A human patient-derived organoid biobank to model tumor heterogeneity and therapeutic vulnerability for oral squamous cell carcinoma

Summary

Oral squamous cell carcinoma (OSCC) remains a significant clinical challenge due to frequent recurrence, metastasis, and therapeutic resistance. Here, we establish a living biobank of OSCC patient-derived organoids (PDOs) comprising 46 lines using optimized culture medium. These PDOs are long-term passaged, cryopreserved, and recovered with stable viability and tumorigenicity. Comprehensive morphological, genomic, and transcriptomic analyses confirm that PDOs faithfully recapitulate the histopathological, genetic, and molecular features of parental tumors. These PDOs enable disease modeling, genetic manipulation, and drug screening. Through transcriptomic profiling and functional assays, we find that CDCP1 mediates cisplatin resistance by modulating Wnt/β-catenin signaling-driven stemness. Notably, we develop a pH-sensitive nanoparticle delivering siCDCP1, which effectively restores chemosensitivity and impairs tumor growth in cisplatin-resistant patient-derived xenograft (PDX) models with favorable safety profile. These findings establish PDOs as robust preclinical models for mechanistic explorations and therapeutics development and highlight CDCP1-targeting strategies as promising approaches to overcome cisplatin resistance in OSCC.

Graphical abstract

Highlights

• •A living OSCC organoids biobank is established from clinical samples and maintained
• •OSCC organoids enable effective chemotherapeutic drug screening
• •CDCP1 drives cisplatin resistance in OSCC via Wnt/β-catenin-mediated stemness
• •Nanoparticle-mediated silencing of CDCP1 restores chemosensitivity in OSCC

Wang et al. establish a living biobank of OSCC organoids that retain original tumor traits for personalized drug screening and identify CDCP1 as a key driver of cisplatin resistance. Therapeutic targeting of CDCP1 with a pH-sensitive nanoparticle restores cisplatin chemosensitivity and impairs tumor growth in cisplatin-resistant PDXs.

Introduction

Oral squamous cell carcinoma (OSCC) arising from malignant transformation of epithelial cells in oral mucosa accounts for the majority of head and neck squamous cell carcinoma (HNSCC).¹ Despite substantial advances in surgical techniques, radiotherapy, and chemotherapy, frequent local recurrence, high cervical node metastasis, and notorious therapeutic resistance remain unresolved clinical challenges, thereby leading to poor treatment outcomes and high mortality.^2,3 Seminal works have established that HNSCC comprises four transcriptomic subtypes (atypical, basal, classical, and mesenchymal) or three integrated multi-omics subtypes (CIN, basal, and immune), each harboring distinct genomic and molecular characteristics and clinical outcomes.^4,5 Prominent intra- and inter-tumoral heterogeneity significantly contribute to the limited values of routine prognostic biomarkers like tumor-node-metastasis (TNM) staging system and inconsistent efficiencies of common therapeutic agents among patients with this malignancy.^6,7 Thus, developing effective strategies for patient stratification, treatment guidance, and response prediction as well as identifying potential therapeutic targets is urgently needed to improve OSCC clinical management and outcomes.

Traditional preclinical models, such as cell lines and cell-derived xenografts (CDXs), have greatly advanced biological understanding of tumorigenesis and facilitated precision oncology; however, they have inherent limitations in capturing the complexity and heterogeneity of human tumors.⁸ Novel, reliable, clinically convenient preclinical models that accurately and better reflect the biological characteristics and tumorigenicity of OSCC might be the keystone to achieve individualized treatment planning and improved survival.⁹ Patient-derived organoids (PDOs) have emerged as promising translational tools in cancer research, offering a three-dimensional (3D) culture system that closely mimics the in vivo parental tumor characteristics.¹⁰ Accumulating evidence has established that these organoids can be derived from clinical samples and retain the genetic and phenotypic features of original tumor cells. Recent studies have demonstrated the potentials of PDOs in modeling various cancers, including colorectal, lung, and breast cancer for drug screening and biomarker discovery.^11,12,13 Moreover, PDOs have advantages in cost, labor, and time requirements, scalability, convenient genetic manipulation, and high-throughput drug screening over patient-derived xenograft (PDX) models.¹⁰ In 2019, Hans Clevers group reported their pioneering results showing that OSCC organoids were successfully established and genetically characterized, with potentials to model personalized patient response to chemotherapy and radiotherapy.¹⁴ These studies highlight the translational promise of cancer organoids to stratify patients who may benefit from specific treatments and to uncover novel therapeutic vulnerabilities.

Platinum-based drugs (cisplatin, carboplatin, oxaliplatin, etc.) are commonly used for chemotherapeutic eradication of OSCC, especially in combination with other agents including 5-fluorouracil, docetaxel, cetuximab, as well as anti-PD-1 in those advanced refractory diseases.^15,16 However, although patients initially respond, most inevitably experience drug resistance and compromised therapeutic effects later. Therefore, proper patient selection based on biomarkers and drug resistance reversal via targeting resistance-associated genes might be effective strategies to address this clinical challenge. Of great importance, PDO treatment with cisplatin followed by systematic profiling of transcriptional and phenotypic responses enables the identification of molecular biomarkers and signaling pathways associated with cisplatin resistance.^17,18,19 Furthermore, PDOs represent a robust platform to evaluate the therapeutic efficacies of individual chemotherapeutic drugs and rationally designed combination regimens targeting drug resistance pathways, thus facilitating the development of novel combination approaches to overcome chemoresistance.^20,21

In this study, we generated a patient-derived OSCC organoid biobank (46 PDOs) from clinical fresh specimens using an optimized culture medium. These organoids faithfully recapitulate the histological and molecular features of their corresponding parental tumors and are successfully utilized for chemotherapeutic drug sensitivity measurements and genetic manipulations in vitro. Notably, we identified CDCP1 as a critical regulator of cisplatin resistance in OSCC by activating the Wnt/β-catenin pathway and in turn enhancing cancer stemness. Nanoparticle-mediated therapeutic targeting of CDCP1 synergized with cisplatin, achieving tumor regression in PDX models. These findings collectively establish PDOs as valuable experimental platforms and resources for both basic and translational research and highlight their potential for the development of personalized therapeutic strategies for OSCC.

Results

Establishment of a living OSCC organoid biobank

To establish a clinically relevant organoid biobank of OSCC, we collected surgically resected/biopsy specimens from 65 patients with primary OSCC spanning diverse ages and tumor stages. The detailed epidemiological and clinicopathological parameters were presented in Table S1. Fresh specimens were immediately split into small pieces and used for histopathological analysis, exome/transcriptional sequencing, organoid culture, and PDX generation (Figure 1A). Dissociated cells were seeded in Matrigel drops and overlaid with the organoid culture medium. To better support growth and ensure long-term expansion, an optimized medium formulation containing FGF2/FGF10, R-spondin, Noggin, A83-01, CHIR-99021, EGF, Wnt3a, and the large tumor suppressor kinase 1/2 (LATS1/2) inhibitor GA-017 was utilized based on well-established genetics and molecular portraits of OSCC (Figures 1B and S1).^14,22,23 Indeed, these elements such as EGF, Wnt3a, and R-spondin were commonly required for the development and growth of various PDOs, irrespective of cancer lineage, which cooperatively promoted cell self-renewal and proliferation in PDOs.^{14,19,24,25,26} Compared to the previously established human oral mucosa organoid protocols (medium 1),¹⁴ our optimized culture medium, medium 2 and medium 3, demonstrated superior growth rates and sustained growth across passages. Notably, medium 3 containing GA-017 yielded optimal organoid formation and better long-term expansion efficiency. As expected, GA-017 potently inhibited LATS1/2 and in turn activated YAP to promote cell proliferation and organoid growth as previously reported (Figures 1C–1E and S2A).²³ Moreover, GA-017 induced cell self-renewal and epithelial-to-mesenchymal transition (EMT) in PDOs as reflected by increased expression of CD44, BMI1, N-cadherin, Vimentin, as well as reduced E-cadherin (Figures 1E, S2B, and S2C). In parallel, similar findings were also recapitulated in OSCC cell lines (Figures S2D–S2H). Intriguingly, this optimized culture medium was also capable to promote the generation and growth of normal oral mucosa-derived organoids (Figure S3).

Figure 1.

Establishment of a living patient-derived organoid biobank from oral squamous cell carcinoma patients

(A) Schematic outline of the oral squamous cell carcinoma (OSCC) patient-derived organoid (PDO) biobank establishment and subsequent experimental analyses.

(B) Overview of OSCC organoid medium components.

(C) Optical microscope images of organoids from different sources (OSCC-1O, OSCC-4O, and OSCC-7O) after 10 days of culture in three distinct media (medium 1, medium 2, or medium 3). Scale bars, 100 μm.

(D) Growth curves of organoid cultures (n = 8) in medium 1, medium 2, and medium 3.

(E) Protein expression of LATS1/2, phosphorylated-LATS1/2 (Thr1079+Thr1041), YAP, phosphorylated-YAP (Ser127), Ki-67, Involucrin, CD44, and BMI1 in OSCC-8O cells cultured with medium 2 or medium 3 (supplemented with GA-017).

(F) Clinical characteristics of OSCC patients corresponding to the successfully established PDOs.

(G and H) Continuous growth conditions (G) and immunofluorescent staining images (H) of three representative organoids with different morphological features. Scale bars, 100 μm.

(I) Representative H&E and immunohistochemical staining results of OSCC markers (Ki67, KRT5/6, and TP63) between parental tumors and organoids with different pathological grades. Correlation analysis of staining intensity quantification for the indicated markers was performed between OSCC tissue samples and their paired organoid samples (n = 20). Scale bars, 100 μm. Data were shown as the mean ± SD.

Statistical analysis for (D) was performed via two-way ANOVA. Statistical analysis for (I) was performed via Spearman correlation analysis. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

Using this optimized medium, we have successfully established 46 organoids from OSCC patients within a 2-week cultivation period, achieving a success rate of 70.77% (46/65) (Figure 1F). These established OSCC organoid lines could be expanded long-term (>10 passages), cryopreserved, and successfully recovered, and they retained stable morphology and transcriptomics for months (Figures S4 and S5). The established OSCC organoid biobank recapitulated key epidemiological features of this disease, corresponding well to male predilection and aged individuals. This biobank encompassed PDOs from diverse anatomical subsites and clinicopathological stages, indicating robust organoid derivation capabilities (Figure 1F). Next, we compared the associations between the success/failure of organoid cultures and the corresponding patients’ epidemiological and clinicopathological parameters. We found that samples from patients with advanced tumor size (T stage) and clinical stage (TNM stage) tended to result in successful PDO establishment (Table S1). Moreover, bright-field microscopy revealed that individual OSCC organoid lines exhibited marked morphological variations, including solid/compact, cyst-like, and solid-cyst-like structures. These organoids displayed varied growth kinetics that recapitulated the histological heterogeneity observed in primary tumors (Figures 1G, 1H, and S6). In aggregate, we successfully created a living OSCC organoid biobank comprising 46 organoid lines through optimized culture medium.

OSCC organoids recapitulate the histological and molecular profiles of parental tumors

To determine whether these OSCC organoids retained the histopathological characteristics in parental tumors, we leveraged morphological and histopathological analyses using hematoxylin & eosin (H&E) staining and immunohistochemistry (IHC) with well-established markers. As shown in Figure 1I, organoid architectures faithfully preserved the features of original tumor specimens, including characteristic patterns of keratin pearl formation and nuclear pleomorphism. Notably, inter-patient heterogeneity in differentiation status (well/poorly/moderately differentiated) was maintained in the organoid biobank, mirroring the clinicopathological diversity of clinical samples. IHC staining further confirmed molecular fidelity between organoid lines and parental tumors as evidenced by highly consistent marker expression including the epithelial differentiation marker (KRT5/6, R = 0.83, p < 0.01), basal cell/stemness marker (TP63, R = 0.85, p < 0.01), and proliferation marker (Ki67, R = 0.68, p < 0.01) (Figure 1I).

Copy number alterations (CNAs) and structural variations (SVs) commonly observed in OSCC samples significantly contribute to tumorigenesis and therapeutic sensitivities.^22,27 We next performed whole-exome sequencing (WES) of 10 paired OSCC organoids and parental tumors and found that typical copy number gains and losses were retained throughout genomes in PDOs, although more distinct copy number signals were detected in organoids compared to original tumors (Figure 2A). This phenomenon is in keeping with previous reports and is presumably attributable to in vitro growth selection, as well as the complex multicellular nature of primary tumors and the high tumor purity of organoids.^17,18,28,29 In addition, most SVs were conserved between OSCC tumor-organoid pairs (Figures 2B and S7).

Figure 2.

OSCC PDO recapitulates genomic landscapes and transcriptional expression pattern of parental tumor tissues

(A and B) Heatmap (A) and circular genome plots (B) illustrating copy number alterations (CNAs) and genomic variations, including duplications and deletions in different organoid lines and their matched parental tumor tissues.

(C) Summary of somatic mutations identified in OSCC PDOs and corresponding tumor tissues. Types of mutations are indicated at the bottom.

(D) Summary of the mutation variations in several cancer-related genes listed in the Cancer Gene Census Tier1 of the COSMIC database.

(E) Uniform manifold approximation and projection (UMAP) plot of paired tissue and PDO samples (n = 9) illustrating their clustering relationships.

(F and G) Heatmap (F) and principal-component analysis (PCA) plot (G) showing molecular subtyping of 27 OSCC PDOs based on bulk RNA-seq.

Given that somatic mutations are key drivers of tumor development,^22,27 we conducted a comparative analysis of mutation profiles between matched tumors and their corresponding organoids (Figure 2C). As expected, these organoid lines preserved the cancer-related genetic alterations observed in the primary OSCC tissues as reflected by frequent mutations of tumor suppressor genes such as TP53 (45%) and TTN (65%), as well as the glycosylated membrane protein MUC16 (50%). These results were in line with the characteristic genomic landscapes previously identified in The Cancer Genome Atlas (TCGA)-HNSC cohort.²² In parallel, these organoids largely maintained the mutational patterns observed in several cancer-associated genes classified under the Cancer Gene Census Tier 1 in the COSMIC database within the parental OSCC tissues, although some discrepancies exhibited between paired organoids and tumors (Figure 2D). This observation suggested that organoids cultured in vitro might occasionally lead to the emergence of new mutations or favor the growth of rare, mutated clones within these cultures. Together, these findings support the genomic fidelity of OSCC organoids in capturing the essential genomic characteristics of primary tumors.

Next, we proceeded to explore the similarity between PDOs and parental tumors at the transcriptomics level. As expected, the transcriptome sequencing results revealed highly consistent profiles between nine paired organoid lines and their corresponding tumors, demonstrating that our PDO models faithfully retained the transcriptomic characteristics of primary OSCC tissues (Figures 2E and S8). In addition, we compared the global transcriptomic profiling data between 6 normal oral mucosa organoids and 27 PDOs cultured with the same media and conditions. As expected, substantial gene expression differences existed, with these differentially expressed genes (DEGs; with 1,887 genes upregulated and 1,912 genes downregulated in PDOs) predominantly enriched in multiple cancer-related signaling pathways and biological processes, such as EMT, KRAS/Wnt/PI3K-AKT signaling, epithelial cell proliferation, and transcriptional misregulation in cancer (Figure S9). Moreover, as illustrated in Figures 2F and 2G, we have calculated the gene set variation analysis (GSVA) scores to determine the molecular subtype among these 27 PDOs using gene-set signatures established for HNSCC molecular subtypes. Most PDOs were assigned to one of the dominant subtypes, and our PDO biobank generally covered these subtypes (15 basal, 9 classical, and 3 atypical), thus suggesting that our PDO biobank well represented transcriptional heterogeneity in OSCC clinical samples.²² Taken together, these findings provide ample support that our PDOs faithfully recapitulate the histological, genomic, and transcriptional characteristics of OSCC tumors.

To verify the oncogenic potential of these established OSCC organoid lines, we performed subcutaneous transplantations into the flanks of NOD/ShiLtJGpt-Prkdc^em26Cd52Il2rg^em26Cd22/Gpt (NCG) mice (Figure 3A). All three PDOs (OSCC-18O, OSCC-19O, and OSCC-32O) resulted in macroscopically visible tumors in at least 75% of animals (n = 4 per line) after a 12-week observation period (Figures 3B–3E). H&E staining of the patient-derived organoid xenografts (PDOXs) demonstrated that the histological features of these PDOXs closely mirrored those of the original patient tumors. Specifically, OSCC-19 and OSCC-32 PDOX tissues preserved the morphological characteristics of low- or high-differentiated malignant epithelial cell, respectively (Figure 3F). Immunohistochemical staining with human-specific Ki-67 and KRT5/6 antibodies showed obvious positive expression in these PDOXs, thus confirming the human origin of tumors. These positive stainings were well concordant between PDOXs and parental tumors, suggesting that these organoids maintained their tumorigenic characteristics and retained molecular features of parental tumors in vivo (Figure 3F).

Figure 3.

Patient-derived organoid xenograft recapitulates histopathological features of OSCC

(A) Schematic illustration of the patient-derived organoid xenograft (PDOX) model construction and analysis.

(B and C) Representative images (B) of PDOX (OSCC-18O, OSCC-19O, and OSCC-32O) and xenografted tumors for each organoid line (C) after 12 weeks of subcutaneous inoculation (n = 4 per group, with the OSCC-18O group being 3). Scale bars, 1 cm.

(D) Tumor growth curves of the 3 different PDOX models (OSCC18O, OSCC19O, and OSCC32O) over 12 weeks (n = 4 per group, with the OSCC-18O group being 3).

(E) The final tumor weight obtained from the indicated PDOX model (n = 4 per group, with the OSCC-18O group being 3).

(F) Histological and immunohistochemical analysis of parental tumors and PDOX tumors, including H&E, Ki67, and KRT5/6 staining. Scale bars, 100 μm. Data were shown as the mean ± SD.

In summary, these data from in vitro assays and in vivo transplantation demonstrate that OSCC organoids retain their tumorigenic potential and form tumors with features that closely resemble parental tumors, thus highlighting PDOs as reliable preclinical models for mechanistic exploration and therapeutic development.

OSCC organoids serve as a platform for genetic manipulation, immuno-oncology, and drug screening

To evaluate whether these OSCC organoids are suitable for genetic manipulation in vitro, we selected FOSL1 for lentivirus-mediated gene silencing and overexpression in organoids, largely because it has established as an oncogenic transcription factor involved in cancer stem cell (CSC) stemness and aggressive behaviors in OSCC.³⁰ Two organoid lines with differential baseline FOSL1 levels were selected: the high-FOSL1 OSCC-14O was subjected to short hairpin RNA (shRNA)-mediated knockdown, while the low-FOSL1 OSCC-32O was used for FOSL1 overexpression (Figure S10A). After 48 h infection, robust transduction efficiency was confirmed by the widespread EGFP (enhanced green fluorescent protein) fluorescence in organoids transduced with either FOSL1-targeting shRNA (Figure S10B) or FOSL1-overexpressing lentivirus (Figure S10F). Functional validations revealed that shRNA-mediated FOSL1 knockdown significantly reduced both mRNA/protein levels (Figures S10C and S10D) and organoid proliferative capacity (Figure S10E). Conversely, enforced FOSL1 overexpression markedly elevated FOSL1 expression (Figures S10G and S10H) and enhanced organoid growth (Figure S10I). These results collectively support the utility of OSCC organoids for genetic manipulation and functional phenotype assessments.

Cancer immunotherapy, particularly immune checkpoint blockade (ICB), has revolutionized therapeutic paradigms for multiple malignancies including OSCC.³¹ However, preclinical models that adequately preserve patient-specific immunogenicity remain scarce. To address this limitation, we developed an in vitro co-culture system combining organoids with autologous T cells. Peripheral blood mononuclear cells (PBMCs) isolated from patients were stimulated in vitro to generate activated T cells, which were then co-cultured with organoids from the same patient (Figure S11A). To dynamically monitor T cell-mediated cytotoxicity against 3D organoids, we performed dual fluorescent labeling: organoids were transduced with EGFP via lentivirus, while activated T cells were labeled with phycoerythrin (PE) dye. Following 10 h co-culture, significant cytotoxic effects were observed as reflected by reduced organoid volume and peripheral cell lysis (Figure S11B). This co-culture platform showcased the capacity to model patient-specific tumor-immune interactions, which might be exploited for testing the efficacy of ICB.

Chemotherapy remains the indispensable pillar in current OSCC treatment paradigm. Cisplatin, docetaxel, and cetuximab, alone or in combinations, are standard care for patients with OSCC.^32,33 To explore the response profile of OSCC organoids to these agents, we randomly selected 20 organoid lines for drug sensitivity assays. These PDOs were treated with drugs under 8 different concentrations for 120 h (Figure 4A). As anticipated, the responses to these drugs dramatically varied among these organoids from different patients as dose-response curves, AUC (area under curve), and Z factor all showed substantial interpatient variabilities in response to single chemotherapy agents (Figures 4B–4D). Thus, these OSCC organoids showed varied responses to different conventional chemotherapeutics or targeted therapy, reenforcing them as effective tools for drug screening.

Figure 4.

Response of OSCC PDOs and PDXs to cisplatin, docetaxel, and cetuximab

(A) Schematic workflow of drug screening using OSCC PDOs. Briefly, PDOs were seeded into 96-well ultra-low attachment (ULA) plate for culture, treated with drugs, and then assayed for cell viability followed by analysis.

(B) Dose-response curves across 20 different OSCC PDOs after administrated with cisplatin, cetuximab, or docetaxel (n = 3).

(C) Representative microscopy images of 4 distinct organoid lines (OSCC-4O, OSCC-10O, OSCC-11O, and OSCC-32O) after treated with vehicle or drugs (cisplatin 10 μM, cetuximab 100 μg/mL, or docetaxel 50 nM) for 120 h. Scale bars, 100 μm.

(D) Heatmaps displaying IC50, area under the curve (AUC) scores, and Z factors for drug responses across 20 PDO lines.

(E) Schematic illustration of patient-derived xenograft (PDX) model workflow for monitoring treatment response of cisplatin or cetuximab.

(F) Images of PDX: OSCC-10T model tissues harvested 18 days after the first treatment with cisplatin, cetuximab, or control (administrated once every 3 days for a sum of 6 doses) (n = 6 per group).

(G) Tumor growth curves of the PDX model over 18 days (left) and the final tumor weight comparison between different treatment groups (right) (n = 6 per group).

(H) Immunofluorescence analysis of Ki-67, γ-H2A, and TUNEL in PDX: OSCC-10T tumors post treatment. Scale bars, 100 μm.

(I) Schematic illustration of PDX model workflow for monitoring treatment response of cisplatin or docetaxel.

(J) Images of PDX: OSCC-4T model tissues harvested 18 days after the first treatment with cisplatin, docetaxel, or control (administrated once every 3 days for a sum of 6 doses) (n = 6 per group).

(K) Tumor growth curves of the PDX model over 18 days (left) and the final tumor weight comparison between different treatment groups (right) (n = 6 per group).

(L) Immunofluorescence analysis of Ki-67, γ-H2A, and TUNEL in PDX: OSCC-4T tumors post treatment. Scale bars, 100 μm. Data were shown as the mean ± SD.

Statistical analysis for (G) and (K) was performed via two-tailed unpaired Student’s t test or two-way ANOVA. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

To validate these in vitro drug screening results, we exploited the PDX models, which were simultaneously developed from same patients, as a proxy in the real-world clinical scenario as PDXs have been demonstrated as reliable models that faithfully reflect the molecular features and drug sensitivities of parental tumors.^34,35 PDX models derived from two representative cases were selected based on their distinct in vitro organoid drug sensitivity profiles: OSCC-10O (cisplatin-resistant/cetuximab-sensitive) and OSCC-4O (cisplatin-resistant/docetaxel-sensitive) (Figures 4E and 4I). In the OSCC-10T PDX model, cetuximab treatment significantly reduced tumor burden compared to cisplatin or controls, with decreased tumor weight/volume (Figures 4F and 4G), impaired cell proliferation (Ki-67, p < 0.01), and increased cell apoptosis (TUNEL, p < 0.01), whereas cisplatin administration showed no noticeable efficacy (Figures 4H and S12A–S12C). In parallel, concordance between organoid-predicted docetaxel sensitivity and in vivo tumor regression was observed in the OSCC-4O PDX model (Figures 4J–4L and S12D–S12F). Collectively, these findings indicate that OSCC organoids are reliable tools for drug screening and response prediction in OSCC.

Transcriptomic analysis identifies CDCP1 involved in cisplatin chemoresistance in OSCC

While cisplatin-based chemotherapy remains a cornerstone in OSCC treatment, its efficacy is frequently compromised by intrinsic and acquired resistance,³ highlighting an urgent need to identify novel targetable vulnerabilities. As anticipated, significant heterogeneity in cisplatin sensitivity among 20 PDOs was observed, thus largely recapitulating clinical response patterns. Based on IC50 values, these PDOs were stratified into cisplatin-sensitive (n = 9, IC50 < 5 μM) and insensitive (n = 7, IC50 > 10 μM) groups (Figures 5A and 5B). The sensitive organoid line (OSCC-16O) exhibited pronounced growth suppression and apoptosis at 5 μM, whereas the insensitive counterpart (OSCC-23O) showed growth inhibition at ≥10 μM (Figures 5C and 5D).

Figure 5.

Transcriptome sequencing analysis demonstrates that CDCP1 is upregulated in cisplatin-insensitive OSCC PDO lines

(A) Area under curve (AUC) values of OSCC organoid lines for cisplatin (n = 20).

(B) The IC50 profile of cisplatin across 20 different OSCC organoid lines. The 9 organoid lines in the blue region were relatively sensitive to cisplatin, whereas the 7 organoid lines in the red region were considered relatively resistant to cisplatin.

(C and D) Analysis of organoid size (bright-field images) and PI-positive cell proportion (fluorescence images) in cisplatin-insensitive (OSCC-23O) versus sensitive (OSCC-16O) PDOs after treatment with varying concentrations of cisplatin (n = 3). Scale bars, 100 μm.

(E) Volcano plot showing differentially expressed genes between the cisplatin-sensitive group (n = 9) and cisplatin-insensitive group (n = 7).

(F) GSEA results between the cisplatin-sensitive and cisplatin-insensitive organoid lines.

(G) Overlapping genes associated with cisplatin resistance from multiple datasets (GSE111585, GSE115119, and our in-house dataset).

(H) Comparison of the CDCP1 expression between cisplatin-sensitive and resistant samples in an OSCC single-cell RNA-seq dataset (GSE117872).

(I) The Kaplan-Meier plot showed the correlation between CDCP1 expression and overall survival in patients receiving platinum-based therapy in the TCGA-HNSC cohort.

(J and K) Western blotting and corresponding quantitative analysis (J) (n = 3) and immunofluorescence staining (K) of CDCP1 protein expression levels in sensitive and insensitive OSCC PDO lines. Scale bars, 100 μm. Data were shown as the mean ± SD.

Statistical analysis for (D) and (H) was performed via two-tailed unpaired Student’s t test. Statistical analysis for (I) was performed via log rank test. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

To explore targets involved in cisplatin sensitivity, we conducted transcriptomic profiling and identified 384 DEGs between insensitive and sensitive groups. Among these, 221 genes were upregulated, including GSDMA, MMP10, FOSL1, CDCP1, and INHBA, which have been previously documented in chemoresistance (Figure 5E).^36,37,38 Gene set enrichment analysis (GSEA) revealed enrichments of chemotherapy resistance-related pathways (e.g., hypoxia and TGF beta signaling) in insensitive organoid lines (Figure 5F). To identify potential targets for overcoming cisplatin resistance, we intersected the upregulated DEGs from our study with those from two OSCC cisplatin-resistance datasets (GSE111585 and GSE115119),³⁹ which yielded 10 overlapped candidates potentially involved in cisplatin resistance (Figure 5G). Among these, CDCP1 emerged as a priority target due to high expression in OSCC and positive association with poor prognosis (Figures S13A–S13C). We re-analyzed two public single-cell RNA sequencing (scRNA-seq) datasets from OSCC with cisplatin chemoresistance (GSE117872 and PRJNA960652)^19,40 and confirmed prominent CDCP1 overexpression in malignant epithelial cells from therapy-resistant specimens (Figures 5H and S14A–S14F). Bulk RNA sequencing (RNA-seq) analyses from OSCC cell lines with cisplatin resistance (GSE168424)⁴¹ also indicated increased CDCP1 expression in those resistant cells (Figure S14G). Complementarily, patients with higher CDCP1 expression portended worse survival in cisplatin-treated patients (hazard ratio [HR] = 1.796, p = 0.067) and significantly associated unfavorable outcomes in platinum-based therapy cohorts in the TCGA-HNSC cohort (HR = 1.833, p = 0.02, Figure 5I). Additionally, CDCP1 overexpression tended to associate with reduced survival in HNSCC patients treated with cisplatin-based therapy as well as in human papillomavirus (HPV)-negative patients treated with platinum-based therapy (Figures S13D and S13E). Moreover, immunoblotting and immunofluorescence validated significantly higher CDCP1 protein levels in insensitive organoids compared with sensitive ones (Figures 5J and 5K).

Taken together, these results highlight CDCP1 as a promising therapeutic target for overcoming cisplatin chemoresistance in OSCC, warranting further investigations into its mechanistic role and translational potential.

CDCP1 promotes cisplatin chemoresistance through modulating Wnt/β-catenin-driven stemness in OSCC

To further verify whether targeting CDCP1 enhances chemosensitivity to cisplatin, we measured cisplatin IC50 values in Cal27 and HN6 after small interfering RNA (siRNA)-mediated CDCP1 knockdown. CDCP1 inhibition markedly increased cisplatin sensitivity in both cell lines but had negligible effects on untransformed HOK cells and normal organoids (Figure S15). Of note, we introduced shRNA lentivirus targeting CDCP1 to a cisplatin-insensitive organoid line (OSCC-23O) and found that CDCP1 ablation reduced cisplatin IC50 from 20.12 to 8.51 μM, accompanied by elevated DNA damage (γ-H2AX) and apoptosis (downregulation of Bcl-2 and upregulation of BAX) (Figures 6A–6D). Similar drug-sensitizing effects induced by CDCP1 depletion were observed in another inherently cisplatin-insensitive PDO line (OSCC-42O; IC50: 17.34 μM), whereas these effects were minimal in inherently cisplatin-sensitive PDOs (Figures S16A–S16C). In keeping with this, our results revealed that CDCP1 knockdown significantly suppressed proliferation of these insensitive PDOs while having negligible effects on those sensitive PDOs (Figures S16D–S16I). Moreover, we found that 8-PN, a chemical inhibitor of CDCP1,^42,43 enhanced anti-proliferative effects of cisplatin and yielded synergistic effects in combination with cisplatin in OSCC cells (Figure S17).

Figure 6.

CDCP1 regulates the cisplatin sensitivity of OSCC PDO lines via stemness maintenance

(A) Merged fluorescence and bright-field images of OSCC PDO lines with or without CDCP1 knockdown.

(B) Effects of cisplatin treatment on organoids in control and CDCP1-knockdown lines. Scale bars, 100 μm.

(C) Sensitivity of OSCC organoid lines with or without CDCP1 knockdown for cisplatin (n = 3).

(D) After cisplatin treatment, western blotting was utilized to analyze apoptosis-related markers (γ-H2A, Bcl-2, and BAX) after PDO lines with or without CDCP1 knockdown.

(E) Correlation between CDCP1 expression and stemness index score in the TCGA-OSCC cohort.

(F) Western blotting against stemness-related markers (CD133, CD44, BMI1, and SOX2) in the cisplatin-sensitive and insensitive PDO lines.

(G) Western blotting against stemness-related markers (CD133, CD44, BMI1, and SOX2) after PDO lines with or without CDCP1 knockdown.

(H) A schematic of the experimental design illustrating genome-wide transcriptional profiling of Cal27 cells infected with or without shCDCP1.

(I) Volcano plot showing differentially expressed genes between shCDCP1 and shNC cells.

(J) Gene set enrichment analysis (GSEA) revealed that CDCP1-regulated genes were significantly enriched in pathways regulating WNT signaling (KEGG and C6: oncogenic signature).

(K) Western blot analyses of CDCP1 and β-catenin expression in Cal27 and HN6 cells with or without shRNA-mediated CDCP1 knockdown.

(L) Nucleoplasm separation assay of β-catenin protein in Cal27 cells with or without shRNA-mediated CDCP1 knockdown.

(M) Western blot analysis of CDCP1, β-catenin, and SOX2 expression in Cal27 and HN6 cells after CDCP1 knockdown alone or combined with LiCl (5 mM) treatment for 48 h.

(N) Cal27 cells with or without CDCP1 knockdown were co-transfection with TOPFlash and Renilla luciferase plasmids, then exposed to LiCl (5 mM) for 24 h before luciferase readout (n = 3).

(O and P) Representative images and quantitative analyses of tumorsphere formation in Cal27 and HN6 cells following CDCP1 knockdown alone or in combination with LiCl (5 mM) treatment (n = 3). Scale bars, 100 μm.

(Q) Western blot analyses of CDCP1, β-catenin, and Cyclin D1 (a direct downstream target of β-catenin) expression in Cal27 and HN6 cells after CDCP1 overexpression.

(R) Nucleoplasm separation assay of β-catenin protein in Cal27 and HN6 cells after CDCP1 overexpression.

(S) Cal27 cells with or without CDCP1 overexpression were co-transfected with TOPFlash and Renilla luciferase plasmids. Luciferase activity was determined 24 h after transfection (n = 3).

(T) Western blot analyses of stemness-related markers (CD44, BMI1, and SOX2) after CDCP1 overexpression.

(U) Tumorsphere formation in Cal27 and HN6 cells with or without CDCP1 overexpression (n = 3).

(V) Cell proliferation difference of Cal27 and HN6 cells with or without CDCP1 overexpression under cisplatin treatment (10 μM; n = 3). Data were shown as the mean ± SD.

Statistical analysis for (E) was performed via Spearman correlation analysis. Statistical analysis for (N), (P), (S), (U), and (V) was performed via two-tailed unpaired Student’s t test. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

Having established that CDCP1 is involved in cisplatin sensitivity in OSCC, we next sought to explore the mechanism by which CDCP1 promotes tumor resistance to cisplatin. Prior reports have implicated cancer stemness intricately linked to chemotherapeutic resistance across human cancers.^44,45 Initially, we conducted bioinformatics analyses and found that CDCP1 expression positively correlated with stemness scores using the bulk RNA-seq dataset (TCGA-OSCC, R = 0.33, p < 0.01, Figure 6E), proteomics dataset (CPTAC-OSCC, p < 0.01, Figures S18A–S18C), and expression of selected stemness markers/scores in two independent scRNA-seq datasets (GSE234933, GSE188737; Figures S18D–S18G).^46,47 These results indicated significant association between CDCP1 expression and stemness and suggested a potential regulatory role of CDCP1 in OSCC stemness. Indeed, CDCP1 silencing suppressed stemness marker expression (SOX2, BMI1, CD44, and CD133), impaired tumorsphere formation, and reduced CD44⁺CD133⁺ CSC ratios⁴⁸ in vitro (Figures S19A–S19E). Importantly, we performed in vivo limiting dilution assay and confirmed that stable CDCP1 knockdown robustly reduced CSC ratios and impaired tumor-initiating capacity in vivo (Figure S19F). In addition, cisplatin-insensitive organoid lines exhibited elevated baseline expression of these stemness markers, which were substantially reduced upon CDCP1 inhibition (Figures 6F and 6G). Together, these results suggested that CDCP1 conferred cisplatin resistance probably by enhancing stemness in OSCC.

To pinpoint the downstream effectors of CDCP1 potentially involved in cisplatin resistance, we next performed global RNA-seq in Cal27 cells treated with shRNA-targeting CDCP1 and found 792 genes upregulated and 1,331 genes downregulated upon CDCP1 depletion (Figures 6H and 6I). These DEGs were significantly enriched in stemness-relevant pathways and Wnt signaling (Figures 6J, S20A, and S20B). Complementarily, re-analyses of bulk transcriptomics dataset revealed that CDCP1 mRNA expression positively correlated with Wnt pathway score and CTNNB1 (encoding β-catenin) mRNA expression; primary OSCC samples with high CDCP1 expression had prominent enrichment of Wnt signaling as compared to those with low CDCP1 (Figures S20C–S20F).

Given that the Wnt/β-catenin signaling pathway critically governs cancer stemness and CDCP1 modulates Wnt pathway in colon cancer,^49,50,51 we hypothesized that CDCP1 affected cisplatin sensitivity probably by modulating Wnt signaling-driven stemness in OSCC. To substantiate this, we leveraged both loss- and gain-of-function approaches to manipulate CDCP1 expression and determine the resulting changes. As shown in Figures 6K and 6L, CDCP1 depletion resulted in prominent reduction of total, cytoplasmic, and nuclear β-catenin protein in both cell lines. Moreover, addition of LiCl, a glycogen synthase kinase 3 inhibitor to activate the canonical Wnt/β-catenin pathway, largely attenuated the inhibitory effects induced by CDCP1 knockdown on β-catenin/SOX2 protein abundance, β-catenin-driven transcriptional activity, and tumorsphere formation in vitro (Figures 6M–6P). Similar phenotypic and protein changes were replicated in cells treated with 8-PN in vitro (Figure S21). Next, we engineered a FLAG-labeled CDCP1 expression vector, which was introduced into cells. As shown in Figures 6Q–6S, ectopic CDCP1 overexpression markedly increased total and nuclear β-catenin protein as well as β-catenin-driven transcriptional activity. Consistently, increased abundance of stemness markers and capacity of tumorsphere formation were observed in cells with CDCP1 overexpression (Figures 6T and 6U). Additionally, upon cisplatin exposure, cells with CDCP1 overexpression appeared to proliferate much faster than those with empty vector (Figure 6V).

In aggregate, these findings establish that CDCP1 activated the Wnt/β-catenin pathway and in turn modulated stemness-driven cisplatin chemoresistance, thus suggesting therapeutic targeting of CDCP1 to restore cisplatin sensitivity in OSCC.

Nanoparticle-mediated CDCP1 silencing reverses cisplatin resistance in OSCC

The paucity of selective potent inhibitors for CDCP1 hinders our efforts to reverse cisplatin resistance by targeting CDCP1. To resolve this, we developed a pH-responsive nanoparticle (NP) delivery system (chitosan/γ-poly-glutamic acid, CS/γ-PGA) to encapsulate siCDCP1 for targeted delivery (Figure 7A). NP characterization confirmed an average size of 107.8 nm, a zeta potential of 18.08 mV, and a uniform spherical morphology under a transmission electron microscope (Figures S22A–S22C). Notably, the NPs exhibited efficient siRNA release under acidic conditions (pH 6.5), with over 80% release within 24 h (Figure S22D). Cell uptake experiments showed rapid internalization of siCDCP1-loaded NPs into the cytoplasm within 9 h, achieving efficient CDCP1 knockdown comparable to conventional transfection reagents (Figures 7B and S22E).

Figure 7.

pH-responsive nanoparticle delivering siCDCP1 to overcome chemoresistance in OSCC

(A) Schematic illustration of the synthesis of pH-responsive nanoparticles (NPs) encapsulated with siCDCP1 and their roles in suppressing CDCP1 expression.

(B) Western blot analysis of CDCP1 expression in Cal27 and HN6 cells after treatment with RNAi regent or siCDCP1-loaded NPs.

(C) Schematic illustration of PDX model workflow for monitoring treatment response of NPs (+) plus cisplatin (Cis). NPs (−) represents nanoparticles encapsulating a non-targeting (nonsense) siRNA; NPs (+) represents nanoparticles encapsulating a CDCP1-targeting siRNA.

(D–F) Image at the end of treatment (D) and growth curves (E) and final tumor weight (F) of xenografted tumors (PDX: OSCC-42T) for each group (n = 6).

(G) Western blot and corresponding quantitative analyses (n = 6) of CDCP1, CD44, BMI1, and β-catenin expression of PDX tumor tissues after indicated treatments.

(H and I) Representative immunofluorescence images of CDCP1 and β-catenin in PDX: OSCC-42T tumors post treatment. Scale bars, 50 μm.

(J) Quantitative analyses of CDCP1, β-catenin, Ki-67, CD44, and BMI1-positive staining area in PDX: OSCC-42T tumors post treatment via ImageJ (n = 6). Data were shown as the mean ± SD.

Statistical analysis for (F) and (J) was performed via two-tailed unpaired Student’s t test. Statistical analysis for (E) was performed via two-way ANOVA. ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

Initially, we performed hemolysis assays by spectrophotometrically measuring hemoglobin release following exposure to varying NP concentrations and detected no significant differences in hemolysis index between the NP and negative saline control (Figure S23). Next, we sought to evaluate the biodistribution and tumor-targeting capacity of these NPs in vivo by administering Cy5-labeled CDCP1-targeting NPs intravenously in the PDX model. As illustrated in Figure S24, these NPs were preferentially accumulated in liver and kidney, consistent with previous reports.⁵² Notably, the CDCP1-targeting NPs were markedly enriched in tumors, while these Cy5-labeled free siRNAs were much less in tumors.

To determine the therapeutic potentials of our nanoplatform in vivo, we administered CDCP1-targeting NPs intravenously to two cisplatin-resistant PDX models (OSCC-23T and OSCC-42T). As displayed in Figures 7C–7F, NP-mediated siCDCP1 monotherapy (NPs(+)) significantly reduced tumor burden in the PDX models compared to NPs(−) bearing non-targeting siRNAs and vehicle-treated controls, thus excluding the effects induced by nanoplatform itself. Moreover, NPs(+) in combination with cisplatin yielded superior efficacy over either cisplatin or NPs(+) alone as tumor regression occurred. Results from western blotting and immunofluorescence staining in samples revealed that combination therapy markedly suppressed expression of CDCP1, β-catenin, Ki-67, CD44, and BMI1, while induced more γH2A and cell apoptosis (TUNEL) (Figures 7G–7J and S25). Similar results from the OSCC-23T PDX model further supported therapeutic effects induced by NPs(+) (Figure S26). These in vivo findings were consistent with those aforementioned in vitro findings regarding CDCP1 involvement in stemness-associated cisplatin resistance and also highlighted the effective on-target therapeutic effects of our NPs.

Lastly, we monitored body weight changes throughout the treatment period and performed detailed analyses, including hematological testing, blood biochemistry, and histopathological examination (H&E staining) of major organs. As shown in Figure S27, we found no significant loss in body weight, no notable differences in hematological or biochemical parameters, and no evident histopathological alterations in major organs examined between NPs(+) alone/in combination with cisplatin and others. Thus, these findings favor that our CDCP1-targeting NP formulation exhibits a favorable biosafety profile in vivo.

In summary, these results demonstrate that NP-mediated siCDCP1 delivery effectively silences CDCP1, reverses cisplatin chemoresistance, and enhances cisplatin efficacy by inhibiting Wnt/β-catenin-driven stemness in OSCC, highlighting its potential as a clinically translatable therapeutic strategy.

Discussion

In this study, we have successfully established a biologically and functionally annotated biobank comprising 46 long-term cultured PDO lines that faithfully recapitulate inter- and intra-tumor heterogeneity in OSCC. These PDOs demonstrated high concordance with original tumors in histopathological characteristics and molecular profiles, largely mirroring key features of human OSCC. This living PDO biobank enables multifunctional applications including genetic manipulation, immuno-oncology research, as well as drug screening. Moreover, we identified CDCP1 as a crucial mediator of cisplatin chemotherapeutic resistance via enhancing Wnt/β-catenin signaling-driven stemness. Notably, NP-mediated CDCP1 silencing effectively restored chemosensitivity in resistant PDX models, demonstrating its translational potential for overcoming cisplatin resistance in OSCC.

The emergence of 3D-cultured organoid models has provided a promising approach to generate clinically relevant preclinical models that preserve heterogeneity and genetic features of tumors in vitro.^11,12,13,53 Previous studies have established OSCC tumor- and normal oral mucosa-derived organoids by serum-free culture conditions, such as adding multiple growth factors and chemicals for organoid growth and long-term expansion. Multiple culture media have been reported for HNSCC PDO generations with varied success rates ranging from 28.4% to 60%.^14,18 To facilitate PDO establishment and expansion, we refined the culture medium for OSCC PDOs by incorporating EGF, Wnt3a, and GA-017 (a LATS1/2 inhibitor) based on their well-established roles for epithelial stemness, cell proliferation/differentiation, as well as clonal growth. This optimization significantly improved organoid proliferative activity and increased the success rate to 70.77%. In particular, addition of GA-017 into the growth factor cocktail significantly enhanced cell proliferation, self-renewal, as well as EMT largely via Hippo signaling inactivation and YAP activation, which is in line with these pioneering reports showing that GA-017 facilitated efficient in vitro expansion of spheroids and organoids by modulating Hippo signaling²³; aberrantly activated Hippo-YAP/TAZ signaling critically promoted cell self-renewal, stemness, and EMT in OSCC.^54,55 On the other hand, FAT1 represents one of the most frequently mutated genes in OSCC (∼30%), and its recurrent mutations disrupt the Hippo pathway via LATS1/2 inhibition, leading to YAP activation and tumorigenesis.⁵⁶ Therefore, we believe that these key elements in our optimized media like EGF, Wnt3a, and GA-017 might cooperatively drive cell proliferation and self-renewal by exerting diverse roles that collectively enhance OSCC PDO generation and expansion.^14,18,23,57 Of course, detailed mechanisms of action behind these elements still warrant further exploration.

Notably, samples from lesions with large size or advanced stage tended to successfully generate PDOs, suggesting that malignant epithelial clones within these advanced diseases had potent proliferative advantages to survive and propagate in vitro. Indeed, we and others reported that PDOs can be generated from small biopsy samples as well as clinical specimens after conventional chemotherapeutics (data not shown). Furthermore, the 2-week turnaround time for OSCC PDO establishment and 1-week drug screen enable them to be well suited for guiding treatment planning such as pre-operative induction chemotherapies and post-operative adjuvant chemotherapy/radiotherapy. Therefore, these advantages together demonstrated the clinical utilities and values of PDOs as preclinical OSCC models.

Subsequent histopathological, genomic, and transcriptomic analyses confirmed that our established PDO models retained the genetic, molecular, and pathological characteristics of OSCC samples. For example, PDO retained similar cell differentiation marker expression as parental tumors and in vivo tumorigenic capacities. High concordance between PDOs and their original tumors in genetic alterations including copy number variation (CNV) and gene mutation was observed. Of note, transcriptomics profiling revealed characteristic patterns of gene expression and pathway activations in OSCC PDOs. Our PDOs were well corresponded to canonical three molecular subtyping of HNSCC at the transcriptional level, thus suggesting that this PDO biobank covered the transcriptional heterogeneity in clinical samples. Moreover, these PDO lines retained growth potentials and maintained transcriptional fidelity after long-term cryopreservation, thus allowing for long-term monitoring and comparing the phenotypical changes of tumors during disease progression or under treatments. Indeed, these transcriptional similarities between these PDOs and paired parental tumors were observed at both bulk and single-cell levels.¹⁷ These findings collectively demonstrated that this optimized model not only improves establishment efficiency but also faithfully recapitulates the molecular complexity of OSCC, making it particularly valuable for studying molecular mechanisms and testing targeted therapeutic strategies.^14,17,18

Current clinical management of OSCC predominantly relies on tumor site, stage, and patient characteristics for treatment planning, highlighting the need for more personalized therapeutic strategies.³ Organoid models have emerged as a superior platform for precision oncology due to their high establishment success rate, rapid expansion, and cost-effectiveness compared to other humanized preclinical models such as PDX.¹⁰ Leveraging the established OSCC organoid biobank, we evaluated three standard-of-care therapeutic agents and observed marked interpatient variabilities in drug responses. These drug response heterogeneities were further validated in vivo using PDX models, confirming the reliability of our organoid-based screening platform. Beyond drug testing, our PDO platform demonstrates functional versatilities including genetic manipulations (gene knockdown, enforced overexpression) as well as applications in immuno-oncology research via the co-culture system. Indeed, previous reports have demonstrated that the PDO system can be exploited for functional genomics through CRISPR-based gene editing and therapeutic resistance modeling through long-term, low-dose drug exposure, which provided insights into the mechanisms of acquired resistance.^20,58 Furthermore, it allows for tumor microenvironment (TME) reconstitution via co-culture systems integrating immune and stromal cells, aiding in the identification of stromal contributions to therapeutic resistance.⁵⁹ Collectively, these applications underscore OSCC organoids as a transformative tool for dissecting tumor biology, accelerating therapeutic discovery, and advancing precision medicine in OSCC.

Cisplatin remains the first-line chemotherapy for OSCC, yet its clinical efficacy is substantially compromised by drug resistance.^32,33 Previous studies have identified several mechanisms contributing to cisplatin resistance, such as the existence of CSCs and EMT.⁶⁰ Despite these efforts, cisplatin resistance remains a significant clinical challenge, necessitating the identification of novel biomarkers for patient stratification and therapeutic targets to improve outcomes. Recently, the Toshiaki Ohteki group has unraveled that autophagy and cholesterol biosynthesis activation are critical for cisplatin resistance using PDOs from tongue squamous cell carcinoma (SCC).¹⁷ Inspired by this, we integrated cisplatin response profiles and transcriptomic profiling data in OSCC PDOs, identifying CDCP1 as a potential regulator of cisplatin resistance by modulating Wnt/β-catenin signaling-driven stemness. Indeed, prior reports have documented that CDCP1 promotes nuclear localization of β-catenin, which in turn enhanced tumor growth in colorectal cancer and nasopharyngeal carcinoma.^51,61 Aberrant activation of Wnt/β-catenin signaling has long been recognized to modulate stemness and therapeutic resistance across human cancer including OSCC.^49,50 Our results from phenotypical changes following genetic and pharmacological inhibition of CDCP1 as well as its ectopic overexpression collectively demonstrated that CDCP1 promoted β-catenin protein expression, nuclear translocation, and transcriptional activation, which in turn mediated cisplatin resistance in OSCC. Complementarily, positive associations between CDCP1 expression and Wnt/β-catenin pathway scores, stemness, and cisplatin resistance across multiple bulk and scRNA-seq datasets provided support for the CDCP1-Wnt/β-catenin axis underlying cisplatin resistance. However, prior studies have documented that CDCP1 functions through the activation of Src-family kinases (SFKs), PI3K/AKT, or RAS/ERK pathways to drive cancer growth, metastasis, and resistance of various chemotherapeutic drugs like carboplatin across multiple cancers.⁶² We cannot rule out that these signaling pathways were also involved in CDCP1-mediated cisplatin resistance. The detailed mechanisms for CDCP1 in modulating the Wnt/β-catenin pathway remain an interesting unresolved question.

Previous studies have demonstrated CDCP1 as a promising target for cancer therapy as evidenced by anti-CDCP1 antibodies disrupted its cell surface expression and downstream signaling, achieving potent therapeutic effects by enhancing sensitivities or reverting resistance of multiple conventional drugs in preclinical models.^63,64 To strengthen the translational potentials of our findings by CDCP1 targeting to reverse cisplatin resistance in OSCC, we developed a pH-sensitive NP to effectively deliver siCDCP1, which successfully inhibited CDCP1 expression and reversed cisplatin resistance in the PDX models. Our results indicated that CDCP1 primarily drives OSCC cisplatin resistance by enhancing Wnt/β-catenin-driven stemness. This aligned with previous reports, which revealed that CDCP1 is a marker of tumor stemness associated with chemoresistance and that chemotherapy selectively enriches CDCP1⁺ CSCs in colorectal carcinomas; targeted inhibition of glycolytic enzyme triosephosphate isomerase (TPI) increased chemosensitivity of CDCP1⁺ CSCs and delayed recurrence.^38,65,66 Importantly, beside the robust therapeutic efficiencies, our comprehensive examinations showcased the satisfactory biosafety profile of this NP design after systematic delivery, thus highlighting the translational potentials. Collectively, our findings establish CDCP1 as a predictive biomarker and effective therapeutic target for overcoming cisplatin resistance in OSCC.

In conclusion, our study establishes OSCC organoids as a robust translational platform for investigating the molecular mechanisms of chemoresistance and advancing personalized therapeutic strategies. We identify CDCP1 as a critical regulator of cisplatin chemoresistance and experimentally validate its targeting as a robust strategy to reverse resistance. These findings underscore the promise of organoid models to bridge the gap between basic research and clinical practice, ultimately improving outcomes for patients with OSCC.

Limitations of the study

Despite these promising results, our study has several limitations. First, although the sample size in this study is substantial, it may not fully capture the extensive heterogeneity of OSCC. Future studies with larger cohorts are needed to validate our findings. Second, the absence of TME components in our organoid models, such as immune cells and cancer-associated fibroblasts, may affect the interpretation of drug response data. Co-culture systems incorporating TME cells and cancer organoids or innovative culture systems (e.g., IPTO, individualized patient tumor organoid)⁶⁷ might better mimic the original tumor ecosystem. Third, the long-term stability of organoid cultures and their ability to retain tumor heterogeneity over multiple passages require further detailed investigations. Finally, clinical validation of CDCP1 as a biomarker to predict cisplatin response and therapeutic target in larger patient cohorts is necessary to confirm its translational potential.

Resource availability

Lead contact

Further information and reagent requests may be directed to and will be fulfilled by the lead contact, Jie Cheng (leonardo_cheng@163.com).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •All data required to support the conclusions of this paper are included within the main text and supplemental materials. The raw data of WES and RNA-seq generated in this study could be found in Genome Sequence Archive (https://ngdc.cncb.ac.cn/gsa-human/) with accession ID HRA011280 (https://ngdc.cncb.ac.cn/gsa-human/browse/HRA011280). The public datasets analyzed in this study were obtained from the Gene Expression Omnibus public database under accession number GSE145057, GSE111585, GSE115119, GSE234933, GSE168424, and GSE188737 or Sequence Read Archiv (SRA) public database under accession number PRJNA960652. The TCGA-HNSC cohort data analyzed in this study were obtained from the UCSC Xena browser (https://xena.ucsc.edu/).
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

The graphical abstract and schematics were created with BioRender.com. We thank Dr. Ming Zhang for his guidance in the synthesis of nanoparticles. This work was financially supported, in whole or in part, by the National Natural Science Foundation of China (82403833, 82573322, 82273446, and 82072991), Natural Science Foundation of Jiangsu Province (BK20240517), Key Research Program in Jiangsu Province-Social Developmental Project (BE2020706), Postdoctoral Fellowship Program of China Postdoctoral Science Foundation (GZC20240739), Jiangsu Funding Program for Excellent Postdoctoral Talent (2024ZB381), China Postdoctoral Science Foundation (2024M751492 and 2025T180495), the Natural Science Foundation of the Jiangsu Higher Education Institutions of China (24KJB320010), research program from Health Commission of Jiangsu Province (ZD2022007), Jiangsu Province Capability Improvement Project through Science, Technology and Education-Jiangsu Provincial Research Hospital Cultivation Unit (YJXYYJSDW4), Jiangsu Provincial Medical Innovation Center (CXZX202227), and Postgraduate Research & Practice Innovation Program of Jiangsu Province (SJCX24_0799).

Author contributions

Yuhan Wang, P.D., and P.L. performed all experiments, data collection, and analyses/re-analyses and wrote the manuscript. N.Q. and Z.W. took part in animal experiments and histological and statistical analyses. J.L., Yanling Wang, and D.W. performed patient inclusion, follow-up, and data collection. Y. Wu and J.C. conceived and supervised the whole project. All authors have read and approved the final manuscript.

Declaration of interests

The authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-P-LATS1/2	MedChemExpress	Cat#HY-P81210; RRID: AB_3103276
Anti-LATS	Cell Signaling Technology	Cat#3477; RRID: AB_2133513
Anti-P-YAP	Cell Signaling Technology	Cat#13008; RRID: AB_2650553
Anti-YAP	Cell Signaling Technology	Cat#14074; RRID: AB_2650491
Anti-GAPDH	Bioworld	Cat#MB001; RRID: AB_3073806
Anti-SOX2	Cell Signaling Technology	Cat#3579; RRID: AB_2195767
Anti-EGFR	Cell Signaling Technology	Cat#4267; RRID: AB_2246311
Anti-Ki67	Proteintech	Cat#27309-1-AP; RRID: AB_2756525
Anti-KRT5/6	Proteintech	Cat#68295-1-Ig; RRID: AB_2935375
Anti-TP63	Proteintech	Cat#12143-1-AP; RRID: AB_10597397
Anti-FOSL1	Santa cruz	Cat#sc-376148; RRID: AB_11012022
PE-CD3	Thermo Fisher Scientific	Cat#12-0037-42; RRID: AB_1272078
Anti-γH2A	Proteintech	Cat#29380-1-AP; RRID: AB_3085345
Anti-CDCP1	Abcam	Cat#ab252947; RRID:AB_3720103
Anti-Bcl2	Proteintech	Cat#12789-1-AP; RRID: AB_2227948
Anti-BAX	Proteintech	Cat#50599-2-Ig; RRID: AB_2061561
Anti-CD44	Cell Signaling Technology	Cat#3570; RRID: AB_2076465
Anti-CD133	Cell Signaling Technology	Cat#64326; RRID: AB_2721172
Anti-BMI1	Cell Signaling Technology	Cat#6964; RRID: AB_10828713
Anti-β-catenin	Proteintech	Cat#51067-2-AP; RRID: AB_2086128
Anti-Cyclin D1	Proteintech	Cat#60186-1-Ig; RRID: AB_10793718
Anti-N-cadherin	Cell Signaling Technology	Cat#13116; RRID: AB_2687616
Anti-E-cadherin	Cell Signaling Technology	Cat#14472; RRID: AB_2728770
Anti-Vimentin	Cell Signaling Technology	Cat#5741; RRID: AB_10695459
Anti-FLAG	Proteintech	Cat#66008-3-Ig; RRID: AB_2749837
Anti-Lamin A/C	Proteintech	Cat#10298-1-AP; RRID: AB_2296961
Anti-Involucrin	Proteintech	Cat#28462-1-AP; RRID: AB_2881148
FITC-CD44	Thermo Fisher Scientific	Cat#11-0441-82; RRID: AB_465045
APC-CD133	Thermo Fisher Scientific	Cat#17-1331-81; RRID: AB_823120
Biological samples
Oral squamous cell carcinoma sample	The Affiliated Stomatological Hospital of Nanjing Medical University	N/A
Chemicals, peptides, and recombinant proteins
DMEM/F12 plus 1× GlutaMAX	Thermo Fisher Scientific	Cat#12634010
HEPES (1 M)	Thermo Fisher Scientific	Cat#15630080
Primocin	InvivoGen	Cat#ant-pm-05
DNase I	Aladdin Scientific	Cat#D406460
Collagenase I	Yeasen Biotechnology	Cat#40507ES
N-acetyl-L-cysteine	Sigma	Cat#A9165-5G
TrypLE Express(1×)	Gibco	Cat#12605010
Red Blood Cell Lysis Buffer	Beyotime Biotechnology	Cat#C3702
B27 Serum-Free Supplement	Thermo Fisher Scientific	Cat#17504044
N-2 Supplement	Thermo Fisher Scientific	Cat#17502048
Recombinant Human FGF10	PerproTech	Cat#HY-P7048
Recombinant Human FGF2	PerproTech	Cat#HY-P7330
Recombinant Human R-spondin 1	MedChemExpress	Cat#HY-112484
Noggin	PerproTech	Cat#120-10C-5UG
A83-01	MedChemExpress	Cat#HY-10432
Forskolin	MedChemExpress	Cat#HY-15371
Prostaglandin E2	MedChemExpress	Cat#HY-101952
CHIR99021	MedChemExpress	Cat#HY-10182
Recombinant Human EGF	PerproTech	Cat#AF-100-15
Wnt3a	MedChemExpress	Cat#HY-P70453C
GA-017	MedChemExpress	Cat#HY-147082
Matrigel	Corning	Cat#356231
Cell Recovery Solution	Corning	Cat#354253
Organoid Cryopreservation Medium	BioGenous	Cat#E238023
Y-27632	MedChemExpress	Cat#HY-10071
Cisplatin	MedChemExpress	Cat#HY-17394
Cetuximab	Selleck	Cat#A2000
Docetaxel	MedChemExpress	Cat#HY-B0011
LiCl	MedChemExpress	Cat#HY-Y0649
8-Isopentenylnaringenin (8-PN)	Enzo Life	Cat# ALX-385-025
Critical commercial assays
TUNEL apoptosis detection kit	Vazyme	Cat#A113
CellTiter-Glo® 3D Cell Viability Assay	Promega	Cat#G9681
Deposited data
Data files for RNA-seq and WES	This study	GSA: HRA011280
Oligonucleotides
shRNA sequence for CDCP1 (5′-3′): GCTCTGCCACGAGAAAGCAACATTA	This study	N/A
shRNA sequence for FOSL1 (5′-3′): CTGTACCTTGTATCTCCCTTT	This study	N/A
Control shRNA sequence (5′-3′): TTCTCCGAACGTGTCACGT	This study	N/A
siRNA1 sequence for CDCP1 (5′-3′): UAAUGUUGCUUUCUCGUGGCAGAGC	This study	N/A
siRNA2 sequence for CDCP1 (5′-3′): AUAGAUGAGCGGUUUGCAAUGCUGA	This study	N/A
Control siRNA sequence (5′-3′): UUCUCCGAACGUGUCACGUUU	This study	N/A
Software and algorithms
FlowJo	FlowJo Software	https://www.flowjo.com
GraphPad Prism	GraphPad Software	https://www.graphpad.com
ImageJ	ImageJ Software	https://imagej.net/ij/
RStudio	RStudio Software	https://www.rstudio.com/

Experimental model and study participant details

Sample collection

Fresh tumor tissues or adjacent normal oral mucosal tissues were obtained from 65 patients (age range: 30–82 years; Gender: 44 male, 21 female; ancestry: Asian) with OSCC receiving surgical resection or biopsy at the Affiliated Stomatological Hospital of Nanjing Medical University from 2022 to 2024. All patients had primary OSCC without prior chemotherapy, radiotherapy, or immunotherapy. This study was approved by the Research Ethics Committee of Nanjing Medical University and conducted in accordance with the Declaration of Helsinki. Written informed consent was obtained from all patients.

Mouse model

Female NCG mice (6 weeks) were purchased from GemPharmatech Co., Ltd. and were housed under specific pathogen-free (SPF) conditions with a 12 h light–dark cycle and had access to food and water ad libitum. All experiments were conducted in accordance with the guidelines approved by the Institutional Animal Care and Use Committee of Nanjing Medical University.

For PDOX model generation, digested OSCC organoids (OSCC-18O, OSCC-19O, OSCC-32O) were injected subcutaneously into the right dorsal flank of 6-week-old Female NCG mice (1×10⁶ cells in 100 μL Matrigel per mouse). Tumor size was measured using a digital caliper every week, and tumor volume (mm³) was calculated using the formula: volume = length × width²/2.

For PDX model generation, fresh primary OSCC sample within 30 min of surgical excision were eliminated of uncancerous tissues and tumor necrosis areas, thoroughly rinsed, trimmed into a 1 mm × 1 mm×1mm cube and transplanted subcutaneously into the flanks of 6-week female NCG mice under strict aseptic conditions. The measurable tumors will appear at the transplantation site within 1–3 months, we named it the first generation. When the diameter of the tumor reaches approximately 1 cm, the tumor will be surgical removal and transplanted using the same methodology to generate the second generation and the third generation. When the tumor size reached 100 mm³, mice were randomly divided into groups. Drugs were administered with cisplatin (i.p., 3 mg/kg), cetuximab (i.p., 10 mg/kg), docetaxel (i.p., 10 mg/kg), or saline every three days for a sum of 6 times. After 18 days of observation, mice were sacrificed after treatment, and fresh tumors were harvested for subsequent analysis.

Patient-derived organoids

The tumor samples or adjacent normal oral mucosal tissues were confirmed by two senior pathologists from Affiliated Hospital of Stomatology of Nanjing Medical University. The fresh surgical specimen was obtained less than 30 min and reserved in tissue storage solution (DMEM/F12 supplemented with 1× GlutaMAX, 1× penicillin/streptomycin/amphotericin B, and 10 mM HEPES) for subsequently patient-derived organoid generation. Then, specimen was minced and incubated at 37°C for 800 rpm with the digestion solution (2 mg/mL collagenase I, 1 mg/mL DNase I, 200 U/mL hyaluronidase). After specimen digestion, the mixture was then filtered by a 100 μm cell strainer and the supernatant was collected and centrifugated. Then cell precipitation was disposed by RBC lysis buffer for 2 min at 4°C to remove the red blood cells and then spun for 5 min at 1000 rpm. The cell precipitation was washed twice using cold advanced DMEM/F12 medium and resuspended in 70% Matrigel. Finally, the Matrigel was added into the bottom of 24-well cell plate and incubated for 30 min at 37°C before adding organoid culture medium. The organoid culture medium contained: advanced DMEM/F12 medium supplemented with 1.25 mM N-acetylcysteine, 10 mM Nicotinamide, 1× B27, 50 ng/mL human EGF, 10 ng/mL human FGF-10, 5 ng/mL human FGF2, 250 ng/mL R-spondin1, 100 ng/mL Noggin, 0.5 μM A83-01, 1 μM Forskolin, 1 μM Prostaglandin E2, 3 μM CHIR99021, 10 μM Y-27632, 100 ng/mL Wnt3a, and 5 μM GA-017. The culture medium was routinely changed every 2–3 days and the organoids were passaged every 10–14 days.

Cell lines

OSCC cell lines Cal27 and HN6 were obtained from China Center for Type Culture Collection (CCTCC, Shanghai, China). All cell lines were routinely cultured in DMEM/F12 with 10% FBS in a 37°C humidified incubator containing 5% CO₂. All the cell lines were validated by short tandem repeat (STR) profiling analysis and were free of mycoplasma contamination.

Method details

Western blot

For western blot analysis, organoids were collected and lysed in ice-cold RIPA buffer supplemented with protease inhibitor cocktail. Protein concentration was determined by Bradford method and equal amounts of proteins were loaded and resolved in 5–10% SDS-PAGE gels. PVDF membranes were blocked with 5% (wt/vol) non-fat dry milk in Tris-buffered saline with Tween 20, incubated with indicated primary antibodies followed by HRP-conjugated secondary antibodies, and finally detected using on SuperSignal West Pico reagents.

Hematoxylin eosin (HE), immunohistochemistry (IHC), and immunofluorescence staining

Tissues or organoids were fixed in 4% paraformaldehyde overnight, dehydrated, and embedded in paraffin. Paraffin blocks were partitioned into 4 μm-thick slides. HE staining was conducted using the standard histological protocol. For immunohistochemistry, after sections were made and hydrated, they were incubated with blocking buffer with H₂O₂ for 15 min and boiled with citrate. After cooling down, sections were treated with pre-blocking buffer and incubated with primary antibodies at 4°C overnight. Sections were incubated with secondary antibodies and DAB stained. For immunofluorescence, sections were permeabilized for 3–5 min by Immunostaining Permeabilization Buffer with Triton X-100 (Beyotime) and blocked in QuickBlock Blocking Buffer (Beyotime). Sections were further labeled at 4°C with primary antibodies overnight followed by incubation with fluorescent-dye-conjugated secondary antibodies. Nucleus was counterstained with 4,6-diamidino-2-phenylindole (DAPI). For TUNEL assay, TUNEL apoptosis detection kit (Vazyme) was used according to the instructions’ guidance.

Drug screening and cell viability assay

Organoids were resuspended in Matrigel and embedded in suspension in a 96 well plate (5000 cells per 10mL Matrigel per well). The cells were allowed to culture for 4–5 days. Medium was replaced with fresh culture medium with varying concentrations of the drugs for another 5 days. For chemotherapy drugs treatments, organoids were cultured with drugs at gradient dilution. The chemotherapy drugs include cisplatin (Med Chem Express), Cetuximab (Selleck), and Docetaxel (Selleck). At the end of treatment, the cell viability was determined by CellTiter-Glo assay (Promega) according to the kit protocol. Dose-response curves were fit to the data using the three-parameter logistic regression with variable slope and constraints at 100% and 0% viability using GraphPad Prism. AUC was calculated using the raw experimental data and normalized uniformly by dividing the AUC by the total maximum area a curve could occupy from 0 to 100% viability over the range of drug concentrations analyzed. Quality of drug screens was assessed using Z factor scores, when a Z factor higher than 0.3 indicates a drug screen of good quality. To calculate Z factor:

$$Zfactor=1-\frac{3\times SD\left(negativecontrol\right)+3\times SD\left(positivecontrol\right)}{Average\left(negativecontrol\right)-Average\left(positivecontrol\right)}$$

RNA extraction

Organoids were extracted from Matrigel using Cell Recovery Solution (Corning). Total RNA from tumor tissue and organoids were extracted using TRIzol (Invitrogen) method, followed by isolation and precipitation in chloroform and ethanol. DNA cleanup was performed using DNA Cleanup Kit (Invitrogen). The quality and quantity of the extracted RNA were assessed using a NanoDrop ND-1000 (ThermoFisher Scientific).

Bulk RNA-seq analysis

For RNA-seq, a total amount of 3 μg RNA per sample was used as input material for the RNA sample preparation. RNA library was prepared and followed by paired-end sequencing on an Illumina sequencing platform. RNA-seq Data Analysis: For alignment RNA-seq data, the pair-end raw reads of RNA-seq were trimmed adaptor and aligned to the reference human genome (hg19) by HISAT2 (version 2.2.1). Aligned fragments were assigned to genes by using featureCounts command in subread python package (version 2.0.6). Differentially expressed genes were identified using the limma R package (version 3.60.2). ClusterProfiler R package (version 4.12.0) was used to conduct Gene Set Enrichment Analysis (GSEA). Kaplan-Meier survival analyses were conducted to evaluate the association between gene expression and clinical outcomes using survminer R package (version 0.4.9).

Single-cell RNA-seq reanalysis

To conduct single-cell RNA sequencing analysis, expression matrices were analyzed with Seurat R package (version 3.1.0). Following quality control, the data underwent normalization through variance stabilization, scaling to mitigate technical variations, and multidimensional visualization employing established Seurat pipelines. Dimensionality reduction was conducted using principal component analysis (PCA) and uniform manifold approximation and projection (UMAP).

DNA extraction

Snap-frozen tumor tissues and organoids were lysed in tris-buffered saline solution with 10% SDS and proteinase K (1 mg/mL) overnight at 55°C. DNA was isolated and eluted on spin columns using proprietary solutions provided by a DNA Extraction Kit (Norgen Biotek). 100 to 200 ng of genomic DNA was used for library preparation (Agilent SureSelect Human All Exon v5 Capture Kit). DNA was sequenced using 125-cycle paired-end protocol and multiplexing to obtain 150× coverage on Illumina Hiseq2500 sequencer.

Whole-exome sequencing and analysis

SOAPNuke (version 2.3.0) was used to filter the data. Sequencing reads were mapped against human genome (hg19) by Burrows-Wheeler Aligner (BWA, version 0.7.17). The BAM files were further processed in terms of duplicate marking using samtools (version 1.3.1). The data were further processed for local realignment and base recalibration using Genome Analysis Toolkits (GATK). We used FACETS (version 0.5.14) to detect somatic copy number variations (CNV) in samples and annotated the identified structural variants using Ensembl VEP. Single-nucleotide polymorphisms (SNP) and small insertions/deletions were identified by providing the normal tissues as a reference and their corresponding tumor or organoid sequencing data to MuTect2 (involved in GATK) with default parameters. BAM-matcher was used to assesses sample relatedness by calculating the concordance of SNP.

Co-culture of tumor organoids and T cells

Fresh peripheral blood was collected with informed consent. The peripheral blood mononuclear cells (PBMC) fraction was isolated from peripheral blood by Ficoll-Paque PLUS density gradient separation, counted and cryopreserved for T cell activation assays. To expand and activate Tcells, PBMCs were seeded on an anti-CD28-coated plate together with organoids dissociated into single cells in 20:1 ratio. T cell culture Medium was refreshed every 2 days, which was composed of RPMI 1640 medium supplemented with 1% penicillin/streptomycin, 1% ultraglutamine I, 10% fetal bovine serum and 200 U/mL IL-2. After 2 weeks of expansion, T cells were ready for coculture assays to evaluate tumor killing effect. Organoids for T cell coculture were released from Matrigel by Cell Recovery Solution (Corning) and resuspended in medium. To facilitate visualization, organoids were pre-stained with EGFP (Thermo Fisher Scientific) and T cells with anti-CD3-PE (eBioscience) for 30 min at 37°C. Activated T cells were cocultured with organoids at a ratio of 5:1 for 10 h. Cells were plated in a 96-well, glass-bottom, anti-CD28-coated plate in 200 mL medium containing 100 mL T cell culture medium and 100 mL organoid culture medium. Living cell imaging were performed at 0h and 10 h after cell seeding by living cells workstation (Zeiss Axio Observer 2).

Lentiviral infection and siRNA transfection

Recombinant lentiviruses expressing shRNAs targeting human CDCP1 (5′-GCTCTGCCACGAGAAAGCAACATTA-3′), FOSL1 (5′-CTGTACCTTGTATCTCCCTTT-3′), and negative control (5′-TTCTCCGAACGTGTCACGT-3′) were synthesized and cloned into GV493 (hU6-MCS-CBh-gcGFP-IRES-puromycin; Genechem). Lentiviral vector that expresses human FOSL1 (NM_005438.5) was constructed using the vector plasmid GV492 (pGC-FU-3FLAG-CBh-gcGFP-IRES-puromycin; Genechem). After sequence verification, 293T cells were co-transfected with the transfer plasmid, psPAX2 and pMD2.G using Lipofectamine 2000 (Thermo Fisher Scientific). Viral supernatants were harvested at 48 h and 72 h, filtered (0.45 μm), concentrated by ultracentrifugation (50 000 × g, 2 h, 4°C), aliquoted and stored at −80°C until use.

For infection, the organoids were digested into single-cell suspensions and counted. Transfection was performed at a multiplicity of infection (MOI) of 10, followed by incubation in a 37°C for 2–4 h. After 4 h, the medium was replaced with fresh medium. GFP expression in cells was visualized 72 h later. At 2–3 days, selection was initiated using organoid medium supplemented with puromycin (5 μg/mL). After 10 days of culture, infection efficiency and gene expression levels were analyzed by Western blot and qRT-PCR. For transient siRNA experiments, cells were transfected with siRNA using Lipofectamine 2000 (ThermoFisher Scientific) according to the manufacturer’s protocols.

TOPFlash assay

Cells were transfected with the TOPFlash reporter plasmid (Beyotime) using Lipofectamine 2000 (Invitrogen), together with a pRL Renilla Luciferase plasmid (Promega) as an internal control. After 24 h of transfection, the culture medium was replaced with fresh medium, and cells were subjected to the treatments for an additional 24 h. Luciferase activities were then measured sequentially using the Dual-Luciferase Reporter Assay System (Promega) according to the manufacturer’s instructions.

Preparation and characterization of CS/siCDCP1/PGA NPs

20 μL siCDCP1 solution (20 μM) was gradually added into 1 mL chitosan solution (CS, 0.2 mg/mL in 1% acetic acid) under pH 5.5, rapidly stirred for 1 h. Subsequently, 1 mL of sodium polyglutamate solution (PGA, 0.2 mg/mL) was dripped slowly into the CS-siCDCP1 complexes while maintaining continuous agitation, with stirring conditions for 1 h. The zeta potential and sizes were determined by dynamic light scattering (Malvern Panalytical Zetasizer). The morphology of the NPs was visualized on a Lorenz Transmission Electron Microscope. To detected siRNA release, NPs loaded with siCDCP1-cy5 were prepared as described above. Subsequently, the NPs were dispersed in 1 mL of PBS (pH 7.4) and then transferred to a Float-a-lyzer G2 dialysis device (Spectrum) that was immersed in PBS (pH 6.5) at room temperature. At a predetermined interval, 10 μL of the NP solution was withdrawn and mixed with 20-fold DMSO. The fluorescence intensity of cy5 was determined using a microplate reader. For cell uptake assay, Cal27 and HN6 cells were seeded into 12-well plates and cultured for 24 h. Thereafter, the medium was replaced with fresh medium, and NPs loaded with siCDCP1-cy5 were added at a siRNA concentration of 50 nm, which was the same concentration of RNAi used. Then the cells were harvested for following fluorescence microscope.

siRNA encapsulation efficiency evaluation

To determine the encapsulation efficiency (EE%) of siRNA, Cy5-labeled siCDCP1 was encapsulated into NPs using the method described above. Next, 10 μL of the NP solution was mixed with 20-fold dimethyl sulfoxide (DMSO). As a standard, 10 μL of naked Cy5-siCDCP1 (1 nmol/mL) was mixed with 20-fold DMSO. The fluorescence intensity of both samples was measured using a multimode microplate reader (Tecan SPARK). The encapsulation efficiency was calculated using the formula: EE% = (FlNP/FlStandard) × 100. NPs with an EE% exceeding 80% were selected for subsequent experiments.

Tumorsphere formation assay

3000 cells per well were seeded and cultured in 24 well ultralow-attachment plates (Corning) with serum-free DMEM/F12 media supplemented with B27 supplement (Gibco), N2 supplement (Gibco), 20 ng/mL recombinant human-derived EGF (PeproTech) and 10 ng/mL bFGF (PeproTech) for 3–5 days. Tumorspheres with diameters larger than 50 μm were counted under a microscope.

Flow cytometry

Cells were trypsinized with EDTA-free trypsin (Gbico) and counted. Single-cell suspension with equivalent cell numbers was Incubated with FITC Anti-Human CD44 (ThermoFisher Scientific) and APC Anti-Human CD133 (ThermoFisher Scientific) for 30 min at room temperature. After washed twice with stain buffer, cells were detected by BD FACSymphony A5 instrument. Data was analyzed using FlowJo.

In vivo limiting dilution tumorigenicity assay

HN6 cells stably expressing CDCP1 shRNA (HN6-shCDCP1) or control shRNA (HN6-shNC) were harvested, and resuspended in serum-free medium mixed with Matrigel (1:1). Cells at four different concentrations (5×10⁵, 1×10⁵, 1×10⁴, and 1×10³ cells) were subcutaneously injected into the flank of 6-week-old female BALB/c nude mice. Three weeks post-inoculation, tumors were surgically excised and tumor incidence was recorded. Data were analyzed by the Extreme Limiting Dilution Analysis (ELDA) software (http://bioinf.wehi.edu.au/software/elda/).

Hemolysis analysis

Approximately 0.5–1 mL of fresh anticoagulated mouse orbital venous blood was collected and centrifuged at 5000 rpm for 10 min at 4°C. The supernatant was discarded, and the erythrocyte pellet was washed five times with physiological saline to obtain a 2% (v/v) red blood cell (RBC) suspension. Saline and 0.1% Triton X-100 solution were used as negative and positive controls, respectively. Equal volumes of the RBC suspension were mixed with various concentrations of the test nanomaterial solution, gently vortexed to ensure homogeneity, and incubated at 37°C for 3 h. After incubation, the samples were centrifuged at 5000 rpm for 10 min at 4°C, and 100 μL of the supernatant from each sample was transferred to a 96-well plate. The absorbance of the supernatant was measured at 540 nm using a microplate reader. The hemolysis rate (%) was calculated according to the formula:

$$\text{Hemolysis}\text{rate}(\%)=\frac{A_{\text{sample}}-A_{\text{negative}}}{A_{\text{positive}}-A_{\text{negative}}}\times 100\%$$

Nanodrug treatment in OSCC PDX models

Two independent cisplatin-resistant PDX models (PDX: OSCC-23T and PDX: OSCC-42T) were established by implanting tumors into the flanks of 6-week-old female NCG mice as described previously. When tumor volumes reached approximately 100 mm³, the mice were randomized into treatment groups (n = 6 per group). All groups received one of the following treatments every three days for a total of six doses: cisplatin (3 mg/kg, i.p.), NPs(+) (siCDCP1-loaded nanoparticles, 10 nM, i.v.), a combination of cisplatin and NPs(+), or saline. To specifically assess the impact of the nanocarrier itself and its potential synergy with cisplatin, the PDX: OSCC-42T model included two additional control groups: NPs(−) (negative control siRNA-loaded nanoparticles; 10 nM, i.v.) or a combination of cisplatin and NPs(−). After 18 days of treatment, all mice were euthanized, and samples of tumors, peripheral blood, and major organs were collected for subsequent analysis.

Hematological and serum biochemical analysis

Blood samples were collected from the mouse orbital venous and divided into EDTA-anticoagulated and non-anticoagulated tubes. Whole blood was used for routine hematological analysis, including red blood cell (RBC), white blood cell (WBC), and platelet (PLT) counts. Serum was obtained by centrifugation of clotted blood and analyzed using an automated biochemical analyzer for various indicators: aminotransferase (ALT), aspartate aminotransferase (AST), UREA, Creatinine (CREA), and creatine kinase (CK) following the manufacturer’s protocols.

Quantification and statistical analysis

Data analyses were performed by GraphPad Prism 10 or R software. The detailed statistical tests were indicated in figures or associated legends where applicable. Unpaired two-tailed Student’s t-tests were used to compare two groups. Differences among three or more groups were compared by Analysis of variance (ANOVA). The results are presented as mean ± SD (standard deviation), and p value less than 0.05 was considered significant.

Published: February 17, 2026