Establishment and characterization of an orthotopic implanted lung cancer model to mimic human tumor structure, microenvironment, and metastatic spread

Abstract

Background

Subcutaneous (SC) lung tumor models are widely used in preclinical studies due to their technical simplicity but fail to recapitulate the complex microenvironment, immune landscape, and metastatic behavior of human lung cancers. These limitations hinder the translational value of such models, particularly in evaluating immunotherapies and metastasis-related mechanisms. There is a critical need for more physiologically relevant in vivo models that better reflect clinical tumor characteristics and disease progression. To address these limitations, we sought to develop a reproducible orthotopic lung cancer (LuO) model that enables detailed study of tumor progression, immune infiltration, and metastatic dynamics.

Methods

We established and characterized a thoracotomy-based LuO model using a panel of human and murine lung cancer cell lines implanted into the pulmonary parenchyma of immunodeficient and syngeneic mice. Tumor progression was monitored longitudinally using bioluminescence imaging (BLI) and micro-computed tomography (CT). Comparative analyses with SC tumors were performed using immunohistochemistry, multiplexed immunofluorescence, transcriptomic and proteomic analyses. Circulating tumor cells (CTCs) and spontaneous metastases were isolated and functionally characterized.

Results

The orthotopic model reliably generated solitary intrapulmonary tumors that closely mimic human lung cancer in growth pattern, vascularization, and progression. Compared to SC tumors, orthotopic tumors exhibited significantly enhanced vascular density, reduced hypoxia and DNA damage, and increased proliferation. Immune profiling revealed enriched and spatially organized infiltrates of CD4⁺, CD8⁺ T cells, dendritic cells (DCs), and myeloid populations in orthotopic tumors, forming structures analogous to those found in patient tumors. Moreover, orthotopic tumors released CTCs capable of forming spontaneous and site-specific metastases to clinically relevant organs. Transcriptomic and proteomic profiling of metastasis-derived cell lines uncovered conserved pro-metastatic signatures and niche-specific adaptations.

Conclusions

This LuO model offers a reproducible, clinically relevant platform that captures important aspects of human lung cancer biology, including immune landscape, tumor microenvironment (TME), and metastatic progression. Its superior anatomical and immunological fidelity makes it a valuable preclinical tool for evaluating therapeutic strategies and dissecting molecular mechanisms of metastasis.

Highlight box.

Key findings

• This study establishes a robust thoracotomy-based orthotopic lung cancer model that provides a more physiologically relevant environment for lung tumor growth than conventional subcutaneous models. The model reliably generates solitary, anatomically localized tumors with enhanced vascularization, reduced hypoxia and stress, and a more physiologically relevant immune microenvironment. It supports spontaneous metastasis to clinically relevant organs and enables isolation of circulating tumor cells (CTCs) and metastatic subclones with patient-like molecular profiles.

What is known and what is new?

• Subcutaneous murine models are widely used in lung cancer research but fail to replicate the native lung environment, limiting their utility for studying tumor progression, immune responses, and metastasis. While genetically engineered and airway-based models exist, they often produce multifocal tumors or fail to capture realistic disease progression.

• This work introduces a technically optimized and reproducible orthotopic model that generates single-site tumors within the lung parenchyma, enabling longitudinal imaging and functional assessment of immune responses and metastatic spread. The study also presents multi-omic characterization of metastasis-derived lines, highlighting conserved and niche-specific metastatic programs reflective of human disease.

What is the implication, and what should change now?

• The findings underscore the need to shift from subcutaneous to orthotopic lung models in preclinical research to improve translational relevance. Adoption of this model can enhance the predictive value of immunotherapy and metastasis studies and support the development of more effective therapeutic strategies. Future research should prioritize orthotopic systems for evaluating drug responses and dissecting tumor-immune interactions in lung cancer.

Introduction

Lung cancer is the leading cause of cancer-related mortality worldwide, representing 12–13% of all new cancer cases and showing the highest overall cancer death rate (almost one in five deaths of cancer) among both sexes (1,2). Alarmingly, two-thirds of patients are diagnosed at advanced stages, with limited curative options and poor survival outcomes. Even in early-stage disease, the risk of relapse remains high, exceeding 20% within 5 years of diagnosis (3). Projections indicate that by 2035, low-income regions will experience a 144% increase in new cancer cases, highlighting the urgent need for more effective prevention and translational research strategies (4).

To achieve meaningful impact at the clinical level, robust and accurate preclinical models are essential tools. Optimal cancer models should yield results which could be readily and successfully translated to patient trials and, eventually, to novel therapies. Murine models are the cornerstone of preclinical cancer research. In lung cancer, three main strategies have been used so far to develop lung cancer in mice: spontaneous tumorigenesis after exposure to carcinogens, genetically engineered modified mice (GEMMs) (5), based on lung targeted alterations in oncogenes or tumor-suppressor genes; and, finally, a variety of strategies based on the engraftment of human or mouse tumor cell lines in different murine strains (immunodeficient or immune-humanized and immunocompetent respectively), constituting allogeneic or syngeneic models (6). Engraftment models are the most frequently used for therapy evaluation. Among the different approaches, the subcutaneous (SC) injection of the tumor cells is widely preferred due to its technical simplicity and analytical reproducibility in assessing therapeutic responses. However, these SC models introduce lung tumor cells into a non-native location, which may alter critical features highly associated to the tumor development such as blood supply, oxygenation, and immune cell interactions-factors together with other key aspects of the native tumor microenvironment (TME) (7). This artificial skin environment differs markedly from the native biological context of the lung parenchyma and airways, where lung tumors originate, grow, and metastasize.

This study aims to determine whether the distinct biological environments influence tumor-intrinsic or microenvironment-dependent properties, potentially limiting the relevance of SC models. This concern is particularly relevant for therapies such as radiotherapy (8) and immunotherapy, which are strongly shaped by immune landscape dynamics (9,10). In response to these potential limitations, the orthotopic lung cancer (LuO) model approach has gained traction as a more physiologically relevant alternative. By implanting tumor cells directly into the lung, the LuO models provide a closer experimental setting to study lung cancer biology, including tumor growth and metastatic spread (11). Although technically demanding and more time-consuming than SC models, the orthotopic approach enables precise tumor localization-typically as a single nodule-allowing controlled assessment of treatment responses. Tumor growth can be monitored via bioluminescence imaging (BLI), using luciferase-transfected cells, or high-resolution micro-computed tomography (CT), which closely parallels clinical practice (12,13). The consistent engraftment of a single tumor nodule in LuO models offers an advantage over GEMMs and carcinogen-induced models, where multiple tumors develop throughout the lungs. In contrast, early-stage human lung cancer usually presents with a single, or occasionally a few, synchronous nodules rather than widespread, simultaneous lesions (14).

Given the increasing need for predictive preclinical models, especially for novel drugs and other therapies evaluation, a comprehensive comparative analysis between SC and LuO models is critical. Such a study will clarify the strengths and limitations of each approach, offering valuable insights for selecting the most appropriate model depending on the research objective. Ultimately, ensuring that preclinical findings reflect clinical realities is key to advancing effective therapies and eventually improving patient outcomes. We present this article in accordance with the ARRIVE reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-871/rc).

Methods

Cell lines and reagents

Human lung cancer cell lines were obtained from the American Type Culture Collection (ATCC). Murine lung cancer cell lines were obtained either from the ATCC, in-house generation or through previous collaborations. Cells were maintained in recommended medium, supplemented with 10% FetalClone serum (Cytiva, Marlborough, MA, USA; Cat. #SH30066.03), 100 U/mL penicillin, and 100 U/mL streptomycin (Gibco, Thermo Fisher Scientific, Waltham, MA, USA; Cat. #15140122). Cultures were maintained at 37 ℃ in a humidified atmosphere containing 5% CO₂. Only mycoplasma-negative cells, as verified using the MycoAlert Mycoplasma Detection Kit (Lonza, Basel, Switzerland; Cat. #LT07-418), were used for experiments.

A total of 16 different cell lines (both human and murine) were used in used in this project to develop their respective orthotopic models and are listed in https://cdn.amegroups.cn/static/public/tlcr-2025-871-1.xlsx.

In vivo studies

Experiments were performed under a project license (062-22 and 067-24) granted by the Ethics Committee for Animal Experimentation of the Universidad de Navarra in compliance with institutional guidelines for the care and use of animals. A protocol was prepared before the study without registration.

Thoracotomy implanted lung orthotopic model

For orthotopic tumor growth, cultured cells were firstly stably transfected with a triple-modality reporter construct encoding green fluorescent protein (GFP), firefly luciferase, and herpes simplex virus thymidine kinase (HSV-TK) (15) or luciferase-reporter plasmid. Six- to eight-week mice (Harlan-Winkelmann) were used. For human cell lines, Rag^−/− IL2Rg^−/− immunocompromised mouse strain was used. For mouse syngeneic cell line transplants, mice from the strains corresponding to the background of the derived cell line were used for the experiments (available online: https://cdn.amegroups.cn/static/public/tlcr-2025-871-1.xlsx). Specifically, for orthotopic Lewis lung carcinoma (LLC) experiment, 1×10⁴–1×10⁵ cells were injected into seven-week C57BL6/J (Harlan-Winkelmann) female mice (n=5) lung parenchyma. Mice were anesthetized with 1.5% isoflurane (Isovet, B. Braun, Melsungen, Germany; Cat. #469860) and positioned in right lateral decubitus. A 1–2 cm skin incision was made below the left scapula to expose the thoracic wall, and 1×10⁴–1×10⁶ luciferase-expressing cells were orthotopically injected into the left pulmonary lobe via a minimal thoracotomy. For cell administration, 3–4 mm of a fine-gauge needle (Micro Fine, BD, Franklin Lakes, NJ, USA; Cat. #324824) was injected in the intercostal space and content was carefully injected. Ketoprofen (Dinalgen, Ecuphar, Barcelona, Spain; Cat. #574099) was administered at a dose of 2 mg/kg via SC injection once daily for three consecutive days, starting immediately after the surgical procedure.

Tumor growth was monitored by BLI every 6 days, and by CT once a week.

For BLI, mice were anesthetized with 1.5% isoflurane and administered 100 µL of D-luciferin (150 mg/mL; 30 mg/kg body weight) (Promega, Madison, WI, USA; Cat. #E1602) intraperitoneally. Images were acquired using the IVIS Spectrum Imaging System (PerkinElmer, Waltham, MA, USA). Signal quantification was performed using Living Image software (PerkinElmer) by drawing regions of interest (ROIs) and expressing photon flux as photons/s/cm²/sr.

To monitor anatomical tumor progression, full-thorax 3D micro-CT scans were performed weekly using the Quantum-GX micro-CT scanner (PerkinElmer). Mice were anesthetized with ketamine (100 mg/kg) and xylazine 10 mg/kg) and scanned under respiratory gating conditions. Reconstructed 3D image datasets were processed and analyzed using the 3D Slicer open-source (https://www.slicer.org/) imaging platform.

Following euthanasia, lungs and total blood were collected. Lungs were excised, fixed in 3.7% to 4.0% formaldehyde (w/v), pH 7, stabilized with methanol (AppliChem, Darmstadt, Germany; Cat. #252931) and embedded in paraffin. 3 µm tissue sections were prepared and stained with hematoxylin and eosin (H&E) for histological evaluation by microscopy.

SC model

1×10⁶–4×10⁶ were subcutaneously inoculated in one flank of eight-week-old mice from strains matching the genetic background of the derived cell line (Harlan-Winkelmann) as previously described (16).

Co-registration of bioluminescence and CT imaging

For spatial alignment of BLI and CT datasets, images were co-registered using either fiducial markers embedded in the IVIS/CT multimodal imaging bed, or by referencing anatomical landmarks visible in both modalities. Image fusion was performed using Living Image Software (PerkinElmer).

Metastatic cell line isolation

At lung orthotopic model’s experimental endpoint, blood was collected by intracardiac puncture under deep anesthesia using EDTA-coated Microvette 500K3E tubes (Sarstedt, Nümbrecht, Germany; Cat. #20.1341). Collected blood samples were processed for CTCs isolation by culturing in complete DMEM medium. After an initial expansion period, GFP-positive cells were isolated using fluorescence-activated cell sorting (FACS) (MoFlo Astrios, Beckman Coulter, Brea, CA, USA).

To develop experimental metastasis models, 1×10⁵ LLC-CTC or A549-CTC cultured cells were injected into the left cardiac ventricle of six-week-old C57BL/6J or Rag^−/− IL2Rγ^−/− female mice (Harlan-Winkelmann) respectively, under isoflurane anesthesia (n=10). Metastatic progression was monitored weekly by BLI using the IVIS Spectrum system.

At study endpoint, animals were euthanized and ex vivo BLI was performed on limbs, adrenal glands, brain, and liver to assess organ-specific metastatic burden. Organs containing detectable metastases were either processed for cell culture or fixed in 4.0% formaldehyde and paraffin embedded. For histological analysis, 3 µm sections were stained with H&E.

To isolate tumor cells from metastatic tissues, organs were incubated in a 2 mg/mL collagenase I (Roche, Basel, Switzerland; Cat# 10103578001) and 100 µg/mL DNAse I (Roche, Cat# 10104159001) digestion buffer for 30 minutes at 37 ℃ (Sigma-Aldrich, Burlington, MA, USA; Cat. #10103578001). Following enzymatic digestion, tissues were mechanically dissociated using a tissue grinder (GentleMacs, Miltenyi Biotec, Bergisch Gladbach, Germany). Cell suspensions were then plated and cultured in complete DMEM medium supplemented with antibiotic—antimycotic agents (ThermoFisher Scientific, Waltham, MA, USA; Cat. # 15240062) for further propagation and analysis.

Tumor microarray (TMA) generation

TMAs were generated from 16 human and mouse lung cancer cell lines implanted either subcutaneously or orthotopically in mice (available online: https://cdn.amegroups.cn/static/public/tlcr-2025-871-1.xlsx). SC tumor cores were subcategorized into central and peripheral regions, while LuO cores were subclassified by lung parenchymal or pleural location when pleural seeding/expansion occurred.

Immunohistochemistry (IHC)

IHC staining of CD31, P21, P16, GLUT1, HIF1-a, KI67, Cleaved Caspase 3, Histone H2AX and Cyclin D1 was performed on formalin-fixed paraffin-embedded (FFPE) sections. After dewaxing, sections were hydrated and endogenous peroxidase activity was blocked with 3% H₂O₂ for 10 min. Heat-mediated antigen retrieval was conducted either with citrate buffer (pH 6) (Vitro S.A., Sevilla, Spain; Cat #MAD-004071R/D) or Tris-EDTA (pH 9) (Vitro S.A., Cat #MAD-004070R/D). Sections were incubated with primary antibodies diluted in a special buffer (Antibody diluent, Dako, Glostrup, Denmark; Cat #S2022) overnight at 4 ℃ in a humidity chamber. After a rinse with tris-buffered saline (TBS), sections were incubated for 30 min with mouse or rabbit EnVision complex (Dako, Cat #K4001 and Cat#K4003). Color was developed with 3,3'-diaminobenzidine (DAB; Dako, Cat #K3468) under microscopic control. Finally, tissue samples were counterstained with hematoxylin and mounted with distyrene, plastisizer, xylene (DPX). Antibodies characteristics and conditions for detection by IHC in FFPE samples are summarized in https://cdn.amegroups.cn/static/public/tlcr-2025-871-1.xlsx.

Multiplexed immunofluorescence

Two fluorescence-based multiplex immunofluorescence (mIF) panels (VECTRA^®; Akoya Biosciences, Marlborough, MA, USA) were used to profile the TME in FFPE tissues from SC and LuO tumors, focusing on lymphoid and myeloid immune cell populations. Multiplex staining was performed according to the manufacturer’s protocol using murine-specific reagents (Akoya, CAT#NEL840001KT), including additional opal fluorophores for enhanced target resolution. Antigen retrieval was performed using citrate buffer (pH 6) or Tris-EDTA buffer (pH 9), depending on the target. Tissue microarrays (TMAs) were processed for multispectral imaging and analyzed accordingly. Detailed information on antibody targets, sources, clones, dilutions, incubation conditions, antigen retrieval buffers, Opal fluorophore assignments, and panel sequences is provided in https://cdn.amegroups.cn/static/public/tlcr-2025-871-1.xlsx.

Image analysis

TMA slides stained by IHC were scanned using the Aperio CS2 brightfield scanner (Leica Biosystems, Barcelona, Spain) at 20× magnification. Whole-slide images were imported into QuPath (version 5.1) for analysis. TMA cores were identified using the dearraying tool, followed by manual adjustment to correct misalignments and ensure accurate core positioning. Tissue detection was applied to each core to exclude areas of artifact, folding or missing tissue. Cell detection was performed using the StarDist algorithm integrated into QuPath, which enables robust and precise nuclear segmentation even in densely packed or heterogeneous tissues. When applicable, tumor regions were segmented manually or semi-automatically in order to restrict marker quantification to tumor areas only. Staining intensity from the DAB chromogen was quantified within cellular compartments using user-defined thresholds.

For mIF analysis, images were acquired using the VECTRA Polaris imaging system (Akoya Biosciences) and spectrally unmixed using inForm software (Akoya Biosciences) to separate individual fluorophores. The resulting multichannel TIFF files were imported into QuPath (version 0.5.1) for analysis.

TMA cores were automatically identified using the dearraying tool in QuPath, as with IHC slides. Nuclear detection was performed using the StarDist algorithm, optimized for the nuclear 4’,6-diamidino-2-phenylindole (DAPI) signal.

Marker expression was quantified within individual cells across all fluorescent channels. To classify cells into phenotypic subpopulations, we used QuPath’s object classification framework based on supervised machine learning. For each marker, representative training data were generated by manually annotating cells as positive or negative based on marker expression. These annotations were used to train separate classifiers for each target marker. Classifiers were built using the Random Trees algorithm within QuPath, and the model performance was evaluated by visually inspecting classification overlays and adjusting the training set as needed to improve accuracy.

Cell classifiers were then applied sequentially across the entire dataset in a batch-processing workflow to ensure consistent application of thresholds and classification rules. The resulting classified cell populations were used for downstream quantification and statistical analysis.

In the comparison of SC and LuO TMAs, each value corresponds to a core. For metastatic derived subclone’s imaging and analysis, multiple ROIs were selected from each sample. Each ROI corresponded to a single field captured at high magnification, enabling quantification of target markers within that specific area. Data were collected from all ROIs on each slide, rather than from a single low-magnification image of the entire tissue.

IHC and IF raw data related to each TMA core can be found in https://cdn.amegroups.cn/static/public/tlcr-2025-871-2.xlsx.

Quantitative spatial profiling of immune phenotypes in TMEs

We applied three complementary methods to quantify spatial interactions among cellular phenotypes in each tumor core: (I) average minimum interphenotype distance; (II) spatial clustering analysis; and (III) lymphoid-myeloid panel registration for integrated spatial metrics.

Average minimum distance

To assess spatial proximity between two phenotypes (A and B), we calculated Euclidean distances from each cell of phenotype A to all cells of B within the same core, selecting the smallest value. This was repeated across all A cells, and the resulting values were averaged. The process was then reversed (B to A), since this metric is directionally asymmetric.

Spatial cell clustering

To detect intra-phenotype spatial clusters, we constructed a k-nearest neighbor (kNN) graph where nodes represented cells and edges reflected centroid distances. The Leiden algorithm (17) was applied to identify high-modularity clusters. Cluster density was defined as the number of cells per unit area within a convex hull enclosing the cluster. To mitigate outlier effects, cells >2 standard deviations (SDs) from the cluster centroid were removed before recalculating hull area and density.

Registration and interpanel spatial analysis

To evaluate spatial interactions between lymphoid and myeloid phenotypes, a rigid registration using DAPI signals (shared between panels) was performed for each sample core (18). The DAPI channel from the lymphoid panel core was used as the fixed reference; the myeloid panel core was rotated and translated to maximize image correlation, and this transformation was applied to all channels.

Post-alignment, we calculated cross-panel average minimum distances and identified multi-phenotype clusters. Seven samples were excluded due to poor registration quality. To quantify potential antigen presentation zones, we analyzed dendritic cell (DC) clusters by drawing a 150 µm radius around each DC and counting neighboring CD4⁺ and CD8⁺ T cells. Local T cell density was estimated as the number of T cells per π·r², reflecting spatially enriched immune interactions.

Transcriptomic analysis

RNAseq was carried out at the Genomics Unit of the Center for Applied Medical Research (CIMA, Universidad de Navarra) as previously described (19).

For transcriptomic studies, LLC and A549 parental, CTC and bone subclones were analyzed.

For data analysis, raw paired-end sequencing reads (FASTQ files) were trimmed to remove adapter sequences and low-quality bases using fastp. Genome indices for both human and mouse were generated from the corresponding Ensembl GRCm39 reference assemblies with STAR, and reads were aligned to their respective genomes generating BAM files. The quality of these files was checked with MultiQC reports. Read-level quantification was then performed by counting reads per gene using featureCounts.

Differential gene expression analysis was performed in R using the EdgeR package. Genes with a false discovery rate (FDR)-corrected P value <0.05 and | log₂ fold change | >2 were considered significantly differentially expressed. Raw data and search results files generated in this study are publicly available in Gene Expression Omnibus (GEO) at GSE301347 and GSE301349.

Proteomic analysis

Proteomic experiments were performed as previously described (20). Briefly, protein extracts from LLC and A549 parental, CTC and bone-derived subclones cell pellets were prepared using lysis buffer with inhibitors, quantified by Bradford assay, and analyzed through shotgun proteomics involving in-solution digestion, LC-MS/MS, and data processing with MaxQuant and Perseus for statistical analysis and visualization. Proteomic data were analyzed with R. Proteins with abs(log₂FC) >0.38 and P value >0.05 were considered differentially abundant proteins. MS data and search results files were deposited in the Proteome Xchange Consortium via the JPOST partner repository (https://repository.jpostdb.org) with the identifier PXD065211 for ProteomeXchange and JPST003884 for jPOST.

Statistical analysis

Sample sizes were determined based on previously published protocols from the authors’ group. When possible, all experiments were performed with a minimum of three independent biological replicates. For comparison of two groups, normality (Shapiro-Wilk test) and variance (Levene test) were assessed. Data from normally distributed samples were compared using a two-tailed t-test, while non-normally distributed data were analyzed using the Mann-Whitney U test (for equal variances) or a median test (for unequal variances). For comparisons involving more than three groups, analysis of variance (ANOVA) was used. The area under the curve (AUC) was calculated to analyze experiments with more than two experimental time points. Data was analyzed with GraphPad Prism 8 software (GraphPad). Values are expressed as median with interquartile range (IQR), and statistical significance was defined as P<0.05. P values greater than 0.05 (P>0.05) are indicated as ns (not significant) in the figures.

Illustrations

All illustrations were created using BioRender.com.

Results

Setting up of a clinically relevant cell line-independent orthotopic implanted lung cancer model with high similitudes to human patient lung cancer

To develop a clinically relevant preclinical lung cancer model, we established an orthotopic murine model by injecting luciferase-expressing lung adenocarcinoma cells (LLC) into the lung parenchyma via thoracotomy (Figure S1A), followed by longitudinal monitoring using both BLI and CT (Figure 1A, Figure S1B). This approach consistently resulted in the formation of tumors which are highly similar to those diagnosed in human patients. They grow as single, well-defined tumor nodules; localized at the injection site within the lung parenchyma (Figure 1B), resembling the presentation of primary lung tumors in patients, which in most cases originate in a single neoplastic lesion. Tumor growth was successfully monitored over time using CT imaging, which revealed progressive volume increase from days 7 to 14 post-injection (Figure 1C). Using both CT and BLI, we were able to detect tumor presence from day 6, confirming tumor cell engraftment in 100% of the animals. The combination of these complementary imaging techniques allowed us to monitor both the tumor mass (via CT) and the metabolic activity of luciferase-expressing cells (via BLI), providing robust evidence of early and consistent tumor establishment. 100% of tumor grew across all tested cell doses (1×10⁴, 5×10⁴, and 1×10⁵), and volumetric analysis confirmed a positive correlation between injected cell number and tumor size (Figure 1D,1E). Interestingly, regardless of the number of cells injected, intrapulmonary tumors invaded the entire affected lung lobe by day 21. Co-registration of CT and BLI enabled precise spatial localization of tumor nodules and real-time tracking of tumor progression (Figure 1F,1G and Video S1). These data demonstrate that, similar to humans, tumors in this model are clearly detectable and quantifiable in live animals using imaging techniques, enabling reliable longitudinal monitoring of tumor progression. Interestingly, BLI signal intensity varied depending on the imaging angle, with significant correlations observed between photon flux and CT-derived tumor volume in ventral and lateral positions, but not dorsally (Figure 1H,1I). Moreover, increased tumor burden correlated with reduced body weight in mice, mirroring cachexia-like symptoms seen in advanced-stage lung cancer patients (Figure 1J). Finally, using CT imaging, we quantified isolated cases of tumor spreads to the contralateral lung lobe, mirroring the dissemination pattern observed in some studies such as SEER lung cancer patients (21). While population-based data suggest that approximately 3–5% of non-small cell lung cancer (NSCLC) patients present with contralateral pulmonary nodules at diagnosis, stratified data for adenocarcinoma are scarce; given that adenocarcinoma accounts for the majority of NSCLC cases, our observation of a ~10% contralateral metastatic rate in animal models is consistent with, and may slightly overrepresent, the human clinical scenario, thereby reinforcing the translational relevance of our experimental system. In contrast, this percentage is overestimated when assessed by BLI (Figure 1K).

Figure 1.

Establishing an orthotopic lung cancer model via thoracotomy with single tumors monitored over time. (A) Schematic representation of the in vivo experimental design. 1×10⁴–1×10⁵ LLC murine lung adenocarcinoma tumor cells were injected into lung parenchyma of 7-week-old C57BL/6 mice through thoracotomy (n=5). Tumor growth was monitored over time using BLI and CT. At the endpoint of the experiment (day 18 after cell injection), total blood was collected though intracardiac puncture for CTC isolation and lungs were collected for tumor tissue analysis following histological processing. (B) Histological section of the left lung lobe from an orthotopic syngeneic mouse model, 8 days post-injection of LLC cells. The image shows a H&E-stained section highlighting a well-defined tumor nodule within the lung parenchyma. LLC cells were orthotopically implanted, and the section illustrates tumor establishment and local infiltration into pulmonary tissue architecture. (C) CT images of an orthotopic lung cancer model at days 7 and 14 post-implantation. Axial, sagittal, and coronal CT planes show the progression of a single tumor nodule in the lung parenchyma (pointed with a red arrow) following orthotopic injection of LLC cells. The images demonstrate tumor growth over time, with increased tumor volume evident at day 14 compared to day 7. (D) Tumor volume of orthotopic LLC tumors quantified from CT images. The graph shows tumor volumes over time in mice injected with 1×10⁴, 5×10⁴, or 1×10⁵ LLC cells into the lung parenchyma. Tumor volumes were calculated from CT scans and are presented as mean ± SEM (n=5 per group). No statistically significant increase in tumor volume was observed with higher cell doses. Statistical analysis was performed using ANOVA. (E) Relative growth of orthotopic LLC tumors at day 14 compared to day 7. Tumor growth is expressed as a percentage of volume increase from day 7 to 14 post-implantation, based on CT measurements. Data are presented as mean ± SEM. Statistical analysis was performed using t-test, with significance defined as P<0.05. ns stands for non-significant (P>0.05). (F) Co-registered bioluminescence and CT image of a mouse with an orthotopic LLC tumor at day 8 post-implantation at coronal plane. The co-registered imaging displays a tumor nodule located in the thoracic region of the mouse, visualized by the overlap of BLI signal and anatomical CT data. This multimodal approach enables precise localization and assessment of tumor burden in vivo. Spatial references are represented using x, y, z coordinates, where negative signs point left, down and front, respectively. (G) Tumor localization reconstruction within a coronal plane of the lung in an orthotopic LLC model at day 14 post-inoculation, based on CT imaging. The tumor’s position within the lung parenchyma was reconstructed using 3D Slicer software, providing a three-dimensional visualization of tumor growth and anatomical context at day 14 post-inoculation. Spatial references are represented using x, y, z coordinates, where negative signs point left, down and front, respectively. (H) Representative BLI and average radiance of orthotopic LLC tumors in dorsal, ventral, and lateral positions. BLI and corresponding average radiance data are shown for three groups of mice injected with 1×10⁴, 5×10⁴, and 1×10⁵ LLC cells into the lung parenchyma (n=5). Images were acquired from the dorsal, ventral, and lateral positions to assess tumor burden across different orientations. Data are presented as mean ± SEM. Statistical analysis was performed using ANOVA. ns stands for non-significant (P>0.05). (I) Correlation between tumor volume (calculated from CT) and bioluminescence in orthotopic LLC tumors. Pearson correlation analysis shows a significant positive correlation between tumor volume and bioluminescence in the ventral (r=0.364, n=15, P=0.01) and lateral (r=0.617, n=15, P=0.0005) positions, but no significant correlation in the ventral position. Data are presented as mean ± SEM. (J) Correlation between mouse weight (g) and tumor weight (g). Pearson correlation analysis reveals a significant negative correlation between animal weight and tumor weight (r=0.328, n=15, P=0.02) at the endpoint of the orthotopic tumor experiment. Data are presented as mean ± SEM. (K) Left: dot plot representing the tumor-occupied volume in the left and right lungs, as quantified from CT scans. Right: corresponding BLI signals for each lung, serving as an independent measure of tumor burden. Data are shown as mean ± SEM. Statistical significance was assessed using a two-tailed t-test (P<0.05). 3D, three dimensional; ANOVA, analysis of variance; BLI, bioluminiscence imaging; CT, computed tomography; CTC, circulating tumor cell; H&E, hematoxylin and eosin; LLC, Lewis lung carcinoma; LuO, orthotopic lung cancer; SEM, standard error of the mean.

Video S1.

Download video file ^{(4.5MB, mp4)}

Co-registered bioluminescence and computed tomography (CT) video of a mouse with an orthotopic LLC tumor at day 8 post-implantation.

These results suggest that the model provides a robust, reproducible platform for studying LuO with high anatomical precision and translational relevance.

Orthotopic lung tumors display enhanced vascularization and proliferative capacity compared to SC models

Based on the premise that tumor biology is niche-dependent, we aimed to investigate and better understand the molecular and functional characteristics of the tumor by comparing the lung orthotopic implanted model with the more commonly used SC model in preclinical research. To this aim, we established syngeneic and xenograft tumors from 16 lung cancer cell lines-representing major histological subtypes (adenocarcinoma, squamous cell carcinoma, and small cell carcinoma) implanted either subcutaneously or orthotopically in mice. Once tumors reached endpoint criteria, tumors were collected and processed for paraffin embedding and TMA construction. TMAs were assembled using representative cores of 1.2 mm diameter from peripheral and central regions of SC tumors, and from parenchymal and pleural areas of LuO tumors. IHC analysis of key cancer hallmarks (22) revealed distinct molecular profiles between the models (Figure S2A). Compared to SC tumors, LuO tumors displayed significantly higher microvascular intratumoral density (CD31, P<0.0001), suggesting a highly vascularized microenvironment. Consistently, LuO tumors showed lower expression of the hypoxia marker HIF-1α (P<0.0001), reduced DNA damage (γH2AX, P<0.0001), and decreased expression of p16 and p21 markers (P<0.0001), indicative of a less stressed and more proliferative tumor context. Cyclin D1 expression was significantly increased in LuO models (P=0.04), supporting enhanced cell cycle progression (Figure 2A-2E). Further stratification of LuO tumors revealed higher vascularization and GLUT1 expression in parenchymal regions compared to pleural sites (P<0.05), underscoring potential specific physiological conditions when the tumor grows in the pleural fluid environment (Figure S2B).

Figure 2.

Differential cancer hallmark profiling of LuO and SC tumor models. (A) Representative IHC images of CD31, HIF1-α, P16, P21, Cyclin D1 and H2AX in LuO (parenchyma and pleura) and SC (central and periphery) in mouse (scale bar: 100 µm). (B) Representative IHC images of CD31, HIF1-α, P16, P21, Cyclin D1 and H2AX in LuO (parenchyma and pleura) and SC (central and periphery) in human (scale bar: 100 µm). (C) Comparison of IHC markers between LuO and SC models in mouse. Each data point represents the quantification of one TMA core. (D) Comparison of IHC markers between LuO and SC models in human. Each data point represents the quantification of one TMA core. (E) Illustration summarizing main differences between LuO and SC models. Compared to SC tumors, LuO models showed significantly higher vascularization (CD31 area, P<0.0001), reduced hypoxia (HIF1-α, P<0.0001), lower senescence marker expression (p16 and p21 H-scores, P<0.0001) increased cell proliferation (Cyclin D1, P=0.04) and reduced DNA damage (H2AX, P<0.0001). IHC, immunohistochemistry; LuO, orthotopic lung cancer; SC, subcutaneous; TMA, tumor microarray.

Together, these findings underscore critical physiological and molecular differences between SC and LuO models and indicate that the LuO model reproduces aspects of the molecular profile seen in patient tumors.

Implanted orthotopic lung tumors show significantly enhanced immune infiltration and specific microenvironmental landscape compared to SC models

To explore how tumor location within the lung parenchyma influences immune landscape, we performed a comparative analysis of the TME in a variety of implanted LuO and SC lung cancer models. Given the crucial role of the TME in shaping responses to therapies such as immunotherapy, chemotherapy and radiotherapy, we applied multiplexed immunofluorescence to assess key immune populations across both models (Figure S3A and available online: https://cdn.amegroups.cn/static/public/tlcr-2025-871-1.xlsx). Compared to the corresponding SC tumors, LuO tumors consistently exhibited a higher degree of immune infiltration, including increased CD4⁺, CD4⁺PD-1⁺ and DC8⁺ T cells (P=0.01, P=0.03 and P=0.05 respectively) (Figure 3A-3E). Interestingly, this enhanced immune ecosystem occurs with a significantly lower programmed death ligand 1 (PD-L1) expression in the tumor cells growing in the lung P<0.001 (Figure S3B). Notably, LuO models also showed a marked enrichment in immunosuppressive myeloid populations, including tumor-associated macrophages and myeloid-derived suppressor cells (MDSCs) (both P<0.0001), suggesting a myeloid-driven immune suppression axis (Figure 3A-3E). Further stratification of tumors revealed a significantly elevated percentage of M2 in central regions of SC tumors compared to peripheral areas (P<0.05), highlighting intra-tumoral heterogeneity within the SC model (Figure S3C). Besides, an increased percentage of M2, DCs, CD4 and CD8 was shown in parenchymal regions compared to pleural sites (P<0.05), underscoring the differences in both locations (parenchymal vs. pleural) when related to immune populations (Figure S3D). Spatial neighborhood analysis via artificial intelligence (AI)-based algorithms revealed significantly reduced distances between CD4⁺, CD8⁺ T cells, and DCs in LuO tumors in comparison to SC counterparts, correlating with the formation of organized immune clusters and denser co-cluster regions-features absent or poorly developed in SC tumors (Figure 3F-3H). Flow cytometry confirmed a higher percentage of both lymphoid and myeloid populations in the LuO LLC model (Figure 3I and Figure S3E-S3F).

Figure 3.

Orthotopic lung tumors display increased immune cell infiltration and spatial organization compared to subcutaneous models. (A) Representative mIF (VECTRA^®) images of differential cell populations between LuO (parenchyma and pleura) and SC (central and periphery) in mouse (scale bar: 50 µm). (B) Representative Multiplex (VECTRA^®) images of differential cell populations between LuO (parenchyma and pleura) and SC (central and periphery) in human (scale bar: 50 µm). (C) Comparison between LuO and SC of immune cell populations analyzed by mIF panels in mouse. Each data point represents the quantification of one TMA core. (D) Comparison between LuO and SC of immune cell populations analyzed by mIF panels in human. Each data point represents the quantification of one TMA core. (E) Illustration summarizing main differences between LuO and SC models. Compared to SC tumors, LuO models immune profiling revealed increased CD4⁺ T-cell infiltration (P=0.01) and elevated CD4⁺PD1⁺ populations (P=0.04). LuO tumors also exhibited more M2 macrophages (P<0.0001), and an increase in MDSCs (P<0.0001). (F) The LuO model shows lower distances than the SC model between cell phenotypes of interest. Average distances between CD8, CD4 and dendritic cells are shown. The line represents the median of the distribution. Outliers were excluded from the analysis, defined as values outside the interval [Q1 − 3×IQR, Q3 + 3×IQR], being Q1 and Q3 the first and third quartiles respectively, and IQR the interquartile range (Welch’s t-test, P<0.05). (G) The LuO model shows a higher number of clusters per core compared to the SC model for CD4, CD8 T and dendritic cells (Welch’s t-test, P<0.05). Left: the average number of CD4, CD8 and dendritic cell clusters per core is shown as a function of the minimum density threshold used to define a cell cluster, with error bars representing the SEM. Right: the distribution of the number of CD4 T, CD8 T and dendritic cell clusters per core is shown at the density threshold at which the P value between the LuO and SC models is minimum (0.098 cells/µm² for CD4 T cells, 0.091 cells/µm² for CD8 T and dendritic cells). (H) The LuO model shows a significative higher number of CD4, CD8 and dendritic cell clusters per core than the SC model (Welch’s t-test, P<0.046). Left: the average number of CD4, CD8 and dendritic cell clusters per core is shown as a function of the minimum density threshold used to define a cell cluster, with error bars representing the SEM. Right: the distribution of the number of CD4, CD8 and dendritic cell clusters per core is shown at the density threshold at which the P value between the LuO and SC models is minimum (0.0025 cells/µm²). (I) Immunophenotyping analysis of LuO and SC models by flow cytometry. Left: the graph shows the percentage of cells corresponding to lymphoid populations. Right: the graph displays the percentage of cells corresponding to myeloid populations. ns stands for non-significant (P>0.05). IHC, immunohistochemistry; LuO, orthotopic lung cancer; MDSC, myeloid-derived suppressor cell; SC, subcutaneous; SEM, standard error of the mean; TMA, tumor microarray.

Collectively, these results indicate that orthotopic lung tumors not only recruit a more diverse and abundant immune infiltrate but also demonstrate a more physiologically relevant immune architecture compared to SC models.

Lung implanted orthotopic tumors release CTCs with metastatic potential to the bloodstream

To recapitulate advanced stages of lung cancer and develop more clinically relevant preclinical tools, we sought to model the full metastatic cascade starting from primary orthotopic tumors. Using the lung orthotopic models, we isolated CTCs from blood samples collected at experimental endpoints and successfully expanded them in culture (Figure 4A,4B). Moreover, these CTCs demonstrated metastatic competence when injected intracardially, giving rise to distal lesions detectable by in vivo and ex vivo BLI in canonical metastatic sites of lung cancer, including the liver, brain, adrenal glands, bone, and mediastinal lymph nodes (Figure 4C-4F). Notably, CTC-derived metastases emerged more rapidly than those formed by parental LLC lines (endpoints at days 19–21 vs. 27–30), indicating enhanced metastatic potential. From these secondary lesions, we established metastatic cell lines, which preserved organ-specific tropism, such as liver, adrenal gland, bone and brain (LLC-derived) or adrenal and bone (A549-derived) (Figure S4). The syngeneic model more accurately mirrored the metastatic percentage distribution seen in patients than the xenograft model, underscoring the value of an immunocompetent system for capturing the selective pressures of metastasis (Figure 4G), and the translational relevance of this implanted orthotopic approach.

Figure 4.

Orthotopic lung tumors release CTCs capable of forming distal metastases. (A) Schematic representation of the in vivo experimental protocol. At the endpoint of the orthotopic experiment, total blood is collected from the animal and cultured in complete medium to isolate CTCs. These CTCs are then injected intracardially, the establishment and progression of metastasis is monitored weekly by BLI and at the endpoint of the experiment, ex vivo BLI is performed to study the presence of distal metastases in specific organs. These distal metastases are paraffin embedded for histology analysis or digested with DNAse I-Collagenase I buffer for metastatic subclones isolation. (B) Inverted microscope images of LLC parental cells and CTCs from LLC and A549 cell lines. Images were captured using an inverted microscope to visualize the LLC parental cells and CTCs from both LLC and A549 cell lines. Scale bar: 100 µm. (C) In vivo BLI of intracardiac CTCs from LLC and A549 cell lines. 1×10⁵ LLC-CTC or A549-CTC cultured cells were injected into the left cardiac ventricle of 6-week-old C57BL/6J or Rag⁻/⁻ IL2Rγ⁻/⁻ female mice respectively, under isoflurane anesthesia (n=10). In vivo BLI was performed to monitor intracardiac CTCs from LLC and A549 cell lines at different time points. LLC CTCs were imaged at days 7 and 20, while A549 CTCs were imaged at days 14 and 28. Images were acquired in both dorsal and ventral positions. Radiance scale is shown for each experiment. (D) Ex vivo BLI at the experimental endpoint of metastatic organs from LLC intracardiac CTCs model: adrenal glands, bone, liver and brain. (E) Ex vivo BLI at the experimental endpoint of metastatic organs from A549 intracardiac CTCs model: adrenal glands, bone, liver and brain. (F) Histological section of lung tumor lesions in a mediastinal lymph node. H&E staining shows a metastatic lesion. Scale bar: 500 µm. (G) Illustration showing the different percentages of metastatic lesions classified by organ in patients and in our LLC-CTC and A549-CTC-based intracardiac mice models (n=10). BLI, bioluminiscence imaging; CTC, circulating tumor cell; LLC, Lewis lung carcinoma.

Altogether, these findings support that the metastatic capacity of cell lines is preserved when implanted orthotopically in the lung. Furthermore, our results show the utility of CTC-based experiments to functionally trace metastatic progression and capture tumor cell evolution in the context of lung orthotopic implanted tumor models.

Orthotopically derived metastatic subclones capture patient-relevant transcriptomic programs for translational research

To molecularly better understand the biological traits of metastasis derived from orthotopic implanted lung tumors, we generated cell lines derived from the different metastatic lesions induced after CTC injection. Transcriptomic profiling revealed that metastasis-derived subclones, especially those isolated from bone, undergo substantial transcriptional reprogramming compared to their parental lines (LLC and A549). Differential gene expression analysis showed a significantly higher number of differentially expressed genes (DEGs) in bone-derived lines than in CTC-derived ones (FDR <0.05, |log₂FC| >2), a pattern conserved across both murine and human models (Figure 5A,5B), and supported by principal component analysis (PCA) and Venn diagram analyses (Figure S5A,S5B). Importantly, these metastasis-derived lines exhibited transcriptional signatures closely resembling those reported in patient-derived samples, supporting their relevance for translational research. Although only 5–7% of DEGs were shared across species, convergence was evident at the signaling pathway level (Figure 5C). Both murine and human metastases showed consistent upregulation of immune-modulatory and pro-metastatic programs, including chemokine signaling (e.g., CXCL1, CXCL2, CCL20), EMT regulation (KLF5, NFKBIZ), extracellular matrix (ECM) remodeling (MMP9, PTGS2, HAS2), and immune evasion (SLPI, GBP2) (23-27) (Figure 5D). Further analysis revealed conserved activation of stromal-interacting and inflammatory axes such as IL-6/STAT3 signaling and the CXCL/CXCR chemokine network, known to drive pre-metastatic niche formation, myeloid recruitment, and CTC survival. These overlapping transcriptional programs suggest that despite differing routes of dissemination (CTC- vs. bone-derived), the metastatic subclones converge on core hallmarks of metastasis: ECM remodeling, immune suppression, EMT, and stress resistance. However, CTC-derived lines showed additional enrichment in stress-adaptive and motility-associated pathways, including upregulation of cytoskeletal regulators and survival-related genes (e.g., CDC42EP3, GFRA1, SLPI), consistent with their transient and invasive nature. In contrast, bone-derived lines showed selective activation of matrix remodeling and bone-invasion mechanisms (e.g., MMP9 upregulation, TIMP3/RECK downregulation, HAS2 induction) (28-30). Significant DEGs from these analyses are listed in https://cdn.amegroups.cn/static/public/tlcr-2025-871-3.xlsx. Thus, metastasis-derived lines from the orthotopic implanted lung model show activation of molecular pathways commonly observed in metastatic lesions from lung cancer patients. These conserved signatures underscore the potential translational value of these tools.

Figure 5.

Patient-like transcriptomic profiles in metastatic lines from an orthotopic lung cancer models. (A) Volcano plot showing log₂FC versus −log₁₀FDR for genes in each derived line compared with parental LLC. At the endpoint of the intracardiac LLC-derived CTC in vivo inoculation experiment, organs showing BLI signal were processed using a DNase I-collagenase I buffer and plated in culture dishes. Transcriptomic analysis was performed on the LLC-CTCs and bone metastasis-derived subclones. Genes meeting significance (FDR <0.05) and magnitude (|log₂FC| >2) thresholds are highlighted in red (up-regulated) or blue (down-regulated); non-significant genes are shown in gray. The vertical dashed lines mark ±2 log₂FC and the horizontal dashed line marks FDR =0.05. Differentially expressed genes between parental and CTC or metastatic cells are listed in https://cdn.amegroups.cn/static/public/tlcr-2025-871-3.xlsx. (B) Equivalent volcano plots for each derived clones versus parental A549, using the same thresholds and color scheme. (C) Comparative venn diagrams of DEGs called at FDR <0.05 and |log₂FC| >2 between LLC and A549 backgrounds in CTC-derived or bone-derived subclones. Numbers in each circle give genes unique to that background (with % of total DEGs), and overlap indicates genes shared by both. (D) Some differentially expressed genes common to both A549 and LLC subclones (CTCs and bone-derived cells) with concordant regulation (same directionality), grouped by the specific metastatic processes to which they contribute. (E) Shared DEGs between A549 and patients. Volcano plot representing the comparison of bone versus primary tumors in patients. Genes that were also differentially expressed in the A549 bone versus parental contrast are highlighted in red and labeled. A total of 406 shared DEGs were identified. (F) Functional illustrated summary of the DEG genes shared across patients and A549, grouped by key metastatic processes. DEG, differentially expressed gene; FC, fold change; FDR, false-discovery rate; CTC, circulating tumor cell; LLC, Lewis lung carcinoma.

Finally, to assess the translational potential of these findings, we compared data from the LuO models with gene expression profiles from actual patient samples. We analyzed the count matrix from study GSE225208, which includes transcriptomic profiles from primary lung adenocarcinoma tumors and their corresponding bone metastases in patients. PCA (Figure S5C) demonstrated clear clustering of biological replicates, supporting sample consistency. We then applied a differential expression threshold (|log₂FC| >2; FDR <0.05) to compare bone metastases with primary tumors and visualized the DEGs in a volcano plot (Figure S5D), highlighting shared genes that were differentially expressed in bone metastasis versus parental contrasts in A549 (Figure 5E), and in all systems (Figure S5E).

Comparative analyses between these human and murine contrasts revealed that 406 DEGs were shared between patients and A549, and a core set of 20 DEGs was common to all three contexts. These shared genes are involved in key metastatic pathways, including cell adhesion, ECM remodeling, motility, and metabolic reprogramming, further supporting the translational relevance of our orthotopic models to the clinical setting (Figure 5F and Figure S5F). Complete gene lists for the three contrasts are provided in https://cdn.amegroups.cn/static/public/tlcr-2025-871-3.xlsx.

Orthotopically derived metastatic lines reveal niche-specific molecular programs

To further elucidate the molecular basis of metastasis, we performed a comparative proteomic analysis of CTC-derived and bone-derived metastatic subclones from both murine (LLC) and human (A549) LuO models. Significantly differentially expressed proteins (DEPs) from these analyses are listed in https://cdn.amegroups.cn/static/public/tlcr-2025-871-4.xlsx. Proteomic profiling revealed broad alterations in protein expression relative to parental lines, with bone-derived subclones showing a greater number of DEPs (|log₂FC| >0.38 and P value <0.05) than their CTC-derived counterparts across both species (Figure 6A,6B). In A549 cells, 1,851 and 2,746 proteins were differentially expressed (P<0.05) in CTC and bone-derived subclones, respectively, with a balanced proportion of up- and down-regulated proteins. Similarly, LLC-derived CTC and bone subclones showed 1,139 and 1,315 DEPs, respectively. Notably, a higher number of DEPs was observed in human lines compared to murine lines, and in bone-derived subclones compared to CTCs. Venn diagram analysis comparing A549 and LLC backgrounds identified 8.5% and 9.15% overlapping DEPs in both, CTCs and bone-derived lines respectively, suggesting potential shared signatures specific to each metastatic context (Figure 6C). Integration with transcriptomic data identified subsets of targets consistently altered at both mRNA and protein levels (Figure 6D). Subclones recapitulate key molecular programs previously described in clinical studies of lung cancer metastasis such as upregulation of immune-modulatory, epithelial-mesenchymal transition (EMT), and ECM remodeling networks—hallmarks of metastatic progression. This highlights the value of multi-omics approaches for uncovering robust metastatic signatures. Finally, when overlapping the shared transcriptomic and proteomic signatures of CTC and bone-derived subclones (Figure 6E), 47 proteins were found exclusively associated with CTCs and 107 with bone-derived lines, defining distinct proteomic signatures independent of species or analytical layer. CTC-derived lines showed enrichment in pathways supporting cellular stress adaptation, survival during dissemination, and transendothelial migration, consistent with their transient and invasive phenotype. This included upregulation of mediators such as PTGS2 (COX-2), LGALS3BP, and Cx43, linked to EMT, immune crosstalk, and intercellular communication (31,32) (Figure S6A). In contrast, bone-derived subclones exhibited pathway enrichment reflective of adaptation to the bone microenvironment, including enhanced matrix degradation (via MMP14) (33,34), cell adhesion, and cytoskeletal remodeling, features that facilitate stromal integration and bone colonization (Figure S6B). Markers such as LGALS3, NRP2, CAPG, MARCKS, and AJUBA (35-40), previously implicated in tumor invasion and bone tropism, were among the proteins enriched in these subclones.

Figure 6.

Comparative proteomic characterization of circulating tumor cells and bone derived subclones from LLC and A549. (A) Volcano plot showing log₂FC versus −log₁₀P value for proteins in each derived line compared with parental LLC. Proteins meeting significance (P<0.05) and magnitude (|log₂FC| >0.38) thresholds are highlighted in red (up-regulated) or blue (downregulated); non-significant proteins are shown in grey. The vertical dashed lines mark ±0.38 log2FC and the horizontal dashed line marks P=0.05. Differentially expressed proteins between parental and CTC or metastatic cells are listed in https://cdn.amegroups.cn/static/public/tlcr-2025-871-4.xlsx. (B) Equivalent volcano plots for each derived line versus parental A549, using the same thresholds and color scheme. (C) Comparative venn diagrams of differentially expressed proteins called at P<0.05 and |log₂FC| >0.38 between LLC and A549 backgrounds in CTC-derived or bone-derived subclones. Numbers in each circle give proteins unique to that background (with % of total DEPs), and overlap indicates proteins shared by both. (D) Comparative venn diagrams of differentially expressed genes (Figure 5C) and proteins (C) in CTC-derived or bone-derived subclones. Each circle shows the number and percentage of DEGs or DEPs unique to each subclone, while overlaps represent shared genes or proteins between both metastatic origins and omics layers. (E) Venn diagrams summarizing the comparison shown in (D). Each circle indicates the number and percentage of differentially expressed proteins unique to each condition, and the overlapping area shows proteins common to both. DEG, differentially expressed gene; DEP, differentially expressed protein; FC, fold change; CTC, circulating tumor cell; LLC, Lewis lung carcinoma.

Together, these findings show that metastasis-derived lines from orthotopic lung models not only preserve patient-relevant molecular programs but also display niche-specific proteogenomic adaptations. This highlights their translational relevance for studying metastatic evolution and for the preclinical testing of anti-metastatic therapies.

Lung cancer metastases exhibit high divergence in cancer hallmarks and immune landscape

To better define the biological properties of distal metastases in the LuO model, we analyzed both tumor-intrinsic characteristics and immune cell infiltration in metastatic tissues from mice bearing secondary lesions in orthotopic syngeneic and xenograft settings (Figure 7A,7B). A panel of hallmark-associated markers was evaluated (Figure S7A,S7B), revealing substantial inter-lesion and inter-model heterogeneity (Figure S7C,S7D). This molecular divergence underscores the complexity of metastatic progression and the influence of microenvironmental context. Notably, however, γH2AX was consistently downregulated in metastatic lesions compared to primary tumors, regardless of the organ site or model (Figure 7C-7E). This points to a potentially conserved activation of the DNA damage signaling during late-stage dissemination, amidst otherwise diverse transcriptional programs. Furthermore, lung cancer metastases seem to be governed by highly heterogeneous biological programs, reflecting both divergent tumor evolution and organ-specific selective pressures.

Figure 7.

Tumor ecosystem study in primary and metastases in LuO model. (A) Histological section (H&E) of primary and metastatic lesions from LLC intracardiac CTCs model including adrenal glands, bone and liver. Scale bar: 100 µm. (B) Histological section (H&E) of primary and metastatic lesions from A549 intracardiac CTCs model including adrenal glands, bone and liver (scale bar: 100 µm). (C) Representative IHC images of H2AX in primary and metastatic lesions (adrenal gland, bone, liver) in the LuO model in mouse (LLC cell line) (scale bar: 100 µm). (D) Representative IHC images of H2AX in primary and metastatic lesions (adrenal gland, bone, liver) in the LuO model in human (A549 cell line) (scale bar: 100 µm). (E) Comparison of H2AX markers between primary and metastasis in the LuO model in mouse and human cell lines. Multiple ROIs were analyzed per slide; each data point corresponds to one ROI. (F) Representative mIF (VECTRA^®) images of differential macrophage and neutrophil cell populations between primary and metastatic lesions from a LuO model in mouse (scale bar: 50 µm). (G) Representative mIF (VECTRA^®) images of differential macrophage and neutrophil cell populations between primary and metastatic lesions from a LuO model in human (scale bar: 50 µm). (H) Comparison between primary and metastatic lesions of immune cell populations analyzed by mIF panels in mouse. ROIs were analyzed per slide; each data point corresponds to one ROI. ns stands for non-significant (P>0.05). (I) Comparison between primary and metastatic lesions of immune cell populations analyzed by mIF panels in human. ROIs were analyzed per slide; each data point corresponds to one ROI. ns stands for non-significant (P>0.05). CTC, circulating tumor cell; H&E, hematoxylin and eosin; IHC, immunohistochemistry; mIF, multiplex immunofluorescence; LLC, Lewis lung carcinoma; LuO, orthotopic lung cancer; ROI, region of interest.

mIF analysis of primary and metastatic lesions further revealed immune heterogeneity shaped by both tumor type and metastatic niche (Figure S7E-S7H). Infiltrating neutrophil abundance was consistently reduced relative to the primary tumor, whereas macrophages were significantly increased in metastatic lesions across all organs and both models (Figure 7F-7I). These findings align with previous reports in patient-derived samples (41) and support the notion that immune suppression is differentially orchestrated in primary versus metastatic tumors, with additional modulation imposed by the metastatic organ microenvironment.

Discussion

Developing in vivo models that more accurately reflect the complexity of lung cancer remains a major challenge in preclinical research. The current models are still not optimal in vivo models, fully representative of human disease, on which high throughput drug tests may be performed with full reproducibility of results in human trials.

The implementation of LuO models has been pursued for a long time. In this regard, GEMMs enable the study of tumorigenesis driven by defined oncogene or tumor suppressor mutations and remain a gold standard for tracking carcinogenic progression for lung cancer (42-45). GEMMs are autochthonous systems in which tumors arise within the native lungs of immunocompetent mice. This setup enables researchers to investigate lung cancer development from its earliest stages. Furthermore, many in vivo models developed in GEMMs naturally metastasize to distant sites such as the liver, adrenal glands, or bones, thereby closely mimicking the aggressive, metastatic behavior of human lung cancer. Despite their strengths, GEMMs also present significant limitations. One major drawback, particularly in the first-generation models, is the need for complex breeding schemes. These crosses are often time-consuming and expensive, which has limited the widespread use of these models in the research community. Additionally, incorporating multiple genetic alterations typically requires several rounds of breeding, further increasing the complexity and cost. This often is accompanied by extended tumor latency periods with additional delays in experimental timelines. An additional limitation is that lung driven GEMM develop multiple simultaneous tumors in the different lobes of the lung, which is very different to what happens in human carcinogenesis. Moreover, the genetic-driven tumor induction fails to recapitulate tumor heterogeneity or the complex mutational burden which are found in human lung cancers (46-48).

Historically, among the most commonly employed approaches as alternatives to GEMMs have been intranasal or intratracheal tumor cell administration to deliver cancer cells within the lung airspace, with the aim of the engraftment of a proportion of them in the lung airways or parenchyma. These models are less invasive than the thoracotomy-based orthotopic model proposed in the present work and retain epithelial integrity, but, as happens with the GEMM models, results in multifocal tumors with high tumor burden and inconsistent localization (49-51). As an alternative, intravenous and intracardiac injection models deliver tumor cells to the lungs via the circulation-through venous return or systemic circulation, respectively. Although these methods are technically simpler, they primarily model metastatic dissemination rather than the initiation of primary tumors. Moreover, many of the tumors are initiated by capillary or other blood vessel entrapment and not within the epithelial tissue. Thus, they are useful for studying tumor bloodstream-based colonization and spread, but they bypass the early stages of tumorigenesis and do not accurately recapitulate the biology of primary lung cancers (16,52).

Among existing strategies, the approach described in this study (thoracotomy-based orthotopic models) offers the advantage of generating a single tumor within the native lung environment. This preserves tissue-specific microenvironmental cues and enables physiological tumor-immune and stromal interactions. In orthotopic lung-cancer models, investigators may choose between a surgical approach typically involving thoracotomy and direct implantation of tumor cells or tissue fragments and a non-surgical, percutaneous method in which cells are injected transthoracically through the intercostal space. Surgical implantation offers precise placement under direct vision but requires cutaneous wounds, prolonged anesthesia and postoperative care, and may perturb the native microenvironment (53). In contrast, the non-surgical technique can be performed in under two minutes per animal, avoids chest-wall incisions, and yields comparable tumor take rates and contralateral metastasis efficiencies from a single injection (54). Moreover, as in patients, tumor progression can be monitored longitudinally using imaging techniques. However, each read-out modality carries distinct strengths and limitations. CT enables non-invasive longitudinal monitoring of tumor growth and mediastinal spread (55) but lacks the cellular resolution of other techniques such as cytometry. In vivo BLI permits real-time tracking of living cells, albeit semiquantitatively and with depth-related signal attenuation. In this regard, our results show potential discrepancies between the CT and BLI measurements at early time points. Specifically, our data show remarkable consistency in tumor volume at day 7 across the three different injected cell quantities, whereas BLI at day 6 indicates substantial differences in luminescence among the groups. This can be explained by the nature of the measurements and tumor development at early stages. CT and BLI are widely used and complementary techniques for assessing tumor burden. However, due to specific limitations of the LuO model used in this study, combining both imaging modalities is necessary to improve accuracy.

Regarding CT imaging at early time points (e.g., day 7), it is important to consider that the initial tumor cell scaffold may act as an imaging artifact. Tumor cells are injected into the lung parenchyma suspended in liquid Matrigel, which solidifies at temperatures above 4 ℃ to help localize tumor growth within the lung. However, this Matrigel plug may appear as a false-positive signal on CT scans, complicating early tumor burden assessment.

BLI also presents specific challenges related to signal variability. First, luciferin substrate distribution can be heterogeneous, meaning that not all tumor cells receive equal amounts of substrate. Second, luciferase expression levels may vary among the injected tumor cells, further contributing to signal inconsistency.

By the second week post-injection, tumor volume surpasses the residual Matrigel, resulting in more realistic inter-mouse variability in CT measurements. Additionally, immune system activity at this stage preferentially targets cells expressing higher levels of the antigen luciferase, preventing their tumor development and stabilizing BLI signals. Therefore, both CT and BLI measurements become more robust from the second week onwards, at which point they exhibit comparable behavior and show good correlation.

The procedure is safe for the mouse, which recovers rapidly from surgery without associated complications such as pneumothorax. This fact is supported by robust statistics from a large number of tested cell lines, with no surgical complications affecting survival. Thus, this model provides the additional advantage of recapitulating a broad spectrum of mutational landscapes at will.

Therefore, the choice of surgical versus non-surgical engraftment and of, CT or BLI read-outs should be guided by the balance between precision, throughput, temporal resolution, and animal welfare.

An additional similarity between the orthotopic model evaluated and human lung cancer is in the realms of metastasis. The implanted intrathoracic model shows CTCs in the bloodstream that can be isolated, as happens in human lung cancer patients, in particular in the later stages, associated to bigger tumor sizes and invasion (56). Moreover, LuO models develop spontaneous metastases to clinically relevant sites such as mediastinal lymph nodes, bone, adrenal gland, brain or liver. The finding of lymph node metastasis confirms the value of this model to study lymphatic spread and regional metastasis while also enabling investigation of primary tumor progression through to the metastatic stage within a single system.

The ability to derive site-specific metastatic cell lines from this model adds further value, enabling detailed investigation of the biological mechanisms underlying organ-targeted metastasis and the development and testing of strategies to prevent or treat it.

We have observed that orthotopic implanted syngeneic models, better replicate the metastatic patterns (in terms of frequency) observed in patients than the xenografts, being well-suited for functional studies and immunotherapy research. Importantly, spontaneous metastasis in these models allows isolation of metastatic subclones without artificial enrichment, preserving natural heterogeneity. This is especially- valuable for molecular mechanistic studies based on multi-omics. While unselected cells offer a more accurate picture of the metastatic cascade, enrichment strategies remain useful to generate high-metastatic cell line variants for functional assays. Furthermore, the possibility to interrogate niche-specific adaptations (e.g., spine vs. limb bone invasion) strengthens the model’s utility, as demonstrated in the intracardiac CTC assays performed in the present study.

Other examples of the utility of the present orthotopic implanted models are the analysis of systemic effects such as cancer-associated cachexia or early microenvironmental remodeling (57-59). Additionally, these models are valuable platforms for spatial transcriptomic or proteomic studies aimed at identifying tumor-host interactions or biomarkers predictive of disease progression or therapy resistance.

The orthotopic model enables the study of both tumor characteristics and the immune landscape of lung cancer within its native location, in xenografts, and when using syngeneic cell lines in immunocompetent models. In this context, the design of the TMA minimizes technical variability allowing parallel assessment under identical processing and staining conditions, and supports high-throughput, multiplexed analysis of immune markers, oncogenic pathways, and therapeutic responses. The rationale for including such a large and diverse panel is that lung cancer is a highly heterogeneous disease, both between and within histological subtypes. By incorporating a broad panel of cell lines representing ADC, SCC, small-cell, and other lung cancer types, we ensure that our findings are not attributable to a single lineage or cell line, but instead reflect a consistent and reproducible effect across this heterogeneity. This design strengthens the conclusions, as it demonstrates that the observed phenomena are attributable to the model system itself rather than to idiosyncratic features of an individual cell line. By capturing spatial and anatomical heterogeneity, this resource enhances the translational relevance of preclinical studies and serves as a foundation for both mechanistic and therapeutic investigations. Using this platform, we observed that tumor location significantly influences both cellular and microenvironmental features.

Tumor cells in the LuO model exhibit features consistent with a more physiologically relevant and less stressed microenvironment compared to their corresponding SC tumors, supporting the translational relevance of LuO for studies of lung tumor biology and therapy. First, LuO tumors display significantly increased intratumoral microvascular density (CD31), a hallmark of tumor angiogenesis commonly documented in NSCLC and associated with tumor progression and the clinical rationale for anti-angiogenic therapies (60,61). Second, LuO tumors show reduced levels of hypoxia and stress-associated markers, including lower HIF-1α protein expression by IHC, reduced γH2AX staining, and decreased expression of the checkpoint proteins p16 (CDKN2A) and p21 (CDKN1A). This pattern is observed in relatively oxygenated, highly proliferative tumor regions that drive expansion in vivo and mirrors the lung’s unique microenvironment, which is naturally exposed to higher oxygen levels compared to more hypoxic tissues (62). Clinically, these markers are highly relevant: γH2AX burden has been used to stratify DNA damage responses and predict radiosensitivity and prognosis in lung cancer patients, making it a meaningful readout for therapy-related DNA damage; meanwhile, low p16 is a frequent event in lung tumorigenesis associated with more aggressive disease in clinical cohorts (63,64). Moreover, the reduced stress burden in LuO tumors avoids artificially inflated treatment responses, enabling more accurate evaluation of genotoxic therapies and radiotherapy (65).

In contrast, SC tumors are characterized by pronounced hypoxia and elevated HIF-1α, which not only promote tumor adaptation but also enhance immune evasion through the induction of PD-L1 expression (9,10). This co-expression of HIF-1α and PD-L1, commonly observed in hypoxic lung tumor regions, correlates with greater tumor necrosis, advanced disease progression, poor prognosis (11), and reduced CD8⁺ T-cell infiltration (12). As a result, SC models may overestimate PD-L1-mediated immunosuppression and the therapeutic efficacy of immune checkpoint inhibitors. Importantly, the lower PD-L1 levels observed in LuO tumors are much more consistent with clinical observations (13).

Finally, the observed upregulation of Cyclin D1 (CCND1) in LuO tumors supports a more proliferative phenotype with reduced senescence signaling, consistent with clinical reports of Cyclin D1 overexpression in NSCLC (66,67). This mirrors early tumor development in native lung tissue (8) and underscores the relevance of LuO for studying cell-cycle deregulation and for the preclinical evaluation of CDK4/6-targeted strategies.

In summary, the LuO model offers a biologically faithful system that recapitulates the complex tumor–host interactions unique to the lung, including oxygen availability, vascular architecture, immune dynamics, and tissue-specific stress responses. By avoiding the artificial hypoxia and stress constraints of ectopic implantation, LuO provides a more physiologically relevant context for tumor evolution and therapeutic testing, enabling accurate assessment of drug efficacy, biomarker dynamics, and resistance mechanisms. This reinforces its adoption as a preferred platform for lung cancer research and translational preclinical studies (65).

mIF analysis revealed distinct immune microenvironments between tumor models. In contrast, SC tumors displayed sparse and disorganized immune cell infiltration, highlighting significant limitations in recapitulating the immunological complexity of the TME. In general, LuO tumors exhibited a more enriched and spatially organized immune infiltrate, characterized by clustered arrangements of CD4⁺ and CD8⁺ T cells alongside DCs. Such triadic cell clusters are critical for effective anti-tumor immunity in lung cancer, as they facilitate coordinated interactions wherein CD8⁺ cytotoxic T lymphocytes receive essential helper signals from CD4⁺ T cells via shared engagement with DCs. This interaction enhances T cell activation, proliferation, and sustained effector function, ultimately promoting more effective tumor clearance (14,15). The presence of these organized immune structures in LuO tumors suggests active immune surveillance and more accurately reflects the immune architecture observed in patient tumors.

An interesting observation in the LuO model was the increased abundance of B cells, as revealed by flow cytometry, compared to the SC model. Although our mIF panel did not include markers to assess their spatial distribution using VECTRA, the elevated presence of B cells may suggest the formation of tertiary lymphoid structures (TLSs). These structures have been described in patient tumors and, to date, primarily in GEMMs (16,17). The potential presence of TLSs in the LuO model warrants further investigation, as their formation could enhance the model’s relevance for studying adaptive immune responses and the tumor-immune microenvironment in lung cancer.

The significant differences in the composition and distribution of immune cells, both in immunocompromised and immunocompetent settings, suggest that, while SC models remain valuable for some applications their immunobiological and metabolic shortcomings limit their translational relevance when assessing therapeutic efficacy, in particular in immune related strategies, but also in other therapies (targeted, radiotherapy, etc.) which are highly dependent on the active role of the immune component of the tumor. Over-reliance on SC models may partially explain the poor clinical success rate of oncology therapeutics, reinforcing the need for more physiologically representative models like LuO in immuno-oncology and other therapy assessment research.

The transcriptomic and proteomic analyses shown provide concrete evidence that the LuO orthotopic model enables in-depth exploration of the molecular mechanisms driving metastasis. It is important to acknowledge the limitation that only one syngeneic and one xenograft model were developed; therefore, the results cannot yet be generalized across all contexts. Despite this, our profiling reveals signatures enriched in pathways known to be involved in metastatic progression. These findings mirror key aspects of what is known from human lung cancer (68), reinforcing the biological relevance of the model. Furthermore, the observed molecular diversity among metastatic lesions reflects the inherent heterogeneity of metastases, driven by clonal evolution and shaped by both tumor-intrinsic hallmarks and interactions with the surrounding TME.

Specifically, the distinct transcriptomic landscapes between CTC-derived and bone-derived sublines align with the clonal evolution model of metastatic formation. The greater transcriptional divergence in bone-derived lines supports the concept that metastatic competence arises through selection and expansion of subclones with enhanced organ-specific fitness. This is consistent with longitudinal studies such as TRACERx, which have demonstrated that NSCLC metastases often originate from late-emerging subclones evolving under selective pressures distinct from those acting on the primary tumor (69).

Collectively, these findings underscore the value of the LuO model in modeling patient-relevant metastatic programs and in investigating the evolutionary dynamics underlying metastatic spread. The integration of multi-omic approaches further highlights key drivers of metastasis and potential therapeutic targets for both CTC-mediated dissemination and bone colonization. Thus, our study provides a strong proof-of-concept for using this orthotopic model to dissect the complex biology of lung cancer metastasis.

In conclusion, this model represents a valuable tool for the in-depth study of human carcinogenesis from multiple perspectives, offering the possibility to investigate tumor evolution over time. However, it also presents certain limitations, including its technical demands and the inter-animal variability in tumor size and location (52,70), which necessitate larger sample sizes and the implementation of standardized protocols to mitigate variability and enhance reproducibility. Moreover, a degree of expertise and an initial period of training are required; nonetheless, once mastered, the model allows for experiments with a sufficient number of animals to meet most research objectives. It is particularly promising for the evaluation of novel therapeutic strategies, and while the transcriptomic and proteomic analyses presented here serve only as a proof of concept, they provide preliminary insights that will need to be expanded through the study of additional cell lines. Overall, this model offers significant advantages over existing systems, especially for longitudinal studies of tumor progression and the assessment of the efficacy of new therapies and drugs, making it highly relevant for preclinical trials conducted by pharmaceutical companies.

Conclusions

We present a refined LuO model that addresses major limitations of traditional SC that better preserves aspects of the anatomical, molecular, and immunological context of lung tumors. The model supports reliable tumor engraftment, enables longitudinal monitoring, and captures hallmark features such as enhanced vascularization, active immune infiltration, and spontaneous metastasis. Notably, it permits the isolation and characterization of CTCs and site-specific metastatic lines, facilitating in-depth study of metastatic evolution and organotropism. Through integrated multi-omics analysis, we demonstrated that these metastasis-derived sublines exhibit both conserved and niche-specific molecular programs, paralleling clinical observations. Thus, this model provides a robust translational platform for preclinical testing, immunotherapy assessment, and metastasis research, with the potential to significantly enhance the predictive accuracy of therapeutic outcomes in lung cancer.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. Experiments were performed under a project license (062-22 and 067-24) granted by the Ethics Committee for Animal Experimentation of the Universidad de Navarra in compliance with institutional guidelines for the care and use of animals. A protocol was prepared before the study without registration.

Data Sharing Statement

Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-871/dss

tlcr-14-11-4868-dss.pdf ^{(124.4KB, pdf)}

DOI: 10.21037/tlcr-2025-871