Transfer of Damaged Mitochondria from Cancer Cells to Cancer-Associated Fibroblasts Promotes Tyrosine Kinase Inhibitor Tolerance in EGFR-Mutant Lung Cancer.

Introduction

EGFR mutations are the most common oncogenic alterations in lung adenocarcinoma (LUAD), particularly among East Asian patients (1). These mutations provide significant benefits to patients treated with EGFR-tyrosine kinase inhibitors (EGFR–TKI; ref. 2). However, the development of acquired resistance limits the clinical effectiveness of these treatments (3). Whereas much research has focused on the genetic mechanisms behind EGFR-TKI resistance, recent evidence highlights the role of a small population of cancer cells that evade cell death by entering a reversible slow-proliferation state known as the drug-tolerant persister (DTP) state (4, 5). This resistance often emerges after an initial positive response to treatment, leading to a stable minimal residual disease (MRD) composed of DTP cells (6). The formation of DTP cells largely depends on reversible mechanisms rather than mutations, including transcriptional changes, epigenetic modifications, and reprogramming of the tumor microenvironment (TME; refs. 7, 8). Targeting DTP cells may offer a promising strategy to combat EGFR-TKI resistance and extend treatment efficacy.
Cancer-associated fibroblasts (CAF) are activated fibroblasts characterized by their diverse phenotypes and functions (9, 10). In lung cancer, CAFs are prevalent in the TME and play a crucial role in resistance to EGFR-TKIs (11). Research has shown that podoplanin-positive CAFs contribute to resistance against the EGFR inhibitor gefitinib by enhancing pERK signaling (11). Additionally, CAF-derived insulin-like growth factor binding proteins (IGFBP) and hepatocyte growth factor significantly inhibit compensatory signaling pathways involving IGF1R, FAK, MAPK, and PI3K/AKT in response to EGFR-TKIs, ultimately improving the effectiveness of gefitinib (8). However, the specific subtypes of CAFs present in patients with LUAD with EGFR mutations and the mechanisms by which these distinct CAF subsets modulate EGFR-TKI resistance remain unclear.
Recent studies have highlighted the intercellular transfer of mitochondria, revealing that cancer cells can hijack mitochondria from stromal cells or T cells to support their growth (12, 13). Additionally, cancer cells can release damaged mitochondria to mesenchymal stem cells (MSC) as a mechanism for mitochondrial clearance (14). The primary mechanisms for mitochondrial transfer include tunneling nanotubes (TNT), gap junctions, and the extrusion of microvesicle-embedded or free-floating mitochondria (12). Whereas previous research has mainly focused on the paracrine role of CAFs in mediating resistance to EGFR-TKIs (15), the physical interactions involving TNTs formed by CAFs remain largely unexplored. Furthermore, there is limited understanding of mitochondrial trafficking and its relationship to drug tolerance in this context.
In this study, we used single-cell RNA sequencing (scRNA) to map the fibroblast landscape in EGFR-mutant LUAD, identifying a previously unrecognized subset of myofibroblasts characterized by the markers RGS5 and MYL9. Our findings revealed that cancer cells treated with osimertinib (third-generation EGFR-TKI) recruited RGS5+MYL9+ CAFs through the chemokine CCL11. This recruitment activated atypical mitochondrial Rho GTPase 1 (Miro1) and Ras homology family member A (RhoA) in cancer cells, which facilitated the formation of TNTs and enabled the transfer of damaged mitochondria from cancer cells to RGS5+MYL9+ CAFs. Reducing the burden of damaged mitochondria was found to promote drug–DTP cell formation and contribute to tumor relapse.
Our results provide a comprehensive understanding of the role that RGS5+MYL9+ CAFs play in fostering sustained drug tolerance in EGFR-mutant lung cancer and promoting tumor recurrence. Notably, combining the EGFR-TKI osimertinib with the FDA-approved Rho kinase inhibitor fasudil effectively reduced tumor relapse and increased overall survival (OS) in mouse models. This indicates a potential therapeutic strategy to combat the development of drug tolerance in this setting.

Materials and Methods

Patient samples
We obtained LUAD tissues from surgical resection specimens at our institution. A total of 15 treatment-naïve patients with EGFR-mutant LUAD were included in this study (eight females and seven males). Patients who had previously received any treatment for EGFR-mutant LUAD were excluded. All cases received histopathologic confirmation by three independent pathologists. Written informed consent was obtained from participants, and collected samples were processed for scRNA-seq, IHC, primary cell isolation, and organoid generation. The characteristics of patients whose tumors were used for isolation of CAF subsets and patient-derived organoids (PDO) were listed in Supplementary Table S1. Informed consent was received from all participating patients, and the study was approved by the Ethics Committee of Jiangsu Cancer Hospital (No. 2020129). All studies were conducted in accordance with the Declaration of Helsinki.

Organoid models and 3D coculture with CAFs
Fresh specimens of EGFR-mutant LUAD were minced and processed using a tissue digestion medium that included Advanced DMEM/F-12 (Gibco), 200 mmol/L GlutaMAX (Gibco), 1 mol/L HEPES (Gibco), 1 mg/mL collagenase IV (Sigma–Aldrich), and 100 μg/mL DNase I (Invitrogen). The dissociated cells were then resuspended in growth factor-reduced Matrigel (Corning) and seeded as 20 μL drops in 24-well plates. They were cultured in 500 μL of organoid growth medium, which consisted of Advanced DMEM/F-12, 200 mmol/L GlutaMAX, 1 mol/L HEPES, 1× B27 supplement, 1 mmol/L N-acetylcysteine (Sigma–Aldrich), 10 μmol/L Y-27632 (Selleckchem), 0.5 μmol/L A83-01 (TargetMol), 0.5 μg/L R-spondin-1 (BioGenous), 100 μg/L Noggin (BioGenous), and 5 μg/L EGF (Novoprotein). PDOs were dissociated with TrypLE (Gibco) and passaged every 7 days. For the PDO–CAF models, organoid dissociations were mixed with matched CAF subsets (derived from the same patient) at a 3:1 ratio and seeded in organoid growth medium supplemented with 100 μg/L FGF10 (Novoprotein).

Cell culture and reagents
HCC827 (RRID: CVCL_2063) and PC9 (RRID: CVCL_XA18) cell lines were obtained from the Cell Bank of the Chinese Academy of Sciences and cultured in RPMI-1640 medium (KeyGene) supplemented with 10% FBS (Corning) and 1% penicillin–streptomycin (KeyGene). To generate DTP cells, the EGFR-mutant LUAD cell lines HCC827 and PC9 were treated with 2 μmol/L osimertinib for 9 days (5). Cells were subsequently infected with ATP sensor ATeam adenovirus (Hanbio Biotech) or FRET-based RhoA-FLARE biosensor (Addgene). Additionally, various reagents such as MitoTracker DeepRed FM (Invitrogen), MitoTracker Green FM (Invitrogen), MtioSOX Green/Red (Invitrogen), CCL11 (TargetMol), anti-CCL11 neutralizing antibody (bertilimumab; RRID: AB_3695259), RhoA Activator (cytoskeleton), 50 μmol/L fasudil (MCE), 50 μmol/L 18-α-GA (MCE), 50 μmol/L dynasore (MCE), or 1 μmol/L cytochalasin D (MCE) were applied according to manufacturer’s protocols at a temperature of 37°C in a 5% CO2 incubator. All cells were authenticated and checked for contamination every month.

Isolation of human primary fibroblasts
Primary fibroblast cultures and matched PDOs were established from 15 treatment-naïve, surgically resected EGFR-mutant LUAD specimens, obtained with approval from the Ethics Committee of Jiangsu Cancer Hospital. Clinical characteristics of the patients from whom these primary cells were isolated are provided in Supplementary Table S1. The specific patient tissues used in each experiment are indicated in the corresponding figure legends. Single-cell suspensions were prepared using a tumor dissociation kit (Miltenyi Biotech), followed by viability staining and antibody-based depletion of CD45+ hematopoietic cells, EPCAM+ (CD326) epithelial cells, and CD31+ endothelial cells (16). RGS5+MYL9+ subsets were FACS-sorted using PE-conjugated anti-RGS5 and BV421-conjugated anti-MACM antibodies. The remaining CAF population after RGS5+MYL9+ cell sorting was collected as RGS5+MYL9+–depleted (RGS5+MYL9+-d) CAFs. Isolated CAFs were maintained in fibroblast medium supplemented with 10% FBS and 1% growth factors (37°C, 5% CO2). All primary CAFs were used within 10 passages without immortalization or transformation. Inflammatory CAFs (iCAF) were characterized by IGFBP6 positivity, whereas antigen-presenting CAFs (apCAF) were identified through CD74 positivity.

Plasmids, siRNA, and virus transduction
Short hairpin RNAs (shRNA) targeting Miro1 were designed and constructed by Realgene. The FRET-based RhoA-FLARE biosensor and Lifeact-mCherry (RRID: Addgene_193723) plasmids were obtained from Addgene. For plasmid transfections, Lipofectamine 3000 was utilized, whereas Lipofectamine RNAiMAX (Thermo) was used for shRNA transfections. The ATP sensor ATeam adenovirus was obtained from Hanbio Tech. To create cells labeled with GFP, RFP, or mitoDsRed (Genomedi Tech), HCC827 and PC9 cell lines, as well as primary CAF subsets, were transfected with lentiviruses expressing these fluorescent proteins.

Cell growth and viability assays
The growth of LUAD cells and PDOs was monitored using the ATP sensor ATeam. LUAD cells or PDOs were infected with the ATeam adenovirus according to the manufacturer’s instructions for 24 to 48 hours before being cocultured with various CAF subsets. The coculture systems were then treated with osimertinib for varying durations, and cell viability was assessed using three-channel FRET imaging (445-nm laser excitation for donor and 515-nm laser excitation for acceptor). To generate corrected FRET images, emissions from each channel were adjusted for bleed-through and used to calculate corrected FRET values (transfer − corrected acceptor − corrected donor). HCC827 or PC9 cells (200 cells/well) were cocultured with specified CAF subsets in 12-well plates and treated as indicated. Following incubation, colonies were fixed with 4% paraformaldehyde (20 minutes, room temperature) and stained with 0.1% crystal violet for quantification.

IC50 value analysis
For IC50 value analysis, single-cell suspensions of LUAD cells infected with the ATP biosensor ATeam adenovirus were seeded in 24-well plates at a density of 2 × 104 cells per well. The cells were treated with osimertinib at concentrations ranging from 0 to 5 μmol/L (0, 0.05, 0.1, 0.5, 1, and 5 μmol/L) for 72 hours. Viability was subsequently assessed using ATP FRET imaging.

RhoA FRET imaging and F-actin dynamics
To assess active RhoA levels, LUAD cells were transfected with the FRET-based RhoA-FLARE biosensor. FRET imaging involved capturing three channels: donor (445-nm laser excitation), acceptor (515-nm laser excitation), and transfer (FRET) channels. Emissions from each channel were corrected for bleed-through to calculate corrected FRET values using the formula Fc = transfer − corrected acceptor − corrected donor. For quantifying F-actin dynamics, LUAD cells were transfected with Lifeact-mCherry, a peptide that binds to F-actin; fluorescence intensity was reported as F/F0.

Cell-cycle analysis
Following specified treatments, cells were harvested and fixed in 70% ethanol at 4°C overnight and then treated with RNase A and stained with propidium iodide (PI; KeyGene). Data acquisition was performed using a BD FACSymphony flow cytometer (BD Biosciences). Cell-cycle distribution (G0/G1, S, and G2/M phases) was analyzed based on PI fluorescence intensity using ModFit LT software. Doublet discrimination was applied to exclude aggregated cells and ensure analysis of single-cell events.

TUNEL assay
Apoptosis in formalin-fixed paraffin-embedded tissue sections was detected using the terminal deoxynucleotidyl transferase dUTP nick-end labeling (TUNEL) assay (KeyGene). Sections were deparaffinized, rehydrated, and treated with proteinase K for antigen retrieval. After permeabilization, samples were incubated with TUNEL reaction mixture containing fluorescently labeled dUTP and terminal deoxynucleotidyl transferase (TdT) enzyme. Nuclei were counterstained with DAPI. TUNEL-positive cells were visualized and quantified using a fluorescence microscope. Appropriate positive and negative controls were included in each experiment to ensure assay specificity and reliability.

PDO drug testing
Both mono- and cocultures of PDOs and PDO/CAFs were established in 24-well plates as drops of Matrigel 20 μL. Osimertinib (Selleckchem) treatment commenced 3 days after PDO seeding and was washed out after 72 hours. Confocal imaging was conducted before drug application or washout at days 3, 6, and 9. For apoptosis analysis, PDOs were stained with calcein/PI (1 μg/mL; Beyotime) 4 hours prior to imaging. To assess cell viability, organoids were infected with ATP sensor ATeam adenovirus before seeding; cell viability was measured using three-channel FRET imaging at days 3, 6, and 9. Mean values from independent replicates were used for analysis.

Mitochondrial status analysis
MitoTracker Green staining was used to assess total mitochondrial mass, whereas MitoTracker DeepRed staining was utilized for evaluating mitochondrial membrane potential. Additionally, MitoSOX Red was used for detecting mtROS. HCC827 and PC9 cells were incubated with these mitochondrial dyes according to the manufacturer’s instructions. In coculture systems, cells were treated with CD326 (RRID: AB_400262) prior to mitochondrial staining to differentiate between LUAD cells and CAFs. For in vivo studies, single-cell suspensions from xenografts were stained with CD326, CD45, and CD31 to distinguish LUAD cells from CAFs, and analyses were conducted using a FACS Celesta flow cytometer (BD Biosciences).

Mitochondrial transfer

In vitro mitochondrial transfer experiments involved coculturing mitoDsRed-labeled HCC827 or CAFs with GFP-labeled CAFs or HCC827 cells in direct coculture well or in chamber setups. Flow cytometry was performed using a FACS Celesta flow cytometer (BD Biosciences), and data analysis was conducted with FlowJo v.10.8.1(RRID: SCR_008520). For in vivo experiments, mice received subcutaneous injections of 1 × 107 mitoDsRed-HCC827 cells alongside 5 × 106 GFP-CAFs. Following tumor establishment, mice were treated with osimertinib (5 mg/kg/day, oral gavage) for 9 days before being sacrificed. Tumors were resected, digested, and analyzed for mitochondrial transfer via flow cytometry.

Chemotaxis assay
Chemotaxis assays were conducted by seeding 1 × 105 HCC827 cells or CAFs in 200 μL of medium into the upper chamber of 8-μm pore transwell inserts (Millipore) within a 24-well plate. In the lower chamber, 600 μL of culture media containing tumor cells or CAFs supplemented with 10% FBS, CCL11, αCCL11, or TEMPO served as chemoattractants. After 24 hours of incubation, cells in the upper chamber were fixed with 4% formaldehyde, stained with crystal violet for 30 minutes, and counted using a light microscope.

Cytokine antibody array
Cytokine profiles from LUAD cells were analyzed using the Proteome Profiler Human XL Cytokine Array Kit (RayBio). The array chips were blocked with 5% BSA at room temperature for 1 hour before incubating overnight with culture supernatants obtained from LUAD-CAF coculture medium. Following this, an antibody cocktail was applied for 1 hour at room temperature, followed by streptavidin–horseradish peroxidase (HRP) incubation for another 30 minutes at room temperature. Chemi Reagent Mix (Thermo Fisher Scientific) was used to detect cytokine spots on the arrays. Images were captured on X-ray film and analyzed using FIJI software (NIH; RRID: SCR_002285).

ELISA
The secretion levels of IL1α, IL13, IL15, MCP-1, MCP-2, CCL11, and GDNF in coculture media of LUAD cells and CAFs were quantified using commercial ELISA kits. HCC827 cells were cocultured with different CAF subsets under specified treatment conditions and durations. Supernatants were collected, centrifuged to remove debris, and analyzed according to the manufacturer’s protocols. Cytokine concentrations were determined based on standard curves, with all samples measured in duplicate to ensure reproducibility.

Animal models
Four-week-old female BALB/c nude mice were obtained from GemPharmatech. All animal studies were approved by the Nanjing Medical University Animal Care Committee. For subcutaneous xenograft models, HCC827 cells (1 × 107) were inoculated alongside different CAF subsets (5 × 106) into the mice. Following tumor establishment, mice received either a vehicle or osimertinib (5 mg/kg/day, oral gavage) for 9 days before drug withdrawal. Body weights and tumor growth measurements continued until endpoints were reached; mice were sacrificed at designated times to collect subcutaneous tumors for flow cytometry and histologic analysis.
In a separate xenograft model, mice received subcutaneous injections of 1 × 107 HCC827 cells combined with 5 × 106 RGS5+MYL9+ CAFs. After tumor establishment, mice were randomly assigned to various treatment groups: Group 1 received vehicle until study completion; group 2 received αCCL11 (5 mg/kg/week, intraperitoneal injection) at the end of the study; group 3 received osimertinib (5 mg/kg/day, oral gavage) at the end of the study; group 4 received osimertinib (5 mg/kg/day, oral gavage) + αCCL11 (5 mg/kg/week, intraperitoneal injection) at the end of the study.
For experiments assessing the effects of osimertinib (5 mg/kg/day, oral gavage) and fasudil (50 mg/kg, 5 times per week, i.p) on MRD formation and tumor relapse, HCC827 cells (1 × 107) were injected alongside different CAF subsets (5× 106). After tumor establishment, groups received treatments as follows: group 1—osimertinib alone for 9 days followed by vehicle; group 2—combination therapy of osimertinib and fasudil for 9 days followed by vehicle; and group 3—combination therapy of osimertinib and fasudil for 9 days followed by fasudil alone. Tumor size was measured using calipers; tumor volume was calculated as 0.5 × L (long diameter) × W2 (short diameter). Survival analyses used Kaplan–Meier methods using GraphPad Prism v.10 (RRID: SCR_002798).

Tissue dissociation and scRNA-seq
Fresh LUAD tissues were collected and washed with ice-cold PBS before dissociation using a tumor dissociation kit (Miltenyi Biotec) to prepare single-cell suspensions. These suspensions underwent cDNA library construction followed by sequencing at Shanghai Biotechnology Corporation. scRNA-seq data were annotated to identify CAF subtypes. The annotated cell type signatures were used as a reference matrix for CIBERSORTx to deconvolute bulk RNA-seq data from The Cancer Genome Atlas (TCGA) LUAD tumor samples. Based on the estimated CAF subtype infiltration proportions, samples were stratified into high- and low-infiltration groups for each CAF subtype using the median value as the cutoff. Kaplan–Meier (RRID: SCR_024521) survival analysis was performed to evaluate the association between CAF infiltration levels and clinical outcomes, providing evidence for the prognostic significance of CAF subtypes in LUAD.

Bulk RNA-seq
Total RNA from 1 × 106 osimertinib-tolerant HCC827 or parental HCC827 cocultured with or without various CAF subtypes was isolated using Trizol (Life Technologies). cDNA libraries were sequenced on an Illumina HiSeq X-ten platform. Differential expression analysis utilized DESeq2 (DESeq, RRID: SCR_000154) v.1.22.2 in R software; differentially expressed genes were identified based on a threshold of fold change >2 and adjusted P < 0.05. Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis (RRID: SCR_012773) was performed using DAVID v.6.7 (http://davis.ncifcrf.gov; RRID: SCR_001881). We performed a two-tiered enrichment analysis integrating both overrepresentation analysis and gene set enrichment analysis (GSEA) to comprehensively evaluate pathway regulation patterns and enhance statistical reliability.

GTP-RhoA pull-down assay
Coculture cells underwent treatment in RPMI-1640 medium containing 2 μmol/L Osimertinib for specified durations before total RhoA levels were evaluated through Western blotting. For GTP-RhoA pull-down assays, cell lysates incubated overnight at 4°C with Rhotekin RBD protein bound to GST beads underwent three washes with PBS prior to separation via SDS-PAGE to assess active RhoA levels; RhoA activity was quantified as the ratio of active RhoA to total RhoA.

Western blot analysis
Cells were washed with PBS before lysis in protein extraction buffer (Thermo Fisher Scientific), supplemented with protease and phosphatase inhibitors (Thermo Fisher Scientific). Cell lysates were subjected to SDS-PAGE electrophoresis before transfer onto polyvinylidene difluoride membranes. Primary antibodies against E-cadherin (RRID: AB_562059), N-cadherin (RRID: AB_10598006), vimentin (RRID: AB_2216132), Miro1, pRHOA [serine 188 (s188); RRID: AB_2816404], total RhoA (RRID: AB_547697), and GST (RRID: AB_10565690) were used; GAPDH served as a loading control, whereas secondary antibodies included goat anti-rabbit HRP or goat anti-mouse HRP.

Immunofluorescence and histology
Cells were fixed with 4% formaldehyde for 15 minutes, permeabilized with 0.1% Triton X-100 for 15 minutes, blocked with goat serum for 60 minutes, and incubated with primary antibodies overnight at 4°C. Alexa Flour 488–labeled, Alexa Four 595–labeled, or Alexa Four 674–labeled secondary antibodies (ProteinTech) were used for immunostaining. F-actin was stained with Alexa Flour 488–labeled phalloidin (ProteinTech). Nuclei were counterstained with DAPI. Images were captured on an LSM 780 laser confocal microscope (Carl Zeiss) and Andor BC43 (OXFORD Instruments).
Paraffin-embedded mouse or human tissues were sliced at 4 μm thickness. Antigen retrieval was performed using 0.01 mol/L citrate buffer (pH 6.0) at 100°C. Sections were then incubated with goat serum for 30 minutes at room temperature. For immunofluorescence (IF), the sections were incubated with antibodies specific for BCHE (RRID: AB_2552008), RGS5 (RRID: AB_2178356), Panck, IGFBP6 (RRID: AB_1545992), CD74 (RRID: AB_2739905), Miro1 (RRID: AB_2147773), and PTGDS (AB_2646107) overnight at 4°C. Alexa Flour 488–labeled, Alexa Four 595–labeled, or Alexa Four 674–labeled secondary antibodies (ProteinTech) were used for immunostaining. Nuclei were counterstained with DAPI. For IHC, sections were incubated with Ki67 antibody overnight at 4°C and subsequently stained with corresponding secondary antibody for 1 hour at room temperature. Finally, sections were stained with DAB and counterstained with hematoxylin. The slides were captured using 3DHISTECH Pannoramic.

Multiplex IF
Paraffin-embedded mouse or human tissues were fixed in 4% paraformaldehyde for 15 minutes, permeabilized with 0.1% Triton X-100 for 10 minutes, and blocked with 5% goat serum for 1 hour at room temperature. Sequential incubation with primary antibodies overnight at 4°C starting with anti-RGS5 (ProteinTech), followed by incubation with HRP-conjugated secondary antibody (ProteinTech) for 1 hour at room temperature. Subsequently, tyramide signal amplification was used for signal amplification, visualizing in green. Three percent H2O2 was used for quenching HRP activity. This process was repeated for anti-MYL9 (ProteinTech, red) and anti-panCK (ProteinTech, Pink). Finally, nuclei were counterstained with DAPI. Samples were mounted using antifade mounting medium for confocal visualization. Images were acquired with Andor BC43 (OXFORD Instruments).

Statistical analysis
All experiments were repeated independently at least three times. Data are shown as the mean ± SD. Statistical analyses were performed using SPSS 22.0 software (SPSS; RRID: SCR_002865). Student two-tailed t tests were used for single comparisons, and one-way ANOVA (RRID: SCR_002427) was used for multiple comparisons. Survival curves were assessed by the Kaplan–Meier (RRID: SCR_024521) method and log-rank test. All P values are *, P < 0.05; **, P < 0.01; and ***, P < 0.001.

Results

Identification of CAF subsets in EGFR-mutant LUAD
To elucidate the mechanisms by which CAFs contribute to EGFR-TKI resistance in LUAD, we performed scRNA-seq on tumor specimens obtained from six treatment-naïve patients harboring EGFR mutations (Fig. 1A). This analysis provided a detailed view of the cellular composition within EGFR-mutant LUAD, with a particular focus on the various fibroblast subtypes (Fig. 1B and C). We identified five distinct clusters of CAFs based on their transcriptomic profiles. In addition to iCAFs and apCAFs, we characterized three novel myofibroblast (myCAF) clusters, labeled as BCHE+ myCAF, PTGDS+ myCAF, and RGS5+ myCAF, with specific markers for each cluster detailed in Fig. 1D. Notably, we observed considerable heterogeneity among the CAF populations across the six patients with EGFR-mutant LUAD studied (Fig. 1E).

RGS5+MYL9+ CAFs is associated with EGFR-TKI resistance and poor prognosis in EGFR-mutant LUAD
To investigate the contribution of CAF clusters to EGFR-TKI resistance, we established a three-dimensional coculture system. This model combined treatment-naïve EGFR-mutant LUAD organoids (EGFR-TKI–sensitive) with matched CAF clusters (PDO-CAF model), enabling direct analysis of stromal–tumor interactions (Fig. 1F; Supplementary Fig. S1A). We found that myCAFs, particularly those expressing RGS5 and MYL9, as well as iCAFs, significantly enhanced the proliferation of EGFR-mutant LUAD organoids (Supplementary Fig. S1B).
To evaluate how distinct CAF subtypes influence EGFR-TKI sensitivity in PDOs, we performed image-based drug assays (17, 18). PDOs were transduced with an ATP biosensor (ATeam adenovirus) for 24 to 48 hours before coculture with patient-matched CAF subsets. Using PDO-CAF models from three treatment-naïve patientswith EGFR-mutant LUAD, we treated the cocultures with osimertinib and quantified PDO viability via three-channel FRET imaging (Fig. 1G–I; Supplementary Fig. S2A and S2B). Apoptosis was concurrently assessed using calcein/PI staining (Fig. 1H; Supplementary Fig. S2A). Notably, RGS5+MYL9+ myCAFs and iCAFs substantially attenuated osimertinib-induced cell death. More importantly, RGS5+MYL9+ CAFs markedly increased the potential for tumor regrowth, as evidenced by a higher cell proliferation ratio and a lower apoptosis ratio following drug washout (Fig. 1H and I; Supplementary Fig. S2A and S2B). The drug washout approach is especially pertinent in EGFR-TKI studies, as it allows evaluation of MRD, detection of preexisting resistant subclones, and discrimination between durable responses and transient growth inhibition. Collectively, these results highlight the crucial contribution of specific CAF subsets in driving EGFR-TKI resistance and facilitating tumor repopulation.
Next, we examined the clinical significance of various CAF subsets in EGFR-mutant LUAD tissues obtained from Jiangsu Cancer Hospital. Multiplex fluorescence staining demonstrated that tumor-infiltrating RGS5+MYL9+ CAFs were more prevalent in osimertinib-resistant LUAD tissues compared with RGS5+MYL9+ depleted CAFs (RGS5+MYL9+-d CAFs; Fig. 1J). Furthermore, patients with LUAD exhibiting higher scores of RGS5+MYL9+ CAFs were predominantly diagnosed at stages III and IV, whereas those with lower scores were primarily at stages I and II (Fig. 1K). Kaplan–Meier analysis indicated that greater infiltration of RGS5+MYL9+ CAFs correlated with poorer outcomes in patients with EGFR-mutant LUAD in TCGA datasets (Fig. 1L).

RGS5+MYL9+ CAFs mitigate EGFR-TKI–induced mitochondrial damage and mtROS production to facilitate DTP formation in vitro
A subset of cancer cells can enter a reversible, slow-cycling state known as the DTP state, enabling survival during therapy and eventual relapse. We observed that coculture with RGS5+MYL9+ CAFs significantly enhanced tumor regrowth following osimertinib withdrawal compared with other CAF subsets (Fig. 1H and I). This led us to hypothesize that RGS5+MYL9+ CAFs promote DTP formation, thereby driving EGFR-TKI resistance.
To test this, we used EGFR-mutant LUAD cell lines (HCC827 and PC9) and patient-derived CAF subsets. Under 2 μmol/L osimertinib treatment, cells cocultured with RGS5+MYL9+ CAFs exhibited increased viability, stemness, epithelial–mesenchymal transition markers, and a predominant slow-cycling phenotype in a time-independent manner (Supplementary Fig. S3A–S3F), consistent with established DTP characteristics. Notably, RGS5+MYL9+ CAFs further reduced osimertinib sensitivity in preestablished DTP cells, which was abolished in transwell chamber assays (Supplementary Fig. S4A–S4F), implicating direct cell-contact dependence.
To elucidate the mechanisms underlying CAF-induced drug tolerance, we performed RNA-seq on HCC827 cells under various culture conditions. KEGG and GSEA analyses demonstrated that coculture with RGS5+MYL9+ CAFs significantly downregulated oxidative phosphorylation pathways and reduced ROS production as compared with RGS5+MYL9+-d CAF coculture (Fig. 2A and B; Supplementary Fig. S5A–S5D).
Given the central role of mitochondria in ROS production and oxidative phosphorylation (19), we established a three-phase model to evaluate mitochondrial dynamics during DTP formation, drug withdrawal, and rechallenge (Fig. 2C and D; Supplementary Fig. S6A and S6B). Transmission electron microscopy (TEM) and functional assays revealed that RGS5+MYL9+ CAFs protected LUAD cells (HCC827 and PC9) from osimertinib-induced mitochondrial damage, as evidenced by preservation of mitochondrial membrane potential (Δψm), reduced mtROS accumulation, and lower levels of dysfunctional mitochondria (MitoTracker Green for mitochondrial mass, MitoTracker Red for Δψm, and MitoSOX for mtROS; Fig. 2C–I; Supplementary Figs. S6A, S6B, S7A–S7C, S7G, and S7I; ref. 20). Importantly, this protective effect was strictly contact-dependent, as it was abolished in transwell coculture systems (Supplementary Fig. S7D, S7E, and S7J). Notably, RGS5+MYL9+ CAFs displayed mitochondrial dysfunction (elevated mtROS and reduced Δψm) when cocultured with TKI-treated LUAD cells, suggesting that RGS5+MYL9+ CAFs may scavenge oxidative stress from neighboring tumor cells, creating a protective niche that promotes drug tolerance.

RGS5+MYL9+ CAFs alleviate mitochondrial damage and mtROS production to support DTP formation in vivo
To investigate the role of mitochondrial damage in the formation of MRD and subsequent relapse, we established HCC827 xenograft models in mice, incorporating either RGS5+MYL9+ CAFs or RGS5+MYL9+ -d CAFs. After the establishment of tumors, the mice were randomly divided into four groups and treated with either vehicle or a high dose of osimertinib (5 mg/kg/day, oral gavage) for 9 days. This treatment was followed by drug withdrawal to allow for tumor regrowth (Fig. 2J). Both RGS5+MYL9+ CAFs and RGS5+MYL9+-d CAFs exhibited similar effects on tumor growth when treated with the vehicle. Notably, in response to osimertinib treatment, RGS5+MYL9+ CAFs enhanced the establishment of osimertinib-regressed MRD tumors. More importantly, RGS5+MYL9+ CAFs significantly promoted tumor regrowth following drug withdrawal (Fig. 2J).
To assess mitochondrial integrity and mtROS production, we used flow cytometry and TEM analysis. As anticipated, MRD isolated from the group coinjected with RGS5+MYL9+ CAFs displayed lower levels of dysfunctional mitochondria and mtROS compared with those from the RGS5+MYL9+-d CAF group. Conversely, RGS5+MYL9+ CAFs exhibited significantly higher levels of dysfunctional mitochondria and mtROS than their RGS5+MYL9+-d CAFs counterparts. Additionally, there were no notable differences in mitochondrial integrity or mtROS levels between tumor cells or CAFs isolated from regrown tumors across both groups (Fig. 2K–M; Supplementary Fig. S8A and S8B).
Multiplex IF staining analysis revealed a greater density of infiltrated fibroblasts in MRD and regrown tumors within the RGS5+MYL9+ CAF group compared with the RGS5+MYL9+-d CAF group. Consistent findings were observed through hematoxylin and eosin (H&E) staining, Ki67 staining, TUNEL analysis, and multiplex IF staining, which indicated that infiltration by RGS5+MYL9+ CAFs inhibited apoptosis induced by osimertinib (Fig. 2N–Q).
These results align with our in vitro findings, supporting the idea that RGS5+MYL9+ CAFs mitigate EGFR-TKI–induced mitochondrial damage and decrease mtROS production by compromising their own mitochondrial integrity. This process seems to facilitate the development of DTP cells within tumors.

Intercellular nanotubes facilitate mitochondrial transfer from LUAD cells to RGS5+MYL9+ CAFs under EGFR-TKI stress
Our findings demonstrate that the formation of DTP cells and resistance to EGFR-TKIs in RGS5+MYL9+ CAFs relies heavily on direct cell interactions, as illustrated in Supplementary Fig. S7A–S7J. Recent studies have reported the intercellular transfer of mitochondria through nanotubes, which not only supports tumor growth by allowing tumor cells to acquire mitochondria from surrounding cells but also aids in the repair of damaged cells (12). Analysis of KEGG pathway enrichment using scRNA-seq data from EGFR-mutated LUAD tumors revealed that RGS5+MYL9+ CAFs exhibit significant activity in pathways related to the regulation of the actin cytoskeleton (Supplementary Fig. S9A). The cytoskeleton plays a crucial role in the formation of intercellular nanotubes.
To investigate whether EGFR-TKI treatment prompts LUAD cells to transfer damaged mitochondria to RGS5+MYL9+ CAFs, we labeled mitochondria in HCC827 cells with MitoTracker Red. After washing away any unbound dye, we cocultured these cells with CAFs and treated them with osimertinib for 9 days to induce a DTP state. By day 9, we observed considerable levels of damaged mitochondria in drug-tolerant HCC827 cells (Fig. 3A). Notably, RGS5+MYL9+-d CAF cocultured drug-tolerant HCC827 cells exhibited clustering of damaged (depolarized) mitochondria near the nucleus, as indicated by TOM20, MitoTracker Red, and mtSOX staining. In contrast, damaged mitochondria in RGS5+MYL9+ CAF cocultured drug-tolerant HCC827 cells were more dispersed throughout the cell (Fig. 3A and B). Additionally, mitoDsRed-labeled HCC827 cells were cocultured with CAFs and treated with osimertinib for 9 days. Phalloidin Red was used to stain F-actin within the nanotubes. We found significant colocalization of MitoTracker-labeled mitochondria within nanotubes connecting drug-tolerant HCC827 cells and cocultured with RGS5+MYL9+ CAFs, whereas few labeled mitochondria were detected between drug-tolerant HCC827 cells and RGS5+MYL9+-d CAFs (Fig. 3C). More importantly, RGS5+MYL9+ CAFs derived from the DTP coculture system exhibited an increased number of nanotubes per cell, particularly those measuring between 30 and 100 μm in length, when compared with RGS5+MYL9+-d CAFs (Fig. 3C).
To further confirm whether mitochondrial transferred between HCC827 cells with RGS5+MYL9+ CAFs during DTP formation, we cocultured EGFP-transduced CAFs with mitoDsRed-transduced HCC827 cells directly or in a chamber supplemented with 2 μmol/L osimertinib (Fig. 3D). After 9 days, we assessed mitochondrial transfer and mtROS levels of the transferred mitochondria using FACS. The results indicated a significantly higher transfer of mitoDsRed-positive mitochondria from HCC827 cells to RGS5+MYL9+ CAFs compared with RGS5+MYL9+-d CAFs. Furthermore, the transferred mitochondria in RGS5+MYL9+ CAFs exhibited elevated mtROS levels compared with those transferred into RGS5+MYL9+-d CAFs. However, no notable transfer was observed using the Boyden chamber assay (Fig. 3E). Importantly, minimal transfer of damaged mitoDsRed-positive donor mitochondria from CAFs to EGFP-positive HCC827 cells was detected (Supplementary Fig. S9B and S9C), suggesting that damaged mitochondrial transfer occurs from cancer cells to RGS5+MYL9+-d CAFs primarily through direct cell-to-cell interaction.
In addition to TNTs, other mechanisms such as macrovesicles and gap junctions have been reported for mitochondrial transfer between cells. To determine whether mitochondrial trafficking depended on intracellular nanotubes, we introduced various inhibitors: 18-α-GA (a gap junction blocker), dynamin inhibitor dynasore (a blocker of macrovesicle endocytosis), and actin polymerization inhibitor cytochalasin D (a blocker of TNT formation) into our coculture system (Fig. 3F). Flow cytometric analysis revealed that cytochalasin D significantly inhibited the trafficking of damaged mitochondria from HCC827 to RGS5+MYL9+ CAFs compared with DMSO controls, whereas treatment with 18-α-GA or dynasore did not affect mitochondrial trafficking, suggesting that intercellular nanotubes facilitated the trafficking of damaged mitochondrial from LUAD cells to RGS5+MYL9+ CAFs.
We also evaluated mitochondrial transfer in xenograft mouse models. Nude mice were received subcutaneous injections of mitoDsRed-transduced HCC827 cells alongside EGFP-transduced RGS5+MYL9+ CAFs or RGS5+MYL9+-d CAFs. Following tumor establishment, mice were treated with osimertinib for 9 days to create osimertinib-regressed MRD tumors. Mice were sacrificed at days 0, 3, and 9 for analysis (Fig. 3G). The results showed that RGS5+MYL9+ CAFs significantly inhibited tumor regression induced by osimertinib treatment (Fig. 3H). Flow cytometric analysis revealed that a substantial proportion of damaged mitochondria were transferred from osimertinib-treated HCC827 cells to RGS5+MYL9+ CAFs (24.2% at day 3% and 17% at day 9; Fig. 3I). Confocal imaging of osimertinib-treated xenografts revealed direct mitochondrial transfer from tumor cells to RGS5+MYL9+ CAFs, evidenced by mitoRed and EGFP double-positive stromal cells. Increased density and proximity of RGS5+MYL9+ CAFs (EGFP+) were found at tumor–stroma interfaces following TKI treatment (Fig. 3J). These findings are consistent with our in vitro observations. H&E staining and IHC revealed that mice with xenografts of HCC827 and RGS5+MYL9+ CAFs exhibited increased tumor volume and higher Ki67 expression, along with reduced TUNEL positivity on days 3 and 9, compared with those with RGS5+MYL9+-d CAFs. These findings suggest that mitochondrial trafficking protects LUAD from dysfunction and apoptosis induced by EGFR-TKI treatment (Fig. 3K–M). Together, our data suggest that intercellular nanotube-mediated trafficking of damaged mitochondria from LUAD cells to RGS5+MYL9+ CAFs promotes a TKI-induced DTP state.

mtROS-upregulated Miro1 leads to the distribution of damaged mitochondria to peripheral locations near RGS5+MYL9+ CAFs
Next, we investigated whether DTP cells form nanotubes with CAF subsets other than RGS5+MYL9+ CAFs. Our findings indicated that RGS5+MYL9+ CAFs were more closely associated with tumor cells compared with other CAF subsets in osimertinib-resistant tumor tissues from both patients and xenograft mice (Supplementary Fig. S10A–S10D). We propose that this proximity facilitates the formation of TNT connections between RGS5+MYL9+ CAFs and DTP cells, as opposed to connections with other CAF subsets.
We then investigated the mechanisms by which damaged mitochondria migrate toward RGS5+MYL9+ CAFs. Mitochondrial Rho GTPase 1 (Miro1) is a key protein involved in regulating mitochondrial movement. To determine the role of Miro1 in mitochondrial migration of LUAD cells under TKI stress, we silenced Miro1 in HCC827 cells. Our results revealed that both RGS5+MYL9+ CAF and RGS5+MYL9+-d CAF cocultured osimertinib-naïve and regrown HCC827 cells exhibited perinuclear clustering of mitochondria. However, during osimertinib treatment, the damaged mitochondria in HCC827 cells cocultured with RGS5+MYL9+ CAFs gradually shifted from the perinuclear region to peripheral areas, which was abrogated by Miro1 silencing. In contrast, damaged mitochondria remained clustered around the nucleus in HCC827 cells cocultured with RGS5+MYL9+-d CAFs throughout osimertinib treatment (Fig. 4A and B).
Western blot analysis showed an increase in Miro1 expression in drug-tolerant HCC827 cells cocultured with RGS5+MYL9+ CAFs but not in those cocultured with RGS5+MYL9+-d CAFs (Fig. 4C). Notably, the elevated Miro1 levels in RGS5+MYL9+ CAF cocultured DTP cells were reversed upon scavenging mtROS using mitoTEMPO (Fig. 4C). Additionally, we observed significant colocalization of Miro1 and MitoTracker Red at peripheral locations within RGS5+MYL9+ CAF cocultured DTP cells, which was also reversed by mtROS scavenging (Fig. 4D and E). Miro1 knockdown in HCC827 cells upregulated the decreased MitoSOX expression in DTP cells cocultured with RGS5+MYL9+ CAFs (Fig. 4F). Moreover, knocking down Miro1 in HCC827 cells reduced mitochondrial transfer from DTP cells to CAFs (Fig. 4G and H), demonstrating Miro1’s essential role in this protective cross-talk. We further confirmed the cellular distribution of Miro1 in LUAD cells from xenograft mouse tumor tissues. Our findings indicated that peripheral distribution of Miro1 was significantly higher in mice injected with RGS5+MYL9+ CAFs compared with those injected with RGS5+MYL9+-d CAFs (Fig. 4I), supporting the hypothesis that Miro1 mediates the transport of damaged mitochondria from LUAD cells to RGS5+MYL9+ CAFs.
Importantly, multiplex IF analysis of osimertinib-naïve and -resistant tissues from patients with EGFR-mutated LUAD revealed distinct mitochondrial distributions: tumor cells from osimertinib-naïve tissues exhibited even distribution of mitochondria from the perinuclear region to peripheral areas, whereas osimertinib-resistant tumor cells displayed predominate peripheral clustering of mitochondria (Fig. 4J). This pattern aligns with our earlier observations of increased infiltration of RGS5+MYL9+ CAFs in osimertinib-resistant LUAD tissues (Fig. 1J).
Collectively, our results demonstrate that EGFR-TKI–induced mtROS drives mitochondrial distribution toward the plasma membrane via upregulation of Miro1, thereby initiating mitochondrial trafficking between LUAD cells and RGS5+MYL9+ CAFs.

RhoA-dependent F-actin accumulation and TNT formation enable mitochondria trafficking and TKI tolerance
TNTs are membrane protrusions supported by F-actin that can extend over considerable distances to connect cells (12, 21, 22). Our observations indicated that F-actin accumulation at the plasma membrane of RGS5+MYL9+ CAFs cocultured with HCC827 cells increased over time, leading to the formation of long, thin extensions into RGS5+MYL9+ CAFs. This accumulation was inhibited by mitoTEMPO (Fig. 5A). RhoA, a member of the Rho GTPase family, is known to be a crucial regulator of F-actin accumulation (23). We used a FRET-based biosensor, RhoA-FLARE, to visualize active RhoA directly (23). As depicted in Fig. 5B, the activity of RhoA in RGS5+MYL9+ CAF cocultured DTP cells was significantly elevated compared with RGS5+MYL9+-d CAF cocultured DTP cells. RhoA phosphorylation at s188 inactivates the RhoA-ROCK pathway (24). To further validate our findings, we conducted GTP-RhoA pull-down assays followed by Western blot analysis. The results demonstrated that coculture with RGS5+MYL9+ CAFs significantly increased levels of GTP-bound RhoA while decreasing phosphorylated RhoA at s188 in drug-tolerant HCC827 cells (Fig. 5C and D). These changes were reversed upon treatment with mitoTEMPO (Fig. 5E).
Moreover, activation of RhoA with RhoA activator enhanced GTP-RhoA activity, F-actin accumulation at the plasma membrane, mitochondrial trafficking via TNTs, and cell viability in drug-tolerant HCC827 cells cocultured with RGS5+MYL9+ CAFs. Importantly, scavenging mtROS with mitoTEMPO abrogated these effects (Fig. 5F–J). Additionally, inhibiting GTP-RhoA activity through either shRNA targeting RhoA or using the selective Rho kinase inhibitor fasudil prevented F-actin accumulation at the plasma membrane of DTP cells; conversely, activating RhoA promoted this effect. Furthermore, treatment with cytochalasin D, an actin polymerization inhibitor, and silencing Miro1 also inhibited F-actin accumulation and membrane protrusions of drug-tolerant HCC827 cells (Fig. 5K–M).
In summary, these findings demonstrate that mitochondria ROS produced by EGFR-TKIs activate RhoA in tumor cells. This activation promotes the formation of membrane protrusions supported by F-actin, creating nanobular connections with RGS5+MYL9+ CAFs. As a result, damaged mitochondria can be transferred from tumor cells to these fibroblasts.

LUAD-derived CCL11 recruits RGS5+MYL9+ CAFs to assist DTP cells in surviving EGFR-TKI–induced oxidative stress
To explore how RGS5+MYL9+ CAFs are recruited and infiltrate minimal residual tumors, we compared the cytokine profiles of HCC827 cells cocultured with RGS5+MYL9+ CAFs and RGS5+MYL9+-d CAFs. Our analysis revealed distinct cytokine secretion profiles, with RGS5+MYL9+ CAFs cocultured with HCC827 cells showing a significant increase in certain cytokines, particularly after treatment with osimertinib for 6 days (Fig. 6A and B). Among the top upregulated cytokines confirmed by ELISA, CCL11 was identified as the most overexpressed in the RGS5+MYL9+ CAF coculture system (Fig. 6C). Notably, mitoTEMPO significantly inhibited CCL11 production by DTP cells (Fig. 6D). Furthermore, using a neutralizing antibody against CCL11 (αCCL11) markedly reduced the ability of RGS5+MYL9+ CAFs to promote osimertinib resistance (Fig. 6E).
Next, we assessed the effect of CCL11 on the migration of LUAD cells and RGS5+MYL9+ CAFs. Our findings indicated that CCL11 significantly enhanced the migration of RGS5+MYL9+ CAFs, whereas αCCL11 notably inhibited this migration (Fig. 6F and G). In contrast, neither CCL11 nor αCCL11 affected the migration of HCC827 cells (Fig. 6H). When RFP-lentivirus–transduced HCC827 cells were cocultured with GFP-lentivirus–transduced HCC827 CAFs and treated with 2 μmol/L osimertinib, microscopic fluorescence imaging showed a significant increase in the accumulation of RGS5+MYL9+ CAFs around HCC827 cells over time. Importantly, treatment with either αCCL11 or mitoTEMPO significantly reduced this accumulation, whereas active CCL11 promoted it (Fig. 6I and J).
We further evaluated the therapeutic potential of inhibiting RGS5+MYL9+ CAFs infiltration using αCCL11 in tumor relapse models. HCC827 xenograft mice were randomly assigned into four groups: vehicle, αCCL11, osimertinib, or a combination of osimertinib and αCCL11 (Fig. 6K). Treatment with αCCL11 alone had minimal effects on tumor growth. Osimertinib treatment alone and its combination with αCCL11 led to substantial tumor regression within 9 days. Notably, the combination therapy significantly delayed tumor relapse and improved OS (Fig. 6L–N). Consistently, the results from immunostaining, TUNEL assays, and multiplex IF analysis demonstrated that combined treatment with osimertinib and αCCL11 markedly reduced tumor-infiltrating RGS5+MYL9+ CAFs while increasing apoptosis and decreasing tumor cell proliferation (Fig. 6O and P).
In summary, these findings indicate that CCL11 secretion from EGFR-TKI–induced LUAD cells recruits RGS5+MYL9+ CAFs, initiating a cascade for scavenging damaged mitochondria. Inhibiting RGS5+MYL9+ CAFs infiltration using αCCL11 enhances sensitivity to EGFR-TKI treatment and delays tumor relapse.

Combined treatment targeting mitochondrial transfer delays tumor relapse in vivo
As inhibiting RhoA-GTP activity reduces the formation of TNTs and DTP cells in vitro, we further explored whether the FDA-approved Rho kinase inhibitor fasudil could prevent tumor relapse and extend remission in vivo. We established xenograft models using HCC827 cells and CAFs as previously described. Following the establishment of these xenografts, mice from control group were treated with osimertinib alone for 9 days, after which, they received a vehicle until the end of the study. Other groups received a combination of osimertinib and fasudil for 9 days, followed by vehicle treatment, or continuous treatment with fasudil throughout the study period (Fig. 7A).
We assessed the effects of fasudil on MRD formation during the initial 9-day treatment phase and monitored its impact over a subsequent 15-day period representing tumor relapse. As anticipated, both osimertinib alone and the combination of osimertinib and fasudil significantly reduced tumor size within 9 days (Fig. 7B). To specifically evaluate fasudil’s role in tumor relapse, osimertinib was withdrawn after day 9. In mice with RGS5+MYL9+ CAF-HCC827 xenografts, those receiving combination therapy followed by continued fasudil treatment exhibited delayed tumor relapse and prolonged OS compared with those treated with osimertinib alone or with combination therapy without ongoing fasudil treatment. Additionally, RGS5+MYL9+-d CAF-HCC827 xenograft mice showed greater sensitivity to osimertinib alone and delayed tumor relapse and improved OS as compared with RGS5+MYL9+ CAF-HCC827 xenograft mice treated with osimertinib alone (Fig. 7B–D). However, the addition of fasudil had minimal impact on RGS5+MYL9+-d CAF-HCC827 xenograft mice compared with those treated with osimertinib alone (Fig. 7B–D). Immunostaining, TUNEL assays, and multiplex IF analysis of osimertinib-regressed HCC827 xenografts revealed that the combination of osimertinib and fasudil with ongoing fasudil treatment significantly reduced the infiltration of RGS5+MYL9+ CAFs, decreased tumor proliferation, and increased apoptosis within tumors (Fig. 7E).
Furthermore, we investigated the effects of fasudil on established MRD. Mice with RGS5+MYL9+ CAF-HCC827 xenografts were initially treated with osimertinib for 9 days to allow MRD emergence. They were then randomly assigned to two groups: one receiving osimertinib alone and another receiving a combination of osimertinib and fasudil (Fig. 7F). Notably, combined treatment significantly delayed the relapse of residual tumors that had regressed following osimertinib treatment. Moreover, mice receiving combination therapy demonstrated significantly improved OS (Fig. 7G–J). Our findings demonstrate that combined osimertinib and fasudil therapy effectively overcomes stromal-mediated resistance by disrupting the protective RGS5+MYL9+ CAF–DTP cell interaction, leading to significant reduction in residual tumor burden and delayed disease relapse in vivo.
Notably, IHC analysis of EGFR-TKI neoadjuvant therapy clinical specimens revealed increased RGS5+MYL9+ CAF infiltration and closer proximity to residual tumor cells versus treatment-naïve samples (Fig. 7K), implicating RGS5+MYL9+ CAF protection in DTP formation and TKI resistance.

Discussion
EGFR–TKIs are the standard treatment for patients with advanced EGFR-mutant LUAD. However, the development of acquired resistance to these therapies is almost inevitable (2, 25). Recent research has increasingly focused on drug-tolerant cells, which survive initial treatments through nonmutational adaptations, ultimately leading to resistance (7). In this study, we present evidence that the infiltration of RGS5+MYL9+ CAFs is correlated with EGFR-TKI resistance and poor prognosis in patients with EGFR-mutant LUAD. Our findings indicate that RGS5+MYL9+ CAFs provide a mechanism for neutralizing ROS, thereby protecting DTP cells from oxidative stress induced by TKIs. Notably, we demonstrate that the intercellular transfer of damaged mitochondria from LUAD cells to RGS5+MYL9+ CAFs via TNTs represents a novel pathway contributing to the formation of EGFR-TKI–dependent DTP cells. Importantly, targeting RGS5+MYL9+ CAF infiltration or combining osimertinib with the FDA-approved Rho kinase inhibitor fasudil effectively reduced tumor relapse in mouse models. Our study elucidates an intercellular communication mechanism involved in drug tolerance, which could be therapeutically targeted to prevent EGFR-TKI resistance and subsequent tumor relapse.
The advent of scRNA-seq has shed light on the functional heterogeneity of CAFs (26, 27), with various subsets implicated in the progression of non–small cell lung cancer (26). However, the specific heterogeneous subsets of CAFs in EGFR-mutant LUAD remain largely unexplored. In this study, we characterize the CAF subsets present in EGFR-mutant LUAD and identify RGS5+MYL9+ CAFs as key players associated with both EGFR-TKI resistance and patient prognosis. Given that PDOs serve as crucial in vitro models for precision medicine (28, 29), we established a coculture system involving EGFR-mutant LUAD-derived PDOs and CAFs to assess sensitivity to EGFR-TKIs. Our results show that RGS5+MYL9+ CAFs significantly promote resistance to EGFR-TKIs and contribute to tumor relapse. Notably, blocking the infiltration of RGS5+MYL9+ CAFs using a CCL11-neutralizing antibody resulted in substantial tumor regression, delayed relapse, and prolonged OS in vivo.
Excessive intracellular ROS can lead to cell apoptosis, and enhancing ROS levels is a critical mechanism through which chemotherapeutic agents induce cancer cell death (30). Recent studies have highlighted mitochondrial ROS release as essential for EGFR-TKI–induced apoptosis (31, 32). Downregulation of mitochondrial ROS production has been associated with resistance to erlotinib (32, 33), a second-generation EGFR-TKI. Furthermore, DTP cells exhibit robust antioxidant systems that enable them to withstand oxidative stress during antidrug treatment (30). Herein, we reveal an intercellular communication mechanism by which RGS5+MYL9+ CAFs protect DTP cells from oxidative stress induced by TKIs. We found that CCL11 released from DTP cells recruits RGS5+MYL9+ CAFs, which act as acceptor cells for damaged mitochondria transferred from DTP cells. By receiving these damaged mitochondria, RGS5+MYL9+ CAFs significantly reduce mitochondrial ROS levels within DTP cells, thus facilitating EGFR-TKI resistance.
Recent studies have demonstrated that cancer cells can acquire healthy mitochondria from stromal cells or tumor-infiltrating lymphocytes to support their growth (12, 34). For instance, research by Wang and colleagues (14) showed that mitochondrial transfer via cell adhesion contributes to chemoresistance in T-cell acute lymphoblastic leukemia cells interacting with MSCs. Various mechanisms have been described for mitochondrial transfer, including gap junctions, macrovesicle extrusion, and notably TNTs. Our findings suggest that elevated mitochondrial ROS levels, triggered by EGFR-TKIs, boost the activity of Miro1 and RhoA in DTP cells. This activation promotes the formation of F-actin–supported membrane protrusions that create nanobular connections with RGS5+MYL9+ CAFs, thereby facilitating the transfer of damaged mitochondria to these CAFs.
Fasudil is a selective inhibitor of Rho kinase approved for treating cerebral vasospasm (35); however, recent studies suggest its involvement in promoting tumor growth as well (36). In our research, we found that combining osimertinib with fasudil significantly delayed relapse in osimertinib-regressed residual tumors and extended OS in xenograft mice models. These findings highlight a complementary therapeutic target for patients with EGFR-TKI–resistant LUAD. Future multicenter clinical trials are necessary to investigate the combination of fasudil and osimertinib for treating patients with advanced EGFR-mutant LUAD.
In summary, our study demonstrates that RGS5+MYL9+ CAFs provide a mechanism for neutralizing ROS and protecting DTP cells from TKI-induced oxidative stress, thereby promoting both EGFR-TKI resistance and tumor relapse in patients with EGFR-mutant LUAD. Additionally, Miro1 and RhoA facilitate TNT formation and the transfer of damaged mitochondria toward RGS5+MYL9+ CAFs—contributing significantly to the establishment of DTP states. Clinically, we propose a combined therapeutic strategy targeting TNT formation using fasudil alongside osimertinib as a potential approach to prevent both EGFR-TKI resistance and tumor relapse.

Supplementary Material
Figure S1Figure S1. Establishment of EGFR-Mutant LUAD-Derived PDO and CAFs Coculture Model. (A) Schematic outlining the development of PDO and CAFs coculture models from patients with EGFR-mutant LUAD. (B) Representative images depicting the morphology and intracellular ATP levels assessed using the ATeam adenovirus. The time course shows the averaged YFP (red) /CFP (Green) emission ratio, with error bars indicating mean ± SD. The primary CAF subsets utilized in this figure were isolated from patients 01 to 03. Relevant clinical characteristics of these patients are summarized in Supplementary Table 1. Data were analyzed using one-way ANOVA followed by Dunnett’s test. Statistical significance is denoted as ns, not significant, *P <0.05, **P <0.01, ***P <0.001, determined by one-way ANOVA followed by Dunnett’s tests.

Figure S2Figure S2. RGS5+MYL9+ CAFs attenuate EGFR-TKI efficacy in patient-derived organoids. (A) Organoid co-cultures with RGS5+MYL9+ CAFs showed preserved viability (Calcein AM+/PI-) and morphology after 5 μM Osimertinib exposure (days 3-6) compared to RGS5+MYL9+-d CAF controls (B) ATeam biosensor analysis revealed maintained ATP levels (YFP/CFP ratio) in RGS5+MYL9+ CAFs-protected PDOs during treatment, with kinetics showing significant metabolic protection. The primary CAF subsets utilized in this figure were isolated from patients 10 to 12. Relevant clinical characteristics of these patients are summarized in Supplementary Table 1. Error bars are presented as mean ± SD, with statistical significance indicated as ns, not significant, *P <0.05, **P <0.01, ***P <0.001 based on one-way ANOVA followed by Dunnett’s tests.

Figure S3Figure S3. RGS5+MYL9+ CAFs Promote EGFR-TKI-Induced DTP. (A-B) ATeam biosensor analysis demonstrates maintained cellular viability (YFP/CFP ratio) in HCC827 (A) and PC9 (B) cells co-cultured with RGS5+MYL9+ CAFs during 2 μM Osimertinib treatment. (C-D) Immunofluorescence staining reveals elevated stemness markers (CD44/CD133) in HCC827 cells protected by RGS5+MYL9+ CAFs under Osimertinib treatment. (E) Western blot shows RGS5+MYL9+ CAF-induced epithelial-mesenchymal transition (EMT) marker expression in Osimertinib treated HCC827 cells. (F) Cell cycle analysis demonstrates RGS5+MYL9+ CAFs-mediated G0/G1 arrest in osimertinib-treated HCC827 and PC9 cells. The primary CAF subsets utilized in this figure were isolated from patients 04 to 06. Relevant clinical characteristics of these patients are summarized in Supplementary Table 1. Data represent mean ± SD (*P<0.05, **P<0.01, ***P<0.001 by one-way ANOVA with Dunnett's test; ns=not significant)

Figure S4Figure S4. RGS5+MYL9+ CAFs foster drug-tolerant persister (DTP) cells through direct cell-cell contact (A-B) Schematic of the establishment of DTP cells (HCC827 and PC9) with or without CAFs in direct contact or separated in a Boyden chamber by a porous membrane. HCC827 and PC9 cells were transduced with ATeam adenovirus before co-culturing with or without different CAF subsets, treated with Osimertinib (2μM) for indicated time. Cell viability was shown by averaged YFP (red) /CFP (Green) emission ratio. (C-D) Relative cell viability of PC9 cells co-cultured (C) (in chamber) (D) with CAF subsets treated with increasing concentration of Osimertinib (0, 0.05, 0.1, 0.5, 1 and 5μM) for 72 hours (n=3). (E) Relative cell viability of HCC827 cells co-cultured (C) (in chamber) (D) with CAF subsets treated with increasing concentration of Osimertinib (0, 0.05, 0.1, 0.5, 1 and 5μM) for 72 hours (n=3). The primary CAF subsets utilized in this figure were isolated from patients 13 to 15. Relevant clinical characteristics of these patients are summarized in Supplementary Table 1. Error bars are presented as mean ± SD, with statistical significance indicated as ns, not significant, *P <0.05, **P <0.01, ***P <0.001 based on one-way ANOVA followed by Dunnett’s tests.

Figure S5Figure S5. RGS5+MYL9+ CAFs mitigate oxidative stress in drug-tolerant persister cells. (A) KEGG pathway analysis of HCC827 cells co-cultured with CAFs reveals significant alterations in oxidative stress-related pathways. (B) KEGG pathway enrichment of between parental HCC827 cells and Osimertinib-induced DTP cells. (C-D) GSEA confirms differential expression of oxidative phosphorylation and reactive oxygen species metabolic pathways in DTP cells compared to parental controls. *P <0.05, **P <0.01, ***P <0.001.

Figure S6Figure S6. RGS5+MYL9+ CAFs facilitate drug-tolerant persister (DTP) formation and tumor regrowth after Osimertinib withdrawal. (A-B) Colony formation assays demonstrate enhanced proliferative recovery of (A) HCC827 and (B) PC9 cells co-cultured with RGS5+MYL9+ CAFs compared to RGS5+MYL9+-d CAFs controls following cyclic Osimertinib treatment (2 μM, 9 days) and withdrawal (n=3). Experiments were performed using paired CAF subsets isolated from three treatment-naïve EGFR-mutant LUAD patients. The primary CAF subsets utilized in this figure were isolated from patients 13 to 15. Relevant clinical characteristics of these patients are summarized in Supplementary Table 1. Data are shown as mean ± SD and were analyzed by a two-tailed unpaired t-test, *P <0.05, **P <0.01, ***P <0.001.

Figure S7Figure S7. RGS5+MYL9+ CAFs protect against EGFR-TKI-induced mitochondrial dysfunction in vitro. (A-F) Schematic of the establishment of drug tolerant HCC827 (DTPs) and rederived DTPs cocultured with either RGS5+MYL9+ CAFs or RGS5+MYL9+-d CAFs. (B-E) Flow cytometry is utilized to measure total mitochondrial mass, damaged mitochondria, and MitoSOX levels in HCC827 cells and their cocultured CAFs. (F) Representative electron micrographs display damaged mitochondria in HCC827 cells cocultured with different CAF subsets. (G-J) Schematic representation describes the establishment of drug-tolerant PC9 cells and their rederived counterparts cocultured with RGS5+MYL9+ CAFs or RGS5+MYL9+-d CAFs. (H) Relative cell viability data for PC9 cells. (I-J) Representative electron micrographs illustrate damaged mitochondria in PC9 cells cocultured directly or within a chamber setup with various CAF subsets. The primary CAF subsets utilized in this figure were isolated from patients 04 to 06. Relevant clinical characteristics of these patients are summarized in Supplementary Table 1. Error bars represent mean ± SD, with significance indicated as ns not significant, *P <0.05, **P <0.01, ***P <0.001 according to independent sample t-tests.

Figure S8Figure S8. RGS5+MYL9+ CAFs protect against EGFR-TKI-induced mitochondrial dysfunction in vivo. (A) Flow cytometry analysis assesses total mitochondrial mass, damaged mitochondria levels in tumor cells and CAFs isolated from xenograft mice in Figure 2J. (B) Flow cytometry analysis assesses MitoSOX levels in tumor cells and CAFs isolated from xenograft mice in Figure 2J. Error bars represent mean ± SD, with significance indicated as ns not significant, *P <0.05, **P <0.01, ***P <0.001 according to independent sample t-tests.

Figure S9Figure S9. Quantification of mitochondrial transfer between RGS5+MYL9+ CAFs and HCC827 cells. (A) KEGG functional enrichment analysis identifies signaling pathways associated with different CAF subsets derived from scRNA−seq. (B−C) Experimental design to quantify mitochondrial transfer from CAFs to HCC827 cells. MitoDsRed-labelled CAFs are cultured alongside GFP-labelled HCC827 cells either in direct contact or separated by a porous membrane within a Boyden chamber setup. (C) Flow cytometry analysis evaluates transferred mitochondria and assesses MitoSOX levels within these transferred mitochondria. The primary CAF subsets utilized in this figure were isolated from patients 07 to 09. Relevant clinical characteristics of these patients are summarized in Supplementary Table 1. Error bars indicate mean ± SD, with statistical significance marked as ns, not significant, *P <0.05, **P <0.01, ***P <0.001 based on independent sample t-tests.

Figure S10Figure S10. Enhanced stromal-tumor proximity in drug-resistant niches. (A-B) Quantitative analysis of CAF-tumor cell distances in (A) treatment-naïve versus (B) Osimertinib-resistant tissues (corresponding to Figure 1J), demonstrating significantly closer association in resistant tumors. (C-D) Spatial analysis of CAF distribution in (C) untreated versus (D) relapsed tumors (from Figure 2Q), revealing preferential peri-tumoral localization of RGS5+MYL9+ CAFs following therapy. Error bars indicate mean ± SD, with statistical significance marked as ns, not significant, *P <0.05, **P <0.01, ***P <0.001 based on one-way ANOVA followed by Dunnett’s tests.

Table S1Table S1