Autophagy-promoted immunogenic cell death elicited by tyrosine kinase inhibitor orchestrates a synergistic immunotherapeutic microenvironment in hepatocellular carcinoma Authors: Shan-Ru Feng, En-Fu Dong, Zheng Gao, Mo-Ran Xu, Pascal Kwangwari, Cheng-Qiang Sun, Jia‑Feng Chen, Ying‑Hong Shi, Wei‑Ren Liu, Wei‑Xun Wu, Xin Zhang, Ao Huang, Xuan‑Ming Luo, Qiang Gao, Guo‑Ming Shi, Ai‑Wu Ke, Jian Zhou, Jia Fan, Xiao-ying Wang, Jing-Bo Qie, Xiu‑Tao Fu, Zhen‑Bin Ding Published: 2026 ABSTRACT AbstractBackgroundThe combination of tyrosine kinase inhibitors (TKIs) like lenvatinib with immune checkpoint inhibitors (ICIs) has become a cornerstone of first-line therapy for advanced hepatocellular carcinoma (HCC). However, this approach faces a significant bottleneck, with objective response rates lingering around 20–36%, largely due to the non-immunogenic ("cold") nature of most HCC tumors. This highlights a critical need for strategies that can remodel the tumor immune microenvironment to enhance therapeutic efficacy.MethodsWe systematically screened existing TKIs for their ability to induce immunogenic cell death (ICD). ICD hallmarks were validated through multiple assays in HCC cell lines and patient-derived organotypic tumor spheroids (PDOTS) from 20 primary and 15 neoadjuvant-treated HCC samples. Mechanistic studies employed RNA-seq, pharmacological inhibitors, and genetic knockdown approaches. A tumor cell-Bone Marrow-Derived Dendritic Cells (BMDC)-CD8+ T cell triple co-culture system was established for functional assays. In vivo efficacy was evaluated in subcutaneous and orthotopic mouse models. Tumor immune profiling was performed using Cytometry by Time-Of-Flight (CyTOF), multiplex immunofluorescence, and spatial proximity analysis.ResultsAmong screened TKIs, anlotinib emerged as the most potent ICD inducer, triggering extracellular ATP release, HMGB1 secretion, and surface calreticulin exposure. RNA-seq and GSEA identified ER stress and autophagy as top enriched pathways. Mechanistically, anlotinib inhibited FGFR-1 phosphorylation, activating ER stress (upregulation of CHOP/ATF4 with ATF4 nuclear translocation) and secondary autophagy (LC3-II conversion). Pharmacological inhibition of ER stress or autophagy, as well as ATG5/LC3B knockdown, abrogated damage-associated molecular patterns (DAMPs) release, confirming autophagy's essential role in ICD execution. Luminex cytokine assay and mass spectrometry identified the key secretome (CCL19, CXCL12) and DAMPs essential for dendritic cell recruitment. In BMDC co-culture experiments, anlotinib-conditioned media enhanced DC migration (blocked by anti-HMGB1), DC activation (CD86/CD40 upregulation), and subsequent CD8+ T cell priming (CD25/CD69/IFN-γ expression). In vivo, anlotinib inhibited tumor growth, enhanced CD8+ T cell and dendritic cell infiltration, and increased spatial interactions between ICD+ tumor cells and T cells. Anlotinib synergized with anti-PD-1 therapy, an effect mediated by CD8+ T cells and DCs. Genetic (shATG5) or pharmacological (3-MA) autophagy inhibition significantly attenuated anlotinib's antitumor efficacy, while autophagy enhancement through ATG5 overexpression or metformin potentiated it. The triple combination of anlotinib, anti-PD-1, and metformin achieved the highest levels of HMGB1 release and immune cell infiltration and maximal tumor suppression.ConclusionThis research pioneers the identification of anlotinib as an ER stress- and autophagy-dependent ICD inducer in HCC. Our comprehensive mechanistic dissection and robust preclinical evidence establish anlotinib's ability to convert immunologically "cold" tumors to "hot" ones through the FGFR-1/ER stress/autophagy/DAMP release axis. The synergy with anti-PD-1 and enhancement by metformin provide a rationale for novel combination strategies to overcome current limitations of targeted-immunotherapy, offering a promising approach to improve response rates in advanced HCC.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s40164-026-00792-2.Keywords: Hepatocellular carcinoma, Immunogenic cell death, Endoplasmic reticulum stress, autophagy, Immunotherapy Introduction Hepatocellular carcinoma (HCC) represents a major global health burden, ranking as the sixth most commonly diagnosed cancer and the third leading cause of cancer-related mortality worldwide [1], with even higher prevalence in China (fourth and second, respectively) [2]. While patients with early-stage HCC can be effectively treated with radical resection combined with systemic therapies, the clinical reality is far more challenging: the majority of HCC patients present with advanced-stage disease that precludes surgical intervention, resulting in poor prognoses and limited therapeutic options [3, 4]. The emergence of immune checkpoint inhibitors (ICIs) targeting programmed cell death protein 1 (PD-1) or its ligand PD-L1 has offered new hope for advanced HCC patients, demonstrating efficacy even in cases with distant metastases affecting the lungs, bones, or brain [5–8]. However, this promise remains largely unfulfilled: even with current standard-of-care approaches such as ICIs alone (response rates < 20%) [9] or combination therapies pairing ICIs with anti-VEGF agents (response rates 20–36%) [10–12], the majority of HCC patients still fail to achieve meaningful clinical benefit [13–15]. This creates an urgent therapeutic need: developing strategies to systematically convert "cold" HCC tumors into immunologically "hot" phenotypes that can respond effectively to immunotherapy [16–19]. To overcome the "cold" tumor challenge, we need therapeutic strategies that can simultaneously: (1) kill tumor cells while (2) alerting and activating the immune system to recognize the tumor as foreign, and (3) creating an inflammatory environment that recruits and activates immune effector cells. Immunogenic cell death (ICD) represents a potentially promising approach that may address all three requirements. Unlike conventional cell death, ICD is a specialized form of regulated cell death that converts dying tumor cells into an in situ source of tumor antigens [20]. The process may work through a coordinated autophagic response that releases tumor antigens, damage-associated molecular patterns (DAMPs), and pro-inflammatory cytokines from dying cells [21]. This molecular cascade initiates a sequential cascade: it promotes dendritic cell (DC) recruitment, facilitates antigen transfer and DC maturation, drives DC migration to lymphoid organs, and ultimately culminates in robust T lymphocyte activation [22]. Multiple therapeutic approaches have demonstrated the ability to induce ICD effectively. Chemotherapeutic agents serve as the most established ICD inducers, including DNA transcription inhibitors (actinomycin D, anthracyclines, lurbinectedin, oxaliplatin) and microtubule-targeting agents (taxanes, vinca alkaloids) [23]. Critically, the clinical efficacy of these agents correlates directly with immune activation: successful treatment is associated with increased DC and cytotoxic T lymphocyte (CTL) infiltration alongside reduced immunosuppressive regulatory T cells [24]. Most importantly, ICD-inducing chemotherapeutics can sensitize previously resistant tumors to PD-1/PD-L1 inhibitors [25, 26]. Oncolytic viruses represent another validated ICD-inducing strategy, exemplified by the FDA-approved talimogene laherparepvec (T-VEC), a genetically modified herpes simplex virus 1 [27–29]. The strategic deletion of the neurovirulence factor ICP34.5 in T-VEC not only ensures safety but also enhances the integrated stress response, contributing to ICD induction [30, 31]. Intriguingly, recent evidence from non-small cell lung cancer and renal cell carcinoma studies suggests that tyrosine kinase inhibitors (TKIs)—originally designed solely to target aberrant protein tyrosine kinases [32], also possess ICD-inducing capabilities [33, 34]. This discovery presents a unique therapeutic opportunity: an optimal TKI could simultaneously inhibit tumor proliferation through direct targeting while enhancing antitumor immunity through ICD induction, effectively converting "cold" tumors into "hot" phenotypes within a single agent. Building on this compelling rationale, we designed a comprehensive study to harness the dual potential of TKIs. Our approach involves systematically screening TKIs currently used in HCC management, including lenvatinib, regorafenib, apatinib, anlotinib, and cabozantinib, for their ICD-inducing potential, validating the most promising candidate in combination with anti-PD-1 therapy, and elucidating the underlying molecular mechanisms. The ultimate goal is to leverage these mechanistic insights to develop an optimized triple-combination therapy regimen that maximizes both direct antitumor effects and immune activation for HCC treatment. Materials and methods Cell culture The human HCC cell line MHCC97-H was developed at our institute. The HCC cell line Hepa1-6 was obtained from the Cell Bank of Type Culture Collection, Chinese Academy of Sciences (Shanghai, China), and cultured in Dulbecco’s Modified Eagle Medium (DMEM, Gibco, NY, USA) supplemented with 10% heat-inactivated fetal bovine serum (FBS, Sigma-Aldrich, MO, USA), 100 U/mL penicillin, and 100 μg/mL streptomycin (Sigma-Aldrich, MO, USA). All cell lines were maintained at 37 °C in a humidified incubator with 5% CO2. Antibodies and reagents The following antibodies were used in the present study: Anti-ATG5 (12,994, 1:1000), anti-LC3A/B (12,741, 1:1000), anti-ERK (9102, 1:1000), anti-pERK (4370, 1:2000), and anti-ATG7 (2631, 1:1000), anti-Calreticulin (12,238, 1:1000) from Cell Signaling Technology (MA, USA); anti-HSP70 (ab181606, 1:1000), anti-HSP90 (ab32568, 1:1000), anti-Calreticulin (ab92516, 1:1000), anti-HMGB1 (ab79823, 1:1000) from Abcam (Cambridge, UK); anti-ATG5 (10,181–2-AP, 1:1000), anti-ATG7 (10,088–2-AP, 1:1000), anti-HMGB1 (10,829–1-AP, 1:1000), anti-CD31 (11,265–1-AP, 1:2000), anti-GAPDH (HRP-60004, 1:5000), anti-CD4 (67,786–1-Ig, 1:200), and anti-CD8 (66868-1-Ig, 1:200) from Proteintech (Wuhan, China). Anlotinib (AL3818) dihydrochloride was obtained from Selleck. 3-Methyladenine (HY-19312), Bafilomycin A1 (HY-100558), Azoramide (HY-18705), and ISRIB (HY-12495A) were obtained from MedChemExpress. InVivoMAb anti-mouse PD-1 (BE0146), InVivoMab anti-mouse CD4 (BE0146), and InVivoMab anti-mouse CD8α (BE0061) were obtained from BioXCell (NH, USA). Cytochrome c (C301837-100 mg) was obtained from Aladdin (Shanghai, China). Doxorubicin (HY-15142, MedChemExpress) was used as a positive control for immunogenic cell death (ICD), as it is a well-established anthracycline chemotherapeutic agent extensively documented to potently induce ICD through ER stress-mediated calreticulin exposure and DAMPs release [35, 36]. This makes doxorubicin an ideal benchmark to evaluate the ICD-inducing potential of novel agents such as anlotinib. Patient samples Patient samples were collected after obtaining informed consent from each individual according to a protocol approved by the Ethics Committee of Zhongshan Hospital. Tumor tissue samples were collected from HCC patients at Zhongshan Hospital, Fudan University, Shanghai, China. The samples included those from patients who underwent direct surgical resection (n = 20) and those who underwent resection following neoadjuvant therapy (n = 15). The effectiveness of TKI-induced ICD was assessed using ex vivo tissue culture models detailed below. Western blotting Proteins from total cell lysates were separated on 10% or 12.5% standard sodium dodecyl sulfate–polyacrylamide gels and then transferred to polyvinylidene difluoride membranes. The membranes were washed, blocked, and incubated overnight at 4 °C with specific primary antibodies, followed by incubation with horseradish peroxidase (HRP)-conjugated species-matched secondary antibodies (Yeasen, China) for 60 min at room temperature. Bands were visualized using a chemiluminescent HRP substrate and an electrochemiluminescence imaging system (Tanon, Shanghai, China). Immunohistochemistry and multiplexed immunofluorescence Paraffin-embedded tissue sections were deparaffinized and rehydrated. Heat-induced antigen retrieval was performed, followed by blocking of endogenous peroxidase activity for immunohistochemistry. Immunohistochemical staining was analyzed using the Caseviewer software. For immunofluorescence staining, images were acquired with a Leica confocal fluorescence microscope and processed using the Leica software. After deparaffinization, endogenous blocking, and antigen retrieval, nonspecific binding was blocked for 30 min. Slides were incubated with corresponding primary antibodies, including anti-CD4 (67786-1-Ig, 1:200), anti-CD8 (66868-1-Ig, 1:200), and anti-HMGB1 (ab79823, 1:1000), followed by secondary HRP-conjugated polymers covalently bound to distinct fluorophores. Additional antigen retrieval steps were used to remove background signals. After sequential reactions, slides were counterstained with DAPI (Sigma) and mounted using a fluorescence mounting medium. Once staining was completed, the slides were scanned using a fluorescence slide scanner (Leica, VERSA8, Germany). ICD staging and spatial analysis were performed on formalin-fixed, paraffin-embedded HCC sections stained for CD4, CD8 and HMGB1. All images were acquired using a 40 × /1.3 NA oil-immersion objective, yielding a pixel size of 0.162 µm. Raw TIFFs were imported into ImageJ/Fiji (v2.3.0) for processing. A uniform scale was set by defining the pixel-to-micron ratio (1 px = 0.162 µm) through the “Set Scale” function. Cells were automatically detected via the “Trainable Weka Segmentation” plugin, with classification based on nuclear size, circularity, and fluorescence intensity thresholds empirically determined from negative and positive controls. Candidate objects were further filtered by area and shape to exclude artifacts. For each CD8⁺ or CD4⁺ reference cell, the three nearest neighboring cells negative for HMGB1 release and the six nearest positive neighbors (three early-ICD and three late-ICD) were identified using the “Nearest Neighbors” macro, which calculates Euclidean distances between object centroids. To minimize observer bias, two trained investigators (A.B. and C.D.) independently reviewed and, if necessary, corrected automated segmentations and morphological classifications. Both observers were blinded to treatment group and time point throughout annotation. Immunofluorescence and confocal microscopy Cells cultured on glass slides were fixed with 4% paraformaldehyde for 15 min, permeabilized with 0.1% Triton X-100 for 15 min at room temperature, washed with phosphate-buffered saline (PBS), and blocked with PBS containing 0.5% bovine serum albumin for 1 h at room temperature. Slides were then incubated with the indicated primary antibodies overnight at 4 °C, followed by incubation with Alexa Fluor-488 or Alexa Fluor-594-conjugated secondary antibodies (Yeasen) for 1 h at room temperature. After rinsing in PBS, nuclei were counterstained with diamidino phenylindole, and images were acquired via confocal microscopy (Olympus FV-3000, Olympus, Japan). qPCR Total RNA was extracted using TRIzol reagent (Invitrogen, CA, USA) and subsequently reverse transcribed using a cDNA synthesis kit (Yeasen). RNA concentration and purity were assessed with a NanoDrop™ One Spectrophotometer (Thermo Fisher Scientific, MA, USA). qRT-PCR was performed using a SYBR Green Master Mix Kit (Takara) according to the manufacturer’s instructions. GAPDH was used as an internal control for normalization. Cell transfection and RNA interference RNA sequences targeting three different sites in MAP1LC3B were designed by Genomeditech. The sense and antisense sequences of small interfering RNAs (siRNAs) are listed in Supplementary Table S1. MAP1LC3B siRNAs were transfected into cells using Lipofectamine 3000 (Invitrogen, USA). Cells were lysed 72 h post-transfection, and protein levels were analyzed by western blotting. Cell proliferation Cell viability was assessed using a Cell Counting Kit-8 (CCK-8, Yeasen) according to the manufacturer’s instructions. Cells were seeded into 96-well plates, and 10 μL of CCK-8 solution was added to each well containing 100 μL of culture medium. Plates were incubated for 1–2 h at 37 °C, and absorbance was measured at 450 nm. ELISAs The levels of human HMGB1 in culture supernatants, as well as HMGB1, IFN-γ, and TNF-α levels in mouse serum and IFN-γ levels in in vitro cultured primary mouse DCs or CD8+ T cells, were quantified using the Quantikine Human ELISA Kit (R&D Systems, Minnesota, USA) according to the manufacturer’s instructions. Briefly, 100 μL of supernatant was added to each well and incubated at room temperature for 2 h. Plates were washed and incubated with cytokine conjugates for an additional 2 h. Substrate solution was then added, and absorbance was measured using a Multiskan™ FC photometer (Thermo Fisher Scientific). Standard curves were used to quantify levels of these molecules in culture supernatants. Luminex cytokine assay Cytokines and chemokines were analyzed using a multiplex magnetic Luminex assay. Supernatants from MHCC97-H cells treated with anlotinib and from control groups were collected. Samples were centrifuged at 4 °C at 2000–3000 × g for 5 min to remove precipitates, and the supernatants were collected. Equal volumes of supernatants were ultrafiltered and concentrated, and the concentrates were reconstituted to the original volume. Each group included three biological replicates. Samples were sent to Shanghai Universal Biotech Co., Ltd., where cytokine and chemokine levels were analyzed using the MILLIPLEX® Mouse Cytokine/Chemokine Magnetic Bead Panel—Immunology Multiplex Assay. Animal models All animal care and experimental protocols were approved by the Animal Care and Use Committee of Zhongshan Hospital, Fudan University. In the subcutaneous HCC model, 5 × 106 Hepa1-6 cells were suspended in 100 μL of PBS and injected subcutaneously into the right flank of 6-week-old male C57BL/6 J mice (Charles River Company). For the orthotopic HCC model, 1 × 106 Hepa1-6 cells were resuspended in 10 μL of PBS, mixed with an equal volume of Matrigel Matrix Basement Membrane (356,234), and kept on ice. This mixture was then horizontally injected into the right liver lobe parenchyma in 6-week-old C57BL/6 J mice using a Hamilton® SampleLock syringe (25 μL). Seven days post-implantation, mice bearing Hepa1-6 tumors were randomly assigned to the designated treatment groups. In studies evaluating anlotinib monotherapy, tumor-bearing mice received either anlotinib [5 mg/kg dissolved in 100 μL, administered intragastrically (i.g.)] or 0.9% NaCl [100 μL (i.g.)] as a control. This dosage was selected based on previous murine studies in other solid tumor models [37–39]. In the combination therapy study, anlotinib and/or a 200 μg dose of an immunotherapeutic monoclonal antibody were administered intraperitoneally following the schedule illustrated in Fig. 6A. Briefly, the immunotherapy antibody was administered three times a week for four doses total, and anlotinib was administered daily for 11 days. For in vivo T cell and DC depletion studies, mice were injected with either αCD8α antibody, αCD4 antibody, both antibodies [200 μg in 100 μL PBS (i.p.)] [40–42], or cytochrome c [5 mg in 100 μL PBS (i.v.)] [43–45], according to the schedule shown in Fig. 6B. All mice were housed under specific pathogen-free (SPF) conditions at the Zhongshan Hospital Animal Center. Mice were weighed daily and routinely monitored to assess tumor growth, normal activity levels, food and water intake, body weight changes, and other clinical signs. All procedures were designed to minimize the number of animals used and their suffering. Anesthesia and euthanasia procedures adhered to widely accepted veterinary best practices (e.g., the use of CO2). At the study endpoint, mice were euthanized, tumors in the liver were excised, and tumor volume was calculated using the formula: length (mm) × width2 (mm) × 0.5. Immunotherapy is potentiated by ICD-driven combination therapy. A Tumor volume comparison in orthotopic hepatocellular carcinoma (HCC) models treated with vehicle control, anlotinib monotherapy, anti-PD-1, or anti-PD-1 combined with anlotinib. n = 4 mice per group; statistical analysis by two-way ANOVA with Tukey's multiple comparisons. B Tumor volume changes in mice receiving various immune combination strategies, including anti-PD-1 with or without anlotinib, anti-CD4, anti-CD8, or cytochrome c (Cyto C). n = 4 mice per group. C Flow cytometry analysis showing the proportions of CD45⁺ leukocytes, CD8⁺ T cells, and dendritic cells (DCs) infiltrating the tumor microenvironment under different treatments. Data represent mean ± SEM from n = 4 mice per group. D Expression levels of calreticulin (CALR), CD80, and CD40 on CD45⁺ immune cells across treatment groups, as indicators of antigen presentation and costimulatory signaling. E Quantification of serum levels of HMGB1, TNF-α, and IFN-γ under each treatment condition, reflecting systemic immunogenic responses. Measured by ELISA; n = 4 mice per group. F Representative magnetic resonance imaging (MRI) of liver tumors before and after treatment with anlotinib (8 mg daily) combined with anti-PD-1 (200 mg every 3 weeks), showing marked tumor regression and regression of portal vein tumor thrombus. Postoperative MRI at one month indicates the resection zone. G Immunohistochemistry staining for HMGB1, CD4, CD8, and CD11c illustrates immune cell distribution in the tumor center and invasive margin, imaged at 40 × magnification To minimize variability introduced during tumor establishment, each treatment group initially consisted of 6–8 mice. Prior to treatment initiation, animals exhibiting abnormal tumor growth were excluded to ensure group homogeneity. Four mice per group with comparable baseline tumor burdens were selected for treatment and subsequent analyses to maintain consistency across experimental arms. For orthotopic models, tumor size was assessed by high-frequency ultrasound before randomization to achieve balanced baseline tumor burdens across groups. Patient-derived organotypic tumor spheroid preparation All fresh human HCC tumor samples used for PDOTS generation were obtained from surgical resections and pathologically confirmed as hepatocellular carcinoma after processing, with any samples diagnosed as non-HCC being excluded [46]. Fresh human HCC tumor samples were immediately placed in pre-chilled RPMI-1640 medium on ice. Under sterile conditions, samples were minced using sterile scalpels and forceps, then transferred to a 10 cm dish on ice. The minced tissue was resuspended in a digestion mixture consisting of RPMI-1640 (Gibco, NY, USA) supplemented with 1:100 penicillin–streptomycin, 10% FBS (Sigma-Aldrich, MO, USA), 100 U/mL collagenase IV (STEMCELL), and 0.08 mg/mL Dispase I (Sigma-Aldrich, MO, USA). After sedimentation, the samples were resuspended in 10–20 mL of medium. Red blood cells were lysed using RBC lysis buffer (Solarbio). The resulting suspension was filtered sequentially through 100 μm and 40 μm strainers to obtain three spheroid fractions: S1 (> 100 μm), S2 (40–100 μm), and S3 (< 40 μm). The spheroids were maintained on culture plates, with the S2 fraction being selected for further in vitro experiments. CyTOF analyses Fresh tumor tissues were processed using the Mouse Tumor Dissociation Kit (Miltenyi Biotec, #130–096-730) according to the manufacturer’s instructions. Dissociated cells were filtered through a 40 μm strainer and subjected to a 40% and 70% Percoll gradient centrifugation to remove debris and enrich immune cells. Residual red blood cells were eliminated using RBC lysis buffer (Solarbio). Cells were then suspended in 100 μL of live cell staining mixture (PBS containing 250 nmol/L cisplatin) and incubated on ice for 5 min. Surface staining was performed with an in-house antibody panel for 30 min, followed by overnight fixation. Intracellular staining was conducted using permeabilization buffer (eBioscience) as per the manufacturer’s protocol. Cells were washed, resuspended, and stored at 4 °C prior to analysis using a Helios mass cytometer (Fluidigm). Detailed information on antibody clones and suppliers is provided in Supplementary Table S2. Mass cytometry data were randomized and normalized before analysis. Data analysis was performed using Cytobank (https://cytobank.cn), applying sequential gating strategies for cells, live cells, singlets, and CD45⁺immune cells. Detailed gating procedures are presented in Supplementary Fig. 4H. Bone marrow-derived dendritic cell (BMDC) culture and stimulation BMDCs were isolated from the femurs of C57BL/6 J mice. After removing surrounding muscles, femurs were rinsed in PBS, and the bone marrow was flushed out using a syringe. Cells were passed through a 70 μm cell strainer and treated with RBC lysis buffer for 5 min. After centrifugation at 500 g for 10 min, cells were resuspended in RPMI-1640 containing 10% FBS, 1:100 penicillin–streptomycin, GM-CSF (10 ng/mL), and IL-4 (10 ng/mL), and cultured for 7 days to generate BMDCs. Differentiated DCs were purified using MojoSort™ Mouse CD11c Nanobeads (BioLegend) and plated in six-well plates. These BMDCs were co-cultured with pre-treated Hepa1-6 cells at a 1:1 ratio for 24 h. Afterward, the cells were harvested for DC activation and antigen presentation assays or further re-isolated and co-cultured with splenic CD8+ T cells at a 1:10 ratio. Supernatants were collected for IFN-γ release assays or for additional flow cytometry analysis. Flow cytometry and CFSE labeling Tumor cells pretreated with different TKIs or harvested from co-culture systems were collected, filtered through a 70 μm cell strainer, and washed with PBS. Cells were stained with various antibodies targeting CALR, HLA-ABC, H2-KD, CD11c, CD86, CD8α, CD25, CD69, and IFN-γ. After staining, cells were centrifuged, resuspended in PBS, and analyzed via flow cytometry. To assess dendritic cell antigen presentation, Hepa1-6 cells were labeled with CellTrace CFSE and co-cultured with BMDCs as described in Sect. "Bone marrow-derived dendritic cell (BMDC) culture and stimulation". Following co-culture, cells were stained for CD11c. The presence of CD11c+CFSE+ cells indicated the uptake of tumor antigens by BMDCs. Bioinformatics analyses RNA sequencing of samples was performed on the Illumina NovaSeq 6000 platform (OE Biotech Co., Ltd., Shanghai, China), generating 150 bp paired-end reads. Differential expression analysis was conducted using DESeq2, with thresholds set at a Q value < 0.05 and fold change > 2 or < 0.5 to identify significantly differentially expressed genes (DEGs). GO and KEGG pathway enrichment analyses were conducted based on the hypergeometric distribution using R software (v3.2.0) to identify significantly enriched terms. Gene set enrichment analysis (GSEA) was performed using the GSEA software. To explore the correlation between HMGB1 expression and DC infiltration in HCC, we utilized the Tumor Immune Estimation Resource (TIMER) (https://cistrome.shinyapps.io/timer/) and Gene Expression Profiling Interactive Analysis 2 (GEPIA2) (http://gepia2.cancer-pku.cn/) tools, both of which reanalyzed data from The Cancer Genome Atlas (TCGA). Statistical analyses Quantitative data were analyzed using Student’s t-test, Mann–Whitney U test, or Kruskal–Wallis test followed by Dunn’s post hoc test, as appropriate. Qualitative data were evaluated with Pearson’s chi-squared test or Fisher’s exact test, depending on data distribution. Survival curves were estimated using the Kaplan–Meier method, and differences were assessed using the log-rank test. All statistical analyses and data visualizations were conducted using appropriate R and Bioconductor software packages. Results High levels of tumor immune infiltration correlate with improved immunotherapy outcomes in HCC patients To investigate the mechanisms underlying differential responses to immunotherapy in HCC, we began by analyzing transcriptomic datasets from immunotherapy-treated cohorts (GSE202069 and GSE153203). In responders from the GSE202069 cohort, differentially expressed genes (DEGs) were significantly enriched in anti-tumor immune pathways, including T cell activation, cytokine production, leukocyte migration, and DC differentiation/chemotaxis (Fig. 1A). Comparative pathway analysis identified 608 upregulated pathways shared between responders, among which 325 anti-tumor immune-related pathways were further validated in the murine immunotherapy dataset GSE153203 (Supplementary Fig. 1A). High levels of tumor immune infiltration correlate with improved immunotherapy outcomes in HCC patients. A GSE202069 shows anti-tumor immune pathway enrichment in HCC patients responding to immunotherapy. Key pathways are highlighted (-Log10(P.adj)). Enrichment analyzed using clusterProfiler v4.0 and org.Hs.eg.db. B Heatmap showing correlations between ICD-DAMP genes and immune checkpoints (CD274, PDCD1) in TCGA-LIHC. Z-scores indicate gene expression. C Correlations between ICD-DAMP genes and immune cell infiltration in TCGA-LIHC. D Heatmap showing ICD-DAMP gene expression and immune cell infiltration in HCC (GSE202069), including patient demographics, ICD subtypes, and RECIST response (responder n = 8, non-responder n = 8; total n = 16). E Bar chart comparing proportions of responders (green) and non-responders (purple) stratified by ICD-DAMP levels (High vs. Low) in GSE202069. Statistical significance assessed by Fisher's exact test. F Kaplan–Meier curves showing OS in HCC patients (GSE202069) by ICD-DAMP levels (High/Low). HR and p-value indicate survival differences. Survival analysis performed using log-rank test. G ROC curves for response prediction to anti-PD-1 and anti-PD-L1 across pan-cancer cohorts. AUC, p-values, optimal cutoffs, and performance metrics (TPR/TNR) are shown. Analysis was performed using ROCplot.org (http://www.rocplot.org), based on publicly available immunotherapy datasets. H Normalized expression of ICD-DAMP genes in our newly generated HCC combination therapy cohort. Each column represents an individual patient (responder n = 9, non-responder n = 10; total n = 19), color-coded by gene expression level Given the critical role of ICD in triggering the release of DAMPs and promoting anti-tumor immunity, we utilized an ICD-DAMP gene signature adapted from a previously published panel of 33 genes [47, 48]. This signature was specifically tailored for hepatocellular carcinoma (HCC) by refining it to 28 genes, primarily by excluding genes with expression profiles less representative of, or divergent in, HCC compared to other malignancies, thereby optimizing the signature's relevance for our study. This signature showed pronounced enrichment in immunotherapy responders across both HCC cohorts (Supplementary Fig. 1B–F). Leveraging TCGA data, we observed strong positive correlations between ICD-DAMP gene expression and the expression of genes encoding immune checkpoint molecules, including CD274 (PD-L1), PDCD1 (PD-1), and 27 additional checkpoints (Fig. 1B, Supplementary Fig. 1G). Furthermore, ICD-DAMP expression correlated with increased infiltration of DCs and cytotoxic T cells in the tumor microenvironment (Fig. 1C). To evaluate the predictive value of this ICD-DAMP signature, we stratified GSE202069 HCC patients (n = 16) into ICD-DAMP-High (n = 8) and ICD-DAMP-Low (n = 8) subgroups based on pretreatment transcriptomic profiles. Distinct immune infiltration patterns emerged: responders exhibited elevated ssGSEA scores for DCs, T helper cells, NK cells, and macrophages, whereas non-responders displayed diminished infiltration of these populations (Fig. 1D). Notably, 75% of ICD-DAMP-High patients achieved favorable clinical responses, compared to 25% in the ICD-DAMP-Low group, underscoring the predictive potential of the signature (Fig. 1E). However, two ICD-DAMP-High non-responders exhibited aggressive disease features (grade III–IV differentiation, ≥ 5 tumor nodules) and reduced baseline infiltration of DC subsets and cytotoxic cells, suggesting that a high tumor burden or pre-existing immune exclusion may compromise immunotherapy efficacy even within this subgroup. A trend toward improved OS was observed in the ICD-DAMP-high group, although this did not reach statistical significance (p = 0.542), likely reflecting the limited cohort size (Fig. 1F). In a pan-cancer analysis, the ICD-DAMP signature demonstrated moderate predictive accuracy for anti-PD-1/PD-L1 therapy (AUC = 0.676–0.684, p < 0.05), but not for anti-CTLA-4 therapy (Fig. 1G). Consistently, our independent uHCC cohort (n = 24) confirmed higher baseline ICD-DAMP expression in responders compared to non-responders (Fig. 1H; Supplementary Fig. 1H). Collectively, HCC patients with high levels of tumor immune infiltration, particularly those in the ICD-DAMP-High group, appear to be more likely to achieve an effective response to immunotherapeutic treatment. ICD induction in HCC is strongly potentiated by specific TKIs Given the potential of ICD to enhance immunotherapy and the central role of TKIs in HCC treatment, we systematically evaluated five clinically used TKIs (lenvatinib, regorafenib, apatinib, anlotinib, and cabozantinib) for their capacity to induce ICD, using doxorubicin as a positive control. Dose–response assays in human (MHCC97-H) and murine (Hepa1-6) HCC cell lines revealed distinct IC50 profiles across the different TKIs (Fig. 2A). All tested TKIs significantly enhanced the expression of hallmark ICD markers. Immunofluorescence and flow cytometry demonstrated increased cell surface exposure of calreticulin (CALR) and cytoplasmic translocation of high-mobility group box 1 (HMGB1), with anlotinib showing the most pronounced effects (Fig. 2B–C). Supernatant Western blotting, ELISA and Mass spectrometry analyses further confirmed that TKIs such as anlotinib and cabozantinib significantly enhanced HMGB1 release and elevated levels of other DAMPs, including HSP70, HSP90, and ATP (Fig. 2D–G, Suplementary Fig. 2A–B). Consistent with these observations, anlotinib also promoted HMGB1 release in a panel of other human HCC cell lines, underscoring the broader applicability of our findings (Supplementary Fig. 2H).A 3D scatter plot integrating ICD markers revealed distinct clustering among treatment groups, indicating that clinically used TKIs exhibit varying capacities to induce ICD. Notably, anlotinib demonstrated a markedly stronger ability to induce ICD, comparable to that elicited by doxorubicin (Fig. 2H). ICD induction in HCC is strongly potentiated by specific TKI. A Dose–response curves of various TKIs in MHCC97-H and Hepa1-6 cells, with calculated IC50 values. Data represent mean ± SEM from three independent experiments. B Immunofluorescence staining of CALR (green) and HMGB1 (red) in MHCC97-H and Hepa1-6 cells following treatment with TKIs and doxorubicin. Arrows indicate CALR exposure. Images captured at 40 × magnification. C Flow cytometry histograms showing surface expression of CALR in MHCC97-H cells treated with TKIs and doxorubicin. Representative data from n = 3 experiments. D Quantification of HMGB1 release in MHCC97-H and Hepa1-6 cells via ELISA after drug treatment. Data represent mean ± SEM; n = 3 biological replicates. E CALR puncta formation on the cell surface of MHCC97-H and Hepa1-6 cells visualized by immunofluorescence; quantification based on membrane-localized CALR + foci per cell. F ATP release assay measuring extracellular ATP levels in MHCC97-H and Hepa1-6 cells under various treatments. Data represent mean ± SEM; n = 3 experiments. G Levels of HSP90, HSP70, and HMGB1 in supernatants of MHCC97-H and Hepa1-6 cells, assessed by immunoblot. EV and EH refer to equal volume and equal total protein loading, respectively. H 3D scatter plot illustrating the combined distribution of extracellular ATP, CALR, and HMGB1 levels across different TKI treatments. Visualization generated using OriginPro (OriginLab, v2024). I Western blot analysis of HMGB1, HSP90, and HSP70 expression in PDOTS models derived from primary HCC (pHCC) and neoadjuvant-treated HCC. J Comparison of supernatant HMGB1 levels between paired PDOTS-pHCC (n = 20) and PDOTS-neoadjuvant (n = 15) models demonstrates treatment-associated differences. Statistical significance assessed by paired t-test. K Luminex-based cytokine profiling of culture supernatants from control and Anlotinib-treated Hepa1-6 cells. Selected chemokines (e.g., MIP-3β/CCL19, SDF-1α/CXCL12) are highlighted. Supernatants were collected after 24 h of treatment with 20 μM Anlotinib. n = 3 experiments. L Chord diagram illustrating the functional enrichment of the supernatant proteome from Anlotinib-treated Hepa1-6 cells. Proteins were identified via integrated Luminex cytokine assay and mass spectrometry (MS), and linked to enriched biological processes and immunology-related pathways. Enrichment analysis performed using clusterProfiler with FDR < 0.05 To further validate anlotinib-induced ICD in HCC, we established primary HCC and neoadjuvant HCC patient-derived organotypic spheroids (pHCC/neoadjuvant HCC-PDOTS) (Supplementary Fig. 2C). In these models, anlotinib similarly increased the release of HMGB1, HSP70, and HSP90, confirming its ICD-promoting effects in heterogeneous human HCC samples with complex treatment backgrounds (Fig. 2I–J). Mechanistic RNA-seq analysis of MHCC97-H-derived samples demonstrated that anlotinib treatment significantly upregulated immune-related signaling pathways, notably the TNF signaling and NF-κB pathways, thereby promoting anti-tumor immune responses (Supplementary Fig. 2D). Increased expression of ICD-related genes and MHC class I molecules further supported the role of anlotinib in boosting immune recognition and responses (Supplementary Fig. 2E). Supernatant Mass spectrometry and Flow cytometry also confirmed enhanced release of HMGB1 and HSP90 and increased MHC class I surface expression (Supplementary Fig. 2B, F). Supernatant Luminex cytokine analysis revealed a marked increase in the levels of key cytokines (CXCL13, CXCL12, IL-2, CCL19) and a reduction in ENA-78 and MCP-1 levels following anlotinib treatment. Proteomic enrichment analysis of released proteins and cytokines linked them to immune pathways and ICD (Fig. 2K–L). Collectively, these findings highlight TKIs—particularly anlotinib—as multimodal agents that synergistically combine anti-angiogenic activity with potent ICD induction, thereby remodeling the tumor-immune interface to enhance immunotherapy responsiveness. The ER stress-autophagy axis mediates anlotinib-induced ICD Gene Set Enrichment Analysis (GSEA) revealed that pathways related to ER stress and autophagy were predominantly enriched among the top 30 pathways activated by anlotinib (Fig. 3A). Enrichment plots showed that anlotinib significantly activated ER stress sensors, the unfolded protein response (UPR) pathway, and transcriptional regulation of autophagy-related genes (Fig. 3B). A volcano plot further demonstrated the significant upregulation of ER stress-, autophagy-, and ICD-related genes in the anlotinib group (Fig. 3C). At the transcriptional level, RNA-seq analysis confirmed that anlotinib treatment significantly upregulated the mRNA expression of genes involved in ER stress (e.g., EIF2AK3, XBP1, ATF4) and key components of the autophagy machinery (e.g., MAP1LC3A, GABARAPL1, ULK1) (Supplementary Fig. 3A). The ER stress-autophagy axis mediates anlotinib-induced ICD. A Gene set enrichment analysis (GSEA) of the top 30 enriched pathways highlights key involvement of ER stress and autophagy-related signaling following Anlotinib treatment. GSEA performed using fgsea package with 1000 permutations. B Enrichment plots for ER stress and autophagy pathways show gene distribution within the ranked gene list. C Volcano plot displaying differential expression of ER stress-, autophagy-, and **ICD-DAMP-**related genes in Anlotinib-treated cells. Significance thresholds: |log2FC|> 1 and adjusted p-value < 0.05. D Western blot analysis of target engagement (p-FGFR1, FGFR1, p-ERK), ER stress (p-eIF2α, CHOP, ATF4), and autophagy (ATG5, ATG7, LC3-I/II) proteins in MHCC97-H and Hepa1-6 cells treated with increasing concentrations of Anlotinib. E, F Immunofluorescence staining of LC3-II (green), HMGB1 (red), and ATF4 (green) in MHCC97-H and Hepa1-6 cells reveals increased autophagy, HMGB1 translocation, and ATF4 nuclear translocation in response to Anlotinib treatment. Images captured at 63 × magnification. G, H Western blot analysis of supernatant DAMPs (HSP90, HSP70, HMGB1) from MHCC97-H and Hepa1-6 cells treated with Anlotinib alone or combined with ER stress inhibitors Azoramide or ISRIB. I, J Western blot analysis of supernatant DAMPs from MHCC97-H and Hepa1-6 cells treated with Anlotinib alone or combined with autophagy inhibitors Bafilomycin A1 or 3-MA. K, L Western blot and quantification of supernatant DAMPs from PDOTS-pHCC models treated with Anlotinib alone or combined with Azoramide or 3-MA. Data represent mean ± SEM; n = 5 patient-derived models. M, N Western blot analysis showing the effect of transient LC3B knockdown (siRNA) and stable ATG5 knockdown (shRNA) on autophagy markers and subsequent release of DAMPs following Anlotinib treatment To validate whether these transcriptional changes translated to the protein level and to elucidate the signaling cascade, we performed Western blotting. The results showed that anlotinib treatment led to a dose-dependent decrease in the phosphorylation of its target FGFR-1 and its downstream effector ERK. Concurrently, anlotinib induced ER stress, as evidenced by the upregulation of p-eIF2α, CHOP, and ATF4. This was coupled with a dose-dependent increase in key autophagy-related proteins, including ATG5, ATG7, and the conversion of LC3-I to LC3-II, in both MHCC97-H and Hepa1-6 cell lines (Fig. 3D, Supplementary Fig. 3B). Immunofluorescence staining confirmed these findings, revealing an increase in LC3-II-positive autophagic vesicles (Fig. 3E), HMGB1 release from the nucleus (Fig. 3F), and nuclear translocation of the transcription factor ATF4 (Fig. 3E), a known transcriptional activator of autophagy genes [49], after anlotinib treatment. To further clarify the causal link between these pathways, we used pharmacological inhibitors. Pre-treatment with the ER stress inhibitors Azoramide or ISRIB significantly attenuated the anlotinib-induced release of DAMPs (HSP90, HSP70, and HMGB1) from both MHCC97-H and Hepa1-6 cells (Fig. 3G, H, Supplementary Fig. 3C). Similarly, inhibiting autophagy with Bafilomycin A1 or 3-MA also markedly reduced the secretion of these DAMPs (Fig. 3I, J, Supplementary Fig. 3D). These findings were recapitulated in the PDOTS-pHCC model, where both ER stress and autophagy inhibitors blunted anlotinib-induced DAMP release (Fig. 3K, L, Supplementary Fig. 3E). To corroborate these findings with genetic evidence, we performed transient knockdown of LC3B (siRNA) and stable knockdown of ATG5 (shRNA). In both human and murine HCC cell lines, silencing these essential autophagy genes significantly impaired the release of HMGB1 and other DAMPs upon anlotinib treatment (Fig. 3M, N, Supplementary Fig. 3F). Collectively, these results suggest that anlotinib induces ICD through a sequential pathway involving target inhibition, ER stress, and the downstream execution of autophagy, which plays a key role in DAMP release. Anlotinib-elicited ICD remodels the tumor immune microenvironment We found that anlotinib-induced ICD demonstrated robust antitumor efficacy in both subcutaneous and orthotopic HCC models. In the subcutaneous model, anlotinib treatment significantly suppressed tumor growth, resulting in a 2.1-fold reduction in tumor volume compared to controls by day 21 (Fig. 4A). Similarly, in the orthotopic model, anlotinib reduced tumor burden by 40% relative to controls, showed no treatment-related mortality during the study period, and markedly decreased the number of pulmonary metastatic nodules (Fig. 4B, Supplementary Fig. 4B–C). To investigate the mechanistic basis underlying this efficacy, we analyzed ICD induction and tumor microenvironment remodeling. In the subcutaneous model, immunohistochemistry revealed that expression of the angiogenesis marker CD31 was significantly reduced in the anlotinib group, while expression levels of the ICD marker CALR, the autophagy marker LC3, and infiltration of CD4+ and CD8+ T cells were significantly increased. These findings indicated that anlotinib inhibits tumor angiogenesis, induces ICD and autophagy activation, and regulates the tumor immune microenvironment (Fig. 4C, Supplementary Fig. 4A). Consistent results were observed in the orthotopic model, with enhanced antitumor immune infiltration and nuclear-to-cytoplasmic translocation of HMGB1 (Fig. 4D). Additionally, histological analyses confirmed a reduction in metastatic nodules and preservation of lung structure; this anti-metastatic effect may be attributed to the inhibition of epithelial-mesenchymal transition and stemness-associated gene expression (Supplementary Fig. 4B–F). Anlotinib-elicited ICD Remodels the Tumor-Immune Microenvironment. A Tumor volume growth curves in subcutaneous HCC-bearing mice treated with Anlotinib or vehicle control. Representative images of tumor-bearing mice are shown. n = 5 mice per group; statistical analysis by two-way ANOVA. B Tumor burden in orthotopic HCC models treated with Anlotinib or control, assessed via tumor volume measurements. n = 4 mice per group. C Immunohistochemical staining of tumor sections for CD31, CALR, CD4, and CD8. Anlotinib treatment increased marker expression, indicating enhanced immune infiltration and ICD. Quantitative analysis is shown below each panel. Images captured at 40 × magnification; quantification from ≥ 5 fields per section. D Immunofluorescence staining of HMGB1, CD8, and CD4 in tumor tissues reveals enhanced immune cell presence and HMGB1 release upon Anlotinib treatment. Images captured at 40 × magnification. E Multiplex immunofluorescence showing nuclei (DAPI, blue), HMGB1 (red), CD8 (pink), and CD4 (green). Insets highlight distinct ICD stages: A pre-ICD cells with nuclear HMGB1 (yellow outline), B early ICD cells with partial translocation (yellow & red outline), C late ICD cells with extracellular HMGB1 (red outline only). Images captured at 63 × magnification with 2 × digital zoom for insets. F Spatial proximity analysis of pre-, early-, and late-ICD cells relative to CD8⁺ and CD4⁺ T cells in tumor sections. Dashed lines represent nearest neighbor distances. Analysis performed using ImageJ with nearest neighbor plugin. G t-SNE plots of tumor-infiltrating immune cells from control and Anlotinib-treated tumors, annotated by cell type. Anlotinib treatment altered the immune landscape. t-SNE performed with perplexity = 30 and 1000 iterations. H Heatmap of functional marker expression across immune cell subsets from CyTOF data. Marker intensity is shown as Z-scores, revealing immune phenotypic shifts. I Quantitative comparison of major immune cell populations in tumors from control and Anlotinib-treated mice, including lymphoid/myeloid distribution and key functional subsets (e.g., MDSCs, activated CD8⁺ T cells, Tregs, M2 macrophages). Statistical significance assessed by unpaired t-test with Welch's correction. J Immunohistochemical staining for CD11c⁺ dendritic cells in tumor and paratumoral regions of murine HCC. Anlotinib markedly increased DC infiltration, as confirmed by staining area quantification (p < 0.001). Images captured at 20 × magnification Higher HMGB1 release has been reported to correlate with increased intratumoral infiltration of antitumor immune cells. To further validate this, we employed multi-color immunofluorescence staining to spatially localize HMGB1 release and divided ICD cells into three stages. First, the pre-ICD stage was defined by complete co-localization of HMGB1 and DAPI, indicating limited HMGB1 release. In the early-ICD stage, HMGB1 was detected in the cytoplasm outside the DAPI-stained nucleus, suggesting partial release while nuclear and membrane structures remained intact. Finally, in the late-ICD stage, HMGB1 was fully released into the interstitial fluid, with fragmentation of the DAPI region, indicating exposure to the tumor microenvironment (Fig. 4E). Notably, no discernible difference in HMGB1 distribution was observed between pre-ICD and normal cells. However, upon sufficient exposure to inducers, pre-ICD cells inevitably transitioned to the early- or late-ICD stages. For statistical analysis of intercellular distances, we further defined pre-ICD and normal cells as HMGB1-release-negative cells, and early-to-late ICD cells as HMGB1-release-positive cells. In the control group, the average distance between HMGB1-release-positive cells and CD8+ T cells was ~ 28 μm, compared to ~ 52 μm for release-negative cells. Conversely, the average distance between CD8+ T cells and HMGB1-release-positive cells was ~ 17 μm, whereas the distance to release-negative cells was ~ 32 μm. In the anlotinib group, these distances decreased to ~ 11 μm and ~ 20 μm, respectively, indicating that HMGB1 released by anlotinib promotes CD8+ T cell aggregation within the tumor microenvironment. Anlotinib also reduced the distance between CD4+ T cells and tumor cells; however, this effect was independent of the ICD stage of tumor cells (Fig. 4F, Supplementary Fig. 4G). Given the regulatory potential of ICD as a mediator of immune cell infiltration, we further characterized the infiltrating immune cells within the TME. To achieve high-dimensional profiling of the TME in Hepa1-6 syngeneic mouse tumors from the two treatment groups, we performed cytometry by time-of-flight (CyTOF) analysis, assessing 37 surface and intracellular immune markers on CD45+ cells from dissociated Hepa1-6 tumors. A panel of metal-tagged antibodies was designed to identify major tumor-infiltrating immune cell populations (Supplementary Table S2), including lineage markers, functional markers (homing and activation receptors), and cytokine-related markers (Fig. 4G, H; Supplementary Fig. 4H for gating strategy). This approach enabled the categorization of CD45+ cells into lymphoid and myeloid compartments, and analysis of lineage marker expression within each compartment (Fig. 4I). Upon manual gating of each major immune lineage across individual samples from different treatment groups, no significant differences were observed in the densities of lymphoid and myeloid compartments, suggesting that specific immune subpopulations may contribute to the antitumor effect of TKI therapy. Subpopulation analysis revealed a shift in immune cell composition, characterized by an increase in CD8+ T cells, CD4+ T cells, DCs, and macrophages, along with a decrease in B cells and MDSCs. Functional assessment further demonstrated an enrichment of activated CD8+ T cells and M1-polarized macrophages, coupled with a reduction in immunosuppressive Treg cells (Fig. 4G–I). IHC analysis further confirmed increased dendritic cell infiltration, indicating enhanced antigen presentation activity in TKI-treated tumors and partially explaining the observed elevation in activated CD8+ T cells (Fig. 4J). Through integrated CyTOF, IHC, and multi-color immunofluorescence analyses in the orthotopic HCC model, these results suggest that immune microenvironment remodeling plays a critical role in the mechanism of ICD-based TKI therapy, significantly enhancing tumor cell killing. ICD-driven DC recruitment and activation enhances T cell responses To further elucidate the mechanism by which ICD-based TKI therapy remodels the immune microenvironment, we integrated Luminex and mass spectrometry data from the supernatants of HCC cells treated with TKI. Using STRING network analysis, we aimed to identify potential immune-inducing functions of secreted proteins. This analysis identified significant changes in protein–protein interactions with high confidence (0.700), highlighting anlotinib's effects in promoting DC dendrite assembly, DC migration, antigen presentation, extracellular matrix remodeling, and T cell migration (Fig. 5A). The key secreted proteins associated with ICD and DC activity were HMGB1, CCL19, and CXCL12. Analysis of the publicly available LIHC database confirmed that the expression of these three proteins was positively correlated with tumor purity and myeloid DC infiltration, and was closely associated with long-term survival in HCC patients, particularly in the TKI-treated population (Supplementary Fig. 5B–D). ICD-driven DC recruitment and activation enhances T cell responses. A Protein–protein interaction (PPI) network of supernatant proteome identified in Anlotinib-treated MHCC97-H cells. Proteins were derived from Luminex cytokine profiling and mass spectrometry of cell culture supernatants. Network analysis performed using STRING database v11.5. B Schematic illustration of the DC–T cell co-culture assay to assess ICD-induced dendritic cell activation and subsequent CD8⁺T cell priming. C Migration assay of dendritic cells treated with supernatants from control, Anlotinib-, or Doxorubicin-treated tumor cells. Anlotinib group was tested with or without HMGB1 blocking antibody. Representative images (4 × and 10 × magnification) and quantification are shown. *p < 0.001. D Assessment of dendritic cell (DC) activation (CD86, CD40 expression) and antigen capture (CFSE +) following exposure to tumor supernatants, analyzed by flow cytometry. Data represent mean ± SEM from n = 5 experiments. E Functional analysis of CD8⁺ T cells stimulated by DCs conditioned with tumor supernatants. Activation markers (CD25, CD69) and effector function (IFN-γ production) were measured by flow cytometry. Representative data from n = 5 donors. F IFN-γ levels in culture supernatants of different co-culture systems. Measured by ELISA; data represent mean ± SEM from n = 3 experiments. G Cytokine and chemokine levels in PDOTS models treated with Anlotinib or control. PDOTS-pHCC (n = 20) and PDOTS-neoadjuvant (n = 15); statistical analysis by paired t-test To further confirm the recruitment of DCs and subsequent activation of CD8+ T cells by TKIs in the induction of HCC cell death, we designed and constructed an in vitro co-culture system (Fig. 5B). BMDCs were successfully generated after 7–10 days of culture and subsequently co-incubated with conditioned media collected from Hepa1-6 cells pretreated with either anlotinib or doxorubicin (Supplementary Fig. 5A). Transwell migration assays demonstrated that conditioned media from anlotinib-pretreated cells significantly promoted BMDC migration compared to the control group, and the effect was comparable to that of the positive control, doxorubicin. Blocking HMGB1 significantly attenuated the chemotactic effect of the supernatant from anlotinib-treated tumor cells on BMDCs (P < 0.001), indicating that HMGB1 is a key chemokine mediating anlotinib-induced DC recruitment (Fig. 5C). After one day of co-incubation with ICD-induced tumor cells, BMDCs were isolated using CD11c-positive selection. CD8+ T cells were then harvested from murine spleens and co-cultured with these conditioned BMDCs. Flow cytometry was utilized to assess antigen capture by DCs and to evaluate the functional activation of both DCs and CD8+ T cells, while IFN-γ levels in cell supernatants were measured by ELISA (Fig. 5D–F). Anlotinib treatment markedly increased the proportion of activated DCs (CD86+ CD11c+) to 12.5%, compared to 8.71% in the control group and 16.4% in the doxorubicin group (P < 0.001). Additionally, the population of antigen-capturing DCs (CD11c+CFSE+) rose to 5.17%, compared to 1.95% in the control and 3.45% in the doxorubicin group, indicating enhanced antigen recognition and processing capabilities (Fig. 5D). In the subsequent DC–CD8+ T cell co-culture system, anlotinib significantly enhanced CD8+ T cell activation, as shown by increased CD25 expression (36.1% vs. 22.0% in the control, P < 0.001), CD69 expression (56.7% vs. 43.1%, P < 0.01), and IFN-γ production (29.1% vs. 11.5%, P < 0.05) (Fig. 5E). Moreover, IFN-γ was assessed as a functional marker of CD8+ T cell activation. In BMDC and Hepa1-6 co-cultures, baseline IFN-γ secretion was low across all groups but significantly increased upon the addition of CD8+ T cells. Co-culturing BMDCs with Hepa1-6 supernatants treated with anlotinib or doxorubicin further elevated IFN-γ levels compared to the control group (Fig. 5F). In primary and neoadjuvant HCC samples using the PDOTS model system, anlotinib treatment significantly increased IFN-γ secretion, further supporting its role in enhancing anti-tumor immunity in HCC patients (Fig. 5G). Overall, these results demonstrate that anlotinib induces ICD in tumor cells, activates DCs, promotes antigen presentation, enhances CD8+ T cell function, and exerts strong immunomodulatory effects, highlighting its potential as a promising TKI for the treatment of advanced HCC. Immunotherapy is potentiated by ICD-driven combination therapy Based on our hypothesis that ICD inducers can reprogram HCC tumors exhibiting low levels of immune infiltration into an ICD-DAMP-enriched phenotype to enhance immunotherapy efficacy, we conducted in vivo studies combining anlotinib with anti-PD-1 therapy. The combination group exhibited the most pronounced tumor suppression, with significantly reduced tumor volumes compared to monotherapy (anlotinib or anti-PD-1 alone) and the control group (P < 0.001) (Fig. 6A). To explore the mechanisms underlying this synergy, we conducted functional depletion studies targeting DCs and T cell subsets. Systemic administration of CD8α-depleting antibodies or cytochrome c (to induce apoptosis in CD8+ DCs) markedly reduced the antitumor efficacy of the combination therapy. While depletion of CD4+ T cells had limited effects, concurrent depletion of CD8+ T cells significantly impaired therapeutic outcomes (Fig. 6B). These findings highlight the dominant roles of CD8+ T cells and DCs in mediating the antitumor effects, with CD4+ T cells playing a lesser role. Flow cytometry analysis revealed that combination therapy significantly increased the proportions of CD45+ immune cells and CD8+ T cells within the tumor microenvironment. Consistent with expectations, anlotinib treatment alone significantly increased DC numbers, and the addition of anti-PD-1 therapy further enhanced this effect and upregulated the expression of antigen presentation molecules, including CD80, CD40, and MHC II (Fig. 6C–D). Moreover, serum levels of HMGB1, TNF-α, and IFN-γ were significantly elevated in the anlotinib group, and were further enhanced by the combination therapy (Fig. 6E). A representative clinical case involved a patient with unresectable HCC due to portal vein tumor thrombus who was enrolled in a clinical trial combining anlotinib with anti-PD-1 therapy. After six months of treatment, MRI revealed marked tumor shrinkage and regression of the thrombus, allowing for curative-intent partial hepatectomy (Fig. 6F). Immunohistochemical analysis of the resected tumor demonstrated increased infiltration of HMGB1, CD4, CD8, and DCs at both the tumor periphery and core, including the necrotic central regions. The pronounced immune cell infiltration at the tumor margins likely contributed to the successful conversion therapy observed in this patient. Thus, the remarkable antitumor effect of the combination therapy appears to be driven by the creation of a highly immune-infiltrated tumor microenvironment induced by TKI, particularly through enhanced infiltration of CD8+ T cells and DCs, thereby fulfilling our aim of converting potential non-responders into responders (Fig. 6G). Autophagy enhancement promotes the efficacy of Anlotinib-based combination immunotherapy in vivo. Building upon our previously established in vitro findings that anlotinib-induced ICD is critically dependent on the ER stress-autophagy axis, we next sought to validate this causal relationship in vivo. To achieve this, we first employed genetic models using an orthotopic HCC model with Hepa1-6 cells stably expressing shNC, shAtg5, or overexpressing Atg5. We found that the antitumor effect of anlotinib was significantly attenuated in Atg5-knockdown tumors but enhanced in Atg5-overexpressing tumors compared to controls (Fig. 7A). These findings provide genetic evidence supporting the promoting role of autophagy in anlotinib's therapeutic efficacy. Autophagy enhancement promotes the efficacy of Anlotinib-based combination immunotherapy in vivo. Autophagy enhancement promotes the efficacy of Anlotinib-based combination immunotherapy in vivo. A Genetic modulation of autophagy impacts Anlotinib efficacy in an orthotopic HCC model. Representative tumor images, tumor volumes, and tumor growth inhibition rates for mice bearing shNC, shAtg5, or Atg5-overexpressing (OE) tumors treated with vehicle or Anlotinib. n = 4 mice per group; statistical analysis by one-way ANOVA with Dunnett's multiple comparisons. B Pharmacological modulation of autophagy impacts Anlotinib efficacy. Representative tumor images and tumor volume growth curves for mice treated with Anlotinib alone or in combination with autophagy inhibitor 3-MA or the autophagy-promoting agent metformin (MET). n = 4 mice per group. C Multiplex immunofluorescence staining of tumor sections from the experiment in (B). Images show nuclei (DAPI, blue), HMGB1 (red), LC3B (green), CD11c (yellow), and CD8 (pink). D Quantification of immunofluorescence staining from (C), showing the number of cells releasing HMGB1 and the staining area of LC3B, CD11c, and CD8 per field of view. Data represent mean ± SEM from ≥ 5 fields per tumor section. E Triple combination therapy enhances antitumor efficacy. Representative tumor images and tumor volumes for mice treated with Anlotinib + anti-PD-1 (Combo) with or without metformin (MET). n = 4 mice per group; statistical analysis by unpaired t-test. F Multiplex immunofluorescence staining of tumor sections from the experiment in (E). G Quantification of immunofluorescence staining from (F). Data represent mean ± SEM from ≥ 5 fields per tumor section Next, we pharmacologically modulated autophagy. In an orthotopic model, co-administration of the autophagy inhibitor 3-MA significantly weakened the tumor-suppressive effect of anlotinib. Conversely, to test the effect of augmenting autophagy, we chose metformin (MET), a widely used agent known to promote autophagy without the profound immunosuppressive effects associated with specific inducers like rapamycin. Combination with metformin potentiated anlotinib's antitumor activity (Fig. 7B). Immunofluorescence analysis revealed that anlotinib treatment increased HMGB1 release, LC3B puncta formation, and the infiltration of CD11c+ DCs and CD8+ T cells within the tumor microenvironment. These molecular and cellular changes were reversed by the autophagy inhibitor 3-MA and further enhanced by the autophagy promoter metformin (Fig. 7C, D). Based on our finding that enhancing autophagy potentiated the antitumor effect of anlotinib, we next investigated whether this strategy could further improve the efficacy of the established anlotinib and anti-PD-1 combination therapy (Combo). Therefore, we explored a triple therapy combining metformin with the anlotinib and anti-PD-1 dual therapy. The triple combination group (Combo + MET) exhibited a further significant reduction in tumor growth compared to the dual combination (Fig. 7E). This enhanced efficacy was associated with markedly increased HMGB1 release, LC3B formation, and infiltration of CD11c+ and CD8+ cells in the tumor microenvironment, as shown by immunofluorescence (Fig. 7F, G). These results indicate that enhancing autophagy can augment the synergistic antitumor effect of anlotinib and PD-1 inhibitors. Discussion ICI therapy targeting PD-1/PD-L1 has shown promise in HCC [50, 51], yet the liver's inherent immune-tolerant environment limits efficacy to a subset of patients with immunologically "hot" tumors [52, 53]. However, the liver's inherent immune-tolerant environment limits the effectiveness of immunotherapy in this cancer type [54]. "Cold" tumors—characterized by low immunogenicity and deficient immune infiltration—remain largely unresponsive to current immunotherapies [55]. This highlights the urgent need for strategies that can systematically convert immune "cold" tumors into "hot" phenotypes to enhance ICI efficacy in HCC [56, 57]. The ICD-DAMP signature provides a precision medicine framework for patient stratification, with 75% of ICD-DAMP-High patients achieving favorable responses versus 25% in the Low group, though the limited sample size warrants validation in larger cohorts. Future optimization through multi-omics integration and machine learning could enhance predictive accuracy by incorporating additional biomarkers such as microsatellite instability status, tumor mutational burden, and HLA expression patterns. This enhanced signature could guide personalized treatment selection: high-scoring patients for standard ICI regimens, while low- and intermediate-scoring patients could benefit from ICD-inducing combinations such as anlotinib plus anti-PD-1 or ICD enhancement strategies including metformin supplementation. Real-world implementation faces challenges including tumor heterogeneity, biopsy-related risks (bleeding and needle tract seeding at < 3% incidence) [58, 59], and the need for alternative approaches like liquid biopsy. While lenvatinib represents the current standard-of-care TKI for advanced HCC [60–62], our systematic screening revealed anlotinib's markedly superior ICD-inducing capabilities. Although both agents target overlapping kinases including FGFR1, anlotinib's unique kinase inhibition profile and potency appear to trigger more robust ER stress signaling, leading to enhanced autophagy and DAMP release. This distinctive downstream effect explains anlotinib's superior performance in promoting HMGB1, CALR, and ATP release compared to other TKIs, despite similar target engagement. Critically, our PDOTS model including neoadjuvant-treated samples demonstrated that anlotinib retained ICD-inducing capacity in treatment-experienced tissues, suggesting potential efficacy in lenvatinib-refractory HCC through enhanced immune activation. We identified a novel mechanistic pathway where anlotinib induces ICD through sequential FGFR-1 inhibition → ER stress (p-eIF2α/ATF4/CHOP activation) → autophagy → DAMP release. This represents a paradigm shift from traditional DNA damage-dependent ICD induction by chemotherapeutics [63, 64]. Our mechanistic validation showed that ER stress or autophagy inhibition significantly suppressed anlotinib-induced ICD, while genetic knockdown of ATG5/LC3B impaired DAMP release. We further confirmed that anlotinib activates key downstream molecules of ER stress, including p-eIF2α, ATF4 and CHOP, and observed nuclear translocation of ATF4, demonstrating that ATF4, as a transcription factor, may directly participate in the activation of autophagy-related genes, thus linking ER stress to the autophagic process. Previous studies have revealed conflicting roles of autophagy in ICD, with some showing that antigen and ATP release during ICD depends on autophagy activation [65], while others demonstrate that disruption of the autophagy-lysosome pathway triggers ICD [66], and autophagy ablation has been linked to ER stress induction and subsequent ICD [67]. This apparent contradiction highlights the complex and context-dependent nature of autophagy's role in ICD, which may vary significantly depending on the specific ICD inducer employed. In our study, we detected LC3 in the cell culture supernatant and hypothesized that HMGB1 might be secreted in complex with LC3 or within autophagosomes. However, our immunofluorescence analysis did not reveal significant co-localization of intracellular HMGB1 and LC3 puncta, indicating that the precise molecular mechanism by which autophagy facilitates HMGB1 release remains unresolved. Future investigations comparing various ICD inducers will be essential to fully elucidate the inducer-specific molecular machinery governing autophagy's role in ICD execution. This targeted approach offers advantages over conventional ICD inducers: improved tolerability, tunable autophagy enhancement, and reduced susceptibility to p53-mediated resistance common in HCC [68]. The conserved nature of ER stress and autophagy machinery across cell types suggests that anlotinib's ICD-inducing mechanism may translate to other cancer contexts. Studies in non-small-cell lung cancer have reported tumor microenvironment remodeling following anlotinib treatment [69], supporting broad therapeutic applicability beyond HCC. The critical role of autophagy in anlotinib's efficacy led us to explore autophagy enhancement. While rapamycin represents the most potent autophagy inducer, its profound immunosuppressive effects—the very reason for its use in preventing organ transplant rejection—directly contradict our immune activation goals. Long-term rapamycin exposure increases infection and malignancy risks [70, 71], making it unsuitable for immunotherapy enhancement [72, 73]. Metformin emerged as an ideal alternative, inducing autophagy through AMPK activation without immunosuppression and potentially offering immunostimulatory properties including enhanced T cell memory formation [74, 75]. Importantly, this combination approach is clinically feasible as anlotinib plus anti-PD-1 has demonstrated acceptable toxicity in Phase III HCC trials [76], while metformin's decades-long safety record in non-diabetic populations minimizes additional risks. Several study limitations warrant acknowledgment. The limited sample size necessitates validation in larger patient cohorts to optimize the ICD-DAMP signature for improved predictive accuracy. The broader applicability of the FGFR-1 → ER stress → autophagy → ICD axis across other cancer types remains to be systematically investigated, though preliminary evidence from lung cancer studies suggests potential translatability [77]. Our preclinical models relied primarily on cell line-derived xenografts, and validation in spontaneous HCC mouse models would strengthen the translational relevance of our findings. While metformin has an established safety profile, its use in this context would constitute off-label application requiring careful regulatory consideration. We also observed enhanced autophagy in infiltrating immune cells after anlotinib or metformin treatment, whose functional consequences remain complex and context-dependent, representing a direction for future research. Furthermore, despite detecting LC3 in cell culture supernatants and hypothesizing its role in HMGB1 secretion, the precise molecular mechanisms governing the unconventional release of HMGB1 through autophagy-mediated pathways require detailed elucidation. Future research should focus on further elucidating the underlying mechanisms, optimizing the ICD-DAMP signature through biomarker integration, exploring combination timing strategies, and validating this approach across diverse HCC populations to maximize clinical impact [78, 79] (Fig. 8). Summary of the findings of this study Conclusion In summary, we identified the multifunctional TKI anlotinib as an agent that can induce ICD in HCC cells, promote the transformation of immunologically “cold” tumors into “hot” tumors, and, when combined with anti-PD-1 therapy, elicit a potent and durable systemic immune response. These findings provide a theoretical basis for effectively eliminating HCC and improving the efficacy of immune checkpoint blockade therapies. In principle, this combination strategy, potentially further enhanced by autophagy modulators like metformin, may be applicable to a wide range of solid tumors, particularly “cold” tumors that respond poorly to ICB therapies, highlighting its potential for future clinical translation. Supplementary Information Additional file 1 (DOCX 2107 KB) Additional file 2 (XLSX 12 KB) Additional file 3 (XLSX 12 KB) Acknowledgements We thank all participants for their support in this study. Author contributions J.F, X.W, X.F, J.Q and Z.D conceived and designed the study. S.F, E.D, Z.G, M.X, P.K, C.S and J.C participated in the experiment. Y.S, W.L, W.W, X.Z, A.H and X.L participated in database search, data collection, and quality assessment. Q.G, G.S, A.K and J.Z performed statistical analysis. All authors read and approved the final manuscript. Funding This work was supported by grants from Shanghai Sailing Program (23YF1433100), the National Natural Science Foundation of China [81972229, 82172610], Natural Science Foundation of Shanghai [20ZR1473100], Natural Science Foundation of Fujian Province, [2024J011443], and Xiamen medical health key project [3502Z20244014]. Data availability The RNA‐seq data can be viewed in the National Omics Data Encyclopedia (NODE database: OEP00006183). The proteomic data have been deposited in the ProteomeXchange Consortium (iProX database: IPX0011935000). Declarations Ethics approval and consent to participate Ethical approval was provided by the Ethics Committee of Zhongshan Hospital, and written informed consent was obtained from each patient. Competing interests The authors declare no competing interests. Footnotes Publisher's Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Shan-Ru Feng, En-Fu Dong, Zheng Gao, and Mo-Ran Xu have contributed equally to this work. Contributor Information Jia Fan, Email: fan.jia@zs-hospital.sh.cn. Xiao-ying Wang, Email: wang.xiaoying@zs-hospital.sh.cn. Jing-Bo Qie, Email: Jingboqie@fudan.edu.cn. Xiu‑Tao Fu, Email: fu.xiutao@zs-hospital.sh.cn. Zhen‑Bin Ding, Email: ding.zhenbin@zs-hospital.sh.cn. References Associated Data Data Availability Statement The RNA‐seq data can be viewed in the National Omics Data Encyclopedia (NODE database: OEP00006183). The proteomic data have been deposited in the ProteomeXchange Consortium (iProX database: IPX0011935000).