In silico design of CLDN4-directed peptide-loaded nanobubbles for ultrasound-assisted delivery, photothermal therapy, immune activation, and multimodal imaging in ovarian cancer Authors: Yi Xu, Yue Hua, Chenyi Zhang, Yusha Zhang, Jiale Wan, Cheng Feng, Enqi Qiao, Wendi Guo, Li Ou, Fang Yang, Yang Shen Published: 2026 ABSTRACT BackgroundOvarian cancer remains among the most lethal gynecologic malignancies, partly because occult peritoneal disease limits complete macroscopic cytoreduction and image-guided intervention. Claudin 4 (CLDN4) is enriched in epithelial ovarian carcinoma but is not tumor-exclusive; therefore, rational CLDN4-directed delivery requires consideration of tumor enrichment, surface accessibility, and ligand engagement. This study aimed to develop a CLDN4-directed peptide–nanobubble phototheranostic platform integrating molecular recognition, ultrasound/fluorescence imaging, ultrasound-associated IP-related delivery/retention, and photothermal treatment.ResultsStructure-guided in silico optimization and surface plasmon resonance validation identified P15 as a CLDN4-binding peptide with nanomolar affinity (steady-state K_D = 11.28 nM). P15 was conjugated with indocyanine green (ICG) to generate an ICG–P15 photothermal conjugate (IP), which was incorporated into sulfur hexafluoride-filled lipid nanobubbles to form IP/NBs. Surface-preserving binding and competition assays, non-permeabilized CLDN4 immunofluorescence, cellular uptake imaging, and patient-derived organoid assays supported CLDN4-associated P15/IP interaction and intratissue penetration in the tested ovarian cancer models. IP/NBs preserved ultrasound visibility and photothermal responsiveness and showed ultrasound-associated changes in IP release and transport-related readouts under the applied acoustic conditions. In vivo imaging demonstrated tumor-associated IP-related fluorescence after IP/NBs administration, with D-NBs serving as a non-targeted dye-loaded nanobubble comparator. Matched post-ultrasound imaging further showed higher tumor-associated fluorescence in the IP/NBs + ultrasound group than in the no-ultrasound group, and an IP/NBs + near-infrared irradiation group without ultrasound showed intermediate tumor suppression compared with the full IP/NBs + ultrasound + near-infrared treatment. The full treatment regimen produced marked tumor suppression and was associated with increased oxidative-stress readouts, calreticulin exposure, ATP/HMGB1 release, cytokine secretion, immune-cell activation, and tumor-associated T-cell infiltration.ConclusionsThis study develops and evaluates a CLDN4-directed, IP-containing nanobubble framework for ovarian cancer phototheranostics. The data support ultrasound-associated enhancement of tumor IP-related delivery/retention, photothermal tumor suppression, and treatment-associated immune activation under the tested workflow, while defining areas for further optimization of peptide stability, nanobubble pharmacology, acoustic dosing, and translational safety.Graphical abstract Supplementary InformationThe online version contains supplementary material available at 10.1186/s12951-026-04675-w. Background Ovarian cancer remains a major clinical challenge because of its occult onset, late-stage presentation, and frequent therapeutic resistance. Over 75% of patients are diagnosed at an advanced stage, contributing to a 5-year survival rate of less than 30% [1, 2]. Within this context, achieving complete macroscopic cytoreduction (R0) remains a primary determinant of both overall and progression-free survival and is emphasized in current clinical consensus guidelines [3, 4]. Nevertheless, intraoperative detection of occult peritoneal implants and microscopic residual disease remains challenging, and even small unresected lesions can contribute to disease recurrence [5, 6]. These inherent clinical limitations highlight a critical need for molecular strategies capable of enhancing precise lesion localization and image-guided intervention throughout the perioperative continuum. Previous efforts to address this challenge have explored the overexpression of folate receptor alpha (FRα) in ovarian carcinoma, including folate-conjugated fluorescence probes for tumor-selective intraoperative imaging during surgical resection [7]. Although clinically informative, this approach is primarily designed for intraoperative lesion visualization and its performance can vary with operative context and lesion characteristics. In addition, endogenous folate and heterogeneous FRα expression may affect tumor-to-background contrast and lesion localization in selected settings. Alternative targeting modalities utilizing antibodies and large protein ligands are similarly constrained by translational hurdles, including restricted tissue penetration, potential immunogenicity, and complex manufacturing protocols [8]. These persistent limitations have driven the investigation of alternative, pharmacologically tractable membrane targets associated with tumorigenesis. Claudins, integral tetraspan proteins fundamental to the maintenance of epithelial barrier integrity [9, 10], are increasingly implicated in malignant transformation through their structural dysregulation. The recent clinical success of CLDN18.2-directed therapeutics, most notably zolbetuximab in gastrointestinal malignancies [11–13] has clinically supported the translational viability of targeting this protein family. Within this family, CLDN4 represents a therapeutically relevant candidate in ovarian cancer because of its frequent enrichment in epithelial ovarian carcinoma and its reported association with aggressive tumor phenotypes [14–17]. Importantly, CLDN4 is also physiologically expressed in several normal epithelial organs; therefore, its translational utility relies not on strict tumor exclusivity, but rather on tumor enrichment, disease-associated junctional remodeling, and the presence of surface-accessible CLDN4 in the malignant context [16]. Accordingly, CLDN4 expression and membrane accessibility should be considered distinct but complementary requirements for ligand-directed delivery. To target CLDN4, the C-terminal domain of Clostridium perfringens enterotoxin (C-CPE), a natural CLDN4-binding ligand, provides a structurally validated scaffold for ligand engineering [18–20]. Relative to bulky macromolecular ligands, peptide derivatives may offer advantages in tissue penetration, potentially reduced immunogenicity, and scalable synthesis [8, 21]. A major barrier, however, is that direct C-CPE-derived peptide fragments often show reduced binding affinity compared with the native enterotoxin. To address this limitation, we employed computer-aided drug design (CADD) as a ligand-optimization strategy [22, 23]. Through rational design and experimental validation, we developed a minimal CLDN4-binding peptide with nanomolar affinity for subsequent imaging and therapeutic studies. In this context, near-infrared photothermal therapy offers a strategy for coupling lesion visualization with localized tumor ablation, with conceptual relevance to image-guided intervention and residual disease management [24, 25]. Leveraging the optimized peptide P15, we chemically conjugated P15 to the clinically used fluorophore indocyanine green (ICG) via a 6-aminohexanoic acid linker to generate IP, a CLDN4-directed photothermal conjugate. To generate an ultrasound-visible formulation, IP was further incorporated into SF₆-filled lipid nanobubbles, yielding IP/NBs (Fig. 1). Within this formulation, the SF₆ gas core provides ultrasound echogenicity, whereas the lipid shell contributes to colloidal stabilization and payload incorporation [26, 27]. More broadly, ultrasound-responsive bubble-based carriers have attracted interest in oncology because, under selected acoustic parameters, they can increase vascular and cellular permeability and thereby facilitate localized transport of associated therapeutic payloads [28]. Ultrasound-triggered bubble activity may also influence the tumor microenvironment and support downstream immune activation, although the extent of these effects depends on formulation properties, acoustic parameters, and tumor context [29–31]. Fig. 1Schematic overview of the CLDN4-directed peptide–nanobubble phototheranostic strategy. (a) Computer-aided peptide discovery workflow, including in silico design/selection of CLDN4-binding motifs and experimental affinity verification by SPR, leading to identification of the lead peptide P15. (b) Incorporation of the ICG–P15 conjugate (IP) into SF₆-filled lipid nanobubbles and the proposed workflow integrating ultrasound-assisted delivery, IP-related tumor retention, photothermal treatment, and treatment-associated immune activation Schematic overview of the CLDN4-directed peptide–nanobubble phototheranostic strategy. (a) Computer-aided peptide discovery workflow, including in silico design/selection of CLDN4-binding motifs and experimental affinity verification by SPR, leading to identification of the lead peptide P15. (b) Incorporation of the ICG–P15 conjugate (IP) into SF₆-filled lipid nanobubbles and the proposed workflow integrating ultrasound-assisted delivery, IP-related tumor retention, photothermal treatment, and treatment-associated immune activation Based on this rationale, we developed a CLDN4-directed, IP-containing nanobubble phototheranostic framework that integrates peptide-guided molecular recognition, ultrasound and fluorescence imaging, ultrasound-assisted IP-related local delivery/retention, and photothermal treatment. We hypothesized that early ultrasound exposure would enhance tumor-associated IP delivery/retention by promoting local transport and release under the tested acoustic conditions, thereby improving the therapeutic response to delayed NIR irradiation. We further evaluated whether this ultrasound-assisted photothermal treatment was associated with ICD-associated danger signaling and downstream immune activation. Using CLDN4-positive ovarian cancer models, we evaluated surface-accessible CLDN4 recognition, P15/IP-associated cellular and organoid penetration, nanobubble physicochemical and imaging performance, ultrasound-assisted IP delivery/retention, photothermal tumor suppression, and treatment-associated immune activation. Methods Materials and instrumentation Fmoc-protected amino acids, 2-chlorotrityl chloride resin, HBTU, HOBt, FITC, and ICG were purchased from GL Biochem (Shanghai, China). DIPEA, TIPS, and trifluoroacetic acid were obtained from Sigma-Aldrich (Shanghai, China). SF₆ gas was supplied by Anhui Qiangyuan Gas Co., Ltd. DSPC and DSPE-mPEG2000 were purchased from Thermo Fisher Scientific. Peptides and dye-conjugated derivatives were synthesized by standard Fmoc solid-phase peptide synthesis and purified by preparative HPLC. Peptide purification was performed using an Agilent 1290 Infinity HPLC system, and analytical HPLC was carried out on an Agilent 1260 system. Peptide and conjugate characterization was performed by ESI-MS on a Finnigan LCQ Advantage spectrometer. SPR measurements were performed on a Biacore T200 instrument. Initial peptide–CLDN4 binding was assessed by equilibrium binding analysis to obtain steady-state K_D values. For kinetic analysis, SPR sensorgrams of P02, P14, P15, and P21 were fitted using a 1:1 Langmuir binding model to obtain k_a, k_d, and derived K_D values, calculated as k_d/k_a. The resulting kinetic parameters were used for comparative analysis of validated candidate peptides. Single-cell transcriptomic analysis of CLDN4 enrichment in ovarian carcinoma Single-cell RNA-sequencing data from eight ovarian carcinoma samples (GEO: GSE130000), including treatment-naïve primary tumors (n = 4), peritoneal metastases (n = 2), and relapsed lesions (n = 2), were analyzed to identify tumor-enriched candidate targets. After quality control, 13,369 high-quality cells were retained by excluding cells with mitochondrial gene content > 20%, fewer than 200 detected genes, more than 6,000 detected genes, or low library complexity (novelty score < 0.8). Downstream analyses were performed using Seurat v4.3. Gene-expression matrices were normalized using SCTransform, and 3,000 highly variable genes were selected for downstream analysis. Principal component analysis was performed on the scaled variable genes, and the top 50 principal components were used for Uniform Manifold Approximation and Projection (UMAP) and graph-based clustering (resolution = 0.8). Cell populations were annotated according to canonical marker-gene expression patterns. To identify genes enriched in the tumor/epithelial candidate compartment, differential expression analysis was performed between the marker-defined tumor/epithelial candidate cells and the remaining cell populations within the dataset, using thresholds of false discovery rate (FDR) < 0.01, |log₂ fold change| > 1, and a minimum expressing-cell fraction > 0.25. Genes meeting these criteria were used to prioritize tumor-enriched candidate targets, including tight-junction-associated genes. To further examine whether CLDN4 expression varied across clinical disease states, a marker-defined tumor/epithelial candidate compartment was extracted based on high EPCAM/KRT19/KRT8/KRT18 expression and low PTPRC/LST1/COL1A1/PECAM1 expression. CLDN4 expression was then summarized at the sample level using pseudo-bulk mean expression, with one value generated per sample. Differences across primary, metastatic, and relapsed lesions were evaluated using the Kruskal–Wallis test. For the exploratory EMT/migration-related analysis, an EMT/migration signature score was calculated within the marker-defined tumor/epithelial candidate compartment. The signature was based on a curated panel of EMT- and migration-associated genes, including VIM, FN1, CDH2, SNAI1, SNAI2, ZEB1, ZEB2, TWIST1, MMP2, MMP9, MMP14, ITGA5, ITGB1, ITGAV, ITGB3, SERPINE1, TGFB1, and TGFBI. Genes not detected in the filtered expression matrix were omitted from score calculation. For each cell, the EMT/migration score was calculated as the mean SCTransform-normalized expression of the detected signature genes. Sample-level EMT/migration scores were then generated by averaging cell-level scores within the tumor/epithelial candidate compartment for each sample. The relationship between sample-level CLDN4 expression and the EMT/migration score was visualized descriptively because only eight samples were available. Design of CLDN4-binding peptide sequences The crystal structure of the CLDN4–C-terminal Clostridium perfringens enterotoxin (C-CPE) complex (PDB: 5B2G) was used as the structural template for peptide design. To identify residues contributing to CLDN4 recognition, per-residue energetic analysis and alanine-scanning mutagenesis were performed using Discovery Studio 2021 (BIOVIA), and residue-level contributions to binding were estimated by MM/GBSA. Based on these analyses, energetically important residues were identified and grouped into two functional regions for peptide-sequence optimization. Saturation mutagenesis was then performed at selected hotspot positions within these regions by systematically substituting each residue with the 20 canonical amino acids. Variants predicted to exhibit more favorable binding free energies than the corresponding parental sequence(s) were retained to construct a focused candidate peptide library. For initial ranking, binding free energies of the resulting variants were evaluated by MM/GBSA using the OPLS4 force field and the VSGB2.0 solvation model. Candidates were ranked according to predicted binding free energy, and top-ranked sequences were advanced for experimental validation. Molecular dynamics simulations Predicted peptide structures generated by AlphaFold3 were docked to CLDN4 using HDOCK to obtain initial peptide–receptor complexes. Molecular dynamics simulations were then performed in GROMACS to evaluate the conformational stability of the docked complexes in an explicit-solvent environment. Each complex was solvated in a TIP3P water box, neutralized with counterions, and subjected to energy minimization and equilibration before production simulation. Production simulations were carried out for 50 ns, and trajectory stability was evaluated by root-mean-square deviation (RMSD) analysis. For comparative prioritization of candidate peptides, MM/GBSA calculations were performed on the resulting 50-ns trajectories, and the calculated binding free energies were used as supportive ranking evidence for candidate selection. Cell culture The human ovarian carcinoma cell line SKOV3 (ATCC HTB-77; RRID: CVCL_0532), OVCAR3 cells (ATCC HTB-161; RRID: CVCL_0465), normal ovarian epithelial cells IOSE-80 (RRID: CVCL_5546), and bEnd.3 endothelial cells (ATCC CRL-2299; RRID: CVCL_0170) were used in this study. SKOV3 cells were cultured in McCoy’s 5 A medium supplemented with 10% fetal bovine serum (FBS; Gibco), 100 U/mL penicillin, and 100 µg/mL streptomycin at 37 °C in a humidified atmosphere containing 5% CO₂. OVCAR3 cells were cultured in RPMI-1640 medium supplemented with 10% FBS, 100 U/mL penicillin, and 100 µg/mL streptomycin. bEnd.3 endothelial cells were cultured in DMEM supplemented with 10% FBS, 100 U/mL penicillin, and 100 µg/mL streptomycin. IOSE-80 cells were maintained under the supplier-recommended culture conditions. All cultures were routinely tested to confirm the absence of mycoplasma contamination. Construction of stable CLDN4-overexpressing cell lines Lentiviral particles carrying human CLDN4 cDNA in the pLV4ltr-Puro expression vector and the corresponding empty-vector control were obtained from Tsingke Biotech (Beijing, China). SKOV3 cells were transduced with the CLDN4-expressing lentivirus to establish a stable CLDN4-overexpressing SKOV3 cell line, referred to as SKOV3 CLDN4⁺/⁺ in this study. Transduced cells were selected in culture medium containing puromycin (2 µg/mL) for 5–7 days. Stable CLDN4 overexpression was verified by flow cytometry before subsequent in vitro and in vivo experiments. Flow cytometric analysis of CLDN4 and immune activation markers Cells were resuspended in ice-cold PBS containing 1% BSA and incubated with a live/dead viability dye where applicable, followed by fluorochrome-conjugated antibodies for 30 min at 4 °C in the dark. For CLDN4 flow-cytometric staining, cells were analyzed as non-permeabilized single-cell suspensions without detergent permeabilization. Data acquisition was performed on a CytoFLEX S flow cytometer (Beckman Coulter), and compensation was established using single-stained controls. Fluorescence-minus-one (FMO) controls were used where appropriate to define gating boundaries. Flow cytometry data were analyzed using FlowJo v10.8.1. The following antibodies were used across the indicated staining panels: FITC anti-human CLDN4 (Miltenyi Biotec, 130-114-869; clone REA898), Brilliant Violet 421 anti-human CD45 (BioLegend, 304032; clone HI30), PE anti-human CD11c (Proteintech, PE-65086; clone 3.9), APC anti-human CD80 (Thermo Fisher Scientific/Invitrogen, MA1-19464; clone MEM-233), FITC anti-human CD86 (Proteintech, FITC-65165; clone BU63), PE/Cyanine5.5 anti-human CD69 (Thermo Fisher Scientific/Invitrogen, MHCD6918; clone CH/4), FITC anti-human CD3 (Proteintech, FITC-65151; clone UCHT1), PE anti-human CD4 (Beckman Coulter, A07751; clone 13B8.2), and PE anti-human CD8α (Beckman Coulter, A07757; clone B9.11). Dendritic-cell maturation and T-cell activation were analyzed using separate staining panels. CD4 and CD8α staining was performed in separate T-cell staining panels where fluorochrome overlap required separate acquisition. Standard ultrasound and NIR treatment parameters Unless otherwise indicated, a single transient ultrasound exposure used for release, cellular treatment, organoid treatment, endothelial transport, matched post-ultrasound fluorescence imaging, and in vivo therapy was applied for 1 min in continuous-wave mode at 1.0 MHz using an unfocused transducer (DeYin DM200B) at 1.5 W/cm², with an estimated PNP of approximately 0.21 MPa and an estimated MI of approximately 0.21. Ultrasound gel was used as the coupling medium, and an approximately 1 cm standoff distance was maintained. For in vivo applications, this ultrasound trigger was initiated at 4 min post-injection and applied to the tumor region for 1 min where indicated. NIR irradiation, where indicated, was delivered using an 808-nm laser at 0.5 W/cm² for 5 min. In the therapeutic workflow, NIR irradiation was applied at 6 h post-injection unless otherwise specified. For ultrasound-only or NIR-only control experiments, the non-applicable modality was omitted. Cellular binding, colocalization, uptake, and ROS Imaging SKOV3 CLDN4⁺/⁺ cells (1 × 10⁵ cells per dish) were seeded in 20-mm confocal dishes (NEST) and cultured for 24 h. For FITC–P15/CLDN4 colocalization analysis, SKOV3 CLDN4⁺/⁺ cells were incubated with FITC–P15 or FITC–Control for 3 h, washed extensively with PBS, fixed, and stained with anti-CLDN4 primary antibody (Proteintech, 16195-1-AP; rabbit polyclonal; 1:200) followed by the corresponding fluorescent secondary antibody. Cells were imaged by confocal laser scanning microscopy, and line-scan analysis was performed across the indicated membrane regions to evaluate the spatial association between peptide-associated fluorescence and CLDN4 signal. For live-cell, non-permeabilized CLDN4 surface immunofluorescence staining, SKOV3 CLDN4⁺/⁺ and OVCAR3 cells were seeded on confocal dishes and cultured overnight. Cells were washed with ice-cold PBS and incubated with anti-CLDN4 antibody (Proteintech, 16195-1-AP; rabbit polyclonal; 1:200) in PBS containing 1% BSA for 30 min at 4 °C before fixation, without detergent permeabilization. After washing to remove unbound antibody, cells were fixed with 4% paraformaldehyde and incubated with the corresponding fluorescent secondary antibody under non-permeabilized conditions. Nuclei were counterstained with DAPI, and cells were imaged by confocal laser scanning microscopy. Membrane-associated CLDN4 fluorescence intensity was quantified from the cell periphery using ImageJ. No detergent permeabilization was performed at any step. For time-course uptake imaging, SKOV3 CLDN4⁺/⁺ cells were incubated with FITC–P15 or FITC–Control (10 µM) at 37 °C for 1, 3, or 6 h. After incubation, cells were fixed with 4% paraformaldehyde at 4 °C for 10 min, stained with membrane dye, washed with PBS, and imaged using an Olympus FV1000 confocal microscope. For surface-preserving peptide-binding and competition experiments, SKOV3 CLDN4⁺/⁺ and OVCAR3 cells were harvested as unfixed, non-permeabilized cells and resuspended in ice-cold PBS containing 1% BSA. Cells were incubated with FITC–Control or FITC–P15 (10 µM) for 30 min at 4 °C in the dark without prior fixation or detergent permeabilization. For anti-CLDN4 antibody pre-incubation experiments, cells were pre-incubated with anti-CLDN4 antibody (Proteintech, 16195-1-AP; rabbit polyclonal; 1:200) for 30 min at 4 °C before FITC–P15 staining. For ligand-competition experiments, cells were pre-incubated with a 10-fold molar excess of unlabeled P15 before FITC–P15 staining. After incubation, cells were washed with ice-cold PBS containing 1% BSA and analyzed by flow cytometry without detergent permeabilization. FITC fluorescence intensity and the FITC-positive fraction, where applicable, were quantified to assess surface-preserving P15 binding and competition under non-permeabilized conditions. For intracellular ROS imaging, SKOV3 CLDN4⁺/⁺ cells were seeded in confocal dishes and cultured overnight. Cells were then treated with PBS, PBS + US + NIR, NBs + US + NIR, D-NBs + US + NIR, IP + US + NIR, IP/NBs + US (− NIR), or IP/NBs + US + NIR at 10 µM (ICG-equivalent where applicable). Ultrasound exposure and NIR irradiation were applied using the standard treatment parameters described above. After treatment, cells were incubated with 10 µM DCFH-DA at 37 °C for 30 min in the dark, washed with PBS, replenished with fresh medium, and imaged immediately in the live state by confocal microscopy. DCF fluorescence intensity was quantified as indicated in the corresponding experiments. Cell Counting Kit-8 (CCK-8) proliferation assay SKOV3 WT and SKOV3 CLDN4⁺/⁺ cells were seeded into 96-well plates at a density of 2 × 10³ cells per well and allowed to adhere overnight. Cell proliferation was assessed using the Cell Counting Kit-8 assay at 12, 24, 48, and 72 h according to the manufacturer’s instructions. At each time point, CCK-8 reagent was added to each well and incubated for 1 h at 37 °C, and absorbance at 450 nm was measured using a microplate reader. Experiments were performed as three independent experiments with replicate wells in each experiment. Transwell matrigel invasion assay Matrigel-coated 24-well Transwell inserts with 8-µm pores were used for invasion assays. SKOV3 WT and SKOV3 CLDN4⁺/⁺ cells (2 × 10⁵ cells per insert) were seeded into the upper chamber in serum-free medium, and medium containing 20% FBS was added to the lower chamber as the chemoattractant. After 24 h, non-invaded cells on the upper membrane surface were gently removed with cotton swabs. Invaded cells on the lower membrane surface were fixed, stained with 0.1% crystal violet, and imaged under a light microscope. The number of invaded cells was quantified using ImageJ from randomly selected microscopic fields. Experiments were performed as three independent experiments. Patient-derived ovarian cancer organoid establishment and culture Fresh ovarian cancer specimens from three independent patients were obtained from Zhongda Hospital following ethical approval (ERC-SEU-ZDH-2023-0147) and written informed consent in accordance with the Declaration of Helsinki. Three independent patient-derived ovarian cancer organoid lines, designated PDO-1, PDO-2, and PDO-3, were established from these individual patient specimens. Fresh tumor tissues were washed with PBS and mechanically dissected into small fragments using sterile microscissors. Tissue fragments were digested in 2.5 mg/mL collagenase II at 37 °C for up to 30 min. Digestion was terminated with ice-cold PBS after sufficient tissue dissociation. The resulting cell suspension was filtered through 100 μm cell strainers and centrifuged at 1000 × g for 5 min at 4 °C. Following erythrocyte lysis where necessary, the cell pellet was resuspended in growth factor-reduced Matrigel (Corning, 356231) and seeded as 25 µL domes in pre-warmed 24-well plates. After polymerization at 37 °C for 30 min, each well was overlaid with 500 µL OncoPro Tumor Organoid Medium (Gibco, A4235601) supplemented with 100 U/mL penicillin and 100 µg/mL streptomycin. Organoids were maintained at 37 °C in a humidified atmosphere containing 5% CO₂, and the culture medium was changed every 72 h. Organoids were passaged by mechanical dissociation when they reached the appropriate size for continued expansion. For organoid penetration imaging, established organoids were incubated with FITC–P15 or FITC–Control at 10 µM, or with IP/NBs at 10 µM (ICG-equivalent where applicable), washed with PBS, counterstained with DAPI where indicated, and imaged by confocal microscopy to evaluate intratissue fluorescence distribution. For organoid treatment-response assays, organoids were treated with PBS + US + NIR, NBs + US + NIR, IP + US + NIR, or IP/NBs + US + NIR using the standard ultrasound and NIR treatment parameters described above. Organoid morphology was monitored at Day 0, Day 3, and Day 5. Experiments were performed using three replicate wells or Matrigel domes per group. Western Blot analysis of CLDN4 protein expression SKOV3 WT, SKOV3 CLDN4⁺/⁺, OVCAR3 cells, and three independent patient-derived ovarian cancer organoid lines (PDO-1, PDO-2, and PDO-3) were lysed in RIPA lysis buffer supplemented with protease inhibitors. Total protein concentrations were determined using a BCA protein assay. Equal amounts of protein (20–30 µg per lane) were separated by SDS-PAGE and transferred onto PVDF membranes. After blocking with 5% non-fat milk in TBST, membranes were incubated with rabbit anti-CLDN4 polyclonal antibody (Proteintech, 16195-1-AP; 1:6000) and mouse anti-GAPDH monoclonal antibody (Proteintech, 60004-1-Ig; 1:10000) overnight at 4 °C. After washing with TBST, membranes were incubated with HRP-conjugated goat anti-rabbit IgG (H + L) secondary antibody (Proteintech, SA00001-2) or HRP-conjugated goat anti-mouse IgG (H + L) secondary antibody (Proteintech, SA00001-1), as appropriate. Protein bands were visualized using enhanced chemiluminescence and imaged using a chemiluminescence imaging system. Densitometric analysis was performed using ImageJ. CLDN4 band intensity was normalized to GAPDH for each lane and expressed relative to SKOV3 WT. Preparation and characterization of nanobubbles Nanobubbles were prepared using a thin-film hydration method. Briefly, DSPC and DSPE-mPEG2000 were dissolved in chloroform at a molar ratio of 9:1, and the solvent was removed by rotary evaporation at 60 °C to form a lipid film. Residual solvent was further removed under vacuum overnight. The dried lipid film was hydrated with the corresponding IP-containing hydration solution to generate IP/NBs. For non-targeted dye-loaded nanobubbles (D-NBs), the lipid film was hydrated with ICG–Control-containing solution under the same conditions. For empty nanobubbles (NBs), hydration was performed using the corresponding aqueous vehicle without dye conjugate. After hydration at 60 °C for 20 min with vortex mixing, the suspension was transferred to glass vials containing an SF₆ headspace, sealed, and mechanically agitated immediately before use to generate gas-filled nanobubbles. The upper buoyant nanobubble fraction was isolated by low-speed centrifugation (210 × g, 10 min) and used for subsequent characterization and biological experiments. The definitions and compositions of PBS, NBs, D-NBs, IP, and IP/NBs used throughout the study are summarized in Supplementary Table S3. For in vivo dosing, dye-containing formulations were dose-normalized according to dye-conjugate content. IP and IP/NBs were normalized to the mass of IP, whereas ICG–Control and D-NBs were normalized to the mass of ICG–Control. Empty NBs contained no dye conjugate and were administered in the same injection volume using a nanobubble/lipid-matched suspension prepared in parallel with the corresponding IP/NBs formulation. Hydrodynamic size distribution and polydispersity index (PDI) were measured by dynamic light scattering (DLS), and zeta potential was determined by electrophoretic light scattering using the isolated buoyant nanobubble fraction. For short-term colloidal stability assessment, IP/NBs were incubated in PBS at 37 °C, and hydrodynamic diameter was measured by DLS at 0, 1, 5, 10, and 20 min. For extended stability assessment, hydrodynamic diameter was monitored at 0, 0.5, 1, and 7 days in serum-containing medium. Morphology was examined by transmission electron microscopy after negative staining with 1.5% phosphotungstic acid. Confocal microscopy was used to evaluate IP-associated fluorescence distribution within the nanobubble formulation. For determination of encapsulation efficiency (EE%) and drug loading (DL%), the isolated IP/NBs fraction was lysed with 2% Triton X-100, filtered, and subjected to HPLC quantification of the characteristic IP peak. EE% was calculated as (amount of IP in the isolated IP/NBs fraction/total amount of input IP) × 100%. DL% was calculated as (amount of IP in the isolated IP/NBs fraction/total amount of lipid used for preparation) × 100%, based on the input lipid amount. Loading content was additionally expressed as micrograms of IP in the isolated IP/NBs fraction per milligram of lipid. In vitro release kinetics of IP from IP/NBs To evaluate the release behavior of physically incorporated IP under aqueous conditions, in vitro release kinetics were assessed in PBS in the absence or presence of ultrasound treatment. Freshly prepared IP/NBs were suspended in PBS and divided into IP/NBs (− US) and IP/NBs (+ US) groups, with PBS alone serving as the blank control. Ultrasound exposure was applied to the IP/NBs (+ US) group using the standard ultrasound treatment parameters described above. Samples were then incubated at 37 °C with gentle shaking. At 0, 0.5, 1, 2, 4, 6, 12, and 24 h, samples were centrifuged at 210 × g for 10 min to separate the buoyant nanobubble-associated fraction from the aqueous release medium. The aqueous release medium was carefully collected without disturbing the buoyant fraction, and an equal volume of fresh pre-warmed PBS was added back to maintain a constant total volume. IP-related fluorescence in the collected release medium was measured using a microplate reader (Ex 780 nm/Em 820 nm). PBS collected and processed in parallel was used for background subtraction. Total releasable IP was determined after complete disruption of IP/NBs using 2% Triton X-100, and apparent cumulative release was expressed as a percentage of total releasable IP after background subtraction against the PBS blank. Data are shown as mean ± SD (n = 3). Photothermal heating measurements For photothermal characterization, PBS or IP/NBs solutions at the indicated ICG-equivalent concentrations were placed in 1.5 mL microcentrifuge tubes (0.4 mL per sample) and irradiated with an 808-nm laser. Temperature changes were recorded in real time using an infrared thermal camera. The initial temperature (T₀) was defined as the temperature immediately before irradiation, and heating curves were plotted as temperature versus time. Power-dependent heating was evaluated using IP/NBs at 10 µM (ICG-equivalent) under 808-nm irradiation at 0.3, 0.5, and 0.7 W/cm² for 5 min. Concentration-dependent heating was assessed using IP/NBs at 5, 10, and 20 µM (ICG-equivalent) under 808-nm irradiation at 0.5 W/cm² for 5 min. Photothermal repeatability was examined using IP/NBs at 10 µM over five laser on/off cycles under 808-nm irradiation at 0.5 W/cm². For estimation of photothermal conversion efficiency, IP/NBs (10 µM, ICG-equivalent; 0.4 mL per sample) were irradiated with an 808-nm laser at 0.5 W/cm² using a 1.0 cm² beam spot, followed by natural cooling after laser shutdown. The heating–cooling curve was analyzed using the standard heating–cooling method. The cooling-time constant was obtained from linear fitting of the plot of − ln(θ) versus cooling time, where θ = (T − T_surr)/(T_max − T_surr). PBS was measured in parallel under identical irradiation conditions for background heat correction. In vitro photothermal cytotoxicity and bliss interaction analysis For in vitro photothermal cytotoxicity assays, SKOV3 CLDN4⁺/⁺ cells were seeded in 96-well plates and allowed to adhere overnight. Cells were then treated with PBS, NBs, IP, or IP/NBs at the indicated concentrations. For the dose-dependent US + NIR cytotoxicity assay, ultrasound exposure and NIR irradiation were applied using the standard treatment parameters described above. After treatment, cells were returned to the incubator, and cell viability was measured using the CCK-8 assay according to the manufacturer’s instructions. Absorbance at 450 nm was measured using a microplate reader, and viability was normalized to the corresponding PBS control. For the photothermal control assay, cells were treated with PBS, NBs, IP, or IP/NBs at the indicated concentration in the absence or presence of NIR irradiation. NIR irradiation, where indicated, was delivered using the standard NIR parameters described above without ultrasound exposure. A Bliss independence analysis was performed as an exploratory single-condition interaction analysis to assess whether the combined IP/NBs + NIR condition produced greater fractional inhibition than expected from the individual IP-associated photothermal and nanobubble-associated components under the tested fixed NIR irradiation condition. The two components were defined as follows: component A, IP + NIR, representing IP-associated photothermal treatment; and component B, NBs + NIR, representing the nanobubble formulation under the same NIR irradiation condition. The combined condition was IP/NBs + NIR. The analysis was performed at the working concentration used for subsequent mechanistic studies (10 µM, ICG-equivalent where applicable) under fixed NIR irradiation at 808 nm and 0.5 W/cm² for 5 min. Cell viability was converted to fractional inhibition using the formula E = 1 − V, where V represents the normalized viable fraction relative to the PBS control. The expected Bliss inhibition was calculated as EBliss = EA + EB − EAEB, where EA and EB represent the fractional inhibition produced by component A and component B, respectively. Bliss excess was calculated as ΔBliss = Eobserved − EBliss, where Eobserved represents the fractional inhibition observed in the IP/NBs + NIR group. A positive ΔBliss was interpreted as suggestive of a positive interaction between IP-associated photothermal treatment and the nanobubble formulation under this fixed-dose NIR irradiation condition. Confidence intervals for ΔBliss were estimated by non-parametric bootstrap resampling of replicate-level viability values using 10,000 bootstrap iterations. This analysis was exploratory and was not intended to represent a full two-dimensional dose–response synergy model. In vitro phantom ultrasound imaging of IP/NBs Ultrasound characterization of NBs and IP/NBs was performed using a VisualSonics Vevo 2100 microimaging platform equipped with an MS-250 transducer to evaluate the intrinsic contrast-generating capacity of the nanobubble formulations. In vitro imaging was carried out using tissue-mimicking phantoms composed of 3% (w/v) agar, 11% (v/v) glycerol, and 86% degassed water, with 2 mL sample wells. After baseline imaging of wells containing degassed water, NBs or IP/NBs suspensions (2.12 × 10⁹ particles/mL) were added to the wells and imaged using three independent samples per group. Images were acquired using the following settings: center frequency 18 MHz, acoustic power 10%, contrast gain 35 dB, imaging depth 20.00 mm, and field width 23.04 mm. Contrast-enhanced ultrasound images were collected at the indicated time points after sample loading, and contrast intensity was quantified for subsequent analysis. In vivo contrast-enhanced ultrasound imaging of IP/NBs In vivo contrast-enhanced ultrasound imaging was performed in SKOV3 CLDN4⁺/⁺ tumor-bearing nude mice to evaluate the acoustic visibility of NBs and IP/NBs after intravenous administration. Mice were anesthetized with isoflurane and placed in the supine position on a temperature-controlled imaging platform. The skin over the tumor-bearing flank was cleaned, and residual hair, if present, was removed before imaging. Ultrasound gel was applied to ensure acoustic coupling. Imaging was performed using a VisualSonics Vevo 2100 microimaging platform equipped with an MS-250 transducer. Baseline tumor images were acquired before injection. Mice were then intravenously injected via the tail vein with NBs or IP/NBs in 100 µL sterile PBS, using nanobubble/lipid-matched doses between groups. In vivo ultrasound imaging was performed with n = 3 mice per group. Contrast-enhanced ultrasound images were acquired dynamically after injection. The time labels indicate elapsed time after completion of intravenous injection. This contrast-enhanced ultrasound imaging procedure was used for signal acquisition and was distinct from the therapeutic ultrasound trigger described above. Images were collected using the following imaging settings: center frequency 18 MHz, acoustic power 10%, contrast gain 35 dB, imaging depth 20.00 mm, and field width 23.04 mm. For quantitative analysis, a region of interest was manually drawn over the tumor area and kept consistent across time points within each animal. Background signal was measured from an adjacent non-tumor region and subtracted from the tumor ROI signal. Ultrasound contrast intensity was calculated as background-subtracted tumor-region contrast mean power and expressed in arbitrary units over time. Ultrasound-assisted endothelial transwell transport assay To evaluate ultrasound-assisted endothelial barrier transport, bEnd.3 endothelial cells were cultured on 0.4-µm pore Transwell inserts until a stable barrier was formed, as monitored by TEER. Free IP, IP/NBs, or free IP + NBs was added to the apical chamber, with IP-containing groups normalized to the same IP-equivalent concentration. Ultrasound exposure was applied for 1 min using the standard ultrasound treatment parameters described above. After ultrasound exposure and incubation for 2 h, medium from the basolateral chamber was collected. IP-related fluorescence in the basolateral samples was measured using a microplate reader (Ex 780 nm/Em 820 nm). Background-subtracted fluorescence intensity was used to assess transendothelial IP-related transport. In vivo fluorescence imaging Tumor-associated fluorescence and ex vivo fluorescence biodistribution of the indicated dye-containing formulations were assessed by near-infrared fluorescence imaging in tumor-bearing nude mice. Mice were intravenously injected via the tail vein with the corresponding dye-containing formulation at a 100 µg dye-conjugate-equivalent dose in 100 µL sterile PBS, according to the formulation-dose definitions described above. For the 3-h in vivo fluorescence comparison, free IP, D-NBs, and IP/NBs were administered as dye-containing groups, with D-NBs serving as the non-targeted dye-loaded nanobubble comparator. Empty NBs were administered at a nanobubble/lipid-matched dose relative to the corresponding IP/NBs formulation in the same injection volume, and PBS controls received 100 µL sterile PBS. Whole-body fluorescence imaging was performed at 3 h post-injection using an IVIS® Spectrum imaging system under near-infrared settings appropriate for ICG detection. For 24-h ex vivo fluorescence biodistribution analysis, mice were euthanized at 24 h post-injection, and tumors and major organs, including the heart, liver, spleen, lungs, and kidneys, were collected for ex vivo fluorescence imaging and quantitative analysis. Images were processed using Living Image® software (PerkinElmer). To obtain a simplified plasma fluorescence-retention profile, tumor-bearing nude mice were intravenously injected via the tail vein with either free IP or IP/NBs at an equivalent dose of 100 µg in 100 µL sterile PBS. At the indicated time points after injection (5, 15, 30, 60, 180, and 360 min), small-volume blood samples were collected into anticoagulant-treated tubes and immediately processed for plasma separation by centrifugation. The resulting plasma fractions were transferred to black 96-well plates, and IP-related fluorescence was measured using a microplate reader under near-infrared settings appropriate for ICG detection (Ex 780 nm/Em 820 nm). Plasma-associated fluorescence values were plotted versus time to generate a simplified plasma fluorescence-retention profile for comparison between free IP and IP/NBs. Data are presented as mean ± SD (n = 3 mice per group). For matched post-ultrasound fluorescence imaging under the therapeutic workflow, tumor-bearing mice were intravenously injected with IP/NBs as described above and assigned to the IP/NBs (no US) or IP/NBs + US group (n = 3 mice per group). For the IP/NBs + US group, a single ultrasound trigger was applied to the tumor region for 1 min at 4 min post-injection using the same acoustic conditions as those used in the therapeutic workflow. Mice in the IP/NBs (no US) group were handled in parallel without ultrasound exposure. Whole-body near-infrared fluorescence images were acquired at 30, 60, 180, and 360 min after intravenous injection using identical IVIS acquisition settings. Tumor regions of interest were defined consistently across time points, and tumor-associated fluorescence intensity was quantified after background subtraction using Living Image® software. For ex vivo validation of the matched post-ultrasound imaging experiment, separate cohorts of mice were euthanized at 3–6 h post-injection. Tumors were excised, processed using a fluorescence-compatible tissue-sectioning protocol, and imaged by fluorescence microscopy to evaluate intratumoral IP-related fluorescence distribution. Images were acquired under identical acquisition settings within each experiment, and ex vivo tumor-section fluorescence intensity was quantified using ImageJ. In vivo anticancer study All animal experiments were approved by the Animal Care Committee of Southeast University (approval no. 20240402301). Female BALB/c nude mice (6 weeks old, 15 ± 2 g) were purchased from Jiangsu GemPharmatech and maintained under specific pathogen-free conditions. Subcutaneous tumor-bearing models were established by injecting 1 × 10⁶ SKOV3 CLDN4⁺/⁺ cells suspended in 100 µL of a 1:1 Matrigel/PBS mixture into the right flank. When tumors reached approximately 100 mm³, mice were randomized into the indicated experimental groups. For the main therapeutic study, tumor-bearing mice were assigned to PBS + US + NIR, NBs + US + NIR, D-NBs + US + NIR, IP + US + NIR, IP/NBs + US (− NIR), or IP/NBs + US + NIR groups (n = 5 mice per group). Dye-containing formulations were administered by tail-vein injection at a 100 µg dye-conjugate-equivalent dose in 100 µL sterile PBS, according to the formulation-dose definitions described above. Empty NBs were administered at a nanobubble/lipid-matched dose relative to the IP/NBs group in the same injection volume, and PBS controls received 100 µL sterile PBS. Where indicated, a single ultrasound trigger was initiated at 4 min post-injection and applied for 1 min using the standard ultrasound treatment parameters described above, and NIR irradiation was applied at 6 h post-injection. Tumor volume and body weight were recorded every 3 days. Tumor volume was calculated as V = length × width²/2. To assess the contribution of ultrasound exposure to the IP/NBs therapeutic regimen, a separate comparison was performed in tumor-bearing mice using three IP-based regimens: IP + US + NIR, IP/NBs + NIR (no US), and IP/NBs + US + NIR (n = 5 mice per group). Formulations were administered by tail-vein injection at the same dose used in the main therapeutic study. In the IP + US + NIR and IP/NBs + US + NIR groups, a single ultrasound trigger was initiated at 4 min post-injection and applied for 1 min using the standard ultrasound treatment parameters described above. In the IP/NBs + NIR (no US) group, ultrasound exposure was omitted, while NIR irradiation was still delivered at 6 h post-injection using the same 808-nm laser parameters. Tumor growth was monitored during treatment, and excised tumors and endpoint tumor weights were collected to compare therapeutic efficacy among these three regimens. For endpoint histological analyses in the main therapeutic study, mice were euthanized on day 16, and tumor tissues were collected for H&E, Ki67, and TUNEL staining. Major organs (heart, liver, spleen, lung, and kidney) were also harvested, fixed in 4% paraformaldehyde, paraffin-embedded, sectioned, stained with H&E, and examined by light microscopy. In a separate survival cohort, mice were monitored until day 48 or until humane endpoint criteria were reached. Human PBMC adoptive-transfer tumor-bearing mouse model For immune-infiltration and splenic T-cell activation analyses, an acute human PBMC adoptive-transfer tumor-bearing mouse model was established using freshly isolated human PBMCs from the same healthy donor cohort used for the in vitro immune activation assay. PBMCs were isolated by Ficoll density-gradient centrifugation after written informed consent under institutional ethics approval ERC-SEU-ZDH-2023-0147, in accordance with the Declaration of Helsinki. For the immune-analysis cohort, mice underwent the same therapeutic workflow described above. Briefly, formulations were administered by tail-vein injection, ultrasound exposure was applied for 1 min at 4 min post-injection where indicated, and NIR irradiation was delivered at 6 h post-injection. At 4 h after NIR irradiation, each mouse received a single tail-vein injection of 1 × 10⁷ human PBMCs suspended in 100–200 µL sterile PBS. Tumor and spleen samples were collected at the treatment endpoint for analysis of intratumoral human T-cell infiltration and splenic human T-cell activation. Assessment of intratumoral IP-related distribution and T-cell infiltration For assessment of intratumoral IP-related fluorescence distribution, tumor tissues were collected at the indicated post-injection time points after intravenous administration of the corresponding IP-containing formulations. Tumor samples were processed using a fluorescence-compatible tissue-sectioning protocol and imaged to evaluate intratumoral IP-related fluorescence distribution. Images were acquired under identical acquisition settings within each experiment, and fluorescence intensity was quantified using ImageJ where indicated. For immune-infiltration analysis, separate tumor samples were collected from human PBMC adoptive-transfer tumor-bearing mice at the treatment endpoint. Tissues were fixed in 4% paraformaldehyde, embedded in paraffin, and sectioned at 5 μm thickness. For immunohistochemical detection of intratumoral CD3⁺ T cells, sections were deparaffinized, rehydrated, and subjected to antigen retrieval in citrate buffer (pH 6.0, 95 °C, 10 min). Endogenous peroxidase activity was quenched with 3% H₂O₂ for 10 min. After blocking with 1% BSA, sections were incubated overnight at 4 °C with anti-human CD3 primary antibody (Proteintech, 84505-4-PBS; clone 242012G2; rabbit recombinant; 1:6400), followed by incubation with the corresponding HRP-conjugated secondary antibody for 2 h at room temperature. Immunoreactivity was visualized using DAB, and sections were counterstained with hematoxylin. For multiplex immunofluorescence co-staining analysis of intratumoral CD4⁺ or CD8⁺ T cells and CD69, deparaffinized tumor sections underwent heat-mediated antigen retrieval in Tris-EDTA buffer (pH 9.0), followed by blocking with 1% BSA. CD4/CD69 and CD8/CD69 co-staining were performed as separate staining panels using a sequential multiplex immunofluorescence protocol. Sections were stained with anti-human CD4 primary antibody (Proteintech, 19068-1-AP; 1:400) or anti-human CD8A primary antibody (Proteintech, 66868-1-Ig; clone 1G2B10; 1:800), followed by anti-human CD69 primary antibody (R&D Systems/Bio-Techne, MAB11691; clone 3190 A). Nuclei were counterstained with DAPI. Images were acquired using a Nikon fluorescence imaging system under identical channel-specific acquisition settings across treatment groups. For representative figure preparation, the DAPI channel was displayed using the same brightness/contrast range across all treatment groups. The integrated fluorescence intensity of the CD4 and CD8 marker channels was quantified using ImageJ by two independent blinded observers. DAPI was used as a nuclear counterstain, and CD69 co-staining was evaluated qualitatively to illustrate the spatial association of intratumoral CD4⁺ or CD8⁺ T cells with an activation-related marker. ICD-associated danger signal and cytokine assays For ICD-associated danger signal assays, SKOV3 CLDN4⁺/⁺ cells were treated at 10 µM (ICG-equivalent where applicable) under the following conditions: PBS + US + NIR, NBs + US + NIR, D-NBs + US + NIR, IP + US + NIR, IP/NBs + US (− NIR), and IP/NBs + US + NIR. Ultrasound exposure and NIR irradiation were applied using the standard treatment parameters described above. Surface calreticulin (CRT) exposure was assessed at 4 h post-treatment by non-permeabilized immunofluorescence staining. Briefly, cells were fixed with 4% paraformaldehyde without detergent permeabilization and incubated with anti-calreticulin primary antibody (Proteintech, 27298-1-AP; rabbit polyclonal) under non-permeabilized conditions, followed by the corresponding fluorescent secondary antibody. No detergent or membrane-permeabilization reagent was used before or during antibody incubation. Surface CRT signal was quantified as integrated fluorescence intensity using ImageJ. Extracellular ATP release was measured in culture supernatants collected at 2 h post-treatment using a luciferase-based ATP assay and expressed as fold change relative to the PBS + US + NIR group. HMGB1 release was quantified in supernatants collected at 24 h post-treatment by ELISA and reported as ng/mL. For cytokine analysis, tumor cells receiving the indicated treatment conditions were co-cultured with human PBMCs using the Transwell co-culture configuration described below. Supernatants were collected after co-culture, and IFN-γ and TNF-α concentrations were quantified by ELISA and reported as pg/mL. All assays were performed as three independent experiments. In vitro immune cell activation assay Freshly isolated human PBMCs from healthy donors were used for the in vitro co-culture assay. SKOV3 CLDN4⁺/⁺ cells were seeded in 6-well plates and cultured overnight for adherence. To assess treatment-associated immune activation in the immune-activation comparator assay, tumor cells were treated with the following matched groups: PBS + US + NIR, D-NBs + US + NIR, IP + US + NIR, and IP/NBs + US + NIR. D-NBs served as the non-targeted dye-loaded nanobubble comparator, IP served as the free CLDN4-directed photothermal conjugate comparator, and IP/NBs served as the CLDN4-directed IP-containing nanobubble formulation. Formulations were applied at 10 µM ICG-equivalent concentration where applicable. Ultrasound exposure and NIR irradiation were applied using the standard treatment parameters described above. After treatment, tumor cells were collected, washed, and resuspended in serum-free RPMI-1640 medium. For co-culture, treated SKOV3 CLDN4⁺/⁺ cells were transferred to Corning Transwell inserts (6.5 mm diameter, 0.4 μm polyester membrane), and freshly isolated human PBMCs were added to the lower chamber at a PBMC: tumor cell ratio of 10:1 in complete RPMI-1640 medium containing 10% FBS. Co-cultures were maintained for 24 h at 37 °C in a humidified atmosphere containing 5% CO₂. After co-culture, cells from the lower chamber were harvested for flow-cytometric analysis. Sequential gating was performed using FSC/SSC, singlet selection, live-cell gating based on the viability dye, and hCD45⁺ human leukocyte gating. Dendritic-cell maturation was assessed within the hCD45⁺CD11c⁺ population by analyzing CD80 and CD86 expression. T cells were identified as hCD45⁺CD3⁺ cells and further subdivided into CD3⁺CD4⁺ and CD3⁺CD8⁺ subsets for analysis of CD69 expression. Dendritic-cell maturation was quantified as the percentage of CD80⁺CD86⁺ cells within the hCD45⁺CD11c⁺ population, and T-cell activation was quantified as the percentage of CD69⁺ cells within the CD3⁺CD4⁺ and CD3⁺CD8⁺ T-cell subsets. Data were analyzed using FlowJo v10.8.1. The corresponding gating strategy is provided in the Supporting Information. Experiments were performed as three independent biological replicates. Statistical analysis Statistical analyses were performed using GraphPad Prism and R. Data are presented as mean ± SD unless otherwise indicated. The number of biological replicates or animals (n) is provided in the corresponding figure legends or text. For comparisons between two groups, a two-tailed unpaired Student’s t-test was used. For comparisons among multiple groups, one-way ANOVA followed by post hoc multiple-comparison testing was used. For tumor growth curves and other time-course comparisons, two-way ANOVA followed by post hoc multiple-comparison testing was applied. Confidence intervals for Bliss excess were estimated by non-parametric bootstrap resampling of replicate-level viability values using 10,000 bootstrap iterations. Survival was visualized using the Kaplan–Meier method. For nonparametric sample-level comparisons of CLDN4 expression across primary, metastatic, and relapsed lesions, the Kruskal–Wallis test was used. Panel-specific statistical methods are indicated in the corresponding figure legends where applicable. A P value < 0.05 was considered statistically significant. Significance was denoted as follows: ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.001. Results CLDN4 Is enriched in ovarian carcinoma cells and associates with invasion-related phenotypes To identify tumor-enriched candidate targets in ovarian cancer, we analyzed a single-cell RNA sequencing dataset (GEO: GSE130000) comprising 13,369 cells from eight ovarian cancer samples, including four primary tumors, two peritoneal metastases, and two relapsed lesions (Fig. 2a). Using a standard Seurat workflow, we resolved five transcriptionally distinct cell populations (Fig. 2b, c). Additional Seurat clustering and marker-based tumor/epithelial candidate definition are shown in Supplementary Fig. S1a, b. Among tight-junction genes, CLDN4 showed preferential expression in the tumor/epithelial candidate compartment (Fig. 2d), supporting its candidacy as a tumor-enriched therapeutic target. Because GSE130000 includes primary, metastatic, and relapsed lesions, we next performed an exploratory sample-level comparison of CLDN4 expression across disease states. To minimize dependence on clustering assignments, CLDN4 expression was quantified within a marker-defined tumor/epithelial candidate compartment (n = 8,232 cells) and summarized as a pseudo-bulk mean for each sample (one value per sample; Primary n = 4, Metastasis n = 2, Relapse n = 2). This exploratory analysis yielded a nominal difference across disease states in this small cohort (Kruskal–Wallis p = 0.048; Supplementary Fig. S1c). Because only eight samples were available and the metastasis and relapse groups each contained two samples, this comparison was interpreted descriptively as supportive evidence for CLDN4 prioritization rather than as definitive evidence of disease-state-specific differences. We also examined the relationship between sample-level CLDN4 expression and an EMT/migration-related signature score within the same tumor/epithelial candidate compartment (Supplementary Fig. S1d). This analysis was likewise considered exploratory and was used to evaluate whether CLDN4 expression showed descriptive concordance with invasion-related transcriptional features. Fig. 2Single-cell analysis, expression validation, and functional characterization of CLDN4.(a) Workflow for GSE130000 ovarian cancer scRNA-seq data integration and analysis. (b) UMAP embedding of GSE130000 cells annotated into major cell types. (c) Marker-selection plots showing representative genes used for cell-type annotation. (d) DotPlot of canonical marker genes across annotated cell types. Color indicates mean expression and dot size indicates the fraction of cells expressing each gene. (e) Proliferation of SKOV3 WT and CLDN4-overexpressing cells (SKOV3 CLDN4⁺/⁺) at 12, 24, 48, and 72 h. (f) Immunohistochemical staining of CLDN4 in paired normal ovarian epithelium and ovarian carcinoma tissues. (g) Quantification of CLDN4 staining by H-score. (h) 3D tumorsphere growth of control and CLDN4-overexpressing SKOV3 cells at Days 0, 3, 6, and 9. (i) Quantification of spheroid growth as relative spheroid volume over Days 0, 3, 6, and 9. (j) Transwell invasion assay at 24 h. (k) Quantification of invaded cells. Data in (e), (g), (i), and (k) are presented as mean ± SD (n = 3 independent experiments for in vitro assays; n = 25 paired tissue samples for IHC H-score analysis in g). Statistical significance in (e), (i), and (k) was determined by two-tailed unpaired Student’s t-test at the indicated time points or endpoint comparisons. Statistical significance in (g) was determined by paired two-tailed Student’s t-test. *P < 0.05; ***P < 0.001. Scale bars are shown in the corresponding panels Single-cell analysis, expression validation, and functional characterization of CLDN4.(a) Workflow for GSE130000 ovarian cancer scRNA-seq data integration and analysis. (b) UMAP embedding of GSE130000 cells annotated into major cell types. (c) Marker-selection plots showing representative genes used for cell-type annotation. (d) DotPlot of canonical marker genes across annotated cell types. Color indicates mean expression and dot size indicates the fraction of cells expressing each gene. (e) Proliferation of SKOV3 WT and CLDN4-overexpressing cells (SKOV3 CLDN4⁺/⁺) at 12, 24, 48, and 72 h. (f) Immunohistochemical staining of CLDN4 in paired normal ovarian epithelium and ovarian carcinoma tissues. (g) Quantification of CLDN4 staining by H-score. (h) 3D tumorsphere growth of control and CLDN4-overexpressing SKOV3 cells at Days 0, 3, 6, and 9. (i) Quantification of spheroid growth as relative spheroid volume over Days 0, 3, 6, and 9. (j) Transwell invasion assay at 24 h. (k) Quantification of invaded cells. Data in (e), (g), (i), and (k) are presented as mean ± SD (n = 3 independent experiments for in vitro assays; n = 25 paired tissue samples for IHC H-score analysis in g). Statistical significance in (e), (i), and (k) was determined by two-tailed unpaired Student’s t-test at the indicated time points or endpoint comparisons. Statistical significance in (g) was determined by paired two-tailed Student’s t-test. *P < 0.05; ***P < 0.001. Scale bars are shown in the corresponding panels To assess whether CLDN4 overexpression is associated with invasion-related phenotypes in vitro, we established a stable SKOV3 CLDN4⁺/⁺ cell line by lentiviral transduction. Flow-cytometric analysis verified increased CLDN4 surface staining in SKOV3 CLDN4⁺/⁺ cells compared with SKOV3 WT cells, with CLDN4-positive fractions of 97.9% and 19.1%, respectively (Fig. S2a). Endogenous CLDN4 staining was also detected in OVCAR3 cells, supporting their use as an endogenous CLDN4-positive ovarian cancer model for subsequent validation experiments (Fig. S2b). CCK-8 assays showed only a modest and non-significant increase in proliferation in CLDN4⁺/⁺ cells (p = 0.07; Fig. 2e). In parallel, immunohistochemical analysis of 25 matched normal and tumor tissues demonstrated marked upregulation of CLDN4 in malignant specimens (H-score: 75.32 vs. 16.7; P < 0.001; Fig. 2f, g). Functionally, enforced CLDN4 expression increased three-dimensional tumorsphere formation by the final observation time point (1.21-fold, + 21.4% at Day 9; p < 0.05; Fig. 2h, i) and enhanced Transwell invasion (1.16-fold, + 16.0% invaded cells; p < 0.05; Fig. 2j, k). Collectively, these data support CLDN4 as a tumor-enriched marker in ovarian cancer and suggest that elevated CLDN4 expression is associated with invasion-related phenotypes in vitro. Iterative in silico screening identifies P15 as a Lead CLDN4-binding peptide with low-nanomolar affinity The overall in silico peptide optimization and validation workflow is summarized in Fig. 3a. To derive a minimal CLDN4-binding motif from the C-CPE–CLDN4 interface, we first analyzed the crystal structure of the CLDN4–C-CPE complex (PDB: 5B2G; Fig. 3b; Supplementary Video 1). Interfacial energy decomposition identified several C-CPE residues with major contributions to receptor recognition, including Phe219, Leu223, Arg227, Leu254, Tyr310, Leu316, and Gln317 (Fig. 3c). Complementary per-residue energy scanning further highlighted Asn218, Asp225, Arg227, Tyr306, Tyr310, and Leu315 as energetically important positions for ligand engagement (Fig. 3d). Fig. 3Structure-guided design and binding validation of CLDN4-binding peptides.(a) Workflow for in silico peptide optimization, synthesis, and experimental validation. (b) Crystal structure of the CLDN4–C-CPE complex (PDB: 5B2G). (c) Per-residue binding energy contributions identifying key C-CPE residues involved in CLDN4 binding. (d) Structural mapping of the CLDN4–C-CPE binding interface and energetically important residues. (e, f) Structural views of representative residues within Regions I and II selected for peptide optimization. (g) MM/GBSA-estimated binding free energies of the designed peptide library, with P15 highlighted as the lead candidate. (h) Predicted binding pose of P15 on CLDN4. (i–l) SPR sensorgrams of the four experimentally validated peptide candidates (P02, P14, P15, and P21), with steady-state K_D values shown in the panels. Steady-state K_D values in (i–l) were obtained from SPR fitting, and comparative kinetic parameters derived from 1:1 SPR kinetic fitting are summarized in Supplementary Table S2. Structure-guided design and binding validation of CLDN4-binding peptides.(a) Workflow for in silico peptide optimization, synthesis, and experimental validation. (b) Crystal structure of the CLDN4–C-CPE complex (PDB: 5B2G). (c) Per-residue binding energy contributions identifying key C-CPE residues involved in CLDN4 binding. (d) Structural mapping of the CLDN4–C-CPE binding interface and energetically important residues. (e, f) Structural views of representative residues within Regions I and II selected for peptide optimization. (g) MM/GBSA-estimated binding free energies of the designed peptide library, with P15 highlighted as the lead candidate. (h) Predicted binding pose of P15 on CLDN4. (i–l) SPR sensorgrams of the four experimentally validated peptide candidates (P02, P14, P15, and P21), with steady-state K_D values shown in the panels. Steady-state K_D values in (i–l) were obtained from SPR fitting, and comparative kinetic parameters derived from 1:1 SPR kinetic fitting are summarized in Supplementary Table S2. Based on these hotspot residues, we performed saturation mutagenesis across two functional regions, Region I (Asn218/Asp225/Arg227) and Region II (Tyr306/Tyr310/Leu315), and retained affinity-favored variants to construct a focused peptide library (Lead-PCL; 20 predicted sequences plus 2 parental sequences; Table S1). Candidate peptides were then ranked by structure-guided docking and MM/GBSA evaluation of 50-ns trajectories, which identified P15 as the top-performing sequence (ΔG = − 48.55 kcal/mol; Figs. 3e–g and S3–S6). Molecular dynamics simulations performed in GROMACS for 50 ns supported the relative conformational stability of the P15–CLDN4 complex during the trajectory (Fig. S5f; Supplementary Video 2). Structural modeling suggested that P15 adopted a favorable binding pose within the extracellular pocket of CLDN4 (Fig. 3h). The top-ranked candidates were synthesized by solid-phase peptide synthesis and confirmed by ESI-MS (Scheme S1; Figs. S7–S28). Surface plasmon resonance analysis showed that four candidates (P02, P14, P15, and P21) bound recombinant CLDN4 with nanomolar-range affinities (Fig. 3i–l). Among them, P15 showed the lowest steady-state K_D value (11.28 ± 0.35 nM), consistent with the in silico ranking and supporting its selection for subsequent binding and functional studies. Kinetic fitting of SPR sensorgrams for the four experimentally validated candidates was performed using a 1:1 binding model. The kinetic-derived K_D values were 90.02 nM for P02, 56.86 nM for P14, 11.03 nM for P15, and 97.13 nM for P21 (Table S2). P15 retained the lowest derived K_D among the four candidates, whereas P14 showed a slower dissociation rate but a lower association rate than P15. These comparative kinetic data further supported the selection of P15 as the lead CLDN4-binding peptide. CLDN4-dependent binding, tumor enrichment, internalization, and penetration of the lead peptide P15 FITC–P15, prepared via a 6-aminohexanoic acid linker, and the scrambled control peptide FITC–Control were confirmed by ESI-MS (Scheme S2; Figs. S29 and S30). Because total CLDN4 expression does not by itself establish extracellular accessibility to exogenous ligands, we first evaluated FITC–P15 binding under surface-preserving, non-permeabilized conditions. In SKOV3 CLDN4⁺/⁺ cells, FITC–P15 produced a clear FITC-positive population compared with FITC–Control, and this signal was reduced after pre-incubation with either anti-CLDN4 antibody or excess unlabeled P15 (Fig. 4a). Similar surface-preserving analysis in OVCAR3 cells, an endogenous CLDN4-positive ovarian cancer model, also showed detectable FITC–P15 binding that was reduced by CLDN4 or ligand competition (Fig. 4a). These results support surface-accessible CLDN4-associated P15 binding under the tested non-permeabilized conditions. To determine whether this binding translated into a practical in vivo tumor-enrichment window for subsequent therapeutic studies, we next prepared indocyanine green conjugates ICG–P15 (termed IP) and ICG–Control (Scheme S3), both of which were confirmed by ESI-MS (Figs. S31 and S32). In mice bearing subcutaneous SKOV3 CLDN4⁺/⁺ xenografts, intravenously administered IP (100 µg per mouse) showed preferential tumor enrichment at 3 h, whereas ICG–Control showed no detectable tumor-associated enrichment (Fig. 4b), supporting a usable time window for downstream in vivo intervention. Fig. 4CLDN4-dependent P15 binding, tumor enrichment, surface accessibility, internalization, and organoid penetration.(a) Representative surface-preserving flow-cytometry histograms showing FITC–P15 binding and competition in SKOV3 CLDN4⁺/⁺ cells and OVCAR3 cells under non-permeabilized conditions for the indicated treatment groups: FITC–Control, FITC–P15, FITC–P15 + anti-CLDN4 antibody pre-incubation, and FITC–P15 + excess unlabeled P15. (b) In vivo fluorescence imaging comparing IP (ICG–P15) and ICG–Control at 3 h after intravenous injection in subcutaneous SKOV3 CLDN4⁺/⁺ xenografts. (c, d) Live-cell, non-permeabilized CLDN4 surface immunofluorescence staining in SKOV3 CLDN4⁺/⁺ and OVCAR3 cells before fixation, with representative images in (c) and quantification of membrane-associated CLDN4 fluorescence intensity in (d). (e) Time-dependent internalization of FITC–P15 in SKOV3 CLDN4⁺/⁺ cells at 1, 3, and 6 h, shown together with DAPI and membrane dye staining, with enlarged views of the indicated regions. (f) Quantification of FITC fluorescence intensity corresponding to (e). (g) Penetration of FITC–P15 and FITC–Control in patient-derived ovarian cancer organoids. The patient-derived ovarian cancer organoid lines used in this study were established from three independent patients, and CLDN4 protein expression in PDO-1, PDO-2, and PDO-3 was verified by western blotting (Fig. S33e, f). Representative flow-cytometry histograms in (a) and representative confocal images in (c), (e), and (g) are shown from three independent experiments. Representative in vivo fluorescence images in (b) are shown from n = 3 mice per group. Data in (d) and (f) are presented as mean ± SD (n = 3 independent experiments). Statistical significance in (d) was determined by unpaired two-tailed Student’s t-test. Statistical significance in (f) was determined by one-way ANOVA with post hoc multiple-comparison testing. Statistical significance is indicated in the panels. Scale bars are shown in the corresponding panels. CLDN4-dependent P15 binding, tumor enrichment, surface accessibility, internalization, and organoid penetration.(a) Representative surface-preserving flow-cytometry histograms showing FITC–P15 binding and competition in SKOV3 CLDN4⁺/⁺ cells and OVCAR3 cells under non-permeabilized conditions for the indicated treatment groups: FITC–Control, FITC–P15, FITC–P15 + anti-CLDN4 antibody pre-incubation, and FITC–P15 + excess unlabeled P15. (b) In vivo fluorescence imaging comparing IP (ICG–P15) and ICG–Control at 3 h after intravenous injection in subcutaneous SKOV3 CLDN4⁺/⁺ xenografts. (c, d) Live-cell, non-permeabilized CLDN4 surface immunofluorescence staining in SKOV3 CLDN4⁺/⁺ and OVCAR3 cells before fixation, with representative images in (c) and quantification of membrane-associated CLDN4 fluorescence intensity in (d). (e) Time-dependent internalization of FITC–P15 in SKOV3 CLDN4⁺/⁺ cells at 1, 3, and 6 h, shown together with DAPI and membrane dye staining, with enlarged views of the indicated regions. (f) Quantification of FITC fluorescence intensity corresponding to (e). (g) Penetration of FITC–P15 and FITC–Control in patient-derived ovarian cancer organoids. The patient-derived ovarian cancer organoid lines used in this study were established from three independent patients, and CLDN4 protein expression in PDO-1, PDO-2, and PDO-3 was verified by western blotting (Fig. S33e, f). Representative flow-cytometry histograms in (a) and representative confocal images in (c), (e), and (g) are shown from three independent experiments. Representative in vivo fluorescence images in (b) are shown from n = 3 mice per group. Data in (d) and (f) are presented as mean ± SD (n = 3 independent experiments). Statistical significance in (d) was determined by unpaired two-tailed Student’s t-test. Statistical significance in (f) was determined by one-way ANOVA with post hoc multiple-comparison testing. Statistical significance is indicated in the panels. Scale bars are shown in the corresponding panels. To directly assess whether CLDN4 was detectable at the cell surface, we next performed live-cell, non-permeabilized CLDN4 surface immunofluorescence staining before fixation in SKOV3 CLDN4⁺/⁺ and OVCAR3 cells. Peripheral membrane-associated CLDN4 signal was observed in both models (Fig. 4c, d). Together with the surface-preserving FITC–P15 competition assay, these results support the presence of surface-accessible CLDN4 capable of engaging exogenous P15 in the tested ovarian cancer models. Additional confocal colocalization analysis showed spatial association between FITC–P15 and CLDN4-positive membrane regions in SKOV3 CLDN4⁺/⁺ cells, whereas FITC–Control exhibited minimal cell association (Fig. S33a–d). Western blot analysis confirmed detectable CLDN4 protein expression in three independent patient-derived ovarian cancer organoid lines (PDO-1, PDO-2, and PDO-3) established from three patients, with heterogeneous expression across lines (Fig. S33e, f). Time-course imaging further showed progressive cellular uptake of FITC–P15 from 1 to 6 h, with a 3.6-fold increase in cell-associated fluorescence, while FITC–Control remained near baseline (Fig. 4e, f; Fig. S34). In patient-derived ovarian cancer organoids, FITC–P15 showed greater intratissue fluorescence distribution than FITC–Control (Fig. 4g). Collectively, these findings support surface-accessible CLDN4-associated P15 binding, cellular uptake, and tissue penetration in the tested ovarian cancer models. Physicochemical and functional characterization of IP/NBs To construct IP/NBs as an ultrasound-responsive CLDN4-directed formulation, IP was incorporated into phospholipid-shelled nanobubbles with an SF₆ gas core (IP/NBs; Fig. 5a). Dynamic light scattering revealed a predominantly submicron, right-skewed size distribution with a major population centered in the ~0.4–0.6 μm range (Fig. 5b). Transmission electron microscopy showed that the particles were predominantly spherical in the dry state (Fig. 5c). The apparent size measured by DLS was larger than that observed by TEM, which is expected because DLS reports hydrodynamic diameter in solution, whereas TEM reflects dehydrated projected size after sample preparation; for SF₆-filled nanobubbles, dehydration and vacuum effects can further reduce the apparent dry-state diameter. Under short-term handling conditions, IP/NBs also maintained stable hydrodynamic size in PBS at 37 °C over 0–20 min (Fig. 5d), supporting colloidal stability during the experimental time window relevant to subsequent imaging and treatment procedures. Fig. 5Characterization of IP-loaded nanobubbles (IP/NBs).(a) Schematic illustration of IP incorporation into phospholipid-shelled nanobubbles. (b) DLS hydrodynamic diameter distribution of IP/NBs in aqueous suspension. (c) TEM image of IP/NBs. (d) Colloidal stability of IP/NBs in PBS at 37 °C over 0–20 min. (e) Ultrasound contrast images of IP/NBs and empty NBs over 0–30 min. (f) Quantification of ultrasound contrast intensity decay for IP/NBs and NBs over 0–30 min. (g) Confocal fluorescence imaging of IP/NBs showing IP-associated ICG fluorescence, bright-field morphology, and merged images. Data in (d) and (f) are presented as mean ± SD (n = 3 independent batches for colloidal stability in d; n = 3 independent measurements for ultrasound contrast quantification in f). Scale bars are shown in the corresponding panels. Characterization of IP-loaded nanobubbles (IP/NBs).(a) Schematic illustration of IP incorporation into phospholipid-shelled nanobubbles. (b) DLS hydrodynamic diameter distribution of IP/NBs in aqueous suspension. (c) TEM image of IP/NBs. (d) Colloidal stability of IP/NBs in PBS at 37 °C over 0–20 min. (e) Ultrasound contrast images of IP/NBs and empty NBs over 0–30 min. (f) Quantification of ultrasound contrast intensity decay for IP/NBs and NBs over 0–30 min. (g) Confocal fluorescence imaging of IP/NBs showing IP-associated ICG fluorescence, bright-field morphology, and merged images. Data in (d) and (f) are presented as mean ± SD (n = 3 independent batches for colloidal stability in d; n = 3 independent measurements for ultrasound contrast quantification in f). Scale bars are shown in the corresponding panels. To assess storage and colloidal stability under protein-rich conditions, we monitored hydrodynamic diameter in serum-containing medium over 0–7 days. IP/NBs remained within the submicron range throughout the observation period, with only a modest increase at later time points and no evidence of pronounced aggregation (Fig. S35a), supporting short-term stability under these conditions. Surface charge analysis further showed that empty NBs carried a mildly negative zeta potential (−13.07 ± 1.03 mV), which shifted toward a less negative value after IP incorporation (−9.17 ± 1.50 mV; Fig. S35b). This zeta-potential shift is consistent with physical association/incorporation of IP within the nanobubble formulation rather than covalent surface functionalization. Further physicochemical characterization showed PDI values of 0.21 for empty NBs and 0.25 for IP/NBs, indicating only a modest increase in dispersity after IP incorporation. HPLC analysis after Triton X-100 lysis showed an encapsulation efficiency of approximately 64% and a drug loading of approximately 2.7% for IP/NBs, corresponding to a loading content of about 27 µg/mg lipid. In vitro release analysis showed time-dependent IP release from IP/NBs in PBS, and ultrasound exposure increased cumulative IP release under the tested conditions (Fig. S35c). A simplified plasma fluorescence-retention assay further showed higher plasma-associated IP-related fluorescence and slower signal decay for IP/NBs than for free IP, supporting improved plasma fluorescence retention associated with nanobubble incorporation under the tested conditions (Fig. S35d). We next evaluated the photothermal performance of IP/NBs under 808 nm irradiation (Fig. S36). Compared with PBS under identical 808-nm irradiation, IP/NBs (10 µM, ICG-equivalent) produced a markedly greater temperature increase at 0.5 W/cm² for 5 min (Fig. S36a). Heating was power-dependent across 0.3–0.7 W/cm² (Fig. S36b) and also increased with formulation concentration across 5–20 µM (Fig. S36c). In addition, IP/NBs maintained reproducible heating behavior over five laser on/off cycles (Fig. S36d), supporting photothermal repeatability under the tested irradiation cycles. Cooling analysis of a representative heating–cooling cycle yielded a slope of 0.3779 min⁻¹ and a system time constant (τs) of 2.65 min. Based on the standard heating–cooling method, the estimated photothermal conversion efficiency of IP/NBs was approximately 36.5% under the present experimental conditions (Fig. S36e, f). On the basis of these results, 808 nm irradiation at 0.5 W/cm² for 5 min was selected for subsequent experiments because it provided robust but controlled photothermal elevation in the IP/NBs group while inducing only minimal heating in PBS. Thus, this irradiation setting was used as a controlled mild-photothermal condition for subsequent experiments. In tissue-mimicking phantoms, SF₆-filled nanobubbles generated strong ultrasound contrast enhancement, and IP incorporation did not measurably attenuate echogenicity relative to empty NBs (Fig. 5e, f). Both formulations retained more than 80% of their initial signal at 15 min before undergoing a more rapid decline. Confocal microscopy further indicated a relatively uniform IP-associated fluorescence distribution within the nanobubble formulation (Fig. 5g). Collectively, these results indicate that SF₆ provides the echogenic gas core responsible for ultrasound visibility, while IP incorporation is compatible with nanobubble stability, photothermal responsiveness, and acoustic performance under the tested conditions. In vitro and in vivo biological performance of IP/NBs IP-loaded nanobubbles (IP/NBs) maintained key fluorescence-associated binding and penetration features after nanobubble formulation. In SKOV3 CLDN4⁺/⁺ cells, IP/NBs-associated fluorescence showed spatial association with CLDN4-positive regions, broadly similar to free IP (Fig. 6a, b). In patient-derived ovarian cancer organoids, IP/NBs also showed detectable intratissue fluorescence penetration (Fig. 6c). Fig. 6In vitro and in vivo biological performance of IP/NBs.(a, b) Confocal analysis showing the spatial relationship between IP- or IP/NBs-associated fluorescence and CLDN4 staining in SKOV3 CLDN4⁺/⁺ cells, with representative images and enlarged regions shown in (a) and line-scan fluorescence intensity profiles across the indicated regions shown in (b). (c) Representative confocal and bright-field images showing penetration of IP/NBs into patient-derived ovarian cancer organoids and intratissue IP-related fluorescence distribution after incubation. (d) Representative B-mode and contrast-mode ultrasound images of subcutaneous SKOV3 CLDN4⁺/⁺ tumors before and after intravenous administration of NBs or IP/NBs. Red dashed circles indicate tumor regions of interest. “Pre” indicates baseline imaging before injection, and 2–8 min indicate elapsed time after completion of tail-vein injection. (e) Quantification of tumor-region ultrasound contrast mean power over time corresponding to (d). (f) Representative in vivo near-infrared fluorescence images at 3 h after intravenous injection of IP, D-NBs, or IP/NBs in subcutaneous SKOV3 CLDN4⁺/⁺ tumor-bearing mice. Red dashed circles indicate tumor regions of interest. D-NBs served as the non-targeted dye-loaded nanobubble comparator, IP served as the free CLDN4-directed conjugate comparator, and IP/NBs served as the CLDN4-directed nanobubble formulation. (g) Quantification of tumor-associated fluorescence at 3 h corresponding to (f). Representative confocal images in (a) and (c) and line-scan profiles in (b) are shown from three independent experiments. In vivo imaging in (d–g) was performed with n = 3 mice per group. Individual data points and group means are shown in (g). Statistical significance in (g) was determined by one-way ANOVA with post hoc multiple-comparison testing and is indicated in the panel. Scale bars are shown in the corresponding panels. In vitro and in vivo biological performance of IP/NBs.(a, b) Confocal analysis showing the spatial relationship between IP- or IP/NBs-associated fluorescence and CLDN4 staining in SKOV3 CLDN4⁺/⁺ cells, with representative images and enlarged regions shown in (a) and line-scan fluorescence intensity profiles across the indicated regions shown in (b). (c) Representative confocal and bright-field images showing penetration of IP/NBs into patient-derived ovarian cancer organoids and intratissue IP-related fluorescence distribution after incubation. (d) Representative B-mode and contrast-mode ultrasound images of subcutaneous SKOV3 CLDN4⁺/⁺ tumors before and after intravenous administration of NBs or IP/NBs. Red dashed circles indicate tumor regions of interest. “Pre” indicates baseline imaging before injection, and 2–8 min indicate elapsed time after completion of tail-vein injection. (e) Quantification of tumor-region ultrasound contrast mean power over time corresponding to (d). (f) Representative in vivo near-infrared fluorescence images at 3 h after intravenous injection of IP, D-NBs, or IP/NBs in subcutaneous SKOV3 CLDN4⁺/⁺ tumor-bearing mice. Red dashed circles indicate tumor regions of interest. D-NBs served as the non-targeted dye-loaded nanobubble comparator, IP served as the free CLDN4-directed conjugate comparator, and IP/NBs served as the CLDN4-directed nanobubble formulation. (g) Quantification of tumor-associated fluorescence at 3 h corresponding to (f). Representative confocal images in (a) and (c) and line-scan profiles in (b) are shown from three independent experiments. In vivo imaging in (d–g) was performed with n = 3 mice per group. Individual data points and group means are shown in (g). Statistical significance in (g) was determined by one-way ANOVA with post hoc multiple-comparison testing and is indicated in the panel. Scale bars are shown in the corresponding panels. In vivo ultrasound imaging further showed that IP/NBs generated immediate tumor contrast within 120 s after intravenous administration, consistent with the echogenicity conferred by the SF₆ gas core, and maintained more than 80% of the initial signal intensity for approximately 480 s before undergoing more rapid decay (Fig. 6d, e). Empty NBs also generated ultrasound signal, allowing comparison of acoustic contrast behavior under the same imaging settings. The shorter in vivo ultrasound-signal persistence compared with the in vitro phantom setting is consistent with reduced acoustic stability under circulatory conditions and a transient in vivo acoustic window. To further assess tumor-associated fluorescence using a dye-matched non-targeted nanobubble comparator, we compared free IP, D-NBs, and IP/NBs at 3 h after intravenous injection in subcutaneous SKOV3 CLDN4⁺/⁺ tumor-bearing mice. D-NBs served as the non-targeted dye-loaded nanobubble comparator, whereas IP served as the free CLDN4-directed conjugate comparator. IP/NBs showed the highest tumor-associated fluorescence, exceeding both free IP and D-NBs (Fig. 6f, g). These data support greater tumor-associated IP-related delivery/retention for the CLDN4-directed IP/NBs formulation than for the free conjugate or a dye-matched non-targeted nanobubble comparator within the tested imaging workflow. Together with the transient in vivo ultrasound contrast window, these findings indicate tumor-associated IP-related fluorescence after nanobubble-assisted transport/release and retention. To further examine whether nanobubble-assisted ultrasound could enhance transvascular transport, we established a TEER-verified bEnd.3 Transwell barrier model. Under ultrasound exposure, IP/NBs produced the highest basolateral IP-ICG signal and significantly exceeded the “NBs + ultrasound + free IP” condition (Fig. S37a, b), consistent with ultrasound–nanobubble-associated enhancement of transendothelial IP-related transport under the tested conditions. Because the therapeutic protocol used early ultrasound exposure followed by delayed NIR irradiation, we performed matched post-ultrasound fluorescence imaging under the same workflow. Compared with the IP/NBs (no US) group, the IP/NBs + US group showed higher tumor-associated fluorescence over time and stronger ex vivo tumor fluorescence at both 3 h and 6 h (Fig. S38), consistent with ultrasound-associated enhancement of local IP-related delivery/retention under the tested workflow. IP/NBs showed relatively limited cytotoxicity toward normal IOSE ovarian epithelial cells, with 73.2 ± 4.1% viability retained even at 500 µM (Fig. S39a). Based on dose–response profiling, 10 µM was selected for subsequent mechanistic and therapeutic studies because it produced a clear treatment gradient, with cell viability of 96.3 ± 2.8% in the PBS + US + NIR group, 81.0 ± 6.5% in the NBs + US + NIR group, 44.3 ± 5.3% in the IP + US + NIR group, and 17.3 ± 5.2% in the IP/NBs + US + NIR group (Fig. 7a). At this working concentration, in a parallel NIR-control assay, no pronounced cytotoxicity was observed in the PBS or NBs groups under NIR irradiation, with cell viability remaining at 101.3 ± 2.5% and 101.7 ± 5.0%, respectively. In contrast, free IP + NIR reduced cell viability to 45.0 ± 6.2%, whereas IP/NBs + NIR further reduced viability to 18.0 ± 6.1% (Fig. S39b). Exploratory interaction analysis showed a positive Bliss excess (ΔBliss = 0.28; bootstrap 95% CI, 0.25–0.31), suggesting a positive interaction between IP-associated photothermal treatment and the nanobubble formulation under the tested fixed-dose irradiation condition (Fig. S39c, d). In the main ROS assay, DCFH-DA imaging and quantification showed increased intracellular ROS-related fluorescence after IP- or IP/NBs-based photothermal treatment, with the strongest signal observed in the IP/NBs + US + NIR group (29.88 ± 0.41 a.u.), approximately 1.30-fold higher than that in the IP + US + NIR group (Fig. 7b, c). In a complementary DCFH-DA live-cell assay, PBS, PBS + US + NIR, NBs + US + NIR, and D-NBs + US + NIR remained near baseline, whereas IP/NBs + US (− NIR) induced only a modest increase; both IP + US + NIR and IP/NBs + US + NIR produced strong DCF signals (Fig. S39e, f). Together, these data indicate increased treatment-associated oxidative stress after IP- or IP/NBs-based photothermal treatment, with DCFH-DA fluorescence used as an oxidative-stress readout. In patient-derived ovarian cancer organoids, IP/NBs induced progressive treatment-associated morphological damage over 5 days, including organoid shrinkage, structural collapse, and increased fragmented/dead-cell-like regions (Fig. 7d). Fig. 7Photothermal effects, organoid response, in vivo therapy, and mechanistic readouts of IP/NBs.(a) Dose-dependent cytotoxicity of SKOV3 CLDN4⁺/⁺ cells treated with PBS, NBs, IP, or IP/NBs under US + NIR conditions. (b, c) Intracellular ROS-related fluorescence after the indicated treatments, with representative DCFH-DA images in (b) and quantification in (c). DCFH-DA fluorescence was used as a treatment-associated oxidative-stress readout. (d) Patient-derived ovarian cancer organoid response after treatment with PBS + US + NIR, NBs + US + NIR, IP + US + NIR, or IP/NBs + US + NIR at Day 0, Day 3, and Day 5. Organoid shrinkage and structural disruption were used as morphological indicators of treatment-associated damage. (e) In vivo treatment workflow, including intravenous formulation administration, ultrasound exposure at 4 min post-injection where indicated, and 808-nm NIR irradiation at 6 h post-injection. (f) Endpoint excised tumors. (g) Final tumor weights. (h) Endpoint tumor volumes. (i) H&E, Ki67, and TUNEL staining of tumor tissues. (j) Ex vivo tumor fluorescence images under the therapeutic workflow; IP + US + NIR and IP/NBs + US + NIR were IP-containing groups, whereas PBS + US + NIR and NBs + US + NIR served as non-ICG background controls. Data in (a), (c), (g), and (h) are mean ± SD (n = 3 independent experiments for in vitro assays; n = 5 mice per group for in vivo therapy). Representative organoid images in (d) are from three organoid replicates per group. Dose-response data in (a) were analyzed by two-way ANOVA; (c), (g), and (h) by one-way ANOVA, with post hoc multiple-comparison testing. Statistical significance is indicated in the panels. Scale bars are shown in the panels. Photothermal effects, organoid response, in vivo therapy, and mechanistic readouts of IP/NBs.(a) Dose-dependent cytotoxicity of SKOV3 CLDN4⁺/⁺ cells treated with PBS, NBs, IP, or IP/NBs under US + NIR conditions. (b, c) Intracellular ROS-related fluorescence after the indicated treatments, with representative DCFH-DA images in (b) and quantification in (c). DCFH-DA fluorescence was used as a treatment-associated oxidative-stress readout. (d) Patient-derived ovarian cancer organoid response after treatment with PBS + US + NIR, NBs + US + NIR, IP + US + NIR, or IP/NBs + US + NIR at Day 0, Day 3, and Day 5. Organoid shrinkage and structural disruption were used as morphological indicators of treatment-associated damage. (e) In vivo treatment workflow, including intravenous formulation administration, ultrasound exposure at 4 min post-injection where indicated, and 808-nm NIR irradiation at 6 h post-injection. (f) Endpoint excised tumors. (g) Final tumor weights. (h) Endpoint tumor volumes. (i) H&E, Ki67, and TUNEL staining of tumor tissues. (j) Ex vivo tumor fluorescence images under the therapeutic workflow; IP + US + NIR and IP/NBs + US + NIR were IP-containing groups, whereas PBS + US + NIR and NBs + US + NIR served as non-ICG background controls. Data in (a), (c), (g), and (h) are mean ± SD (n = 3 independent experiments for in vitro assays; n = 5 mice per group for in vivo therapy). Representative organoid images in (d) are from three organoid replicates per group. Dose-response data in (a) were analyzed by two-way ANOVA; (c), (g), and (h) by one-way ANOVA, with post hoc multiple-comparison testing. Statistical significance is indicated in the panels. Scale bars are shown in the panels. We next evaluated therapeutic efficacy in vivo using ultrasound exposure at 4 min post-injection followed by NIR irradiation at 6 h (808 nm, 0.5 W/cm², 5 min; Fig. 7e). The in vivo therapeutic study included PBS + US + NIR, NBs + US + NIR, D-NBs + US + NIR, IP + US + NIR, IP/NBs + US (− NIR), and IP/NBs + US + NIR groups to compare irradiation-only, empty nanobubble, non-targeted dye-loaded nanobubble, free IP, NIR-omission, and full-treatment conditions under the same therapeutic workflow. The irradiation-only, empty nanobubble, and non-targeted dye-loaded nanobubble controls showed limited tumor suppression, whereas IP + US + NIR produced partial tumor inhibition and IP/NBs + US + NIR yielded the smallest mean endpoint tumor volume and tumor weight (10.6 ± 14.0 mm³ and 0.097 ± 0.128 g, respectively), corresponding to an approximately 93.0% reduction in mean tumor weight relative to PBS + US + NIR, without overt body-weight loss (Fig. 7f–h; Fig. S40a). In a separate survival cohort, all mice in the PBS + US + NIR group reached the predefined survival endpoint within the observation window, whereas the IP/NBs + US + NIR group showed the most favorable descriptive survival pattern (Fig. S40b). We further compared three IP-based regimens to evaluate the contribution of early ultrasound exposure to the therapeutic outcome: IP + US + NIR, IP/NBs + NIR (no US), and IP/NBs + US + NIR. The IP/NBs + NIR (no US) group produced an intermediate tumor-suppressive effect, whereas IP/NBs + US + NIR further reduced the mean day-16 relative tumor volume from 52.8 ± 5.6% to 34.5 ± 5.0% and reduced mean endpoint tumor weight from 0.450 ± 0.060 g to 0.300 ± 0.040 g, corresponding to a 33.3% reduction in mean endpoint tumor weight compared with IP/NBs + NIR (no US) (Fig. S41). These results support a contribution of the early ultrasound trigger to therapeutic efficacy under the tested workflow, beyond the effect of IP/NBs plus delayed NIR irradiation alone. Histological analyses were consistent with these findings: the IP/NBs + US + NIR group showed more extensive tumor damage and necrosis, reduced Ki67 staining, and increased TUNEL positivity (Fig. 7i) relative to the principal control groups, whereas IP/NBs + US (− NIR) did not show comparable histological changes. Representative ex vivo tumor fluorescence images under the therapeutic workflow showed ICG-channel fluorescence in the IP-containing groups, with PBS + US + NIR and NBs + US + NIR serving as non-ICG background controls (Fig. 7j). Consistent with the body-weight data, H&E staining of major organs across the main therapeutic comparator set, including D-NBs + US + NIR and IP/NBs + US (− NIR), did not reveal overt treatment-associated histopathological injury in the heart, liver, spleen, lung, or kidney at day 16 (Fig. S42). Biodistribution, safety, and ICD-associated immune activation of IP/NBs Biodistribution analysis at 24 h post-administration showed detectable tumor-associated ICG-channel fluorescence in both the IP and IP/NBs groups (Fig. 8a, b), indicating persistent tumor-associated IP-related signal at this terminal sampling time point. Organ-level ICG-channel signals in the dye-containing groups were dominated by hepatic uptake, consistent with the expected reticuloendothelial/hepatobiliary clearance pattern of lipid- and ICG-containing formulations. PBS + US + NIR and NBs + US + NIR showed near-background ICG-channel signals, as expected for non-ICG control groups. The ex vivo tumor-fluorescence difference between IP and IP/NBs was modest at 24 h, whereas earlier matched post-ultrasound imaging showed a clearer ultrasound-associated difference in tumor-associated fluorescence. Fig. 8Biodistribution, safety, and immune activation associated with IP/NBs-mediated treatment.(a, b) Ex vivo fluorescence imaging of tumors and major organs collected 24 h after intravenous administration of the indicated formulations, with representative ICG-channel images in (a) and quantification in (b). PBS + US + NIR and NBs + US + NIR served as non-ICG background controls. (c) Ex vivo tumor cell–PBMC co-culture workflow. (d–i) Flow-cytometric analysis of PBMC co-cultures after stimulation with treated tumor cells, including PBS + US + NIR, D-NBs + US + NIR, IP + US + NIR, and IP/NBs + US + NIR. Quantification of CD80⁺CD86⁺ dendritic cells, CD69⁺CD4⁺ T cells, and CD69⁺CD8⁺ T cells is shown in (e), (g), and (i), respectively. D-NBs served as the non-targeted dye-loaded nanobubble comparator, IP as the free CLDN4-directed photothermal conjugate comparator, and IP/NBs as the CLDN4-directed formulation. (j, k) CD3 immunohistochemistry of tumor sections from human PBMC adoptive-transfer tumor-bearing mice, with representative images in (j) and quantification in (k). (l–n) Immunofluorescence co-staining of intratumoral CD4⁺ or CD8⁺ T cells and CD69, with representative images in (l) and CD4/CD8 quantification in (m, n). In (l), CD4 or CD8 is green, CD69 is yellow, and DAPI is blue. Quantification in (m, n) used CD4 and CD8 marker channels; DAPI served as a nuclear counterstain, and CD69 was evaluated qualitatively. Quantitative data are mean ± SD (n = 3 mice per group for b, k,m, n; n = 3 independent experiments for e, g,i). Representative images are from n = 3 mice per group, and representative flow-cytometry plots are from three independent experiments. Statistical significance was determined by one-way ANOVA with post hoc multiple-comparison testing and is indicated where applicable. Scale bars are shown in the panels. Biodistribution, safety, and immune activation associated with IP/NBs-mediated treatment.(a, b) Ex vivo fluorescence imaging of tumors and major organs collected 24 h after intravenous administration of the indicated formulations, with representative ICG-channel images in (a) and quantification in (b). PBS + US + NIR and NBs + US + NIR served as non-ICG background controls. (c) Ex vivo tumor cell–PBMC co-culture workflow. (d–i) Flow-cytometric analysis of PBMC co-cultures after stimulation with treated tumor cells, including PBS + US + NIR, D-NBs + US + NIR, IP + US + NIR, and IP/NBs + US + NIR. Quantification of CD80⁺CD86⁺ dendritic cells, CD69⁺CD4⁺ T cells, and CD69⁺CD8⁺ T cells is shown in (e), (g), and (i), respectively. D-NBs served as the non-targeted dye-loaded nanobubble comparator, IP as the free CLDN4-directed photothermal conjugate comparator, and IP/NBs as the CLDN4-directed formulation. (j, k) CD3 immunohistochemistry of tumor sections from human PBMC adoptive-transfer tumor-bearing mice, with representative images in (j) and quantification in (k). (l–n) Immunofluorescence co-staining of intratumoral CD4⁺ or CD8⁺ T cells and CD69, with representative images in (l) and CD4/CD8 quantification in (m, n). In (l), CD4 or CD8 is green, CD69 is yellow, and DAPI is blue. Quantification in (m, n) used CD4 and CD8 marker channels; DAPI served as a nuclear counterstain, and CD69 was evaluated qualitatively. Quantitative data are mean ± SD (n = 3 mice per group for b, k,m, n; n = 3 independent experiments for e, g,i). Representative images are from n = 3 mice per group, and representative flow-cytometry plots are from three independent experiments. Statistical significance was determined by one-way ANOVA with post hoc multiple-comparison testing and is indicated where applicable. Scale bars are shown in the panels. To evaluate downstream immune activation associated with IP/NBs-mediated therapy, we established an ex vivo co-culture system incorporating treated tumor cells and human PBMCs (Fig. 8c). In the main immune activation panels, a matched four-group comparator set was used, including PBS + US + NIR, D-NBs + US + NIR, IP + US + NIR, and IP/NBs + US + NIR. Relative to the PBS + US + NIR and D-NBs + US + NIR controls, IP-containing treatments showed higher immune-activation marker frequencies, with IP/NBs + US + NIR producing the strongest response. This was reflected by increased frequencies of CD80⁺CD86⁺ dendritic cells, activated CD3⁺CD4⁺CD69⁺ T cells, and activated CD3⁺CD8⁺CD69⁺ T cells, reaching 39.63 ± 5.36%, 53.43 ± 3.80%, and 46.63 ± 3.30%, respectively, in the IP/NBs + US + NIR group (Fig. 8d–i; gating shown in Fig. S43). These results are consistent with stronger downstream immune activation in the IP/NBs + US + NIR group than in the non-targeted dye-loaded nanobubble comparator group in this co-culture setting. To further assess ICD-associated danger signaling, we quantified surface calreticulin exposure together with extracellular ATP and HMGB1 release after the indicated treatments. IP/NBs + US + NIR showed increased surface calreticulin signal (36,015 ± 8,895 a.u.) and was accompanied by elevated extracellular ATP and HMGB1 release, reaching 4.50 ± 0.35-fold and 95.7 ± 11.5 ng/mL, respectively (Fig. S44a–d). Consistent with these danger-signal readouts, PBMC co-culture supernatants from the IP/NBs + US + NIR group showed increased secretion of IFN-γ and TNF-α, reaching 1250 ± 180 pg/mL and 673 ± 97 pg/mL, respectively (Fig. S44e, f). These data indicate that IP/NBs-mediated therapy was associated with ICD-associated danger signaling and downstream immune activation under the tested assay conditions. In human PBMC adoptive-transfer tumor-bearing mice, IP + US + NIR and IP/NBs + US + NIR increased intratumoral T-cell infiltration relative to the principal control groups (Fig. 8j, k), consistent with treatment-associated immune activation under the tested conditions. Immunofluorescence co-staining of tumor sections further showed increased CD4 and CD8 marker-specific signals in treated tumors, with the strongest signals observed in the IP/NBs + US + NIR group (Fig. 8l–n). CD69 co-staining was evaluated qualitatively, whereas quantitative analysis was based on the CD4 and CD8 marker channels; DAPI served only as a nuclear counterstain. Quantification confirmed significantly higher intratumoral CD4 and CD8 fluorescence signals in the IP/NBs + US + NIR group than in the control conditions, consistent with the ex vivo PBMC activation results. We next examined whether the treatment regimen was associated with splenic T-cell activation in vivo. Spleens were harvested from tumor-bearing mice after PBMC infusion and the indicated treatments, dissociated into single-cell suspensions, and analyzed by flow cytometry. The results showed increased activation-marker positivity among both CD4⁺ and CD8⁺ T cells in vivo, with the strongest response observed in the IP/NBs + US + NIR group (Fig. S45). Collectively, these findings indicate that the IP/NBs + US + NIR regimen exhibited antitumor effects in the tested ovarian cancer models and was associated with increased immune-cell activation and tumor-associated T-cell infiltration under the tested conditions. Discussion Ovarian cancer remains a major clinical challenge because of its insidious early progression, frequent advanced-stage presentation, and the strong dependence of survival on complete macroscopic cytoreduction. These clinical realities continue to motivate molecularly targeted strategies that can improve lesion visualization, local delivery, and therapeutic precision. In this study, we developed a CLDN4-directed, IP-containing nanobubble phototheranostic framework that integrates peptide-guided molecular recognition, ultrasound/fluorescence imaging, ultrasound-associated IP-related delivery/retention, and photothermal treatment in CLDN4-positive ovarian cancer models. A central premise of this work is that CLDN4 may serve as a therapeutically relevant and ligand-accessible target candidate in ovarian cancer. CLDN4 should not be viewed as tumor-exclusive; its translational value depends on tumor enrichment, disease-associated junctional remodeling, and the presence of surface-accessible CLDN4 for external ligand engagement. Our single-cell and tissue-level analyses supported CLDN4 enrichment in the ovarian tumor/epithelial compartment, while invasion-related assays suggested an association between CLDN4 expression and aggressive phenotypes. These findings provide target-prioritization support but remain associative. Importantly, surface-preserving FITC–P15 binding/competition assays and non-permeabilized CLDN4 immunofluorescence provided experimental evidence that CLDN4 was accessible to exogenous ligand engagement in the tested models, thereby strengthening the rationale for P15-mediated recognition. Through structure-guided computational screening and experimental validation, P15 was identified as the lead CLDN4-binding peptide and used to generate the ICG-conjugated photothermal probe IP. Kinetic SPR analysis of the four validated candidates further supported P15 selection, as P15 retained the lowest kinetic-derived KD among the tested peptides. These data provide binding-context information beyond steady-state affinity estimation and support the use of P15 as the targeting component of the IP/NBs platform. The incorporation of IP into SF₆-filled lipid nanobubbles enabled the construction of an ultrasound-visible formulation with photothermal responsiveness. Within this formulation, the IP component provides CLDN4-directed molecular recognition, whereas the SF₆ gas core contributes to ultrasound visibility. The in vivo acoustic signal persisted for a shorter period than the in vitro phantom signal, indicating a transient acoustic window under circulatory conditions. Therefore, delayed fluorescence signals at later time points are best interpreted as tumor-associated IP-related signals following nanobubble-assisted transport/release and retention, rather than as evidence of prolonged persistence of intact nanobubbles. This interpretation is supported by the matched post-ultrasound fluorescence imaging experiment, in which IP/NBs + US showed higher tumor-associated fluorescence over time and higher ex vivo tumor fluorescence at 3 h and 6 h than IP/NBs without ultrasound. In the IP-based therapeutic comparison, IP/NBs + NIR without ultrasound produced an intermediate antitumor effect, whereas IP/NBs + US + NIR showed the greatest tumor suppression among the tested IP-based regimens. Together, these findings support a contribution of early ultrasound exposure to local IP-related delivery/retention and therapeutic efficacy under the tested workflow. The immune-related findings further suggest that IP/NBs-mediated treatment was associated with ICD-associated danger signaling and downstream immune activation. In the main immune activation panels, D-NBs served as the non-targeted dye-loaded nanobubble comparator, IP served as the free CLDN4-directed photothermal conjugate comparator, and IP/NBs served as the CLDN4-directed formulation. Relative to the PBS + US + NIR and D-NBs + US + NIR control groups, CLDN4-directed IP-containing treatments increased dendritic-cell maturation and T-cell activation, with the strongest responses observed in the IP/NBs + US + NIR group. Increased calreticulin exposure, ATP and HMGB1 release, and elevated IFN-γ and TNF-α further supported the presence of treatment-associated ICD-related danger signaling and immune activation. These data indicate an immune-activation association under the tested conditions, while further studies will be needed to distinguish the individual contributions of peptide recognition, nanobubble transport, ultrasound exposure, and photothermal injury. Several limitations should be considered when interpreting these findings. The linear peptide format of P15 may have limited systemic durability, and future structural optimization, such as cyclization or incorporation of non-natural amino acids, may improve plasma stability and in vivo half-life. The current evidence supports surface-accessible CLDN4 and P15-mediated ligand recognition in the tested models but does not establish broad CLDN4 accessibility across all ovarian cancer subtypes or clinical specimens. Further validation in endogenous CLDN4-high models, additional patient-derived organoids, orthotopic or metastatic ovarian cancer models, and larger tissue cohorts will be important. In addition, although the endothelial Transwell assay supported ultrasound-assisted barrier transport, cavitation events were not directly visualized or quantitatively measured. Future studies incorporating cavitation monitoring, acoustic dose mapping, and spatially resolved vascular permeability analysis would help define the ultrasound-mediated transport mechanism more precisely. Additional translational questions also remain. Although no overt histopathologic toxicity was observed in major organs, long-term safety, off-target binding, receptor occupancy, immunogenicity, and pharmacokinetic behavior were not comprehensively characterized. The nude mouse/human PBMC adoptive-transfer model provides useful information on human immune-cell activation but only partially approximates human antitumor immunity and does not fully recapitulate an intact immune microenvironment. Future studies should therefore extend validation to immune-competent or more advanced humanized models where appropriate. Finally, although IP/NBs maintained submicron size characteristics over 7 days under the tested serum-containing conditions, comprehensive shelf-life evaluation of SF₆-filled nanobubbles, including gas retention, ultrasound contrast activity, payload integrity, ligand accessibility, and batch-to-batch reproducibility, will be required before translational development. Conclusions This study developed a CLDN4-directed, IP-containing nanobubble phototheranostic framework integrating peptide-guided recognition, ultrasound/fluorescence imaging, ultrasound-associated IP-related delivery/retention, and photothermal treatment in CLDN4-positive ovarian cancer models. P15 supported CLDN4-associated ligand recognition, and IP/NBs preserved ultrasound visibility and photothermal responsiveness while enhancing tumor-associated IP-related retention under the tested workflow. IP/NBs + US + NIR produced marked tumor suppression and was associated with ICD-associated danger signaling, immune-cell activation, and tumor-associated T-cell infiltration. These findings support further optimization of CLDN4-directed ultrasound-assisted peptide–nanobubble phototheranostics for ovarian cancer. Supplementary Information Supplementary Material 1. Supplementary Material 1. Supplementary Material 2. Supplementary Material 2. Supplementary Material 3. Supplementary Material 3.