PURPOSE:Nonmetastatic tumor-draining lymph nodes (TDLNs-) are central to initiating and sustaining antitumor immunity. The impact of their surgical removal on immunotherapy efficacy in recurrent biliary tract cancer remains unclear. We investigated the effect of TDLNs- dissection extent on treatment outcomes in this population. EXPERIMENTAL DESIGN:This real-world study retrospectively analyzed clinical and survival data from 101 patients with recurrent biliary tract cancer who received immunotherapy across five Chinese hospitals (2018-2023). Patients were stratified by extent of TDLNs- dissection (≤6 vs. >6). Multiplex immunofluorescence (mIF) analysis of lymph node immune microenvironments was performed in a representative subset of patients (n = 20) from Sun Yat-sen Memorial Hospital. RESULTS:Patients with ≤6 TDLNs- dissected (n = 59) achieved significantly longer progression-free survival (PFS) than those with >6 dissected (n = 42; HR, 0.48; 95% confidence interval, 0.31-0.73; P = 0.001), although there was no statistical difference in overall survival. The mIF staining analysis revealed that TDLNs- contained higher densities of TCF-1+PD-1-CD8+ tumor-specific memory T cells and CD11c+ conventional dendritic cells and lower proportions of FOXP3+CD4+ regulatory T cells and TCF-1-PD-1+CD69+CD8+ terminally exhausted T cells compared with metastatic TDLNs (TDLNs+). Among immunotherapy responders, TDLNs- exhibited greater TCF-1+PD-1-CD8+ T-cell and CD11c+ cell densities than those in nonresponders. Importantly, a higher proportion of TCF-1+PD-1-CD8+ T cells in TDLNs- correlated with improved PFS, whereas extensive dissection of TDLNs- diminished this benefit. CONCLUSIONS:Excessive removal of TDLNs- compromises the efficacy of immunotherapy in recurrent biliary tract cancer. A selective lymphadenectomy approach that prioritizes clearance of TDLNs+ while preserving TDLNs- may optimize outcomes for patients receiving postoperative immunotherapy. See related commentary by Vonderheide, p. 2326.

BACKGROUND Hepatocellular carcinoma (HCC) is a top primary liver malignant tumor, and still is hard to control after invasive growth and spread have already occurred. A big part of its biological difference in properties still cannot get explanation on the molecule level. Transmembrane emp24 domain-containing protein 4 (TMED4) it is one cargo-trafficking protein that belongs to the p24/TMED family. Albumins inside this grouping take part in endoplasmic reticulum-Golgi carrying, and can change signal plans which are related to tumor action, hence the function of TMED4 in HCC has not been clearly given definition. Owing to the fact that nuclear factor kappa-B (NF-kappa B)/snail pathway has close connection with proliferation, movement ability, invasion capability, and epithelial-mesenchymal transition, hence it is a reasonable path that TMED4 may use it to produce effect on HCC cells. AIM To describe the expression situation of TMED4 in HCC cells, we therefore confirm whether the forced expression of TMED4 can change cell growth, migration and invasion ability, and we also check whether these appearance changes are accompanied by the change of expression quantity of NF-kappa B p65 and snail. METHODS The abundance of TMED4 transcription product in LO2 cells and SMMC-7721 cells has been measured through the method of quantitative real-time polymerase chain reaction. Through plasmid transfection, we have constructed a TMED4 overexpression SMMC-7721 model, the empty vector is used as the control by us. The increment of cell number was assessed through cell counting kit-8 test at continuous time points. Transwell chambers that have no or have Matrigel covering were utilized to measure migration and invasion, respectively. We have carried out the operation of Western blot for the detection of the protein expressions of TMED4, NF-kappa B p65 and snail. RESULTS Endogenous TMED4 messenger RNA expression level in SMMC-7721 cells is lower than that in LO2 cells (P < 0.05). After the transfection operation, the group with TMED4 overexpression displayed an obvious elevation in TMED4 messenger RNA when compared with the control group (P < 0.05). Optical density numerical values went up through time in two groups, but the increase was made weaker in TMED4-overexpressing cell groups (P < 0.05). When we make comparison with the control group, the overexpression group was found to have smaller numbers of cells that had migrated and invaded (P < 0.05). The experiment which used Western blot method has discovered that in the cells which have overexpression of TMED4, the protein content of TMED4 and NF-kappa B p65 is higher, while snail expression level is lower (P < 0.05). CONCLUSION The overexpression of TMED4 has the ability to suppress the proliferation, migration and invasion of HCC cells, therefore it may restrain the biological behaviors of HCC through the regulation of NF-kappa B/snail signaling pathways.

e16279 Background: Inoperable extrahepatic cholangiocarcinoma (eCCA) carries a poor prognosis. This study evaluates the efficacy and safety of photodynamic therapy (PDT) combined with biliary drainage (BD) versus BD alone. Methods: We retrospectively analyzed 86 eCCA patients from June 2021 to June 2025. Patients received PDT+BD (n = 39) or BD alone (n = 47). A 1:1 propensity score matching (PSM) was performed to balance baseline characteristics, including age, gender, tumor location, and TNM stage. Primary endpoint was overall survival (OS); secondary endpoints were complications and stent removal rate. Results: After PSM (n = 22/group), all baseline variables were well-balanced ( p > 0.05). The PDT+BD group showed significantly superior median OS compared to the BD group (17.8 vs. 6.4 months; p = 0.011). The 1-year OS rate was 60.5% (95% CI: 48.5–72.5) for PDT+BD versus 16.4% (95% CI: 8.1–24.7) for BD ( p < 0.001). 10.3% (4/39) in the PDT group achieved drainage independence due to lumen relief, versus 0% in the BD group ( p = 0.039). Subgroup analysis confirmed that the survival benefit remained significant even for patients receiving only a single PDT session (mOS 17.8 vs. 6.4 months; p = 0.022). The overall complication rate post-PSM was identical between the PDT+BD and BD groups (18.2% vs. 18.2%; p = 1.000). Conclusions: PDT combined BD significantly prolongs OS in eCCA patients without increasing procedural risks. The higher rate of drainage independence highlights PDT's efficacy in local tumor control. Characteristics and clinical outcomes of patients before and after propensity score matching. Variable Pre-PSM PDT (n=39) Pre-PSM BD (n=47) P -value¹ Post-PSM PDT (n=22) Post-PSM BD (n=22) P -value² Tumor Location, n (%) <0.001 0.750 - Hilar Cholangiocarcinoma 16 (41.0%) 36 (76.6%) 15 (68.2%) 14 (63.6%) - Distal Cholangiocarcinoma 23 (59.0%) 11 (23.4%) 7 (31.8%) 8 (36.4%) TNM Stage, n (%) 0.188 0.728 - Stage III 30 (76.9%) 30 (63.8%) 17 (77.3%) 16 (72.7%) - Stage IV 9 (23.1%) 17 (36.2%) 5 (22.7%) 6 (27.3%) Median Overall Survival 17.8 mo 5.7 mo <0.01 17.8 mo 6.4 mo 0.011 1-Year Survival Rate, % 59.5% 22.0% <0.001 60.5% 16.4% <0.001 Stent Removal Rate, % (n/N) 10.3% (4/39) 0% (0/47) 0.039 — — — 1 P -value for comparison between PDT and Stent groups before PSM. 2 P -value for comparison between PDT and Stent groups after PSM.

Background/Objectives: Gallbladder cancer (GBC) is an aggressive biliary tract malignancy with poor prognosis. Lymph node metastasis is a major determinant of adverse outcome in patients with GBC. Gene mutations play an essential role in tumorigenesis; however, it remains uncertain whether genetic mutations play a substantial role in lymph node metastasis in GBC. Methods: In this study, transcriptome and whole-exome sequencing (WES) were used to analyze gene mutations and expression in GBC tissues, focusing on lymph node metastasis. Bioinformatics tools identified differentially expressed genes (DEGs) and significantly mutated genes (SMGs), followed by pathway enrichment and survival analyses. Results: In total, 669 DEGs were identified between metastatic and non-metastatic GBC tissues. Through protein-protein interaction (PPI) network analysis of these DEGs, GPT and NR1I2 were identified as candidate genes associated with metabolic reprogramming in lymph node metastasis. Prognostic analysis revealed 22 DEGs associated with patient survival, and significant differences in overall survival, clinicopathological features (e.g., N-stage and positive lymph node count) were observed between cluster 1 and cluster 2. Mutation analysis identified 55 SMGs, primarily related to immune and inflammatory responses. By integrating DEGs and SMGs, PLCL2 was identified as a candidate gene potentially associated with both lymph node metastasis and prognosis. GSEA enrichment analysis suggested that PLCL2 was potentially linked to immunity, inflammation, and cellular processes, which may imply its possible involvement in GBC metastasis pending experimental validation. Conclusions: Based on integrative transcriptomic and exomic analyses, we identified PLCL2 as a candidate gene potentially associated with lymph node metastasis in GBC. These hypothesis-generating findings provide a preliminary basis for future mechanistic validation and biomarker exploration.

Representativeness of study participants.

Immunotherapy efficacy in patients with recurrent BTC.

Background To develop a clinical-prompt-enhanced deep learning framework that noninvasively predicts gallbladder polypoid lesion (GBPL) pathology across cholesterol polyps, adenomas, and carcinomas using preoperative ultrasound and routinely available clinical text, addressing overtreatment associated with size-based assessment. Methods We retrospectively enrolled 516 cholecystectomy patients with histologically confirmed GBPLs (January 2015–January 2023). Patients were randomly stratified into training set (n = 362) and an independent test set (n = 154). Lesions were segmented using nnU-Net v2 to construct region-of-interest (ROI)-guided inputs, achieving a mean Dice score of 0.797 ± 0.02. We developed GB-DualmodalNet, which converts clinical variables into natural-language prompts via semantic mapping and fuses with ROI-guided VGG16 image features using cross-modal attention strategy. Performance was assessed on the independent test set using macro-averaged area under the receiver operating characteristic curve (AUC), accuracy, and standard diagnostic metrics. A two-reader study on the independent test set was performed to evaluate changes in clinician performance with artificial intelligence (AI) assistance. Results In the test cohort, the ROI-guided image-only model achieved macro-AUC 0.840 (95% CI 0.787–0.890). GB-DualmodalNet achieved the highest macro-AUC of 0.852 (95% CI 0.801–0.902), with accuracy 70.1% and specificity 79.2%. Subtype-specific performance yielded AUC 0.933 for carcinoma, 0.758 for adenoma, and 0.865 for cholesterol polyps. In 10–15 mm gray-zone subgroup (n = 48), GB-DualmodalNet correctly classified 19 of 22 benign lesions as low risk, corresponding to a specificity of 86.4%. Grad-CAM + + highlighted lesion margins and the attachment base, and SHapley Additive exPlanations (SHAP) identified long diameter, short diameter, and age as the most influential clinical features. AI assistance improved pooled clinician macro-AUC from 0.782 to 0.865. Conclusion GB-DualmodalNet provides an automated, interpretable approach for preoperative three-class GBPL prediction and improves performance over image-only and clinical-only baselines. Multimodal deep learning can improve GBPL risk prediction and reduce unnecessary cholecystectomies, though prospective multicenter validation is required before clinical use.

Proportional hazards assumptions checked by schoenfeld residuals.

Baseline characteristics of the entire cohort.

Bootstrap distribution of the optimal cutoff value for dissected TDLNs−.

Reagents and antibodies used in the study.

Validation of the prognostic threshold for TDLNs− and assessment of the model robustness.

Objective Gallbladder cancer (GBC) is a high aggressive malignancy prone to postoperative early recurrence (ER). However, reliable tools for preoperative risk stratification remain limited. This study aimed to develop and validate a noninvasive predictive model integrating clinical variables and contrast-enhanced computed tomography-based radiomic features for ER prediction in GBC. Methods A retrospective cohort of 204 patients with pathologically confirmed GBC was included and randomly divided into a training cohort (n = 142) and a validation cohort (n = 62). Feature selection was performed using least absolute shrinkage and selection operator regression. Six machine learning algorithms were applied to construct clinical, radiomic, and combined models. Model performance was evaluated using the area under the receiver operating characteristic curve, and interpretability was assessed using Shapley additive explanations (SHAP). Results Among 204 patients, 67 (32.8%) experienced ER. Multivariate analysis identified CEA > 5.0 ng/mL, albumin-to-γ-glutamyl transpeptidase ratio, and imaging lymph node metastasis as independent predictors of ER (all P < 0.05). A total of 6 clinical features and 11 radiomic features (comprising first-order statistics, texture, and higher-order wavelet features) were selected. In the validation cohort, the clinical SVM model (AUC = 0.830) and radiomic XGB model (AUC = 0.786) achieved the best individual performances. XGB-based fusion models outperformed unimodal models, yielding AUCs of 0.865 for feature-level fusion and 0.879 for decision-level fusion. Conclusion The proposed multidimensional model demonstrates good performance in preoperatively identifying patients at high risk of early recurrence, which may assist individualized treatment planning and postoperative surveillance strategies.

In the global phase 3 KEYNOTE-966 study (NCT04003636), pembrolizumab plus gemcitabine and cisplatin (pembrolizumab group) showed a statistically significant, clinically meaningful improvement in overall survival (OS) versus placebo plus gemcitabine and cisplatin (placebo group) without any new safety signals in participants with advanced biliary tract cancer (BTC). This analysis focused on the subgroup of participants from KEYNOTE-966 enrolled in China. Adults with previously untreated advanced BTC were randomly assigned (1:1) to receive pembrolizumab 200 mg or placebo intravenously every 3 weeks plus gemcitabine 1000 mg/m2 and cisplatin 25 mg/m2 intravenously on days 1 and 8 of every 3-week cycle. Primary endpoint was OS. Secondary endpoints were progression-free survival (PFS), objective response rate (ORR), and duration of response (DOR), all per RECIST v1.1 by blinded independent central review, and safety. One hundred fifty-eight participants were enrolled in China (75, pembrolizumab group; 83, placebo group). Median time from randomization to data cutoff (December 15, 2022) was 20.5 (range, 15.0–28.8) months. Median OS was 14.1 (95

Peripheral nerve invasion (PNI) is an early and decisive step in gallbladder cancer progression that strongly predicts poor postsurgical outcome. The tumor-neuron interactions that drive PNI could represent potential targets and biomarkers to improve treatment of gallbladder cancer. In this study, we demonstrated that gallbladder cancer provoked necroptosis of neurons to enable PNI. Gallbladder cancer cells transferred extracellular vesicles (EV) containing O-GlcNAcase (OGA) to neurons, which activated RIPK1-dependent necroptosis. Mechanistically, EV-derived OGA suppressed RIPK1 glycosylation while enhancing its phosphorylation, thereby activating the RIPK1/RIPK3/MLKL axis to trigger neuronal necroptosis. Subsequent neuronal release of HMGB1 engaged RAGE on gallbladder cancer cells, establishing a loop that accelerated PNI. Moreover, the RAGE antagonist FPS-ZM1 synergized with gemcitabine to suppress tumor progression. Collectively, these findings uncover an EV-mediated cross-talk between gallbladder cancer cells and neurons in which RIPK1-dependent necroptosis and its effector HMGB1 drive PNI, positioning the HMGB1-RAGE axis as a tractable therapeutic target.Significance: Tumor-derived extracellular vesicles trigger neuronal necroptosis that fuels peripheral nerve invasion, creating a tumor-neuron signaling loop that could be leveraged for liquid biopsy and personalized therapy strategies in neurotropic cancers.

Distribution of dissected TDLN counts.

Forest plots of PFS by subgroup for the TDLNs− ≤6 group versus the TDLNs− >6 group.

Background/Objectives: Malignant biliary stricture (MBS) remains difficult to diagnose accurately despite digital single-operator cholangioscopy (DSOC). We developed a deep learning (DL)-based computer-aided detection (CADe) and diagnosis (CADx) system for DSOC-based MBS assessment. Methods: This retrospective multicenter study included 149 patients from one center for model development and internal validation and 25 patients from two independent centers for external evaluation. CADe used a You Only Look Once version 11 (YOLOv11) architecture to localize irregular mucosa, abnormal vasculature, and nodular protrusions defined by the Carlos Robles-Medranda and Mendoza criteria. CADx used a Residual Network-18 classifier with gradient-weighted class activation mapping for interpretability. Results: CADe achieved a mean average precision at 50% intersection-over-union of 91.2%, with a precision of 92.0% and recall of 87.0%. CADx achieved a frame-level area under the receiver operating characteristic curve (AUC) of 0.960 in internal validation and 0.843 in external validation. External frame-level sensitivity was 52.0% and specificity was 95.2%. For the patient-level external endpoint, sensitivity was 85.7%, specificity was 94.4%, accuracy was 92.0%, and AUC was 0.881. Conclusions: The two-stage system combines localization of predefined cholangioscopic features with interpretable diagnostic classification. The small external cohort and marked reduction in frame-level sensitivity preclude firm conclusions regarding generalizability; prospective multicenter and live-procedure evaluation is required.

e16272 Background: Locally advanced, initially unresectable gallbladder cancer (GBC) is associated with a poor prognosis. Although immunotherapy combined with chemotherapy is recommended by NCCN guidelines, its potential to enable conversion surgery remains underexplored. This study evaluated the efficacy and safety of an intensified regimen combining PD-1/PD-L1 inhibitors with gemcitabine-based chemotherapy. Methods: This single-center real-world retrospective study enrolled 47 patients with locally advanced, initially unresectable GBC treated between February 2021 and July 2025, with follow-up through December 31, 2025. The cohort comprised 33 females (70.2%) and 14 males (29.8%). Baseline unresectability was confirmed by a multidisciplinary team (MDT) for all patients. All patients received at least two cycles of immunochemotherapy consisting of PD-1 or PD-L1 inhibitor combined with gemcitabine-based regimens including GAP (n = 32), AG (n = 5), GC (n = 4), GS (n = 3), GEMOX (n = 2), and GCS (n = 1). The primary endpoint was the conversion to surgical resection rate. Secondary endpoints included objective response rate (ORR), R0 resection rate, tumor regression grade (TRG), treatment-related adverse events (TRAEs), and survival outcomes. Results: Patients received a median of 2 conversion therapy cycles (range, 2-10) and were followed up for a median of 15.8 months (range, 4-46 months). 40.4% of patients (19/47) achieved conversion to surgical resection, with an R0 resection rate of 94.7% (18/19). Immunochemotherapy resulted in a median progression-free survival (PFS) of 14.5 months (1-year rate 51.2%, 2-year rate 30.9%) and median overall survival (OS) of 17.3 months (1-year rate 68.6%, 2-year rate 39.4%). The objective response rate (ORR) was 46.8% (22/47), including 1 complete response (CR), 21 partial responses (PR), 15 stable disease (SD), and 10 progressive disease (PD). Surgical patients demonstrated a significant survival benefit compared with non-surgical patients, with superior median OS [not reached (NR) vs. 11.4 months, P < 0.001] and PFS (NR vs. 6.4 months, P < 0.001). In the surgical cohort, a major pathologic response (TRG 0-1) was observed in 15.7% (3/19) of surgical patients. Two patients experienced Clavien-Dindo grade ≥3 postoperative complications, with no postoperative deaths. Grade 3-4 TRAEs occurred in 48.9% (23/47) of all patients, with similar incidence between surgical and non-surgical cohorts. Conclusions: The intensified multi-agent immunochemotherapy regimen demonstrated encouraging survival outcomes and manageable toxicity as conversion therapy for initially unresectable locally advanced GBC, supporting its potential as a therapeutic strategy.