ANGPTL4 contributes to osimertinib resistance through NDRG1-associated extracellular matrix remodeling in EGFR-mutant NSCLC organoid models

Abstract

Background

As a first-line treatment for EGFR-mutated non-small cell lung cancer (NSCLC), acquired resistance to osimertinib has become a major clinical challenge. Traditional two-dimensional cell culture models have significant limitations in simulating the tumor microenvironment, making it difficult to fully elucidate complex resistance mechanisms. In recent years, three-dimensional organoid models have emerged as a crucial platform for studying resistance mechanisms due to their superior ability to preserve tumor tissue architecture and heterogeneity. Angiopoietin-like protein 4 (ANGPTL4), a secreted glycoprotein, has been implicated in regulating extracellular matrix remodeling and is associated with chemotherapy resistance in multiple cancers. However, its specific role in osimertinib resistance remains unclear. Therefore, this study investigates the mechanism of ANGPTL4 in lung cancer resistance using organoid models.

Methods

This study established a three-dimensional osimertinib-resistant organoid model that highly mimics the tumor microenvironment. Transcriptome sequencing and bioinformatics analysis were used to screen key resistance mediators. Functional gain-of-function/loss-of-function experiments validated candidate gene functions. Immunoprecipitation, molecular docking, and single-cell RNA sequencing data were employed to explore the underlying mechanisms in depth. Finally, virtual screening was conducted to identify potential inhibitors targeting this resistance pathway.

Results

This study revealed that ANGPTL4 was upregulated in osimertinib-resistant models and functionally contributed to reduced osimertinib sensitivity. Mechanistically, ANGPTL4 interacted with NDRG1 and was associated with activation of extracellular matrix-related programs. Knockdown of ANGPTL4 restored osimertinib sensitivity, whereas ANGPTL4 overexpression promoted a resistant phenotype. Additional validation in H1975-derived models further supported the reproducibility of ANGPTL4-associated resistance. Knockdown of ANGPTL4 in resistant models restored drug sensitivity, whereas overexpression induced resistance. NDRG1 is an essential molecule for ANGPTL4-regulated ECM pathway activation, exhibiting significant colocalization with ANGPTL4 in clinical samples. Knocking down NDRG1 in ANGPTL4-overexpressing drug-resistant models effectively restored drug sensitivity. Furthermore, virtual screening of 2.08 million small molecules yielded five lead compounds with high binding affinity for ANGPTL4.

Conclusions

This study supports ANGPTL4 as an important contributor to osimertinib resistance in EGFR-mutant NSCLC and suggests that ANGPTL4/NDRG1-associated ECM remodeling may represent a relevant resistance-associated pathway. Computational virtual screening further nominated candidate ANGPTL4-binding molecules, which require future biochemical and functional validation. It not only provides novel insights into resistance mechanisms but also identifies potential therapeutic candidates for overcoming resistance, highlighting the significant value of 3D organoid models in investigating resistance mechanism.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12935-026-04400-5.

Introduction

The worldwide prevalence of epidermal growth factor receptor (EGFR) mutations in non-small cell lung cancer (NSCLC) is approximately 10–15% [1–3], with rates reaching up to 50% in Asian populations [4]. Osimertinib has emerged as the first-line therapeutic option for patients with EGFR-mutant NSCLC, demonstrating significant improvements in survival; however, acquired resistance remains a serious challenge, with approximately 50% of patients experiencing disease progression within 20 months [5]. The mechanisms of drug resistance are complex and diverse, among which known mechanisms such as MET amplification (30% -50%) [6] and C797S mutation (10% -26%) [7] can only explain some cases, and the driving factors of drug resistance in nearly half of patients are still unclear [8, 9]. After drug resistance, the prognosis of patients deteriorates sharply, and the effectiveness of existing treatment plans is limited. The objective remission rate of chemotherapy is only about 30% [10, 11], and there is an urgent need to explore new drug resistance mechanisms and effective treatment strategies.

Angiopoietin-like protein 4 (ANGPTL4) is a secreted glycoprotein that It binds specifically to the extracellular matrix (ECM) via a heparin/heparan sulfate proteoglycan-dependent mechanism [12, 13]. Concurrently, it plays multiple roles in the initiation and progression of cancer [14], previous studies have found that abnormal expression of ANGPTL4 has been implicated in the initiation and progression of NSCLC [14]. Research indicates that ANGPTL4 contributes to the malignant progression of lung cancer by promoting tumor cell proliferation, invasion, and metastasis [15–17]. Within lung cancer research, the ECM has been identified as a key driver of acquired resistance to EGFR tyrosine kinase inhibitors [18]. Recent evidence indicates that the ECM can modulate tumor cell resistance to EGFR TKIs in a dose-dependent and reversible manner through its core component collagen and its corresponding receptor integrin-β1 [1]. Based on this, we hypothesize that ANGPTL4 may play a crucial role in osimertinib resistance in non-small cell lung cancer.

Conventional drug screening methods using cell lines often fail to identify latent drug resistance targets, as their simplified conditions may not adequately replicate key in vivo features, such as the complex tumor architecture and the microenvironment [19]. Organoid models serve as a promising platform for uncovering drug-resistance pathways by recapitulating the dynamic evolution of intratumoral heterogeneity under treatment pressure and mirroring the multicellular crosstalk and interactions [19–23]. Based on this, we used organoids to investigate the mechanism of ANGPTL4 in the resistance of osimertinib.

This study constructed an organoid model of Osimertinib resistance in lung cancer, and systematically explored the core driving factors and molecular mechanisms of Osimertinib resistance by combining multi omics analysis and clinical data. In this study, we used lung cancer organoid models combined with transcriptomic, single-cell, biochemical, and pharmacological analyses to investigate the contribution of ANGPTL4 to osimertinib resistance. We further validated ANGPTL4-associated resistance in an additional H1975-derived model and explored its relationship with NDRG1-associated ECM remodeling. Finally, virtual screening was performed to nominate candidate ANGPTL4-binding molecules for future experimental validation.

Methods and materials

Cell culture and cell line construction

The human non-small cell lung cancer PC9 cell line (EGFR 19-del mutation) was purchased from the EnzymeResearch Biotechnology Co., Ltd., Shanghai, China. Cells were cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin at 37 °C in a 5% CO₂ incubator.

To establish an osimertinib-resistant cell line (PC9-OSI), a stepwise drug concentration escalation method was employed. PC9 cells were initially treated with a low concentration (10 nM) of osimertinib. After cell growth resumed, the drug concentration was progressively increased (up to 1 µM) over a period exceeding 6 months, ultimately yielding a resistant cell line capable of stable growth in 1 µM osimertinib. The human NSCLC H1975 cell line, harboring EGFR L858R/T790M mutations, was obtained from American Type Culture Collection (ATCC) and cultured in RPMI-1640 medium supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin at 37 °C in a 5% CO₂ incubator. To generate osimertinib-resistant H1975 cells (H1975-OSI), parental H1975 cells were exposed to gradually increasing concentrations of osimertinib using a stepwise dose-escalation strategy, as described for PC9 cells. The final resistant cells were maintained in 0.5 µM osimertinib and cultured in drug-free medium for 3 days before functional assays.

Stable cell lines were established using a lentiviral transfection system. shRNA targeting ANGPTL4 (shANGPTL4), overexpression plasmids (OE-ANGPTL4), and corresponding negative controls (shNC, VectOSI) were synthesized and packaged by Gema Gene Company. Viral infection was performed according to the manufacturer’s instructions, followed by selection with puromycin to obtain stable knockdown or overexpression cell lines.

Establishment of 3D lung cancer organoid model

Cells in the logarithmic growth phase were digested with 0.25% trypsin-EDTA, then counted and assessed for viability using the trypan blue staining method to ensure viability exceeded 95%. Next, resuspend cells in pre-chilled Matrigel, adjusting cell density to 1 × 10⁶ cells/mL with a final Matrigel concentration of 70%. Using pre-chilled pipette tips, precisely dispense 50µL/well of the cell-Matrigel mixture into the center of a 24-well plate. Incubate at 37 °C for 15–20 min to allow complete solidification. Finally, slowly add 500 µL of organoid-specific medium (Advanced DMEM/F12 basal medium supplemented with 1× B27 Supplement, 20 ng/mL EGF, and 10 ng/mL FGF) along the well walls. Incubate at 37 °C in a 5% CO₂ incubator, maintaining a medium change every 3 days. Regular microscopic observation revealed that after 7–10 days of culture, typical three-dimensional structures with diameters of 100–200 μm formed with a formation rate exceeding 90%. Survival rates after 7 days of culture exceeded 85%. Furthermore, this model maintained stable morphological characteristics through at least 5 stable passages.

Transcriptome sequencing and bioinformatics analysis

Collect PC9 sensitive/resistant cells (2D and 3D models) and PC9-OSI cells treated with shANGPTL4 or controls. Extract total RNA using TRIzol reagent. After sample quality control approval, entrust BGI to perform mRNA library construction and sequencing on the Illumina NovaSeq platform.

Raw data underwent quality control and alignment (reference genome GRCh38), followed by gene counting using featureCounts. Differential expression analysis was performed with the DESeq2 R package, setting screening criteria as |log2(Fold Change)| > 1 and adjusted p-value (adj. p-val) < 0.05. Differentially expressed genes underwent GO functional annotation and KEGG pathway enrichment analysis (clusterProfiler R package).

Single-cell transcriptomic data were obtained from the integrated NSCLC single-cell dataset and processed using the Seurat R package. Malignant epithelial cells were stratified into ANGPTL4-high and ANGPTL4-low groups according to the median ANGPTL4 expression level. To evaluate ECM pathway activity, the KEGG ECM-receptor interaction gene set, KEGG_ECM_RECEPTOR_INTERACTION, was used as the predefined ECM-related signature. Gene symbols in this pathway were matched to the single-cell expression matrix, and genes not detected in the dataset were removed before downstream analysis. Pathway activity scoring was performed using the AUCell R package. In addition, differential expression analysis between ANGPTL4-high and ANGPTL4-low malignant cells was followed by functional enrichment analysis to assess whether ECM-related pathways were enriched in ANGPTL4-high malignant cells.

Real-time fluorescent quantitative PCR

Total cellular RNA was extracted using the TRIzol method and reverse transcribed into cDNA. GAPDH was used as the housekeeping gene, and qPCR reactions were performed on the QuantStudio 6 Flex system using SYBR Green premix. Gene expression levels were calculated using the 2^(-ΔΔCt) method.

Knockdown and overexpression experiments

Gene manipulation was performed using a lentiviral vector system. Three specific shRNAs targeting ANGPTL4 and a corresponding scramble control shRNA were designed and synthesized. In parallel, an ANGPTL4 overexpression vector and an empty vector control were constructed. The recombinant lentiviral plasmids were then packaged into lentiviral particles by GeneChem (Shanghai, China).

For functional validation, ANGPTL4 was overexpressed in parental PC9 and H1975 cells/organoids, while ANGPTL4 was knocked down in osimertinib-resistant PC9-OSI and H1975-OSI cells/organoids. Briefly, cells/organoids were infected with lentiviruses carrying the ANGPTL4 overexpression construct, shANGPTL4, or the respective negative controls. Stable lines were established by puromycin selection for two weeks and subsequently used for drug sensitivity and Western blotting assays.

Western blotting analysis

Total cellular proteins were extracted using RIPA lysis buffer, and protein concentrations were determined by the BCA method. Equal amounts of protein were separated by SDS-PAGE and transferred to PVDF membranes. After blocking with 5% nonfat milk, membranes were incubated overnight at 4 °C with primary antibodies. Primary antibodies used included: anti-ANGPTL4 (1:1000, Abcam), anti-NDRG1 (1:1500, CST), anti-Collagen I (1:1000, Proteintech), anti-LAMC1 (1:1000, Santa Cruz), and anti-GAPDH (1:5000, Proteintech). Subsequently, incubation with the corresponding HRP-labeled secondary antibody was performed at room temperature for 1 h. Images were developed using an ECL chemiluminescence kit and captured with a Bio-Rad imaging system.

Immunohistochemistry

Paraffin sections were processed through dewaxing, hydration, and antigen retrieval. Endogenous peroxidase was blocked, followed by overnight incubation with the primary antibody at 4 °C. The HRP-labeled secondary antibody was incubated at room temperature for 1 h. DAB staining was performed, followed by hematoxylin counterstaining and neutral resin mounting.

Multiple immunofluorescence staining

Multiplex staining was performed using tyramine signal amplification technology. The procedure involved sequential steps of antigen retrieval, primary antibody incubation, HRP-conjugated secondary antibody incubation, Opal fluorescent dye signal amplification, and microwave antibody stripping. Multiple target labels were cycled through this process, followed by DAPI counterstaining and final sealing with an anti-fluorescence quenching mounting medium. Images were observed and captured using a confocal microscope.

Co-immunoprecipitation

Collect cells and lyse them using NP-40 lysis buffer. Reserve a portion of the lysate as the Input control. Incubate the remaining lysate overnight at 4 °C with slow rotation using anti-ANGPTL4 antibody or isotype IgG control antibody. Subsequently, add Protein A/G agarose beads and continue incubation for 4 h. Centrifuge to collect beads, wash five times with pre-chilled lysis buffer, then add 1× SDS loading buffer and boil before proceeding to Western blot analysis.

Cell viability detection (drug sensitivity test)

Cell viability was assessed using the Cell Counting Kit-8 (CCK-8). For H1975-derived validation experiments, parental H1975-vector, H1975‑OE‑ANGPTL4, H1975‑OSI‑shNC, and H1975‑OSI‑shANGPTL4 cells were seeded at a density of 3,000–5,000 cells per well in 96‑well plates. After attachment, cells were treated with serial concentrations of osimertinib or corresponding concentrations of DMSO (control), and a blank control group was included. After 72 h of drug treatment, 10 µL of CCK‑8 solution was added to each well and incubated for an additional 2–4 h. Absorbance at 450 nm was measured using a microplate reader. Cell viability (%) = (Experimental group OD − Blank group OD) / (Control group OD − Blank group OD) × 100%. IC₅₀ values were calculated by nonlinear regression using GraphPad Prism 9.

Molecular docking and virtual screening

The three-dimensional structure of the ANGPTL4 protein was obtained from the Protein Data Bank or constructed via homology modeling. The small-molecule ligand library was sourced from the ZINC20 database. Molecular docking was performed using AutoDock Vina software. The protein underwent preprocessing steps including hydrogenation and charge calculation, and a docking box encompassing potential active sites was defined. For the top candidate molecules selected, more refined molecular dynamics simulations were conducted to evaluate binding mode stability and binding free energy.

Molecular dynamics simulation

The simulation system was constructed using the OpenMM toolkit (version 8.0): protein-ligand complexes were parameterized with the AMBER14 force field, the solution environment was modeled with the TIP3P water model, and physiologically concentrated ions were added to maintain overall charge neutrality. Under custom Python script control, the following steps were performed sequentially: energy minimization, NVT ensemble temperature ramping (to 310 K), and NPT ensemble equilibration. A 10 ns production simulation was then executed, employing a Langevin integrator for temperature control and Monte Carlo methods for pressure regulation. Trajectory analysis was performed entirely using the MDAnalysis library (version 2.4.2), including: - Calculating root-mean-square deviation (RMSD) to assess stability of the protein-ligand complex - Analyzing residue root-mean-square fluctuations to examine conformational changes - Quantifying hydrogen bond occupancy and hydrophobic contact area to evaluate interaction strength - Performing MM-GBSA binding free energy calculations and energy decomposition analysis based on trajectory data.

Statistical analysis

All in vitro experiments were independently replicated at least three times. Data are expressed as mean ± standard deviation. Statistical analysis was performed using GraphPad Prism 9 and R software. Comparisons between two groups were conducted using Student’s t-test, while comparisons among multiple groups were performed using one-way analysis of variance (ANOVA). Survival analysis employed the Kaplan-Meier method and log-rank test. Correlation analysis employed Pearson or Spearman correlation coefficients. Differences were considered statistically significant at p < 0.05.

Results

Screening and identification of ANGPTL4 as a resistance-associated candidate in 3D organoid model

In order to overcome the limitations of traditional two-dimensional culture in simulating the tumor microenvironment, we first successfully constructed a patient-derived and cell-derived non-small cell lung cancer 3D organoid model. This system adopts matrix gel embedding culture method. In a special medium containing growth factors, primary lung cancer cells or tumor tissue fragments can self-organize to form a typical gland-like three-dimensional structure within 0–7 days ( Fig. 1A ). Compared with the traditional 2D monolayer culture, lung cancer organoids are closer to the original tumor in histological features, showing complete chamber structure and cell polarity, while retaining the expression profile of key molecular markers of the original tumor.

Fig. 1.

Identification of ANGPTL4 as a key resistance related gene based on 3D organoid model screening. (A) Cultivation of 3D organoids for lung cancer. The microscopic images of PC9 cell lung cancer organoids (CON) and PC9 osimertinib resistant lung cancer organoids (OSI) at 0.2.4.6 days were presented separately (B) Sensitivity of PC9 (A) and PC9-OSI (B) to osimertinib under 2D and 3D culture conditions C. Characterization of PC9 and PC9-OSI lung cancer cells under microscopy using H&E D. Differential gene volcano plot between 2D cultured PC9 and 2D cultured PC9-OSI cells E. Differential gene volcano map between 3D cultured PC9 and 3D cultured PC9-OSI cells F. Differential gene volcano plot between 3D cultured PC9-OSI and 2D cultured PC9-OSI cells G. Intersection of the three Venn diagram H. Consistently upregulated genes in this intersection I. Intersection difference gene String protein mapping J. Prediction of Osimertinib Sensitivity and ANGPTL4 Expression Scatter Plot in GSE130160 Dataset K. The relationship between clinical prognosis and ANGPTL4 in TCGA database (median differentiation). Statistical significance was determined using Student’s t-test or one-way ANOVA, as appropriate. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

Through functional validation, we observed that lung cancer organoids exhibit drug response characteristics more consistent with clinical practice. 3D organoids demonstrated significantly enhanced resistance (Fig. 1B), with IC50 values exceeding those of 2D cultured cells by more than threefold. The stability of this drug-resistant phenotype was maintained during continuous passage and could be repeatedly verified in organoid models from different patient sources.

Further histological analysis showed that there were obvious structural changes in drug-resistant organs, including increased deposition of basement membrane components and disordered cell spatial arrangement ( Fig. 1C ), which were highly similar to the pathological changes of tumor tissues in clinical drug-resistant patients [24]. Additionally, proteomic analysis of organoid culture supernatants detected abnormal secretion of multiple extracellular matrix proteins, indicating that the 3D culture system successfully recreated key molecular interactions within the tumor microenvironment.

To systematically identify the core molecules driving osimertinib resistance in conditions closer to the in vivo environment, we conducted multiple comparative transcriptomic analyses. Specifically, we established the following three groups of comparisons: 2D drug resistance mechanism profile: 2D-cultured PC9 osimertinib-resistant cells vs. 2D-cultured PC9 parental sensitive cells (Fig. 1D). 3D in vivo-mimicking drug resistance mechanism profile: 3D-cultured PC9 osimertinib-resistant organoids vs. 3D-cultured PC9 sensitive organoids (Fig. 1E).Microenvironment-Influenced Resistance Mechanism Profile: 3D-cultured PC9 osimertinib-resistant organoids vs. 2D-cultured PC9 osimertinib-resistant cells (Fig. 1F). aiming to isolate resistance-associated genes induced or amplified by the 3D microenvironment itself.

By performing a Venn diagram analysis on these three sets of differentially expressed genes, we obtained the intersection of the three gene sets (Fig. 1G). We focused specifically on genes consistently upregulated in this intersection (Fig. 1H), as these are most likely core resistance drivers stably functioning across different microenvironments, rather than noise attributable to specific culture conditions. To identify the most clinically valuable and functionally promising targets from this candidate gene pool, we implemented a rigorous two-step screening process. First, we input the identified differentially expressed genes into STRING for protein interaction network analysis, identifying key protein molecules (Fig. 1I). The second step involved drug sensitivity prediction screening. We further employed computational biology tools like OncoPredict to analyze correlations between candidate gene expression levels and cancer cell sensitivity to osimertinib (IC50)(GSE130160) (Fig. 1J) [25]. Concurrently, we utilized public databases such as TCGA for clinical prognostic association screening, performing survival analyses on lung cancer patient data to identify genes whose high expression was significantly associated with reduced overall survival (Fig. 1K) [26]. Through this stepwise screening process, ANGPTL4 emerged as a resistance-associated candidate. Elevated ANGPTL4 expression was positively correlated with osimertinib resistance in our internal dataset and public datasets and was associated with poorer clinical outcome. Therefore, ANGPTL4 was selected for subsequent functional and mechanistic validation.

Functional validation of the association between ANGPTL4 and osimertinib resistance

To confirm the critical role of ANGPTL4 in Osimertinib resistance across multiple dimensions, we conducted a systematic validation series spanning in cellular, organoid, single-cell, and clinical sample analyses.

We first detected ANGPTL4 expression in osimertinib-resistant PC9 cells (PC9-OSI) and their parental sensitive counterparts. Real-time quantitative PCR revealed that ANGPTL4 mRNA transcription levels were significantly elevated by approximately 15-fold in PC9-OSI cells compared to sensitive cells (Fig. 2A). Consistent with this, Western Blotting confirmed strong enrichment of ANGPTL4 at the protein level (Fig. 2B). This finding preliminarily establishes a direct molecular link between ANGPTL4 and acquired resistance.

Fig. 2.

ANGPTL4 expression and functional validation in PC9- and H1975-derived osimertinib-resistant models (A) ANGPTL4 mRNA expression in parental PC9 and osimertinib-resistant PC9-OSI organoids. (B) ANGPTL4 protein expression in parental PC9 and PC9-OSI organoids, as determined by Western blotting and densitometric quantification. (C) Representative ANGPTL4 staining in paired osimertinib-sensitive/pre-treatment and osimertinib-resistant/progressive NSCLC tissue samples. (D) UMAP visualization of ANGPTL4 expression in integrated lung cancer single-cell transcriptomic data. (E) PERCEPTION-predicted osimertinib resistance scores in ANGPTL4-high and ANGPTL4-low malignant cells. (F) Correlation between sample-level ANGPTL4 expression and predicted osimertinib resistance scores. G–H. Validation of ANGPTL4 overexpression in parental PC9 organoids by Western blotting and qPCR. I. Dose-response curves and IC50 analysis showing that ANGPTL4 overexpression decreased osimertinib sensitivity in PC9 organoids. J–K. Validation of ANGPTL4 knockdown in PC9-OSI organoids by Western blotting and qPCR. L. Dose-response curves and IC50 analysis showing that ANGPTL4 knockdown partially restored osimertinib sensitivity in PC9-OSI organoids. M. ANGPTL4 expression in parental H1975 and H1975-OSI cells. N. Validation of ANGPTL4 overexpression in H1975 cells and ANGPTL4 knockdown in H1975-OSI cells. O. Dose-response curves and IC50 analysis showing that ANGPTL4 knockdown reduced osimertinib resistance in H1975-OSI cells. P. Dose-response curves and IC50 analysis showing that ANGPTL4 overexpression increased osimertinib resistance in H1975 cells. Data are presented as mean ± SD from three independent experiments. Statistical significance was determined using Student’s t-test or one-way ANOVA, as appropriate. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

To translate our discovery to clinical relevance, we collected a cohort of paired non-small cell lung cancer (NSCLC) patient tissue samples, including tumors from before osimertinib treatment (sensitive) and after disease progression. Through immunohistochemical staining and semi-quantitative scoring analysis, we observed significantly elevated ANGPTL4 protein expression in the vast majority of patient tissues exhibiting drug resistance (Fig. 2C). In contrast, ANGPTL4 expression was generally weak or negative in pre-treatment or sensitive-phase tissue samples. This strongly demonstrates that high ANGPTL4 expression is closely associated with osimertinib resistance in clinical patients.

To explore the role of ANGPTL4 at a finer cellular resolution, we integrated and analyzed six independent single-cell transcriptomic datasets from non-small cell lung cancer [27]. Cell clustering analysis revealed that ANGPTL4 is not uniformly expressed across all tumor cells but is specifically enriched in a distinct malignant cell subpopulation (Fig. 2D). Notably, resistance scoring predictions for this subpopulation using the PERCEPTION algorithm (a tool for predicting therapeutic response based on single-cell transcriptomics) [28] revealed that these ANGPTL4 + cells also exhibited elevated resistance scores (Fig. 2E). Integrating the predicted osimertinib resistance scores at the sample level with ANGPTL4 expression revealed a positive correlation between the two (Fig. 2F). These findings provide robust evidence at the single-cell level that ANGPTL4-high tumor cell subpopulations are key drivers of clinical resistance and disease progression.

We next examined the functional contribution of ANGPTL4 to osimertinib resistance using gain- and loss-of-function experiments. In parental PC9 organoids, ANGPTL4 overexpression increased ANGPTL4 expression and reduced osimertinib sensitivity, as reflected by an increased IC50 value (Fig. 2G–I). Conversely, in PC9-OSI resistant organoids, shRNA-mediated ANGPTL4 knockdown partially restored osimertinib sensitivity and decreased the IC50 value (Fig. 2J–L).

To determine whether this ANGPTL4-associated resistant phenotype was restricted to the PC9 background, we further performed validation in H1975-derived models. ANGPTL4 expression was increased in H1975-OSI organoids compared with parental H1975 organoids (Fig. 2M). ANGPTL4 overexpression in H1975 organoids increased the osimertinib IC50, whereas ANGPTL4 knockdown in H1975-OSI organoids decreased the IC50 and partially restored osimertinib sensitivity (Fig. 2N–P). These findings provide additional support that ANGPTL4-associated osimertinib resistance can be observed in more than one EGFR-mutant NSCLC model.

ANGPTL4 promotes ECM remodeling in osimertinib-resistant organoids

To investigate the mechanism of ANGPTL4 in osimertinib resistance in non-small cell lung cancer, we first performed whole-transcriptome sequencing on PC9 osimertinib-resistant organoids stably knocked down for ANGPTL4 (shANGPTL4) and their controls (Supplementary Fig. 3A). Differential expression analysis revealed that hundreds of genes exhibited significantly altered expression following ANGPTL4 knockdown compared to the control group (set threshold: |log2FC| > 1, adj. p < 0.05) (Fig. 3A).To interpret the biological functions of these differentially expressed genes, we performed GO functional annotation and KEGG pathway enrichment analysis. Results consistently showed that downregulated genes were significantly enriched in cellular components and biological processes such as “extracellular matrix organization” and “collagen fiber organization,” as well as key signaling pathways including “ECM-receptor interactions” and “adhesion spots” (Fig. 3B-C). These findings suggest that ANGPTL4 is closely associated with ECM pathway remodeling in resistant organoids.

Fig. 3.

ANGPTL4 mediates osimertinib resistance by activating the ECM pathway. A Volcano plot of PC9-OSI organoids with ANGPTL4 knockdown versus control transcriptome B KEGG enrichment plot of differentially expressed genes in PC9-OSI organoids after ANGPTL4 knockdown versus control transcriptome C ECM pathway GSEA of differentially expressed genes in PC9-OSI organoids after ANGPTL4 knockdown versus control transcriptome D ECM pathway activation in ANGPTL4-high and ANGPTL4-low subclusters from integrated lung cancer single-cell data (median-based clustering) E ECM pathway activation following simulated ANGPTL4 knockout in integrated lung cancer single-cell data F mRNA expression of ECM pathway markers (LAMC1 and Collagen I) in PC9 organoids and PC9-OSI organoids G-H Protein expression of ECM pathway markers (LAMC1 and Collagen I) in PC9 organoids and PC9-OSI I Multiplex immunofluorescence staining of lung cancer sections from osimertinib-resistant and sensitive patients J organoids Protein expression of ECM pathway markers (LAMC1 and Collagen I) after ANGPTL4 knockdown(sh) or overexpression (OE)in PC9-OSI organoids K. Pyrintegrin rescued osimertinib resistance in ANGPTL4-knockdown PC9-OSI organoids. L. Cilengitide blocked ANGPTL4-overexpression-induced osimertinib resistance in PC9 organoids. Statistical significance was determined using Student’s t-test or one-way ANOVA, as appropriate. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

To verify these findings at the single-cell level, we conducted an in-depth analysis using integrated single-cell transcriptomic data from lung cancer. To verify these findings at the single-cell level, we conducted an in-depth analysis using integrated single-cell transcriptomic data from lung cancer. The KEGG ECM-receptor interaction pathway gene set, KEGG_ECM_RECEPTOR_INTERACTION, was used as the predefined ECM-related signature. Using the AUCell algorithm, we calculated ECM-receptor interaction pathway activity scores for malignant cells. The results showed that ANGPTL4-high malignant cells exhibited significantly higher ECM-receptor interaction pathway activity than ANGPTL4-low malignant cells (Fig. 3D). Consistently, differential expression followed by functional enrichment analysis also showed enrichment of ECM-related biological processes and pathways in ANGPTL4-high malignant cells, including extracellular matrix organization, collagen-containing extracellular matrix, and KEGG ECM-receptor interaction (Supplementary Figure S2B-D) [29]. Results revealed significantly higher ECM pathway activity scores in tumor cell subpopulations expressing high levels of ANGPTL4 (Fig. 3D). More importantly, we used the gene perturbation analysis algorithm, scTenifoldKnk, to perform ' in vitro simulated knockout ' of ANGPTL4 at the single-cell level [30]. It was predicted that knocking out ANGPTL4 would significantly suppress the activity of ECM-related pathways (Fig. 3E). These computational biology verifications on independent single-cell data sets are consistent with our in vitro sequencing results, which greatly enhance the reliability of the conclusion that ANGPTL4 regulates the ECM pathway.

Based on the results of omics analysis, we verified the key markers of the ECM pathway at the molecular level. Through Western Blot and qPCR detection, we found that the mRNA and protein expression levels of core ECM components Collagen I and laminin γ1 ( LAMC1 ) [31] in PC9 drug-resistant organoids were significantly higher than those in sensitive cells ( Fig. 3F-H ). Following ANGPTL4 knockdown, expression of both key proteins was effectively downregulated (Fig. 3J), whereas overexpression conversely enhanced Collagen I and LAMC1 expression (Fig. 3J).

Consistently, in H1975-derived validation models, ANGPTL4 overexpression increased the expression of ECM-related markers, including Collagen I and LAMC1, whereas ANGPTL4 knockdown in H1975-OSI cells reduced their expression (Supplementary Fig. 2E). These results further support the association between ANGPTL4 and ECM remodeling across EGFR-mutant NSCLC models.

To establish clinical relevance, we performed immunohistochemical staining on previously collected patient tissue sections. Results confirmed markedly increased deposition of Collagen I and LAMC1 in osimertinib-resistant patient tumors, positively correlated with ANGPTL4 expression patterns(Fig. 3I). This confirmed the existence of the ANGPTL4-ECM axis from the clinical sample level.

To further test whether ECM pathway activation functionally contributes to ANGPTL4-associated osimertinib resistance, we performed rescue experiments. In ANGPTL4-knockdown drug-resistant organoids, we introduced an ECM pathway activator (Pyrintegrin) [32]. Results showed that exogenously activating the ECM pathway reversed the osimertinib sensitization effect caused by ANGPTL4 knockdown, causing the cells’ IC50 values to rise again (Fig. 3K).

In sensitive organoids overexpressing ANGPTL4, we used an ECM pathway inhibitor (Cilengitid) [33]. Results demonstrated that inhibiting the ECM pathway effectively blocked ANGPTL4-induced resistance, restoring cellular sensitivity to osimertinib (Fig. 3L).

ANGPTL4 interacts with NDRG1 and cooperates with NDRG1 in ECM pathway regulation

Although we have clarified that ANGPTL4 plays a role through the ECM pathway, its direct target and molecular mechanism are still unclear. We attempted to identify potential interacting proteins by screening gene targets showing significant co-downregulation with ANGPTL4 expression. Through combined screening using the String protein interaction network and publicly available ANGPTL4 immunoprecipitation-mass spectrometry (ip-MS) data [34]. N-myc Downregulated Gene 1 (NDRG1)—a multifunctional protein playing key regulatory roles in cellular stress responses and tumor progression [35]—caught our attention. Its expression was significantly downregulated upon ANGPTL4 inhibition (Fig. 4A), suggesting potential co-localization within the same regulatory axis or complex. Both the String protein network interactions and ANGPTL4 IP-MS data confirmed their interaction (Fig. 4B-C).

Fig. 4.

ANGPTL4 interacts with NDRG1 and jointly regulates the ECM pathway. A Volcano plot showing NDRG1 expression in PC9-OSI organoids with ANGPTL4 knockdown versus controls B String protein interaction network of differentially expressed genes in PC9-OSI organoids with ANGPTL4 knockdown versus controls C Volcano plot of ANGPTL4 IP-MS data D Computer simulation of ANGPTL4-NDRG1 protein interaction E Multiplex immunofluorescence staining results in patient lung cancer tissues F IP-WB results using ANGPTL4 antibody G Relationship between NDRG1 and ANGPTL4 expression in integrated lung cancer single-cell datasets H Correlation among NDRG1 expression, ANGPTL4 expression, and ECM pathway activation in integrated lung cancer single-cell datasets I NDRG1 expression following ANGPTL4 knockdown and overexpression in osimertinib-resistant lung cancer organoids J Changes in ECM pathway following NDRG1 knockdown after ANGPTL4 overexpression in PC9 lung cancer organoids. Statistical significance was determined using Student’s t-test or one-way ANOVA, as appropriate. *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001

On this basis, using the crystal structure of ANGPTL4 as the receptor and the predicted structure of NDRG1 as the ligand, we computationally simulated their binding mode. The results indicate that ANGPTL4 and NDRG1 can form a stable complex, with a binding conformation exhibiting a low binding free energy.(Fig. 4D). This suggests a thermodynamically spontaneous and stable interaction.

To confirm this prediction at the biochemical level, we performed immunoprecipitation experiments in PC9 cells. Following immunoprecipitation with the ANGPTL4 antibody, Western Blotting analysis clearly detected co-precipitated NDRG1 protein (Fig. 4F). Conversely, precipitation using the NDRG1 antibody also pulled down ANGPTL4 protein. This demonstrates that ANGPTL4 and NDRG1 exhibit direct physical interaction within cells.

We also performed immunofluorescence co-localization analysis in tissue sections. In clinical lung cancer tissue sections, both ANGPTL4 and NDRG1 were labeled simultaneously. Confocal microscopy images revealed extensive colocalization of the two proteins within the cytoplasm, indicating a high degree of spatial overlap in their intracellular distribution. This further supports their close relationship (Fig. 4E).

Finally, we regressed to single-cell transcriptomic data to validate the functional association between ANGPTL4, NDRG1, and ECM pathways within the complex tumor microenvironment. Correlation analysis revealed a significant positive correlation between ANGPTL4 and NDRG1 mRNA expression levels across all malignant epithelial cells (Fig. 4G). More importantly, when we clustered cells based on ANGPTL4 and NDRG1 expression levels, we found that the “double-positive” cell subpopulations expressing high ANGPTL4 and high NDRG1 exhibited significantly higher ECM pathway activity scores than “single-positive” or “double-negative” cells (Fig. 4H). These findings support a model in which ANGPTL4 and NDRG1 are functionally associated, and ANGPTL4/NDRG1-high malignant cells display higher ECM pathway activity.

Notably, knocking down NDRG1 in PC9 cells overexpressing ANGPTL4 resulted in a significant reduction in the protein expression levels of key markers of the ECM pathway (including Collagen I and LAMC1 ) (Fig. 4J). This finding directly demonstrates that NDRG1 plays an indispensable role in ANGPTL4-mediated activation of the ECM pathway.

Virtual screening nominates candidate ANGPTL4-binding molecules

Given the critical role of the ANGPTL4 protein in mediating drug resistance, we identify it as a potential therapeutic target and aim to conduct large-scale drug screening through virtual molecular docking to nominate small molecules with predicted binding affinity to ANGPTL4 (Fig. 5A).

Fig. 5.

Computational nomination of candidate ANGPTL4-binding molecules

We downloaded about 2.08 million ' drug-like ' small molecules from the ZINC20 small molecule database to form the initial library [36]. First, we performed high-throughput preliminary molecular docking (using AutoDock Vina), prioritizing binding energy as the primary screening metric for each small molecule’s interaction with the active pocket of the ANGPTL4 protein [37]. By ranking the binding energies of all molecules, we selected the top 50 small molecules with the lowest binding energies and most stable predicted interactions to advance to the next round of refined screening. (Fig. 5B-C)

On this basis, 10 ns molecular dynamics simulations were conducted on the initial candidate molecules. MDAnalysis was used to assess trajectory stability [38] evaluate the complex’s root mean square deviation, binding interface fluctuations, and interaction characteristics. Finally, the MM-GBSA method was used to calculate the binding free energy [39], and key binding residues were identified through binding energy decomposition analysis. Five candidate molecules with favorable predicted binding affinity and molecular dynamics stability were selected for future experimental validation. Because these molecules have not yet been evaluated in biochemical or cellular assays, they should be considered computationally nominated ANGPTL4-binding candidates rather than experimentally validated ANGPTL4 inhibitors. (Fig. 5D-F)

A Computer-Aided Virtual Screening Workflow for ANGPTL4 Targeted Drugs B Distribution Map of Binding Energies for 2.08 Million Small Molecules from Initial Screening C Top 50 Small Molecules by Binding Energy from Initial Screening D Molecular Dynamics Simulation of TOP50 Small Molecules, MM-GSBA Simulation Results E Final Selection of Top 5 Small Molecules F Binding Details of Top 5 Molecules with ANGPTL4 (Yellow: Binding Pocket, Pink: Surface Atoms, Green: Small Molecule Drugs).

Discussion

Elucidating the mechanisms of osimertinib resistance is a critical scientific challenge for improving long-term survival in patients with EGFR-mutant NSCLC [8]. In this study, we established and characterized 3D lung cancer organoid models to investigate osimertinib resistance and identified ANGPTL4 as an important contributor to the resistant phenotype. Together with additional H1975 validation, our data support a model in which ANGPTL4 promotes osimertinib resistance partly through NDRG1-associated ECM remodeling.

The primary innovation lies in successfully establishing a 3D lung cancer organoid platform that highly mimics the in vivo microenvironment. Compared to traditional 2D cultures, this model not only morphologically recreates the tumor’s three-dimensional architecture and cellular heterogeneity but, more importantly, functionally reproduces drug-resistant phenotypes closer to clinical reality [40, 41]. This discovery holds dual significance: it validates the critical role of the tumor microenvironment in drug response while explaining why many compounds effective in 2D models underperform in clinical trials—because they fail to address microenvironment-mediated resistance mechanisms [42]. Notably, the activation of ECM pathways observed in our organoid model—a phenomenon rarely detectable in conventional 2D cultures—demonstrates this platform’s unique value in uncovering novel mechanisms. The establishment of this model system provides an ideal technical platform for future research into the interactions between the tumor microenvironment and targeted therapeutics.

Through organoid-based screening and functional validation, we identified ANGPTL4 as a reproducible contributor to osimertinib resistance. This discovery holds significant scientific implications: First, as a secreted protein, ANGPTL4’s mechanism of action inherently relies on cell-cell interactions [43], explaining why its importance was difficult to detect in 2D models. Second, we revealed for the first time a novel mechanism whereby ANGPTL4 activates the ECM pathway through interaction with NDRG1. As a multifunctional stress response protein, NDRG1’s interaction with ANGPTL4 may constitute a critical mechanism for tumor cells to counteract drug pressure [44]. The discovery of this signaling axis not only deepens our understanding of osimertinib resistance mechanisms but, more importantly, links cell surface signaling to extracellular matrix remodeling, offering a novel perspective on the role of the tumor microenvironment in drug resistance.

Notably, the correlation between ANGPTL4 expression and PERCEPTION-predicted resistance scores was significant but modest. This suggests that ANGPTL4 is unlikely to be the sole determinant of osimertinib resistance. Instead, ANGPTL4-associated ECM remodeling may cooperate with additional resistance programs, such as integrin-mediated survival signaling, EMT-like state transition, hypoxia/stress responses, altered drug penetration, and EGFR-independent bypass signaling. Therefore, the ANGPTL4/NDRG1/ECM axis should be considered an important component of a broader resistance network.

The findings of this study provide compelling evidence for the critical role of the tumor microenvironment in the development of drug resistance [45]. Activation of the ECM pathway is not merely a consequence of drug resistance but an active process that sustains the resistant state [46]. We discovered that in a 3D environment, tumor cells secrete ANGPTL4 to interact with NDRG1, thereby remodeling the extracellular matrix. This remodeling may promote drug resistance through multiple mechanisms: creating physical barriers that impede drug penetration [47], activating integrin-related survival signaling, and altering cellular states via mechanosignaling [48]. This finding suggests that future strategies to overcome drug resistance should focus more on targeting the tumor microenvironment, rather than solely the tumor cells themselves.

At the translational medicine level, this study demonstrates clear application prospects. The candidate ANGPTL4-binding molecules nominated by virtual screening provide a computational starting point for future drug-discovery efforts. However, their direct binding to ANGPTL4, ability to disrupt ANGPTL4/NDRG1-associated signaling, effects on ECM remodeling, and capacity to restore osimertinib sensitivity require biochemical and functional validation. Collectively, these findings establish a comprehensive research paradigm spanning mechanism exploration to clinical application, offering a multi-tiered solution for overcoming osimertinib resistance.

Several limitations should be acknowledged. First, although the additional H1975-derived validation provides supportive evidence that ANGPTL4-associated osimertinib resistance is not restricted to the PC9 background, further validation in additional EGFR-mutant models, patient-derived organoids, and in vivo PDX models is still needed. Second, although the organoid models showed stable morphology and drug-response patterns during serial passaging, the SNV and CNV stability analyses were inferred from bulk RNA-seq data and should be further confirmed using whole-exome sequencing or targeted DNA sequencing. Third, although our data support a physical and functional association between ANGPTL4 and NDRG1, the precise regulatory level by which ANGPTL4 affects NDRG1 and the downstream transcriptional mechanisms leading to ECM gene activation remain incompletely defined. Future studies using protein stability assays, ubiquitination assays, rescue experiments, and transcription-factor profiling will be required to clarify this molecular cascade. Fourth, the specific effector molecules downstream of ECM remodeling and their contribution to osimertinib sensitivity require further investigation.

Finally, the candidate ANGPTL4-binding molecules nominated by virtual screening have not yet been experimentally validated. Their direct binding to ANGPTL4, ability to disrupt ANGPTL4/NDRG1-associated signaling, effects on ECM remodeling, pharmacokinetic properties, safety, and in vivo efficacy should be evaluated in future studies.

Conclusion

In conclusion, this study identifies ANGPTL4 as an important contributor to osimertinib resistance in EGFR-mutant NSCLC organoid models. Functional validation in PC9- and H1975-derived models supports the reproducibility of ANGPTL4-associated resistance. Mechanistically, our data suggest that ANGPTL4 interacts with NDRG1 and promotes resistance partly through ECM remodeling. In addition, virtual screening nominated candidate ANGPTL4-binding molecules that warrant further biochemical and functional validation. These findings highlight the value of 3D organoid models for investigating microenvironment-associated mechanisms of osimertinib resistance.

Author contributions

Conceptualization, Xin-xin Zeng, Xu-Hui Zhang; Data curation, Dai-yan Zhou; Formal analysis, Pei Xie; Funding acquisition, Dai-yan Zhou, Pei Xie; Methodology, Dai-yan Zhou; Project administration, Dai-yan Zhou, Wei-Lu Liu, Meng-Di Hao, Zi-Yi Wang; Validation, Dai-Yan Zhou; Writing, Dai-Yan Zhou, Si-Yuan Wang.

Funding

This work was supported by the Guangdong Senior-care and Health Charity Foundation (Grant No. JZ2026013, JZ2026014, JZ2025064 and JZ2025063), the Zhongguancun Precision Medicine Foundation (Grant No. 2024J2Y4367), and the Clinical Research Fund of Guangdong Second Provincial General Hospital (Grant No. LCYJ-2025002).

Data availability

The data presented in this study is included within the article and its supplementary files.

Declarations

Institutional review board statement

The study was approved by the Institutional Guangdong Second People's Hospital and performed under the ethical guidelines of the Helsinki Declaration (2025-KY-SQ-227). All participants provided written informed consent prior to their enrollment in the study.

Competing interests

The authors declare no competing interests.