Ubenimex synergizes with the PD-L1 blockade in gastric cancer by competitively binding LAP3 with UBE3A

Abstract

Immune checkpoint inhibitors (ICIs), particularly PD-1/PD-L1 blockade, represent a cornerstone of treatment for advanced gastric cancer (GC). However, their clinical efficacy is hampered by low response rates and the development of both primary and acquired resistance, underscoring the need for innovative combination therapies. In this study, we investigate the potential of Ubenimex, an immunomodulator and inhibitor of leucyl aminopeptidase 3 (LAP3), in enhancing the therapeutic efficacy of PD-L1 blockade in GC. Using a syngeneic GC mouse model, we demonstrate that Ubenimex significantly augments the efficacy of anti-PD-L1 therapy. We further explore the role of LAP3 in GC progression and find that it is highly expressed in both GC tissues and cells, with elevated LAP3 levels correlating with poor prognosis. Functionally, LAP3 facilitates immune evasion through impaired CD8⁺ T cell infiltration and cytotoxicity in the GC tumor microenvironment (TME). Notably, our findings reveal that LAP3 enhances PD-L1 expression by binding to UBE3A, an E3 ubiquitin ligase. Ubenimex disrupts the LAP3-UBE3A interaction, leading to restored UBE3A-mediated ubiquitination and degradation of PD-L1. This mechanism reinvigorates CD8⁺ T cell infiltration and cytotoxic activity within the TME, thereby overcoming resistance to anti-PD-L1 therapy. In conclusion, our study provides a strong rationale for the synergistic potential of Ubenimex in combination with PD-1/PD-L1 blockade, offering a promising strategy to overcome current limitations of ICIs therapy in GC patients.

Introduction

China, with one of the highest gastric-cancer incidence and mortality rates worldwide, bears the heaviest disease burden. Its poor prognosis is attributed to marked tumor heterogeneity and intricate immune-evasion mechanisms [1, 2]. Recent breakthroughs in immunotherapy, particularly ICIs, have significantly altered the therapeutic landscape, yielding notable improvements in patient outcomes. For instance, the KEYNOTE-811 trial established a combination of immunotherapy and chemotherapy as the first-line treatment for HER2-positive GC patients with a PD-L1 combined positive score (CPS) ≥ 1 [3]. Similarly, the COMPASSION-15 trial demonstrated that PD-1/CTLA-4 bispecific antibodies provide benefits even for HER2-negative advanced GC, including in PD-L1-low/negative populations [4]. Despite these advances, immunotherapy’s efficacy remains limited due to both primary and acquired resistance, which occur in a majority of patients [5, 6].

Primary resistance arises from tumor-intrinsic mechanisms such as impaired antigen presentation, dysregulated signaling pathways (e.g., MAPK, PI3K, WNT, and IFN), and constitutive PD-L1 overexpression [7, 8]. These mechanisms prevent immune recognition and negate the therapeutic potential of immunotherapy. Acquired resistance, on the other hand, develops progressively during treatment through processes such as further antigen loss, compensatory upregulation of alternative immune checkpoints (e.g., TIM-3, LAG-3), sustained PD-L1 expression, T-cell exhaustion, and tumor microenvironment (TME) remodeling. Consequently, immunotherapy, though initially effective, ultimately fails to maintain durable clinical responses [8–10].

Current clinical strategies aim to overcome these resistance mechanisms by targeting both primary and acquired resistance factors. For primary resistance, agents such as HDAC inhibitors or IFN-γ enhance antigen presentation, leading to an objective response rate (ORR) of approximately 21% when combined with PD-1 inhibitors [11]. PI3Kα inhibitors extend progression-free survival (PFS) to 5.7 months in PTEN-deficient patients [12], while CLDN18.2-targeted therapies have shown ORRs as high as 48% [13, 14]. For acquired resistance, multi-target immune checkpoint blockade (e.g., combining TIM-3/LAG-3 inhibitors with PD-1 inhibitors) improves ORR to 18–32%. However, these therapies face significant challenges, including severe adverse effects (≥ Grade 3 adverse events occurring in 30–50% of patients) and transient efficacy (median PFS typically < 7 months) [11–14]. These limitations underscore the urgent need for novel combination therapies, particularly those driven by biomarker-guided precision medicine, to overcome GC immunotherapy resistance.

Ubenimex, a clinically approved chemoradiotherapy adjuvant [15–19], has emerged as a promising candidate to enhance immune responses in cancer treatment. It exerts significant immunomodulatory effects by increasing the susceptibility of GC cells to cytotoxic T-cell-mediated killing [20], and enhancing T/B cell responsiveness as well as macrophage activation [21]. Notably, its efficacy in augmenting the effectiveness of immunotherapy has been documented in acute non-lymphocytic leukemia, particularly in elderly patients [21]. Despite these encouraging results, the potential of Ubenimex to synergize with PD-L1 blockade has not been fully explored.

In this study, we observed robust inhibitory effects of Ubenimex in combination with anti-PD-L1 in a syngeneic GC mouse model. This promising synergy motivated us to investigate the underlying mechanisms. Ubenimex functions as a specific inhibitor of leucine aminopeptidase 3 (LAP3), a key enzyme involved in protein processing and cellular homeostasis [22]. LAP3, an exopeptidase that cleaves N-terminal leucine residues, has been implicated in regulating crucial cellular processes such as antigen presentation, vesicular trafficking, meiosis, and oxidative stress responses [22–24]. Moreover, recent studies suggest that LAP3 may act as an oncogenic driver in multiple cancers [25–31], although its role in GC remains poorly understood. Given this background, we aim to address two critical questions: how LAP3 contributes to GC progression, and how Ubenimex enhances anti-PD-L1 efficacy through LAP3 targeting. By investigating these molecular interactions, our study seeks to provide valuable insights into LAP3’s role in GC pathogenesis and offer potential avenues for the development of more effective, biomarker-driven immunotherapies.

Methods

Plasmids, cells and animals

Human full-length LAP3 cDNA was amplified from GES-1 cells by RT-PCR and inserted into pEGFP-C1-3×Flag to generate the GFP-LAP3 overexpression construct. Short hairpin RNAs (shRNAs) targeting human LAP3 were annealed and cloned into plvx-zsGreen-scramble. The primers used for cloning are listed in Table S1. LAP3-specific siRNAs, mCherry-UBE3A, HA-PD-L1, c-Myc-Ub, murine LAP3 overexpression and knockdown lentiviruses were synthesized and validated by Tsingke Biotechnology (Beijing, China).

The human gastric-cancer lines AGS (RRID: CVCL_0139) and HGC-27 (RRID: CVCL1279), together with the immortalized normal gastric epithelial line (GES-1 RRID: CVCLEQ22), were purchased from the National Collection of Authenticated Cell Cultures (Shanghai, China) in March 2022. The human GC line NCI-N87 (RRID: CVCL1603) and the murine MFC (RRID: CVCL5J48) line were obtained from the Cell Bank, Chinese Academy of Sciences (Beijing, China), on 11 September 2021. Upon receipt, all lines were verified mycoplasma-free and authenticated by short-tandem-repeat (STR) profiling (BIOWING, Shanghai).

Male 615 strain mice (4 weeks old) were obtained from the Chinese Academy of Medical Sciences Institute of Hematology (Tianjin, China) and maintained under specific pathogen-free (SPF) conditions with controlled light/dark cycles. Mice were randomly allocated to treatment groups and provided ad libitum access to food and water.

Bioinformatics analysis

RNA-seq and matched clinical data for 375 GC samples and 32 normal tissues were obtained from The Cancer Genome Atlas - Stomach adenocarcinoma (TCGA-STAD, https://portal.gdc.cancer.gov). 375 GC samples were divided into LAP3-High and LAP3-Low expression groups according to the median LAP3 expression levels. Differentially expressed genes (DEGs) were then performed with the limma R package (|log₂FC | > 1, FDR < 0.05), and annotated by Gene Ontology (GO) / Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment via clusterProfiler v4.8.3 (hypergeometric test). LAP3-associated processes were further investigated with Gene Set Enrichment Analysis (GSEA) (1000 permutations; FDR q-value). Single-cell sequencing data were analyzed from four GEO datasets-GSE163558 (three primary gastric cancers, one adjacent normal tissue, and six metastatic lesions), GSE184198 (one tumor and one matched normal tissue), GSE167297 (superficial and deep layers from five diffuse-type gastric cancers along with matched normal tissues), and GSE268238 (38 gastric cancers and 16 adjacent normal gastric mucosa). Immunophenoscore [32] (IPS) scores from TCIA (http://tcia.at/) were used to compare LAP3-high versus LAP3-low patients.

Clinical samples and immunohistochemistry (IHC)

Fifty-six paired GC and adjacent paracancerous formalin-fixed, paraffin-embedded specimens (collected between January 10 and December 29, 2016) were obtained from Heping Hospital Affiliated to Changzhi Medical College. IHC staining was performed to evaluate LAP3, PD-L1, CD8⁺ T cells, CD4⁺ T cells, Tregs, and M2 macrophage infiltration under standard procedure. Detection was performed using HRP-polymer and DAB. Staining was evaluated independently by two blinded pathologists using a semi-quantitative scoring system that combines the percentage of positive cells (1, <25%; 2, 25–75%; 3, >75%) and staining intensity (0, negative; 1, weak; 2, moderate; 3, strong). The final IHC score (range 0-9) was calculated by multiplying these values, with samples categorized as high expression (6-9) or low expression (0-5). Informed consent was obtained from all patients. This study has been approved by the Ethics Committee in Heping Hospital (2020018) and the Institutional Animal Care and Use Committee of Shanxi Medical University (SYD2024048), and antibody details are provided in Table S2.

Cell culture

HGC-27, NCI-N87, and MFC cells were maintained in RPMI 1640 medium (Boster, Wuhan, China; Cat# PYG0006), while GES-1 and AGS cells were cultured in DMEM (Gibco, Grand Island, NY, USA; Cat# C11995500BT) and F12K medium (Boster; Cat# PYG0036), respectively. All culture media were supplemented with 10% fetal bovine serum (Moregate Biotech, Melbourne, Australia; Cat# FBSF-500) and 100 U/mL penicillin-streptomycin. Cells were incubated at 37°C in a humidified atmosphere of 95% air and 5% CO₂.

Plasmid and lentiviral transfection

For plasmid transfection, Cells were seeded in 6-well plates and grown to 70-80% confluence. A transfection mixture containing 2.5 µg plasmid DNA and Lipofectamine 8000 (Cat# C0533-1.5 mL, Beyotime, Shanghai, China) in Opti-MEM was added dropwise. After 24–48 h incubation, cells were harvested for assays or subjected to 0.4 µg/mL G418 (Cat# HY-108718, MCE, NJ, USA) selection for stable line generation.

For lentiviral transduction, cells were infected at a Multiplicity of Infection (MOI) of 20 in the presence of 8 µg/mL polybrene. After 48 h later, 0.5 µg/mL puromycin (Cat# P8230, Solarbio, Beijing, China) was added for 7–10 days to select stable pools, which were then maintained in puromycin-containing medium and validated by RT-qPCR and western blotting.

RNA isolation and quantitative reverse transcription-polymerase chain reaction

Total RNA was extracted using an RNA extraction kit (CWBIO, Jiangsu, China; Cat# CW0597S) and reverse transcribed into cDNA using EasyQuick RT MasterMix (CWBIO; Cat# CW2019M) according to the manufacturer’s instructions. mRNA expression levels were quantified using SYBR Green-based quantitative reverse transcription PCR (RT-qPCR) with the QuantiNova SYBR Green PCR Kit (QIAGEN, Düsseldorf, Germany; Cat# 208054). Gene expression was calculated using the 2^(-ΔΔCt) method, with normalization to glyceraldehyde-3-phosphate dehydrogenase (GAPDH). Primer sequences are listed in Table S1.

Western blotting (WB)

GC cells and tissues were lysed in RIPA buffer (Cat# P0013B, Beyotime), sonicated and centrifuged. Protein concentrations were determined with the BCA kit (Cat# EC0001-B, SparkJade). Proteins were separated on 10% SDS-PAGE gels (Cat# PG212, EpiZyme), transferred to PVDF membranes (Cat# ED0004, SparkJade), blocked with 5% non-fat milk (Cat# AR0104, Boster) for 2 h at room temperature, incubated overnight at 4°C with primary antibodies and then for 1 h at room temperature with HRP-conjugated secondary antibodies. All antibodies are listed in Table S3. Signals were visualized using the ChemiDoc™ Imaging System (Bio-Rad, Hercules, USA).

In vivo ubiquitination assay

Ubiquitination assay was used to examine the ubiquitin levels of PD-L1. Cells were co-transfected for 48 h with HA-PD-L1 and c-Myc-Ub together with GFP-LAP3-3×FLAG or empty vector. After two PBS washes, cells were lysed in 2% SDS and heated at 95°C for 10 min to inactivate deubiquitinases. Lysates were sonicated, then diluted 10- to 40-fold in native lysis buffer (10 mM Tris-HCl pH 8.0, 150 mM NaCl, 2 mM EDTA, 1% Triton X-100, plus phosphatase and protease inhibitors). Following 1 h rotation at 4 °C and centrifugation, 10% of the supernatant was kept as input; the remainder was incubated overnight at 4°C with Pierce anti-HA magnetic beads (Thermo Scientific). Beads were washed three times with wash buffer (10 mM Tris-HCl pH 8.0, 1 M NaCl, 1 mM EDTA, 1% NP-40) and eluted in 2 × SDS loading buffer for WB.

Co-immunoprecipitation (Co-IP) and mass spectrometry (MS)

Co-IP followed by MS was used to identify LAP3-interacting proteins. Briefly, cells were transfected with GFP-LAP3-3×FLAG plasmids for 48 h. After washing three times with PBS, 2 × 10⁷ cells were lysed in 500 µL IP buffer (Cat# P0013, Beyotime, Beijing, China). Ten percent of the lysate was saved as input. The remainder was incubated with 20 µL GFP-Trap® magnetic beads (gtma-20, ChromoTek/Proteintech, Wuhan, China) for 3 h at 4°C. Beads were washed four times with wash buffer (10 mM Tris-HCl pH 7.5, 150 mM NaCl, 0.05% NP-40, 0.5 mM EDTA, 0.6 mM PMSF, 1× protease inhibitors). Finally, the eluted proteins were subjected to MS analysis at MetWare (Wuhan). All antibodies used are listed in Table S3.

Co-IP and WB

Co-IP was performed to assess the interactions of LAP3-UBE3A and UBE3A-PD-L1. Cells were lysed in 500 µL IP buffer (Cat# P0013, Beyotime, Beijing, China) containing 1× protease inhibitor cocktail. Ten percent of the lysate was reserved as input; the remainder was incubated overnight at 4°C with Pierce Protein A/G magnetic beads (Cat# 88802, Thermo Fisher, MA, USA) pre-crosslinked to the appropriate IP antibodies. After five washes with TBST, the beads were eluted in 80 µL 2 × SDS loading buffer and boiled for WB. All antibodies are listed in Table S3.

Truncation body construction

All UBE3A truncation mutants were synthesized by Tsingke Biotechnology (Beijing, China) and cloned into pcDNA3.1. Each construct contains an N-terminal HA-tag (nucleotide sequence: ATGTACCCATACGACGTACCAGATTACGCT) followed by the truncated UBE3A sequence, with a TAA stop codon at the C-terminus.

Immunofluorescence (IF) assay

IF was used to examine UBE3A-LAP3 co-localization. Cells grown on chamber slides were fixed with 4% PFA (10 min, RT), permeabilized with 0.5% Triton X-100 (10 min), and blocked with 5% FBS (1 h, RT). Primary antibodies were applied overnight at 4°C; after PBS washes, species-matched Alexa-Fluor secondary antibodies (1:150, Abcam) were added for 1 h in the dark. Slides were mounted in DAPI-containing anti-fade medium (Beyotime) and stored at 4°C. Images were acquired on an Olympus FV3000 or Leica TCS SP8 STED confocal microscope under identical settings; fluorescence signals were quantified in ImageJ (NIH) using FV31S-SW/FV31S-DT (Olympus) and LAS X (Leica) software. The antibodies used in this study are listed in Table S4.

T-cell killing and cytotoxicity assay in vitro

Tumor-bearing 615-line mice were euthanized by cervical dislocation. Splenocytes were harvested, red cells lysed and removed by washing, and CD8⁺ T cells were purified using the EasySep™ Mouse CD8⁺ T Cell Isolation Kit (STEMCELL, Vancouver, Canada). Effector CD8⁺ T cells and target cells (stable shLAP3 or control MFC cells) were added at an E:T ratio of 30:1 and co-cultured for 24 h at 37°C. Target-cell lysis was quantified using the CytoTox 96® Non-Radioactive Cytotoxicity Assay kit (G1780, Promega, Madison, WI, USA). CD8⁺ T-cell cytotoxicity (%) was calculated as: [(Experimental − Effector Spontaneous − Target Spontaneous) / (Target Maximum − Target Spontaneous)] × 100.

Enzyme-linked immunosorbent assay (ELISA)

The secretion of T-cell cytokines was measured using ELISA kits for mouse TNF-α (Cat# abs551104, Absin, Shanghai, China) and IFN-γ, according to the manufacturer’s instructions. The co-culture cell suspension was collected and centrifuged at 12,000 × g at 4°C for 5 min. The supernatant was then transferred to a new 1.5 mL microcentrifuge tube. RPMI 1640 medium served as the negative control. Enzyme activity was determined by measuring absorbance at 450 nm using a microplate reader.

Mouse models

All animal experiments were approved by the Institutional Animal Care and Use Committee of Shanxi Medical University (SYD2024048). This method employed the 615 inbred strain (brown) and its syngeneic mouse gastric carcinoma cell line, MFC. The 615 mouse was developed by the Institute of Hematology in Tianjin, China, through hybridization of Kunming mice with the C57BL strain and subsequent continuous sibling mating. It remains the sole fully immunocompetent gastric carcinoma model mouse created in China. The MFC cell line was derived from forestomach squamous cell carcinoma tissue isolated from this same strain. Inoculating 615 mice with MFC cells reliably establishes a transplanted gastric cancer model.

For the LAP3 knockdown study, 4-6-week-old 615 mice were randomly divided into shNC or shLAP3 groups (n = 6) and subcutaneously injected with 2.5×10⁶ cells in 200 μL PBS. Tumor volumes were measured every 3 days using the formula V = (L×W²)/2, and mice were euthanized by cervical dislocation under 2.5% isoflurane when tumors reached 1500 mm³, in compliance with AAALAC and GB/T 35892-2018 guidelines [33]. For therapeutic evaluation, 16 mice bearing LAP3-overexpressing tumors (±150 mm³) were randomized (Using Randomize R) into four groups (n = 4) [34–36]: PBS + DMSO, anti-PD-L1 (200 μg/mouse, i.p. every 3 days) [34, 35], Ubenimex (5 mg/kg, daily oral gavage), or combination therapy. During the 16-day treatment, two mice died from anesthesia complications and two from fighting. Survival studies used six mice per group following the same protocol.

Flow cytometry

Mouse spleens were dissociated into single-cell suspensions, incubated with a fixable viability dye, and blocked with Fc-receptor antibody for 15 min at RT. Cells were then stained with fluorophore-conjugated antibodies against CD45, CD3 and CD8 surface markers. After fixation and permeabilization, intracellular IFN-γ and granzyme B were labeled with the corresponding antibodies and quantified by flow cytometry. All flow-cytometry antibodies (details in Table S5) were obtained from Thermo Fisher Scientific (Massachusetts, USA).

Molecular docking

Three-dimensional structures of UBE3A, LAP3 and Ubenimex were retrieved from UniProt (https://www.uniprot.org, accessed 28 June 2024) and MedChemExpress (MCE, https://www.medchemexpress.cn). Each protein was protonated and optimized (hydrogen count, charge) using the Protonate3D protocol. Semi-flexible docking was performed in Molecular Operating Environment with UBE3A treated as flexible and LAP3 as rigid. Docking poses were analyzed in PyMOL 2.4 (Schrödinger, NY, USA).

Statistics and reproducibility

Data were analyzed with GraphPad Prism. Results are presented as mean ± SD from ≥ 3 independent biological replicates. Statistical tests are detailed in the figure legends. Briefly, two-group comparisons used Student’s t-test; multi-group comparisons employed one-way ANOVA followed by Bonferroni’s post hoc test, or two-way ANOVA followed by Tukey-Kramer post hoc test; all tests were two-sided, with P < 0.05 considered significant. Representative IF, IHC and WB images derive from three independent experiments. Gene-expression values were normalized to β-actin or GAPDH. Associations between LAP3 levels and clinicopathological variables were evaluated with Pearson’s χ² test. Survival curves were generated by Kaplan-Meier analysis; log-rank tests compared survival between LAP3-high and LAP3-low groups.

Results

Ubenimex sensitizes anti PD-L1 blockade in syngeneic mouse GC models

We first evaluated the anti-tumor efficacy and safety profile of Ubenimex in combination with PD-L1 antibody. The 615 mice bearing MFC-LAP3-overexpressing tumors were treated with Ubenimex and/or PD-L1 antibody for four cycles (Fig. 1A). Significant tumor regression was observed in all treatment groups compared to the control (n = 3 per group; Fig. 1B–E). Specifically, Ubenimex alone reduced tumor volume by 37.5%, PD-L1 antibody monotherapy by 69.8%, while the combination therapy achieved complete regression (100% reduction; Fig. 1F).

Fig. 1. Ubenimex sensitizes the anti-tumor effect of PD-L1 antibody in the syngeneic GC mouse models.

A Schematic illustration of the treatment protocols for each experimental group: A total of 615 mice were subcutaneously inoculated with MFC oeLAP3 cells. When subcutaneous tumors reached 150 mm³, mice were divided into four treatment groups: DMSO + PBS (control), anti-PD-L1 monotherapy, Ubenimex monotherapy, and anti-PD-L1 combined with Ubenimex. Anti-PD-L1 (200 μg/mouse) was administered intraperitoneally on days 8, 11, 14, and 17, while Ubenimex (5 mg/kg) was given daily via oral gavage. B Tumor growth curves in MFC-bearing 615 mice across treatment groups (n = 3/group). C–E Representative images of excised tumors and quantitative analysis of tumor weight (g) and volume (mm³). Red dashed circles highlight complete tumor regression in treated mice (n = 3/group). F Tumor growth inhibition ratio, calculated as (V⁰−V^e)/V[⁰], where V[⁰] = baseline volume and V^e = endpoint volume (n = 3/group). G Images of Spleens and splenic index (spleen weight/body weight × 100%) across groups. H Quantification of IFN-γ⁺ CD8⁺ and GZMB⁺ CD8⁺ of four groups. I Flow cytometry analysis of splenocytes stained with mAbs against CD45, CD8, IFN-γ, and GZMB. J Survival curves of four treatment groups (n = 6). Data are presented as mean ± SD (n = 3 mice per group for A–H; n = 6 mice per group for J. Statistical comparisons were performed with two-way ANOVA B–H and the log-rank test I. *P < 0.05, **P < 0.01, ***P < 0.001.

Flow cytometric analysis of splenic CD8⁺ T cells showed that the combination group exhibited significantly elevated proportions of IFN-γ⁺CD8⁺ and GZMB⁺CD8⁺ T cells compared to either monotherapy (P < 0.05, Fig. 1G–I). In contrast, no significant differences were observed in IFN-γ⁺CD4⁺ T cells, GZMB⁺CD4⁺ T cells, macrophages, or Tregs (Fig. S1A). We also aimed to analyze the proportions of immune cells (CD8⁺ T cells, CD4⁺ T cells, macrophages, and Tregs) in subcutaneous tumors. However, in the combination group, pronounced tumor shrinkage or complete regression after four treatment cycles (as shown in Fig. 1C) left an insufficient number of tumors (n = 0) to perform flow cytometry. Nevertheless, even in the absence of data from the combination therapy group, our results demonstrate that both monotherapies significantly increased the proportions of intratumoral IFN-γ⁺ CD8⁺ T cells and GZMB⁺ CD8⁺ T cells, suggesting a potential synergistic effect when combined (Fig. S1B).

A survival study confirmed that anti-PD-L1 monotherapy prolonged survival, while the combination therapy conferred a marked synergistic benefit (Fig. 1J). Taken together, these findings indicate that Ubenimex enhances PD-L1 antibody-mediated anti-tumor immunity by synergistically activating and potentiating the function of CD8⁺ T cells, providing a novel mechanistic basis for the observed therapeutic synergy in this mouse GC model.

Of note, body weight, ALT, and AST levels did not differ significantly (P > 0.05; Fig. S1C, D), demonstrating good tolerability in all treatment groups throughout the study.

LAP3 is upregulated in GC and positively correlated with poor prognosis

Due to Ubenimex is the specific inhibitor of LAP3 [22], we sought to investigate the role of LAP3 in GC. We first analyzed LAP3 expression using the TCGA database, confirming its upregulation across multiple tumor types, including GC (Fig. 2A, B). To validate the bioinformatic findings, we assessed LAP3 expression in GC cell lines and tissues. We observed higher mRNA and protein levels in GC cells and tissues compared to GES-1 cells (Fig. 2C, D) or para-tumor tissues (Fig. 2E). IHC analysis of tumor tissues from GC patients confirmed the elevated expression of LAP3, which was consistent with TCGA data (Fig. 2F). Additionally, survival analysis further supported its association with malignant progression in GC (Fig. 2G).

Fig. 2. LAP3 expression is upregulated in GC and strongly correlated with patient prognoses.

A Pan-cancer analysis of LAP3 expression levels in TCGA database. B Expression of LAP3 in TCGA STAD cohort. C, D Human normal stomach mucosal epithelial cells (GES-1) and GC cell lines (NCI-N87, AGS, HGC-27) were subjected to expression analysis of LAP3 using WB and quantitative reverse transcription PCR (RT-qPCR). E LAP3 protein expression was determined in GC tissues by WB. F The LAP3 level scored by intensity (0-3) and area (1-3) of the IHC staining in tissue microarrays constructed from 56 GC tissues and paired adjacent paraneoplastic tissues, with statistical analysis. Black scale bar = 1.25 mm. G Kaplan-Meier curves for overall survival of 56 GC patients based on LAP3 expression. H Tacked bar chart showing LAP3 expression in relation to T stage and lymph-node status (n = 56). I Quantitative analysis of tumor growth curves. J Representative images J, final tumor volume K, and weight L of a syngeneic 615 mouse model established with control or LAP3-knockdown MFC cells (n = 6/group). LAP3 knockdown significantly inhibited both tumor growth and weight. Data are presented as mean ± SD with three replicates. Two-tailed unpaired Student’s t-test was used in C, D, F, K, L ((*P < 0.05, **P < 0.01, ***P < 0.001). The log-rank test was used in G.

To investigate the correlation between LAP3 expression and clinicopathological indexes of GC, univariate analysis revealed a significant association between LAP3 expression and both pT and pN stages (P = 0.003 and 0.035) (Fig. 2H, Table 1). However, no significant correlations were found between LAP3 expression and patients’ age, gender, degree of differentiation, position, size, or TNM stage (Table 1). Furthermore, Cox univariate analysis indicated that LAP3 expression level was an independent prognostic factor for GC patients (Hazard ratio 2.078, 95% CI (1.040–4.153), P = 0.038; Table 2; Fig. S2A).

Table 1.

Correlation between LAP3 expression and clinicopathological characteristics.

Variable		total(n = 56)	LAP3-High expression(n = 30)	LAP3-Low expression(n = 26)	χ 2	P-Value
Age	<60	19 (33.9)	12 (40.0)	7 (26.9)	1.063	0.399
	≥60	37 (66.1)	18 (60.0)	19 (73.1)
Gender	Male	35 (62.5)	18 (60.0)	17 (65.4)	0.172	0.785
	Female	21 (37.5)	12 (40.0)	9 (34.6)
Differentiation	low-grade	17 (30.4)	11 (36.7)	6 (23.1)	1.321	0.581
	middle-grade	35 (62.5)	17 (56.7)	18 (69.2)
	high-grade	4 (7.1)	2 (6.7)	2 (7.7)
Position	upper	38 (67.9)	21 (70.0)	17 (65.4)	0.455	0.920
	middle	9 (16.1)	4 (13.3)	5 (19.2)
	low	9 (16.1)	5 (16.7)	4 (15.4)
Size	≤3	20 (35.7)	9 (30.0)	11 (42.3)	1.084	0.581
	3-5	23 (41.1)	14 (46.7)	9 (34.6)
	≥5	13 (23.2)	7 (23.3)	6 (23.1)
pT	T1-T2	13 (23.2)	2 (6.7)	11 (42.3)	9.926**	0.003
	T3-T4	43 (76.8)	28 (93.3)	15 (57.7)
pN	N0	29 (51.8)	10 (33.3)	19 (73.1)	8.508*	0.035
	N1	8 (14.3)	6 (20.0)	2 (7.7)
	N2	7 (12.5)	5 (16.7)	2 (7.7)
	N3	12 (21.4)	9 (30.0)	3 (11.5)
TNM stage	I-II	39 (69.6)	18 (60.0)	21 (80.8)	2.842	0.892
	III-IV	17 (30.4)	12 (40.0)	5 (19.2)

^*P < 0.05, **P < 0.005.

Table 2.

Univariate analysis of related factors affecting survival time of patients with GC.

Variable	Hazard Ratio(95%CI)	P-Value
Age(<60)	1(Reference)
≥60	0.598(0.302-1.186)	0.141
Gender(Male)	1(Reference)
Female	1.217(0.615-2.41)	0.573
Differentiation(low-grade)	1(Reference)	0.126
middle-grade	0.734(0.349-1.544)	0.415
high-grade	2.247(0.709-7.119)	0.169
Position(upper)	1(Reference)	0.314
middle	0.457(0.138-1.519)	0.201
low	1.303(0.561-3.026)	0.538
Size(≤3)	1(Reference)	0.679
3-5	0.835(0.398-1.753)	0.634
≥5	0.654(0.251-1.707)	0.386
PT(T1-T2)	1(Reference)
T3-T4	0.94(0.438-2.016)	0.874
PN(N0)	1(Reference)	0.180
N1	0.408(0.122-1.362)	0.145
N2	0.799(0.276-2.316)	0.680
N3	0.366(0.126-1.064)	0.065
TNM stage(I-II)	1(Reference)
III-IV	0.737(0.333-1.631)	0.452
expression level(low-expression)	1(Reference)
high-expression	2.078(1.04-4.153)	0.038

^*P < 0.05.

Using animal studies, we also confirmed the tumor-promoting function of LAP3. To this end, we monitored tumor growth in 615 mice bearing control or LAP3 knockdown MFC cells (Fig. 2J). A marked reduction in both tumor size and weight was observed in the LAP3-knockdown group (Fig. 2I–L), demonstrating that LAP3 depletion effectively suppresses gastric cancer cell proliferation in vivo.

LAP3 suppresses CD8⁺ T cells infiltration and cytotoxicity in GC

By promoting the processing and presentation of antigenic peptides, LAP3 often activates immune cells [22]. To clarify the immunomodulatory role of LAP3 in GC, we firstly used CIBERSORT algorithm to analyze the effects of LAP3 expression on CD8⁺ T cells infiltration in STAD in TCGA database and observed a positive correlation between LAP3 expression and CD8⁺ T cells infiltration (Fig. S3A). This apparent discrepancy prompted us to interrogate four independent single-cell RNA-seq cohorts (GSE163558, GSE184198, GSE167297, and GSE268238), which revealed a significant inverse relationship between LAP3 levels and CD8⁺ T cells infiltration (Fig. 3A). Therefore, we further explored the actual impact of LAP3 on CD8⁺ T cells infiltration. Consecutive IHC sections from a tissue microarray of 56 GC specimens was employed to assess the immune score of LAP3 expression versus T-cell infiltration, confirmed that high LAP3 expression was associated with markedly reduced CD8⁺ T cells infiltration (Fig. 3B). Concordantly, syngeneic mice bearing LAP3-knockdown MFC tumors exhibited a pronounced increase in CD8⁺ T cells, whereas CD4⁺ T cells, M2 macrophages, and Treg cells remained unchanged (Fig. S3B). Collectively, real-world data converge on the conclusion that LAP3 expression is negatively correlated with CD8⁺ T cells infiltration in GC.

Fig. 3. LAP3 suppresses CD8⁺ T cells infiltration and cytotoxicity in GC.

A Single-cell analysis of four independent datasets (GSE163558, GSE184198, GSE167297, GSE268238) revealed that high LAP3 expression groups exhibited reduced CD8⁺ T cell infiltration. B Representative IHC images of one case from each group and statistical analysis of LAP3, CD4, CD8, and CD163 were analyzed in LAP3 low-expression group (upper) (n = 22) and LAP3 high-expression group (lower) (n = 34) from 56 GC samples. Scale bar = 1.25 mm (2×), 100 μm (20×). C Knockdown of LAP3 was confirmed by WB in MFC cells. D Schematic of magnetic beads sorted CD8⁺ T cells for co-culture with MFC LAP3-knockdown and control cells. E Flow cytometry analysis of the positive expression rate of CD8⁺ T cells after sorting with magnetic beads. F CD8⁺ T cells cytotoxicity was determined by CytoTox96^® Non-Radioactive Cytotoxicity Assay kit at effector-to-target (E:T) ratio of 30:1. (E:T = 30:1). G–H Co-culture supernatants were subjected to ELISA to measure IFN-γ and TNF-α. Data are presented as mean ± SD from three independent replicates F–H. Statistical significance was determined by two-tailed unpaired Student’s t-test (*P < 0.05, **P < 0.01).

To determine whether LAP3 modulates CD8⁺ T-cell effector function, we established a direct co-culture system. Briefly, CD8⁺ T cells were isolated from the spleens of tumor-bearing 615 mice (Fig. 3C, D). After confirming a sorting efficiency >90% by flow cytometry (Fig. 3E), the cells were plated at 2 × 10⁵ cells per well in 96-well flat-bottom plates. LAP3-knockdown (shLAP3) or control (shNC) MFC cells were then added at a 30:1 effector-to-target ratio (Fig. 3D, E). After 24 h of co-incubation at 37 °C, supernatants were collected and analyzed for target cell (MFC) death using a lactate dehydrogenase (LDH) release assay (Fig. 3F and Fig. S3B, C), while IFN-γ and TNF-α levels quantified by ELISA (Fig. 3G–H). This direct cell–cell contact model allows robust detection of T-cell–mediated cytokine release and target-cell killing. Furthermore, the secretion of IFN-γ and Granzyme B (GZMB)-key markers of CD8⁺ T cell activation and cytotoxicity-was markedly elevated in co-cultures with LAP3-knockdown cells, as measured by flow cytometry (Fig. 3G, H). These findings indicate that LAP3 suppression robustly enhances CD8⁺ T cell killing capacity. Our results support a model in which LAP3 promotes immune escape in GC by both impeding CD8⁺ T cell infiltration and impairing their cytotoxicity.

LAP3 stabilizes PD-L1 by preventing its polyubiquitination

To investigate the mechanism of regulatory role of LAP3 in CD8⁺ T cells infiltration, we re-analyzed GO and KEGG enrichment of DEGs between LAP3-high and LAP3-low TCGA-STAD samples. GO data showed predominant immune-related signatures (Fig. 4A). Intriguingly, KEGG highlighted PD-L1 expression and PD-1 checkpoint pathways (Fig. 4B). Furthermore, correlation analysis demonstrated a strong positive association between LAP3 and CD274 (PD-L1) (Fig. 4C), and IHC confirmed the positive correlation between LAP3 and PD-L1 expression (Fig. 4D–F). Additionally, IPS analysis revealed that patients with high LAP3 expression exhibited enhanced responsiveness to PD-1 inhibitor therapy (Fig. 4G). These findings collectively suggest a functional association between LAP3 and PD-L1.

Fig. 4. Association between LAP3 and PD-L1.

A GO analysis of differential genes based on median value of LAP3 expression from TCGA-STAD database. B KEGG enrichment analysis of the DEGs between LAP3-high and LAP3-low GC samples from TCGA-STAD. C Correlation analysis of LAP3 expression with 47 immune checkpoint molecules from TCGA-STAD dataset. D Representative IHC staining of LAP3 and PD-L1 in one case from each group, derived from a cohort of 56 gastric cancer (GC) specimens. Scale bar=1.25 mm (2×), 100 μm (20×) E Significant positive correlation between LAP3 and PD-L1 positivity rates in GC tissues (n = 56; Pearson’s correlation). F Percentages of specimens exhibiting low or high LAP3 expression were correlated with PD-L1 levels. Two-sided χ² test. G The IPS predicted a positive association between high LAP3 expression and responsiveness to anti-PD-1/PD-L1 therapy.

Therefore, we investigated whether LAP3 regulates PD-L1 expression. WB and RT-qPCR results demonstrated that PD-L1 protein levels, but not mRNA levels altered upon LAP3 expression (Fig. 5A–F). These finding were further validated by IF analysis in AGS cells (Fig. 5G), confirming LAP3 as a regulator of PD-L1 protein levels.

Fig. 5. LAP3 stabilizes PD-L1 by preventing its polyubiquitination.

A–F WB and RT-qPCR evaluated the expression of PD-L1 in GC cells transfected with vector or oeLAP3 plasmids A, B, E/control or shLAP3 plasmids C, D, F. G IF assays demonstrate the levels of PD-L1 (red) upon LAP3 overexpression (green) in AGS cells, with DAPI nuclear counterstain (blue). H WB of PD-L1 in GC cells with stable LAP3-knockdown, overexpression of LAP3 or control following treatment with 100 µM CHX for the indicated time points. I WB of PD-L1 in LAP3-knockdown or control AGS cells following treatment with 20 µM of MG132 for 4 h or 50 µM of CQ for 12 h. J, K IP analysis of polyubiquitination of PD-L1 in LAP3-overexpression J, LAP3-knockdown K or control GC cells treated with MG132 (20 µM, 4 h). Data are presented as mean ± SD with three independent replicates A–I. Statistical significance was determined by two-tailed unpaired Student’s t-test (*P < 0.05, **P < 0.01, ***P < 0.001, ***P < 0.0001).

To clarify how LAP3 governs PD-L1 protein levels, we performed cycloheximide (CHX)-chase assays. The result revealed that PD-L1 half-life was markedly shortened upon LAP3 knockdown (Fig. 5H; Fig. S4A) and prolonged following LAP3 overexpression (Fig. 5H; Fig. S4B). WB indicated that both lysosomal and proteasomal routes contribute to PD-L1 turnover in GC cells (Fig. S4C). To further determine the degradation pathway of LAP3, LAP3-knockdown AGS cells were treated with the proteasome inhibitor MG132 or the lysosomal inhibitor chloroquine (CQ). The results showed that both MG132 and CQ could stabilize PD-L1. However, only MG132 treatment partially rescued PD-L1 expression in LAP3-knockdown cells (Fig. 5I), indicating that LAP3 stabilizes PD-L1 by inhibiting its ubiquitin-proteasome degradation pathway.

Next, we examined whether LAP3 affects the ubiquitination level of PD-L1. IP revealed that LAP3 overexpression markedly decreased PD-L1 polyubiquitination (Fig. 5J, Fig. S4D), whereas LAP3 knockdown had the opposite effect (Fig. 5K, Fig. S4E). While, LAP3 did not influence PD-L1 SUMOylation or neddylation. Together, these findings suggest that LAP3 stabilizes PD-L1 by inhibiting its polyubiquitination, thereby preventing proteasome-mediated degradation.

LAP3 directly binds to UBE3A, an E3-ligase for PD-L1

We then sought to analyze whether LAP3 could bind to PD-L1. Co-IP assays confirmed that LAP3 does not directly interact with PD-L1 (Fig. 6A), suggesting the involvement of additional mediator in this regulatory process. Therefore, we established a stable HGC-27 cell line expressing GFP-LAP3. Using GFP-IP- MS, we identified 2608 proteins that interact with LAP3 (Table S6). By comparing these LAP3 interactors with known E3 ligases involved in the ubiquitination process (Table S7), we narrowed down the candidates to five overlapping proteins: UBE3A, STUB1, FBXO22, HUWE1, and CBL (Fig. 6B–D). Reciprocal Co-IP assays further confirmed that LAP3 may bind to three proteins (UBE3A, CBL, and STUB1) in GC cells. Among these, UBE3A demonstrated the highest peptide score (3.425) in the MS analysis and showed the strongest binding affinity to LAP3 in the CO-IP assay (Fig. 6E), a finding consistent with their correlation analysis (Fig. S5A); thus, we identified it as the most promising candidate for further study. Notably, neither LAP3 overexpression nor knockdown affected UBE3A protein levels (Fig. S5B). Additionally, IF assays revealed that LAP3 and UBE3A predominantly colocalize in the cytoplasm of AGS cells (Fig. 6F). These findings indicate a direct interaction between LAP3 and UBE3A.

Fig. 6. LAP3 directly binds to UBE3A and UBE3A is an E3-ligase of PD-L1.

A Co-IP analysis of the interaction between endogenous PD-L1 and GFP-LAP3 in AGS and HGC-27 cells. B GFP -LAP3 plasmid was transfected into HGC-27 cells for GFP-IP following by the SDS-PAGE and silver staining. The images of silver staining were displayed, and the locations of target proteins were marked with black arrows. C The MS analysis to explore different LAP3-binding E3-ligases. Represent the peptide-spectrum matches of five E3-ligases by MS in the independent respective GFP-IP groups. D The secondary peptide fragments of LAP3 and UBE3A in the mass spectrum were displayed. E Reciprocal CO-IP analysis for UBE3A and GFP-LAP3. F IF images for colocalization of LAP3 and UBE3A. G WB of PD-L1 in HGC-27 cells with overexpression of UBE3A or control following treatment with CHX (100 µM) for the indicated time points. H WB of PD-L1 in HGC-27 cells with different amount of mcherry-UBE3A (0, 1, 2.5, 5 μg plasmids) with or without treatment with MG132 (20 µM, 4 h). I IP analysis of polyubiquitination of PD-L1 in UBE3A-overexpression or control HGC-27 cells treated with MG132 (20 µM, 4 h). J The interaction between HA-PD-L1 and UBE3A was examined by reciprocal Co-IP.

Next, we investigated whether PD-L1 serves as a substrate of UBE3A. To this end, we first demonstrated that CHX treatment reduced PD-L1 levels in a time-dependent manner with or without UBE3A overexpression (Fig. 6G, Fig. S5C). Furthermore, UBE3A overexpression dose-dependently decreased endogenous PD-L1 levels, an effect reversed by MG132 treatment (Fig. 6H). An IP assay further confirmed that UBE3A increased the ubiquitination level of PD-L1 (Fig. 6I). The above findings indicate that UBE3A promotes the proteasome-dependent degradation of PD-L1 by increasing its ubiquitination. In addition, Co-IP assay suggested that endogenous UBE3A interacts with HA-PD-L1 (Fig. 6J), confirming that PD-L1 is a substrate of UBE3A. Therefore, we proposed that LAP3 may reduce PD-L1 ubiquitination by interacting with UBE3A.

LAP3 stabilizes of PD-L1 by competitively binding to UBE3A

To test the above hypothesis, WB was conducted to show that LAP3 overexpression counteracted the effect of UBE3A, while LAP3 depletion exacerbated the effect, indicating that LAP3 and UBE3A antagonistically regulate PD-L1 degradation (Fig. 7A, B). Further IP assay supported that UBE3A overexpression increased PD-L1 polyubiquitination, an effect that was reversed upon LAP3 upregulation (Fig. 7C). Notably, Co-IP results revealed that LAP3 overexpression reduced the interaction between UBE3A and PD-L1, whereas LAP3 knockdown enhanced this interaction (Fig. 7D, E). These findings suggest that LAP3 competes with PD-L1 for binding to UBE3A (Fig. 7F). In summary, LAP3 is upregulated in GC and interacts with UBE3A, thereby impairing UBE3A-mediated ubiquitination and degradation of PD-L1, ultimately stabilizing PD-L1. High PD-L1 expression impedes the infiltration and cytotoxic function of CD8⁺ T cells, which finally drives GC immune evasion and malignant progression.

Fig. 7. LAP3 stabilizes PD-L1 by competitively binding to UBE3A.

A, B WB of PD-L1 in LAP3-overexpression A, LAP3-knockdown B, or control of HGC-27 cells transfected with empty vector or mcherry-UBE3A plasmids. C IP assay of polyubiquitination of HA-PD-L1 in LAP3-GFP overexpression or empty-vector HGC-27 cells transfected with mcherry-UBE3A or empty vector. D, E Co-IP for binding affinity between UBE3A and HA-PD-L1 in HGC-27 cells with LAP3 knockdown or overexpression. F The Schematic Diagram of LAP3 inhibiting UBE3A-mediated PD-L1 ubiquitination.

Ubenimex restores PD-L1 polyubiquitination by disrupting LAP3-UBE3A interaction

Based on the above findings, we further explore the details of how Ubenimex exerts its synergistic effects with PD-L1 inhibitors. We first predicted the binding region between LAP3 and UBE3A via molecular docking (Fig. 8A). Ubenimex, a compound previously reported to inhibit LAP3 activity by binding to its active pocket [27], was also docked to LAP3 (Fig. 8B). Interestingly, a comparison of the binding regions revealed complete overlap, despite the distinct binding sites of LAP3 with UBE3A and LAP3 with Ubenimex. To verify the LAP3-UBE3A molecular docking predictions, different truncated UBE3A variants were constructed (Fig. 8C), followed by Co-IP assays to verify the binding domain of UBE3A to LAP3. The disordered region, E6-binding domain, and HCV core-protein interaction domain (amino acids 175-517) were collectively identified as the LAP3-interacting segment, consistent with the molecular-docking predictions (Fig. 8D). These findings underscore the potential of Ubenimex to disrupt the LAP3-UBE3A interaction.

Fig. 8. Ubenimex restores PD-L1 polyubiquitination by disrupting LAP3-UBE3A interaction.

A, B Molecular docking analysis of LAP3-UBE3A and LAP3-Ubenimex interactions. C Schematic diagram for UBE3A truncated domains. D Co-IP analysis for the interaction of different truncations of HA-UBE3A with LAP3 in HGC-27 cells. E–G WB and CO-IP for the effects of Ubenimex on the stability of HA-PD-L1 E, the binding affinity between PD-L1 and UBE3A F, and the ubiquitination level of PD-L1 G in HGC-27 cells. H Flow cytometry analysis of the effects of LAP3 on membrane PD-L1 expression in MFC and HGC-27 cells transfected with shLAP3 or oeLAP3 plasmids. I Flow cytometry for the effects of Ubenimex on membrane PD-L1 expression in MFC and HGC-27 cells transfected with oeLAP3 plasmids. Data are presented as mean ± SD with three independent replicates H, I. Statistical significance was determined by two-tailed unpaired Student’s t-test (**** P < 0.0001).

Next, we evaluated the effect of Ubenimex on the binding affinity between UBE3A and LAP3. WB analysis revealed that Ubenimex counteracted the LAP3 overexpression-induced upregulation of PD-L1 (Fig. 8E). Co-IP further demonstrated that Ubenimex enhanced the interaction between UBE3A and PD-L1 (Fig. 8F, Fig. S6A). Similarly, IP assays confirmed an increase in PD-L1 ubiquitination upon Ubenimex treatment (Fig. 8G, Fig. S6B), consistent with the Co-IP findings. Notably, flow cytometry analysis confirmed that Ubenimex suppressed LAP3 overexpression-induced PD-L1 upregulation on the cell membrane in both MFC and HGC-27 cells (Fig. 8H, I). These results establish that as a selective LAP3 inhibitor, Ubenimex competitively disrupts the LAP3-UBE3A interaction. By liberating UBE3A, Ubenimex promotes PD-L1 ubiquitination and degradation, thereby reducing PD-L1 abundance and reversing CD8⁺ T cells suppression. This mechanism underpins the observed synergy between Ubenimex and PD-L1 blockade (Graphical abstract).

Discussion

In the present study, we identified that the upregulated LAP3 expression correlates to worse GC prognosis. A central finding of our investigation is that LAP3 acts as a driver of immune escape in GC. Furthermore, we uncovered a novel mechanism through which ubenimex synergizes with anti-PD-L1 therapy by disrupting the LAP3-UBE3A interaction. These results significantly enhance our understanding of resistance to anti-PD-L1 treatment and provide a robust mechanistic rationale for combining of ubenimex with anti-PD-L1 in GC treatment.

A key finding of this study is the identification of LAP3 as a promoter of GC progression. We demonstrate that high LAP3 expression is strongly associated with poorer overall survival (OS) in GC patients. This result is consistent with prior mechanistic studies that have implicated LAP3 in promoting tumor aggressiveness across multiple malignancies, including glioma, esophageal squamous cell carcinoma, and breast cancer [24–31]. These collective findings provide a novel theoretical basis for molecular subtyping and targeted therapeutic strategies in GC.

The secondary major contribution of this study is the elucidation of the mechanism by which Ubenimex enhances anti-PD-L1 therapy. Ubenimex disrupts the LAP3-UBE3A interaction, restoring UBE3A-mediated ubiquitination and proteasomal degradation of PD-L1. This reduces PD-L1 protein levels, alleviating CD8⁺ T-cell suppression and enhancing the efficacy of anti-PD-L1 therapy. Ubenimex, approved in Japan since 1987 as an adjuvant in tumor chemoradiotherapy, is known for its inhibitory effects on LAP3 and its antitumor and immunomodulatory properties [18, 19, 28, 37]. To the best of our knowledge, this study provides the first evidence that ubenimex significantly augments the efficacy of anti-PD-L1 therapy in GC mouse models. Although a phase II trial (TORG2241) is currently evaluating the combination of Ubenimex and pembrolizumab in advanced squamous non-small-cell lung cancer, no efficacy data have been reported thus far [38].

Furthermore, our study reveals a negative correlation between LAP3 expression with CD8⁺ T-cell infiltration and killing activity. This finding contradicts a recent report by Zhou et al., which identified LAP3 as a hub gene positively correlated with immune-cell infiltration in Epstein–Barr virus-positive (EBV⁺) GC [39]. While Zhou’s study focused on TCGA-STAD samples enriched for EBV-associated tumors, our cohort consists exclusively of EBV-negative tumors. This leads us to hypothesize that distinct molecular mechanisms underlie the development of EBV⁺ and EBV ^- GC subtypes. The apparent contradiction of the positive LAP3/CD8⁺ T-cell correlation in EBV⁺ GC may be attributed to the potent inflammatory milieu induced by the virus. This excessive inflammatory response is driven by two principal factors: EBV infection and co-infection with other bacterial. Firstly, EBV-infected GC cells expressing viral antigens (VCA and EA) undergo immune-mediated necrosis, igniting vigorous, persistent, dense lymphocyte infiltration-mediated inflammation [40]. Moreover, EBV frequently co-infects with Helicobacter pylori, further escalating the inflammatory cascade [41]. Consequently, EBV⁺ GC often exhibit lymphoepithelioma-like carcinoma (LEC), distinguished by prominent infiltration of CD8⁺ T cells, CD4⁺ T cells, and dendritic cells [42]. In contrast, within our EBV⁻ cohort, 53.57% of GC patients exhibited high LAP3 expression, where its immunosuppressive function—antagonizing CD8⁺ T-cell recruitment and activity—became predominant. In the remaining 46.43% with low LAP3 expression, immune evasion is likely driven by alternative mechanisms, such as lactate accumulation, TP53 gene status, and mutations in genes associated with antigen processing [43–46].

The persistent of PD-L1 overexpression is a key mechanism of acquired resistance to immune checkpoint inhibitors. Even with escalating doses of anti-PD-L1 antibodies, full saturation of the ligand is not achieved, enabling tumor cells to re-suppress CD8⁺ T-cell function. To address this, it is crucial to identify the drivers of sustained PD-L1 up-regulation, discover novel combination targets, and improve anti-PD-L1 efficacy to restore CD8⁺ T-cell infiltration and function. The constitutive overexpression of PD-L1 arises from three main mechanisms: gene amplification, sustained transcriptional activation, and post-translational stabilization of PD-L1 [7–10]. Among the post-translational modifications, ubiquitination plays a critical role in regulating PD-L1 stability [47–53]. Here we report the novel finding that LAP3 elevates PD-L1 abundance by inhibiting its ubiquitination and identify UBE3A as a new E3 ligase for PD-L1. While both UBE3A and the previously identified HUWE1 share the HECT domain, they operate in distinct cellular compartments. HUWE1 mediates PD-L1 poly-ubiquitinates at Lys281 within the ER, leading to ER-associated degradation [50]. Whereas UBE3A, being a cytoplasmic protein, is presumed to target PD-L1 during post-ER trafficking prior to its insertion into the plasma membrane. We further propose that UBE3A ubiquitinates PD-L1 in the vesicular transport phase, creating spatially segregated checkpoints for PD-L1 degradation. LAP3 competes with PD-L1 for binding to UBE3A, thereby inhibiting PD-L1 ubiquitination. Interestingly, UBE3A does not ubiquitinate and degrade LAP3, as LAP3 interacts with the N-terminal region (residues 175-517) of UBE3A, rather than its HECT domain. Supporting our findings, Singh et al. determined the crystal structure of UBE3C, which shares a similar architecture with UBE3A. In both proteins, an N-terminal segment preceding the HECT domain allosterically modulates its activity, thereby attenuating substrate ubiquitination [53]. This explains why LAP3’s interaction with UBE3A influences PD-L1 ubiquitination. Collectively, Ubenimex acts as a LAP3-selective inhibitor, competing with UBE3A to bind LAP3. This releases UBE3A to ubiquitinate and degrade PD-L1, reducing cell-surface PD-L1 levels, relieving CD8⁺ T-cell suppression, and synergistically enhancing anti–PD-L1 therapy in GC.

In conclusion, our study demonstrates that Ubenimex can significantly enhance the inhibitory effect of anti-PD-L1 on GC and elucidates the underlying mechanisms. These findings pave the way for overcoming clinical resistance to immune checkpoint inhibitors and advancing precision combination therapy in GC, particularly in patients with aberrant LAP3 expression. Nevertheless, further clinical trials are warranted to validate its definite efficacy. Moreover, the generalizability of Ubenimex combined with PD-1 blockade requires further analysis.

Clinical Value of LAP3 as a Predictor for Precision Therapy

Our findings suggest that patients with elevated LAP3 expression are more likely to benefit from the combination of ubenimex and anti-PD-L1, establishing LAP3 as a potential predictive biomarker. Its detection via immunohistochemistry is feasible and could guide personalized treatment strategies, facilitating the clinical translation of this combination therapy.

Author contributions

Tao Yang and Lijun Yang: conception, design, writing-review, editing, supervision, funding acquisition, resources, and validation. Caixia Zhao, Jiaxin Li, Jinxiu Zheng, Ming Chi, and Jie Dai: data curation and investigation. Caixia Zhao, Jiaxin Li, Jinxiu Zheng, Ming Chi, and Likun Zan: methodology. Caixia Zhao: formal analysis, software, visualization, and writing-original draft. Jiaxin Li and Caixia Zhao: writing-original draft. Chunxia Chen: software and visualization. Ying Shao, Yuxin Che, Wenjing Chen, Xiaoning Li, Yu Shi, Yujing Duan: supervision. Ying Shao, Yuxin Che, Wenjing Chen: writing-review and editing. All authors read and approved the final manuscript.

Funding

The current study was supported by a grant from Science and Technology Innovation Team of Shanxi Province to Tao Yang (202204051002030); Shanxi Province Higher Education “Billion Project” Science and Technology Guidance Project to Tao Yang (BYJL008).

Data availability

All biometics data used in this work can be acquired from the TCGA database (http://cancergenome.nih.gov/), Gene Expression Omnibus (GEO) datasets (https://www.ncbi. nlm.nih.gov/geo/), GTEx project (https://gtexportal.org/home/). Immunopheno-score [32] (IPS) scores from TCIA (http://tcia.at/). Experimental data supporting the findings of this study are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

Ethics statement

All animal experiments complied with the regulations and guidelines of the ARRIVE Guidelines 2.0. (PLoS Bio 8(6), e1000412,2010). The protocols were approved by the Animal Research Ethics Committee at the Shanxi Medical University (SYD2024048).

Supplementary information

The online version contains supplementary material available at 10.1038/s41419-026-08509-3.