxiliary domains in RNA-binding proteins are crucial for their context-dependent target recognition and phase separation, warranting future structural investigations.

Importantly, the identification of this splicing-centric regulatory axis offers a targetable vulnerability for GBC immunotherapy. Traditional pharmacological interventions often struggle to drug RNA-protein interactions or specific splicing isoforms. ASOs are increasingly utilized to correct pathogenic splicing dysregulation for therapeutic purposes [43]. In this study, we designed splice-switching ASOs specifically targeting the SRSF2-regulated hnRNPD exon 7 splicing junction. These ASOs demonstrated robust suppression of PD-L1 protein levels in preclinical gallbladder cancer models, providing a novel therapeutic avenue for gallbladder cancer immunotherapy by reprogramming immune checkpoint expression. However, while these findings highlight the therapeutic potential of splicing modulation, substantial translational challenges, including pharmacokinetic optimization, toxicity profiles, and validation of human-specific splicing conservation must be rigorously addressed before clinical translation, as current evidence is limited to murine in vivo studies.

Collectively, our work defines a complete regulatory pathway from SRSF2-mediated splicing to PD-L1 modulation and immune evasion in GBC. The discovery of this pathway, and its susceptibility to ASO-based intervention, reveals a new targetable axis for cancer immunotherapy.

Author contributions

WG, Y-JS, L-JP, and PD conceived the project, secured funding, supervised the research, and critically revised the manuscript. CZ, LJ, M-YW, and Q-YT designed and performed the majority of the experiments, analyzed the data, and wrote the original draft of the manuscript. J-WZ and Z-YY provided technical support for the in vitro and in vivo functional assays. JG and X-LS assisted with clinical sample collection, processing, and bioinformatic analyses. WG and CZ reviewed and edited the final manuscript. All authors read and approved the final manuscript.

Funding

This study was supported by the National Natural Science Foundation of China (NO. 82472646, 82473043, 82403040, 82173081, 82403527). Shanghai Anticancer Association EYAS PROJECT (SACA-CY23C11). Talent Program of Shanghai Health Care Commission (2022YQ002). Shanghai Jiao Tong University School of Medicine PhD Student Scientific and Technological Innovation Development Fund (25KCPYZD03).

Data availability

All data generated or analyzed during this study are included in this published article. The processed data are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Ethics approval and consent to participate

All murine experimental procedures were ethically reviewed and authorized by the Ethics Committee of Xinhua Hospital Affiliated to Shanghai Jiao Tong University School of Medicine (Approval number: XHEC-NSFC-2024-058). All methods were performed in accordance with the relevant guidelines and regulations, including the institutional guidelines for the care and use of laboratory animals.

Consent for publication

All authors have seen and approved the manuscript and consent publication.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41388-026-03898-3.