Graphical abstract graphic file with name fx1.jpg [45]Open in a new tab Highlights * • The iso-palmitoylation of CLDN4 promotes its stability within lipid rafts * • CLDN4-CNTN1-NOTCH axis drives hepatic-to-biliary transition (HBT) in HCC * • High levels of CLDN4 and HBT correlate with lenvatinib resistance in HCC patients __________________________________________________________________ Xu et al. identify CLDN4 palmitoylation as a key driver of hepatic-to-biliary transition (HBT) and lenvatinib resistance in HCC. Targeting CLDN4 with salvianolic acid B (SalB) reverses HBT and restores lenvatinib sensitivity, offering a promising therapeutic strategy for resistant HCC patients. Introduction Lineage transition is an intricate and poorly understood process in which the original lineage of a cancer undergoes alteration, leading to the emergence of a new cancer lineage. Due to the dynamic nature of cancer, tumor cell subpopulations evolve continuously throughout malignant progression and systemic treatment. This ongoing evolution results in tumor heterogeneity, characterized by diverse populations of cancer cells with unique molecular and cellular attributes.[46]^1 Tumor-initiating cells, also known as cancer stem cells, exhibit significant plasticity, enabling phenotypic transitions. These transitions may involve the adoption of drug-resistant, slow-cycling states or differentiation into various cell lineages to evade drug toxicity.[47]^2^,[48]^3 Moreover, terminally differentiated cells have the potential to undergo dedifferentiation or transdifferentiation into alternative lineages, thereby reducing their dependence on signaling pathways targeted by therapies.[49]^4 Among patients with epidermal growth factor receptor-mutant lung adenocarcinoma who develop resistance to osimertinib, approximately 7%–9% exhibit pathological characteristics of lung squamous cell carcinoma upon secondary biopsy.[50]^5 Similarly, the adenocarcinoma-to-squamous transition can contribute to resistance against Kirsten Rat Sarcoma Viral Oncogene Homolog-targeted therapies in LKB1-mutant lung cancer.[51]^6 Loss of TP53 and RB1 function may also facilitate the phenotypic transition of prostate cancer from androgen receptor-dependent luminal epithelial cells to androgen receptor-independent basal-like cells, leading to resistance to the anti-androgen drug enzalutamide.[52]^7 Alterations in cellular plasticity are evident in normal hepatic development and the oncogenic transformation associated with hepatocellular carcinoma (HCC).[53]^8 Under specific activation of developmental signaling pathways, hepatocytes and cholangiocytes possess the ability to transdifferentiate, resulting in a mixed-lineage phenotype in HCC patients.[54]^9^,[55]^10 Our prior retrospective study indicated that patients with HCC exhibiting this mixed lineage phenotype experienced poor prognostic outcomes and reduced responsiveness to therapeutic interventions.[56]^11 However, the mechanisms underlying lineage transition in HCC during targeted therapy remain largely unexplored. Lenvatinib, an oral multi-target receptor tyrosine kinase inhibitor targeting VEGFR1-VEGFR3, FGFR1-FGFR4, PDGFRα, KIT, and RET, is widely recommended as a first-line treatment for HCC globally, particularly for patients in intermediate to advanced stages.[57]^12^,[58]^13 It has also been employed in several prospective clinical trials as part of combination therapy for patients with intrahepatic cholangiocarcinoma (ICC).[59]^14 Despite its widespread application, the efficacy of lenvatinib in treating advanced HCC remains limited, as evidenced by objective response rates below 20% in the phase III clinical trial LEAP-002.[60]^15 Thus, there is an urgent need to identify clinical features or molecular markers capable of predicting the response to lenvatinib treatment in HCC to reduce healthcare costs and advance precision medicine. The present study identified Claudin-4 (CLDN4) as a potential key target for overcoming lenvatinib resistance in HCC, specifically by facilitating hepatic-to-biliary lineage transition (HBT). Palmitoylation occurring at cysteine residues C104 and C107 of CLDN4 promoted the lipid raft anchoring of CLDN4 and drive contactin-1 (CNTN1) to lipid rafts, fueling Notch pathway activation and HBT. Salvianolic acid B (SalB), a screened CLDN4 antagonist, could effectively alleviate HBT and treatment resistance in lenvatinib-resistant patient-derived xenografts (LR-PDXs). Moreover, for HCC patients with HBT, combination chemotherapy may yield significant benefits. Results CLDN4 serves as a potential target to mitigate lenvatinib resistance in HCC To identify key molecules involved in lenvatinib resistance, we established a spontaneous HCC model and acquired potentially resistant cells using an in vivo induction approach,[61]^16 specifically by continuous lenvatinib exposure to induce drug tolerance in HCC mice, with the increased drug tolerance of primary HCC cells being validated through both in vivo and in vitro experiments ([62]Figures 1A–1E). Primary cells exhibiting tolerance to lenvatinib were collected for proteomics analysis. Additionally, tumor samples from three pairs of patients, categorized as either resistant or sensitive to lenvatinib, were collected and subjected to single-cell sequencing (Fu-LR Sc cohort). CLDN4 emerged as the most significantly upregulated molecule in resistant patients and the only molecule that is simultaneously upregulated in both resistant cell lines and patients ([63]Figures 1F–1H). After stringent quality control, 93,776 cells were retained for single-cell RNA sequencing (scRNA-seq). Following the normalization of gene expression data, we employed principal-component analysis and uniform manifold approximation and projection (UMAP) for dimensionality reduction and clustering. Subsequently, copy number variation analysis was used to distinguish malignant from nonmalignant cells. The resulting cell population was classified into nine distinct cell types based on established marker genes: epithelial and tumor cells (24,463 cells, 26.1%, marked by EPCAM, ALDH1A1, and ALB), B cells (4,341 cells, 4.6%, marked by MS4A1 and CD79A), T cells (36,264 cells, 38.7%, marked by CD3D and CD3E), natural killer cells (2,168 cells, 2.3%, marked by FGFBP2), neutrophils (2,459 cells, 2.6%, marked by Integrin Subunit Alpha M and CD33), macrophages (12,624 cells, 13.5%, marked by CD68, CD163, and CD14), monocytes (5,898 cells, 6.3%, marked by Integrin Subunit Alpha X), endothelial cells (4,120 cells, 4.4%, marked by PECAM1),; and smooth muscle cells (1,439 cells, 1.5%, marked by PECAM1 and von Willebrand factor). Notably, CLDN4 was localized in malignant cells ([64]Figures 1I and [65]S1A). A previous study assessed the sensitivity of multiple HCC cell lines to lenvatinib, categorizing them into lenvatinib-resistant lines, including SUN449, JHH1, Huh6, SUN182, PLC/PRF/5, SK-HEP1, MHCC97H, and HepG2 cells, and lenvatinib-sensitive lines, including SUN398, Huh7, and Hep3B cells.[66]^17 Our findings indicated that CLDN4 expression was significantly elevated in most lenvatinib-resistant cell lines compared to their sensitive counterparts ([67]Figures 1J and 1K). To further investigate the role of CLDN4 in lenvatinib resistance, we collected residual tumor tissues from 55 patients who had undergone preoperative lenvatinib treatment and untreated tumor tissues from 55 patients who had undergone upfront hepatectomy (Fu-LR cohort).[68]^18 Immunohistochemical revealed that CLDN4 was localized to the membranes of tumor cells ([69]Figures 1L and 1M). HCC patients expressing CLDN4 or evaluated as having stable disease were more prevalent in the residual cohort than in the untreated cohort ([70]Figure 1N). Additionally, elevated CLDN4 expression was correlated with a less favorable postoperative prognosis ([71]Figure 1O; [72]Table S1). To explore whether CLDN4 expression could serve as a predictor of the therapeutic efficacy of lenvatinib in HCC, pretreatment biopsy samples were collected from 12 patients. The findings demonstrated that the expression of CLDN4 in tumor tissues, as assessed by immunofluorescence, was significantly lower in patients who were responsive to lenvatinib (n = 4) than in those who exhibited resistance to the drug (n = 8) ([73]Figure 1P). These findings suggest that CLDN4 could be a therapeutic target to mitigate lenvatinib resistance in HCC. Figure 1. [74]Figure 1 [75]Open in a new tab CLDN4 serves as a potential target to mitigate lenvatinib resistance in HCC (A) Schematic showing the establishment of spontaneous HCC mouse models. After confirming the successful establishment of spontaneous HCC, the mice were administered lenvatinib (20 mg/kg/day) or mock via oral gavage for 1 month. The mice were euthanized, and primary HCC cells were isolated. (B) Representative gross appearance and bioluminescence imaging (left), liver/body weight (middle), and body weight (right) of spontaneous HCC models (n = 3). Data were analyzed using Student’s t test. (C) CCK-8 assay of the lenvatinib (Len) group and mock groups treated with Len at the indicated concentrations for 72 h. (D) Colony formation assay of the Len and mock groups treated with Len (30 μM) for 3 weeks (n = 3). Representative images (bottom) and average numbers of colonies (top) are shown. Data were analyzed using Student’s t test. Scale bar: 5 mm. (E) Kaplan-Meier survival graph for C57BL/6 and severely immunodeficient mice (NOD-Prkdc^scidIl2rg^em1/Smoc, NSG) mice treated with Len or mock treatment (n = 12). Data were analyzed using log rank test. (F) Proteomic sequencing analysis was used to analyze the differentially expressed proteins in Len- and mock-group cells. The top 25 differentially expressed proteins are presented in a heatmap, including 10 upregulated and 15 downregulated proteins in the Len group. (G) Pre- and post- Len treatment images of six HCC cases in (H). (H) UMAP plots for CLDN4 in the Len-resistant and Len-sensitive patients. (I) Dot plot showing the average expression of CLDN4 across the nine cell types. The dot size indicates the fraction of cells expressing CLDN4 in each cluster. (J) Western blot of CLDN4 in insensitive, sensitive, Len-group and mock-group cell lines. (K) Western blot of CLDN4 in Len-resistant and Len-sensitive patients. (L) Panoramic images of H&E staining (top) and CLDN4 immunohistochemistry (IHC) (bottom) for the Fu-LR cohort. (M) IHC staining and H score analysis of CLDN4 expression in the Fu-LR cohort. Scale bar: 500 μm. (N) Frequency of CLDN4^+ or CLDN4^− in residual tissues versus untreated tissues (top) and in residual tissues evaluated as partial response (PR) versus stable disease (bottom). Data were analyzed using the chi-square test. (O) Kaplan-Meier survival analysis of CLDN^+ or CLDN4^− HCC patients evaluated by IHC in the Fu-LR cohort. Data were analyzed using the log rank test. (P) Representative images (left) and statistical analysis (right) of CLDN4 protein (green) detected using immunofluorescence in biopsy tissues between responders and non-responders before Len treatment. Scale bar, 100 μm. Data were analyzed using Welch t test. Data in (B)–(D) and (P) are presented as the mean ± SEM of three independent experiments or triplicates. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. ns, not significant; HCC, hepatocellular carcinoma; i.g., intragastric. Elevated levels of CLDN4 are associated with HBT and lenvatinib resistance in HCC To explore the characteristics of CLDN4 in lenvatinib resistance in HCC, the Fu-LR Sc cohort dataset was assembled along with two additional published studies[76]^19^,[77]^20 ([78]Figures S1B and S1C). We conducted rigorous quality control and filtering, followed by unsupervised clustering across each dataset to distinguish malignant cells from other cell types, retaining clusters that exhibited a high expression of typical malignant cell marker genes. Subsequently, we classified 69,711 malignant cells derived from 56 patients in our scRNA-seq analysis and clustered them into 12 distinct subpopulations ([79]Figure 2A). All 12 subclusters exhibited varying levels of characteristic hepatic or biliary markers, which were determined by their distinct expression profiles in HCC versus ICC.[80]^9^,[81]^21^,[82]^22 Interestingly, we observed that CLDN4 expression highly correlates with markers of the biliary lineage, in contrast to markers of the hepatic lineage ([83]Figures 2B, 2C, and [84]S1D). To elucidate the lineage transitions and differentiation potential among malignant cell subtypes, we employed Monocle 2 and CytoTRACE algorithms to derive pseudo-temporal trajectories and differentiation levels for various malignant cell subtypes ([85]Figures 2D–2F). As CLDN4 expression increased, a gradual upregulation of biliary markers, such as EPCAM, SOX9, and CECAM1, was observed, while hepatic markers, including Apolipoprotein E, Albumin (ALB), Fibrinogen Gamma Chain, and Fibrinogen Alpha Chain showed a corresponding decline in expression, supporting the existence of biliary lineage transition in patients with HCC ([86]Figures 2G and 2H). Gene set enrichment analysis (GSEA) plots demonstrated that the gene signatures were positively related to the cholangiocarcinoma (CHOL) gene set and negatively related to the liver hepatocellular carcinoma (LIHC) gene set in patients with high CLDN4 expression ([87]Figures 2I, [88]S1E, and S1F). qPCR analysis confirmed that CLDN4 expression peaked on embryonic day 16 in mice ([89]Figure 2J), gradually decreasing from early developmental stages to mature hepatocytes ([90]Figure 2K). Furthermore, the expression of CLDN4 progressively increased from distant normal liver through peritumoral liver to HCC in patient tissues and even further in lenvatinib-resistant tumor tissues ([91]Figure 2L). Notably, CLDN4 levels were positively correlated with the pathological grade of the tumor ([92]Figure 2M). Overexpression of CLDN4 upregulated the levels of the entire panel of biliary lineage markers and inhibited the levels of hepatic lineage markers ([93]Figures S1G and S1H). In addition, the knockdown of CLDN4 markedly decreased the resistance of HCC cells to lenvatinib in vitro and in vivo ([94]Figures 2N–2P and [95]S1I). These findings indicate that CLDN4 overexpression enhances HBT in HCC, which may confer lenvatinib resistance in HCC. Figure 2. [96]Figure 2 [97]Open in a new tab Elevated levels of CLDN4 are associated with HBT and Len resistance in HCC (A) tSNE plots showing the malignant cell subtypes in HCC tissues. (B) Violin plots showing the hepatic or biliary marker genes in the malignant HCC cell subtypes. (C) tSNE plots of the hepatic or biliary marker genes in the malignant cells subtypes in HCC tissues. (D) CytoTRACE analysis of the subtypes of malignant cells. The boxplot shows the comparison of the CytoTRACE score between different malignant cell subtypes (top). tSNE plots show the CytoTRACE score, malignant cell subtypes, and expression levels of CLDN4 (bottom). (E) Trajectory of differentiation process colored by states, pseudotime, cell type and CytoTRACE score. (F) Heatmap showing upregulated and downregulated hepatic and biliary marker genes in HCC tumor differentiation. (G) tSNE plots showing pseudotime scores in the malignant cells subtypes in HCC tissues. (H) Gene trends in hepatic and biliary marker genes across the pseudotime by Monocle2. (I) GSEA data showing the enrichment of the LIHC gene set and the CHOL gene set peaks in patients with high CLDN4 expression compared to those with low CLDN4 expression in TCGA-LIHC and the Fu-LR Sc cohort. (J) Relative expression of CLDN4 in the liver of mice at different embryonic days. (K) Relative expression of CLDN4 at different developmental stages from the hepatocyte differentiation model. Data were analyzed by one-way ANOVA. (L) Relative expression of CLDN4 detected by qPCR in different regions of patients with untreated and Len-resistant tumors (LRTs) from Len-resistant cases. Data were analyzed by one-way ANOVA. (M) Relative expression of CLDN4 in patients from TCGA-LIHC with different histologic grades. Data were analyzed using one-way ANOVA. (N) Colony formation assay of Huh7 and MHCC97H cells treated with Len (5 μM) in 12-well plates for 3 weeks (n = 3). Data were analyzed using Student’s t test. Scale bar: 5 mm. (O) CCK-8 assay of Huh7 and MHCC97H cells treated with Len at the indicated concentrations for 72 h. (P) Tumor images and volume growth curve after 5 weeks of Len treatment. Subcutaneous tumors were constructed from cells with the indicated treatment. Data were analyzed using two-way repeated measures ANOVA. Data in (K), (L), (N), and (O) are presented as the mean ± SEM of three independent experiments or triplicates. Data in (P) are presented as the mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. NES, normalized enrichment score; ESC, embryonic stem cell; EN, endoderm; LP, liver progenitor cell; PH, premature hepatocyte; HEP, mature hepatocytes; DL, distant liver; PT, para tumor; T, tumor; HBT, hepatic-to-biliary lineage transition. Palmitoyl acyltransferase zDHHC5 confers CLDN4 S-palmitoylation at evolutionarily conserved cysteine residues C104 and C107 Some studies have suggested the involvement of lipid modifications in claudins, particularly in signal transduction and tight junction formation.[98]^23 For example, compared to nonpalmitoylated mutants, palmitoylated CLDN proteins are more efficiently distributed into the lipid bilayer, indicating the role of lipid modifications in TJ assembly.[99]^24 To investigate the lipid modifications of CLDN4, we extracted membrane fractions from HCC cell lines and conducted palmitoylated proteomic sequencing ([100]Figures 3A–3C). Two assays, the click chemistry reaction to biotin-azide and acyl-biotin exchange (ABE), were used to confirm palmitoylation in HCC ([101]Figures 3D, [102]S2A, and S2B). Interestingly, treatment with 2-BP, an inhibitor of palmitoylation, resulted in a significant decrease in CLDN4 expression on the plasma membrane in the two HCC cell lines ([103]Figures 3E and [104]S2C). Conversely, enhanced palmitoylation by selective inhibitors of acyl protein thioesterase (APT), palmostatin B, and ML348 increased CLDN4 expression in MHCC97H cells ([105]Figures S2D and S2E). Sequence alignment showed that the C104, C107, C182, C183, and C185 sites of CLDN4 are conserved across different species ([106]Figures 3F and 3G). We employed the motif-based predictor CSS-palm 4.0 ([107]http://csspalm.biocuckoo.org/), which suggested that the region containing C182, C183, and C185, specifically the C tail of CLDN4, has a higher likelihood of undergoing palmitoylation. However, through ABE-mass spectrometry, we confirmed that the palmitoylation sites of CLDN4 in HCC were located at C104 and C107, which reside in the transmembrane region (TR) of CLDN4 ([108]Figures 3H, 3I, and [109]S2F–S2H). This highlights the discrepancy between the palmitoylation modification sites predicted by the algorithms and actual sites. Figure 3. [110]Figure 3 [111]Open in a new tab Palmitoyl acyltransferase zDHHC5 confers CLDN4 S-palmitoylation at evolutionarily conserved cysteine residues C104 and C107 (A) MHCC97H cells were collected, and plasma membrane (PM) samples were prepared. The samples were then subjected to immunoprecipitation using high-capacity neutravidin beads, followed by mass spectrometry of the precipitated constituents. (B) Data obtained in mass spectrometry (A) were further analyzed for numbers of palmitoylated sites, peptides, and proteins. (C) Relative intensity of the indicated amino acid residues flanking the sites of palmitoylated cysteine. (D) MHCC97H cells were cultured in medium containing 100 μM palmitic acid-azide for 6 h, and the cell lysates were prepared for the Click-iT assays (top). MHCC97H cells were transfected with FLAG-CLDN4 and then subjected to ABE palmitoylation assays (bottom). (E) MHCC97H cells were treated with 2-BP (50 μM) or DMSO for 24 h. Then, cells were fixed and immunostained for CLDN4 (red). Scale bars, 25 μm. (F) Transmembrane topology of CLDN4. (G) The C104, C107, C182, C183, and C185 residues of the CLDN4 gene are conserved in different species. (H) CLDN4 was identified to be palmitoylated at C104 and C107 by liquid chromatography-mass spectrometry (LC-MS) in MHCC97H cells. (I) HEK293 cells were transfected with FLAG-CLDN4-C104A, C107, C182A, C183A, or C185A and then subjected to an ABE palmitoylation assay. (J) Molecular docking was performed for CLDN4 and zDHHC5 based on AlphaFold3, with special annotations on the palmitoyl acids at corresponding positions (top left). Shown are a schematic illustrating the palmitoylation modification of CLDN4 by zDHHC5, where the characteristic DHHC sequence of zDHHC5 is located at amino acid positions 131–134 (bottom left) and a schematic of dynamic palmitoylation modification of CLDN4 (right). (K) zDHHC5^C134A reduces palmitoylation of CLDN4. The palmitoylation of CLDN4 was detected by ABE assay in MHCC97H cells expressing FLAG-tagged CLDN4 and HA-tagged WT or mutant zDHHC5 as indicated. (L) Palmitoylation of CLDN4 is downregulated in zDHHC5 knockdown MHCC97H cells. FLAG-tagged CLDN4 was expressed in zDHHC5-depleted MHCC97H cells, followed by ABE assay. (M) Bacterially purified FLAG-CLDN4 protein, glutathione S-transferase (GST), and GST-zDHHC5 proteins were subjected to a GST pull-down assay. (N) MHCC97H cells were transfected with HA-zDHHC5 and FLAG-CLDN4-WT, TR, C-tail, ΔN-tail, or ΔC-tail and then subjected to immunoprecipitation (IP) analysis. (O) Pearson correlation of CLDN4 expression levels and zDHHC5 protein levels by IHC of HCC patients from the Fu-LR cohort. (P) Kaplan-Meier survival analysis of HCC patients from the Fu-LR cohort with low or high zDHHC5 expression. Data were analyzed by log rank test. (Q) Co-localization between zDHHC5 and CLDN4 is shown by immunofluorescence (IF) of CLDN4 (red) and zDHHC5 (green) in HCC tissue sections. Scale bar: 50 μm. WCL, whole-cell lysates; HAM, hydroxylamine; EV, empty vector; WT, wild type; TR, transmembrane region. Protein palmitoylation is typically mediated by specific palmitoyltransferases, with selectivity determined by the enzyme’s substrate specificity, subcellular localization, regulatory signals, and structural features of the substrate. For instance, palmitoyltransferases localized in different cellular compartments only modify substrates with which they physically interact. In HCC, elevated expression of zDHHC5, zDHHC7, zDHHC9, zDHHC16, zDHHC17, zDHHC18, zDHHC20, and zDHHC21 was correlated with poor prognosis.[112]^25 To identify the predominant palmitoyltransferases for CLDN4 in MHCC97H cells, hemagglutinin (HA)-tagged zinc-finger DHHC domain-containing proteins (zDHHCs) were individually co-expressed with FLAG-tagged CLDN4 in HEK293 cells. Immunoprecipitation combined with immunoblotting demonstrated that zDHHC1, zDHHC5, zDHHC20, zDHHC21, and zDHHC24 could bind to CLDN4 ([113]Figure S3A). Subsequently, we observed that the knockdown of zDHHC5 led to a decrease in endogenous CLDN4 palmitoylation ([114]Figure S3B). Therefore, we hypothesized that zDHHC5 is involved in the dynamic palmitoylation of CLDN4 and modeled their molecular docking using AlphaFold3[115]^26 ([116]Figure 3J). ABE assay results demonstrated that zDHHC5 facilitated CLDN4 palmitoylation at residues C104 and C107 ([117]Figures 3K–3N and S3C–S3E). High zDHHC5 expression was correlated with high CLDN4 expression and poor prognosis of HCC cases within the Fu-LR cohort ([118]Figures 3O, 3P, [119]S3F, and S3G). zDHHC5 levels were also significantly upregulated in lenvatinib-resistant patients and cell lines ([120]Figures 3Q, [121]S3H, and S3I). After zDHHC5 knockdown, the resistance of HCC cells to lenvatinib significantly decreased in vitro and vice versa ([122]Figures S4A–S4H). These results demonstrate that zDHHC5 facilitates CLDN4 S-palmitoylation at evolutionarily conserved residues C104 and C107. LYPLA2 depalmitoylates CLDN4 To identify the potential physiological APT for CLDN4, we co-expressed Myc-tagged Abhydrolase Domain-Containing (ABHD), LYPLA1, LYPLA2, PPT1, and PPT2 proteins with FLAG-tagged CLDN4 in HEK293 cells. We found that exogenous CLDN4 readily co-immunoprecipitated with LYPLA1, LYPLA2, and ABHD17A ([123]Figure S5A). Further confirmation assays showed that zDHHC5-mediated CLDN4 palmitoylation could be decreased more effectively by LYPLA2 but not by LYPLA1 or ABHD17A ([124]Figures S5B and S5C). Immunofluorescence staining revealed the apparent co-localization of LYPLA2 and CLDN4 in MHCC97H cells ([125]Figure S5D). As expected, knockdown of LYPLA2 in HEK293 and MHCC97H cells also increased CLDN4 palmitoylation ([126]Figures S5E and S5F). In MHCC97H co-transfected with C104 and C107 mutations, Myc-tagged LYPLA2 further reduced the CLDN4 palmitoylation levels ([127]Figure S5G). These results demonstrated that LYPLA2 depalmitoylates CLDN4. CLDN4 palmitoylation is required for lipid-raft localization CLDNs are generally localized within functional membrane domains, such as lipid rafts, where they perform their roles.[128]^27^,[129]^28 We investigated whether palmitoyl conjugation could promote CLDN4 lipid-raft anchoring. In MHCC97 cells treated with 2-BP or that underwent zDHHC5 knockdown, we observed a significant decrease in the levels of CLDN4 present on lipid rafts, which could be rescued by the endocytosis inhibitor Dynasore and lysosomal inhibitor chloroquine (CQ) rather than the proteasomal inhibitor MG132 ([130]Figures 4A and 4B). We conducted mass spectrometry analysis of the immunoprecipitates of CLDN4, and the results suggested interactions between CLDN4 and both clathrin interactor 1 (CLINT1) and lysosomal-associated membrane protein 2 (LAMP2) ([131]Figure S6A). Therefore, we hypothesized that CLDN4 undergoes clathrin-mediated endocytosis (CME) and is transported to lysosomes for degradation. The results validated the interactions of CLDN4 with Clathrin Heavy Chain (CLTC) and LAMP2 ([132]Figure S6B). Intriguingly, we discovered a direct interaction between CLDN4 and CLINT1 ([133]Figures 4C–4E, [134]S6C, and S6D). As expected, knockdown of CLINT1 efficiently rescued the degradation of CLDN4 induced by depalmitoylation ([135]Figure 4F). In contrast, the proteasome inhibitor MG132 did not slow down the degradation rate of CLDN4 protein ([136]Figure S6E). The immunofluorescence results suggested that the co-localization of CLDN4 and CLTC or LAMP2 significantly increased in cells treated with 2-BP and shzDHHC5, while it decreased markedly in cells treated with shCLINT1 ([137]Figures 4G and 4H). However, we have yet to explain why palmitoylation of CLDN4 can influence the CME process, which has been reported to be affected by dynamic ubiquitination and deubiquitination.[138]^29 Additionally, palmitoyl groups have been reported to block ubiquitin binding, suggesting potential crosstalk and competition between these two modifications at the molecular level.[139]^30 The results indicated that inhibiting palmitoylation significantly increased the interaction between CLDN4 and ubiquitin or CLINT1 ([140]Figure 4I). To reveal the E3 ubiquitin ligase responsible for the ubiquitination of CLDN4, we investigated the association between CLDN4 and tripartite motif-containing 33 (TRIM33), which was screened from the immunoprecipitates of CLDN4. Notably, TRIM33 knockdown increased CLDN4 protein levels and decreased the accumulation of ubiquitinated CLDN4 in MHCC97H ([141]Figures 4J–4M), while MG132 did not affect the ubiquitin levels bound to CLDN4 ([142]Figure S6F). Moreover, as seen by combining the prediction results from BDM-PUB (computational prediction of protein ubiquitination sites with a Bayesian discriminant method, [143]http://bdmpub.biocuckoo.org/results.php), the K103 site may be a potential monoubiquitination site of CLDN4 influenced by palmitoylation. K48-linked ubiquitination typically mediates degradation via the proteasomal pathway, whereas K63-linked ubiquitination mediates processes such as endocytosis and DNA damage repair.[144]^31 We performed simultaneous transfection of HA-tagged K63 and FLAG-tagged K103 mutant CLDN4 into MHCC97 cells. The result suggested that ubiquitination at the K103 site of CLDN4 was mediated by a K63-linked ubiquitin chain ([145]Figures 4N and 4O). Taken together, the palmitoylation occurring at cysteine residues C104 and C107 of CLDN4 promotes the lipid-raft anchoring of CLDN4 ([146]Figure 4P). Figure 4. [147]Figure 4 [148]Open in a new tab CLDN4 palmitoylation is required for lipid-raft localization (A) MHCC97H cells were pretreated with 2-BP or shzDHHC5, and then lipid rafts were extracted. The levels of CLDN4 were detected by immunoblotting at the indicated time points. (B) MHCC97H cells were pretreated with control, 2-BP, Amil + 2-BP, MG132 + 2-BP, Dynasore + 2-BP, and CQ + 2-BP and then incubated with cycloheximide. The levels of CLDN4 in lipid rafts were detected by immunoblotting (left) and quantified (right) at the indicated time points. Data were analyzed using Student’s t test. (C) Reciprocal coIP of CLDN4 and CLINT1 showing the physical interaction between endogenously expressed CLDN4 and CLINT1 in MHCC97H cells. (D) Two-step coIP (TIP) of CLDN4 and CLINT1-CLTC, showing the physical interaction between endogenously expressed CLDN4 and CLINT1-CLTC in MHCC97H cells. (E) Bacterially purified FLAG-CLDN4, GST, and GST-CLINT1 proteins were subjected to GST pull-down assay. (F) Western blot showing the involvement of CLINT1 in depalmitoylation-induced degradation of CLDN4. The process of clathrin-mediated endocytosis (CME) was disturbed by knockdown of CLINT1. (G and H) Representative images and quantification of IF staining for CLDN4 (red) and CLTC or LAMP2 (green) co-localization in MHCC97H from control, 2-BP, shzDHHC5, shCLINT1, Dynasore, or CQ treated group. n = 5; scale bar: 25 μm. (I) IP assay in MHCC97H cells showing the effects of de-palmitoylation by 2-BP treatment or knocking down zDHHC5 on CLDN4 ubiquitination and the interaction between CLDN4 and CLINT1. (J and K) Western blot analysis of CLDN4 (J) and ubiquitination levels of CLDN4 (K) in control and TRIM33 knockdown cells. (L and M) Western blot analysis of CLDN4 (L) and ubiquitination levels of CLDN4 (M) in TRIM33 knockdown cells transfected with WT or ligase-dead (LD) TRIM33. (N and O) HEK293 cells were transfected with the K103R plasmid (N) to measure K63-linked (O) ubiquitination of CLDN4. (P) Palmitoylation by zDHHC5 blocks CLDN4 ubiquitination and prevents its interaction with CLINT1 and internalization to the clathrin and lysosome, thereby promoting the lipid raft anchoring of CLDN4. Data in (B) are presented as the mean ± SEM of three independent experiments. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. CQ, chloroquine; ECM, extracellular matrix. CLDN4 alters tumor cell plasticity by activating the Notch pathway Synthesis of peptides that competitively bind to the palmitoylation region of a target molecule is a strategy for developing specific inhibitors.[149]^30^,[150]^32 We then synthesized a competitive peptide, CPP-S4, against CLDN4 palmitoylation, which effectively inhibits the membrane stabilization of CLDN4 and enhances the cytotoxic effect of lenvatinib ([151]Figures 5A–5E and [152]S7A–S7F). MHCC97H cells were treated with CPP-S4 peptides, and HBT genes were reduced to varying degrees ([153]Figures 5F, 5G, [154]S7G, and S7H). The transcriptional profile after treatment also indicated a shift from lenvatinib-resistant characteristics to lenvatinib-sensitive transcriptional features ([155]Figure S7I). Notably, pathway enrichment analysis revealed that the Notch-related signaling pathway was the most significantly inhibited, confirmed by immunoblot analysis and qPCR ([156]Figures 5H–5I and [157]S7J). Importantly, the increased levels of HBT markers resulting from CLDN4 overexpression were significantly reduced following treatment with the γ-secretase inhibitor RO4929097 ([158]Figures 5J and 5K). We then determined whether the NICD (Notch intracellular domain)-RBPJ (Recombination signal Binding Protein for immunoglobulin kappa J region) binding sites were located in the promoter regions of the HBT markers ([159]Figures 5L and 5M). Chromatin immunoprecipitation (ChIP)-PCR and ChIP assays were performed, and significant recruitment of NICD-RBPJ to the promoter region of HBT markers was observed ([160]Figures 5N and 5O). Probes with wild-type and mutant NICD-RBPJ sites were designed for DNA affinity precipitation to identify the regulatory elements in the HBT gene promoter. The results demonstrated that NICD-RBPJ bound to oligonucleotides spanning the wild-type but not the mutant promoter regions of HBT markers ([161]Figure 5P). The dual-luciferase reporter assay further revealed that NICD-RBPJ was required for the transcriptional activity of the promoter of HBT markers and for HBT gene expression ([162]Figure 5Q). Collectively, these results indicated that CLDN4 induced the expression of HBT markers via the Notch signaling pathway. Figure 5. [163]Figure 5 [164]Open in a new tab “Anchoring” CLDN4 through palmitoylation alters tumor cell plasticity by interfering with Notch signaling (A) Schematic of the fusion motif GFP-S1, including GFP and CLDN4 (99–112), and GFP-S2, including GFP and CLDN4 (104A and 107A). (B) MHCC97H cells were transfected with either GFP-S1 or GFP-S2 plasmids and examined using confocal microscopy. Scale bars, 25 μm. (C) ABE assay detection of palmitoylation of GFP-S1 and GFP-S2 expressed in MHCC97H cells. (D) MHCC97H cells were transfected with GFP-S1 for 48 h and incubated with the selective lysosomal inhibitors CQ, MG132, Dynasore, or Amil. Then, the cells were analyzed for CLDN4 expression. (E) Schematic of CPP-S1, CPP-S2, CPP-S3, and CPP-S4 peptides. The different residues are shown in red. (F) The transcriptome analysis of MHCC97H cells treated with CPP-S4 and mock. (G) GSEA of Hallmark gene sets in CPP-S4-treated MHCC97H cells. Heatmaps show NES of significantly (false discovery rate < 0.05, Kolmogorov-Smirnov test) altered gene sets. (H) Western blot showing the effects of CLDN4 on the Notch pathway in the indicated cells. (I) qPCR analyses of the transcriptional levels of downstream target genes of the Notch pathway in the indicated cells. Data were analyzed by one-way ANOVA. (J) qPCR analyses of the transcriptional levels of HBT genes in the indicated cells. Data were analyzed by one-way ANOVA. (K) Western blot showing the effects of RO4929097, a specific inhibitor of the Notch pathway, in the indicated cells. (L) Schematic of putative NICD-binding sites in the promoter region of the biliary marker gene. (M) The sequence logo and frequency matrix of RBPJ (NICD) obtained from the JASPAR website. (N) Binding intensity of NICD to the biliary marker gene promoter relative to the IgG control by ChIP-qPCR analysis. Data were analyzed using Student’s t test. (O) ChIP analysis was conducted to identify NICD-binding sites within the promoter region of the biliary marker gene in MHCC97H cells. GAPDH was used as the negative control. (P) Exogenous NICD and Lamin B1 (control) protein pulled down by biotin-labeled oligonucleotides containing the WT or mutant HBT gene promoter. (Q) Relative luciferase activity of reporters containing the WT or mutated sequence of the biliary marker gene promoter. Data were analyzed using Student’s t test. Data in (I), (J), (N), and (Q) are presented as the mean ± SEM of three independent experiments or triplicates. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. CLDN4 drives CNTN1 to lipid rafts, fueling Notch activation Using cell membrane co-immunoprecipitation studies in MHCC97H cells, we observed the interaction of CLDN4 with CNTN1 rather than with JAG1 ([165]Figures 6A and 6B), both of which have been reported to activate Notch signaling.[166]^33^,[167]^34 Co-expression of immunofluorescence and data from TCGA-LIHC suggested consistency in the expression of CNTN1 and CLDN4 ([168]Figures 6C and 6D). Our next question was how CLDN4 binding promotes CNTN1 activity. CNTN1 is a neural cell adhesion protein consisting of six N-terminal immunoglobulin (Ig) domains followed by four fibronectin-like repeats.[169]^35 Fibronectin type III-like and Ig-like C2 type 1 mutants fused to a FLAG tag for CNTN1 were generated ([170]Figure 6E). Fibronectin type III of CNTN1 was found to be involved in the CLDN4-CNTN1 interaction ([171]Figure 6F). Owing to its connection with glycosylphosphatidylinositol (GPI) on the plasma membrane, CNTN1 is almost exclusively localized within lipid rafts[172]^36 ([173]Figure 6G). Therefore, we hypothesized that palmitoylated CLDN4 recruits CNTN1 to lipid rafts, thereby activating the Notch pathway. The results indicate that CLDN4 did not affect the overall expression level of CNTN1 but significantly affected its distribution within lipid rafts ([174]Figures 6H and 6I). Methyl-β-cyclodextrin (MβCD), an inhibitor of lipid rafts, significantly reduced the co-localization of CLDN4 and CNTN1 on the membrane and inhibited the Notch signaling and HBT ([175]Figures 6J–6N). CNTN1 downregulation or MβCD supplementation diminished the changes in the IC[50] (half maximal inhibitory concentration) values of lenvatinib observed in HCC cells overexpressing CLDN4 and vice versa ([176]Figure 6O). Subcutaneous tumor experiments also showed the effect of MβCD or RO4929097 on the antitumor effect of lenvatinib ([177]Figure 6P). These findings emphasize that palmitoylated CLDN4 drives CNTN1 to lipid rafts and amplifies the oncogenic pathway of Notch. Figure 6. [178]Figure 6 [179]Open in a new tab CLDN4 drives CNTN1 to lipid rafts, fueling Notch activation (A) Peptide information of CNTN1 in the LC-MS/MS of CLDN4 in MHCC97H cells. (B) IP of CLDN4 and JAG1 or CNTN1 in MHCC97H and PLC/PRF/5 cells. (C) Endogenous CNTN1 (green) and CLDN4 (red) were stained by immunofluorescence in MHCC97H cells. Scale bar: 25 μm. (D) Pearson correlation of CLDN4 expression levels and CNTN1 expression levels in HCC from TCGA-LIHC patients. Data were obtained from the Timer database ([180]http://timer.cistrome.org/). (E) Schematic of the CNTN1 structure and its truncated forms. (F) HEK293 cells were transfected with HA-CNTN1-WT, IgΔ, GPIΔ, FNΔ or Ig and FLAG-CLDN4 and then subjected to IP analysis. (G) The levels of CLDN4 and CNTN1 were detected by immunoblotting in WCLs and lipid rafts. (H) The levels of CNTN1 were detected by immunoblotting in WCLs with the indicated treatment. (I) Immunoblots representing CLDN4, CNTN1, and CAV1 of the 10 fractions of density gradient separation from CLDN4-EV (left) and CLDN4-KD (right) MHCC97H cells. The amount of CNTN1 in the lipid raft isolates from CLDN4-EV was considerably higher when compared with CLDN4 knockdown (KD). (J) Image of co-localization between CNTN1 and CLDN4 in MHCC97H cells with MβCD treatment. Scale bar: 25 μm. (K) Western blot showing the effects of different expression of CNTN1 and CLDN4 on the Notch pathway in the MHCC97H and Huh7 cell lines with MβCD treatment. (L) qPCR analyses of the transcriptional levels of downstream target genes of the Notch pathway in the indicated cells. Data were analyzed by one-way ANOVA. (M and N) Western blot (M) and qPCR analyses (N) showing the effects of different expression of CNTN1 and CLDN4 on HBT genes in the Huh7 and MHCC97H cell lines with MβCD treatment. Data were analyzed using one-way ANOVA. (O) CCK8 and colony formation assays of MHCC97H and Huh7 cell lines with indicated treatment. Data were analyzed using one-way ANOVA. (P) Tumor images, volume growth curve, and tumor weight after 5 weeks of Len treatment. Subcutaneous tumors were constructed from cells with the indicated treatment. Data were analyzed by two-way repeated measures ANOVA. Data in (L), (N), and )O) are presented as the mean ± SEM of three independent experiments or triplicates. Data in (P) are presented as the mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. Discovery of a CLDN4 antagonist by virtual screening We next attempted to find a clinically relevant method for inhibiting CLDN4. First, we came up with the idea of using a monoclonal antibody targeting CLDN4 or an antibody-drug conjugate approach to deliver a cytotoxic drug to CLDN4^+ cells, which has already been applied to other claudins.[181]^10^,[182]^37 However, these two methods are not feasible for targeting lenvatinib-resistant HCC because the expression level of CLDN4 in normal epithelial tissues, such as the esophagus and gastric mucosa, is very high. This implies that the direct targeting of CLDN4 for killing would result in significant gastrointestinal side effects. Targeting the palmitoylation modification of CLDN4 in HCC may be an effective strategy. We collected 17,681 small-molecule drugs (Topscience, T001) and performed molecular docking to the 3D model of the CLDN4-zDHHC5 binding domain. After virtual screening and toxicity evaluation, we identified SalB as an effective and safe CLDN4 antagonist ([183]Figures 7A, 7B, and [184]S8A–S8J). To further explore the effect of SalB on HCC pathogenesis, we established a spontaneous HCC model in Zdhhc5^−/− mice ([185]Figure S9A). Compared with their wild-type (WT) counterparts, Zdhhc5^−/− mice exhibited slower progression of spontaneous tumors, as evidenced by decreased liver/body weight, and less liver injury, as suggested by decreased Alanine Transaminase (ALT) and Aspartate Transaminase (AST) values ([186]Figure S9B). The expression level of CLDN4 in the HCCs of Zdhhc5^−/− mice was significantly decreased, coupled with decreased CLDN4 palmitoylation ([187]Figures S9C and S9D). Accordingly, SalB reduced tumor growth in WT mice without significant liver toxicity ([188]Figures S9E–S9H). These results suggest that SalB is an effective CLDN4 antagonist. Figure 7. [189]Figure 7 [190]Open in a new tab CLDN4-induced HBT is associated with poor survival in HCC (A) Binding of the zDHHC5-CLDN4 binding domain and the top 3 compounds with the best efficacy: magnesium lithospermate B (left), salvianolic acid B (SalB) (middle), and resveratrol 4′-(6-galloylglucoside) (right). (B) Binding of CLDN4 and the zDHHC5 substrate domain (left) and of CLDN4 and the zDHHC5 substrate domain in the presence of SalB (right). (C) Molecular docking to the 3D model of the SalB-zDHHC5 binding domain. (D) Schematic showing establishment of orthotopic HCC mouse models. After confirming the successful establishment of orthotopic HCC, the mice were administered Len or mock at 20 mg/kg/day via oral gavage for 20 days. Following the treatment, the mice were euthanized, and orthotopic HCC was collected. The unit of the signal was Ps^−1 cm^−2 sr^−1. (E) Liver/body weight of orthotopic HCC tumors with the indicated treatment (n = 5). Data were analyzed by Student’s t test. (F) Schematic showing the establishment of subcutaneous PDX models in NSG mice (left) and tumor images (right) with the indicated treatment. (G) The volume growth curve of subcutaneous PDX tumors in (D). Data were analyzed by two-way repeated measures ANOVA. (H) The weight of NSG mice from the control, SalB, Len, and Len and SalB combined treatment group. Data were analyzed by one-way ANOVA. (I) Panoramic scan images of multiple immunofluorescence of CLDN4, zDHHC5, KRT7, EPCAM, ALB, and Transthyretin (TTR) for the Fu-LR cohort. (J) Representative IF staining of CLDN4, zDHHC5, KRT7, EPCAM, ALB, and TTR in the Fu-LR cohort. In the process of tissue cytometry analysis, each cell in the microarray corresponds to a point in the flow cytometry plot. Scale bar: 500 μm. (K) The proportions of CLDN4^+ or CLDN4^− cells as well as the proportions of zDHHC5^+ or zDHHC5^− cells in HBT tissues. Data were analyzed by Mann-Whitney U test. (L) Frequency of biliary lineage (BL)-HCC or hepatic lineage (HL)-HCC in residual tissues versus untreated tissues of the Fu-LR cohort. Data were analyzed by chi-square test. (M) The HL scores and corresponding BL scores for each spot in the microarray were subjected to logarithmic transformation for data processing. (N) Kaplan-Meier survival analysis of BL-HCC or HL-HCC of the Fu-LR cohort after resection. Data were analyzed by log rank test. (O) Anchoring CLDN4 by palmitoylation induces HBT via the Notch signaling pathway, contributing to Len resistance and poor prognosis of HCC patients. A specific CLDN4 inhibitor, SalB, may achieve significant benefits for Len therapy of CLDN4^+ HBT patients. Data in (K) are presented as the mean ± SEM of three independent experiments. Data in (E), (G), and (H) are presented as the mean ± SD. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001. i.p., intraperitoneal injection; i.g., CDX, cell-derived xenograft; PDX, patient-derived xenograft. CLDN4-induced HBT is associated with poor survival in HCC To extend our findings in in vivo models, we established orthotopic models by injecting Hepa1-6 cells into the liver of C57BL/6 mice. Tumors became detectable after 10 days, and lenvatinib, SalB, or vehicle was administered for 20 days. Significantly smaller tumors were observed in the combo treatment group ([191]Figures 7C–7E). CLDN4^+ LR-PDXs were constructed, and SalB significantly enhanced the antitumor effect of lenvatinib in LR-PDXs, with no significant toxicity observed ([192]Figures 7F–7H). Since HCC patients, unlike ICC patients, typically do not undergo chemotherapy, there is no clear consensus on whether patients with mixed-type liver cancer should receive chemotherapy. We investigated whether HCC patients with HBT would benefit from chemotherapy. The results suggested that the combination of gemcitabine/cisplatin and lenvatinib achieved the best therapeutic effect for CLDN4^+KRT7^+EPCAM^+ PDXs, although this was accompanied by a significant decrease in mouse body weight ([193]Figures S10A–S10D). Notably, SalB administration partially reversed HBT, with treated subjects showing markedly lower HBT marker expression versus controls ([194]Figure S10E). For CLDN4^−KRT7^−EPCAM^− PDXs, combined chemotherapy did not yield significant benefits ([195]Figures S10F–S10J). To verify the relationship between HBT and lenvatinib resistance, we performed multiplexed immunohistochemistry (mIHC) on the Fu-LR cohort and conducted tissue flow cytometry-like analysis, enabling quantitative fluorescence analysis of each single cell in each spot of the tumor microarray.[196]^38 The prevalence of biliary-lineage HCC was significantly higher in the residual cohort and vice versa ([197]Figures 7I–7L, [198]S11A, and S11B). Additionally, the expression patterns of hepatic-lineage markers and biliary-lineage markers were diametrically opposed, suggesting that biliary characteristics and hepatic characteristics in HCC are mutually constrained and undergo transitions between each other ([199]Figure 7M). Compared with HCC cases without HBT, those with HBT exhibited significantly shorter overall survival and recurrence-free survival ([200]Figure 7N). In summary, HBT is associated with lenvatinib resistance and poor survival in HCCs while also showing potential benefits from chemotherapy ([201]Figure 7O). Discussion It is common for tumor cells to undergo lineage transitions to enhance their defense against drug-induced cytotoxicity, underscoring the significance of identifying potent lineage transition promoters to improve drug resistance. In this study, we provide conceptual validation that CLDN4 is not only a tumor-associated antigen in HCC but also a functional regulator of tumor lineage plasticity. Overexpression of CLDN4 leads to changes in cellular identity, conferring resistance to lenvatinib. We identified that the CNTN1-Notch pathway located on CLDN4 lipid rafts is a powerful driver of HBT in HCC. The results also emphasized the role of palmitoylated CLDN4 in bolstering the oncogenic signaling of CNTN1, thereby activating the Notch pathway. These events are crucial for triggering transdifferentiation of HCC. The Notch signaling pathway has been implicated in controlling hepatocyte fate and phenotypic plasticity in mature hepatocytes.[202]^39 Poorly differentiated advanced HCC often exhibits overexpression of the Notch pathway and features phenotypes similar to the biliary lineage.[203]^40^,[204]^41 Inhibition of the Notch signaling pathway can restore hepatocyte differentiation in advanced HCC, leading to tumor regression. However, considering the role of Notch signaling in normal development, the potential cytotoxicity of Notch inhibitors should be carefully considered before entering clinical trials. Direct targeting of CLDN4-expressing tumor cells may be an alternative to systemic Notch inhibition. Nevertheless, non-specific expression of CLDN4 presents challenges for direct targeting. Based on the role of palmitoylation modification in membrane stability and signal transduction of CLDN4, our synthesized palmitoylation-competitive peptides and a small-molecule inhibitor targeting the interaction domain between CLDN4 and its palmitoylation enzyme zDHHC5 markedly enhanced the efficacy of lenvatinib therapy. CLDN4 is highly expressed in some digestive tracts and urothelial epithelia, making targeting CLDN4 inherently non-selective. However, SalB targets the CLDN4-modifying enzyme, which is highly expressed in the liver, ensuring that CLDN4 inhibition does not cause obvious organ damage, hypothermia, abnormal blood parameters, or weight loss in mice. This suggests that patients exhibit good tolerance when treated with an appropriate dose of SalB. Our data from HCC cell lines and PDX tumors strongly indicated the antitumor efficacy of combining SalB with lenvatinib for the treatment of HCC. Considering that CLDN4 is only expressed in a subset of patients, we recommend conducting preliminary clinical trials of combination therapy with SalB for CLDN4^+ HCC patients who develop resistance to lenvatinib after treatment. It is crucial to screen patients with HBT and tailor treatment regimens accordingly, as the distinct characteristics of hepatic and biliary lineages often necessitate different therapeutic approaches and have various outcomes. Therefore, reversing HBT may be an effective strategy to overcome lenvatinib resistance. A reasonable extension of this strategy holds great potential for lineage shifting toward HCC in combined or mixed HCC-ICC and even in separate cholangiocarcinoma. Unlike ICC, HCC is typically not managed with chemotherapy.[205]^42 Our findings indicate that chemotherapy may also provide substantial benefits for HCC with HBT and for combined or mixed HCC-ICC, the latter of which previously lacked a definitive consensus regarding use of chemotherapy. Nevertheless, the application of combined chemotherapy necessitates a cautious approach that carefully weighs the expected benefits against inherent risks. Identifying patients with HCC and HBT presents significant challenges. Thus, we propose employing biopsy, imaging omics, peripheral blood markers, other accessible features, and machine learning as streamlined approaches to evaluate HBT with the objective of achieving precision treatment in HCC. In summary, our study highlights the role of CLDN4 in promoting lineage plasticity and drug resistance in HCC, specifically by facilitating HBT. By targeting palmitoylation of CLDN4 and developing tools to identify HBT in HCC patients, we propose actionable strategies to overcome lenvatinib resistance and enhance precision medicine approaches in HCC. Future studies should focus on elucidating the molecular mechanisms underlying HBT and validating the efficacy of our proposed strategies in clinical settings to improve personalized treatment for patients with advanced HCC. Limitations of the study In this context, although we have identified the crucial role of CLDN4 in HBT, we believe that the transition between HCC and ICC still requires the involvement of other epigenetic regulatory mechanisms. The tight junction components formed by CLDN4 to some extent reshape the arrangement and signaling between HCC cells. However, the intricate signaling transduction within cells are subject to precise, flexible, and dynamic regulation by the microenvironment, which includes factors such as cytokines, growth factors, and the extracellular matrix. Moreover, the broad-spectrum pharmacological activity of SalB may stem from its regulation of multiple molecular targets. Although this study identifies CLDN4-zDHHC5 as the core pathway in CLDN4-associated lenvatinib resistance, other mechanisms (e.g., nuclear factor κB pathway inhibition by SalB) may synergistically contribute to the overall effects. Future studies should employ conditional gene knockout or pharmacological tools to further dissect the relative contributions of different targets. Resource availability Lead contact Requests for further information or reagents should be directed to the lead contact, Yinghao Shen (syh12268@163.com). Materials availability All reagents generated in this study are available upon request from the [206]lead contact with a completed materials transfer agreement. Data and code availability * • The raw single-cell sequencing data generated by this study have been deposited in the China National GeneBank (CNGB) Sequence Archive (CNSA: CNP0006650). * • This paper does not report original code. * • Any additional information required to reanalyze the data reported in this paper is available from the [207]lead contact upon request. Acknowledgments