Aberrant Activation of Wound-Healing Programs within the Metastatic Niche Facilitates Lung Colonization by Osteosarcoma Cells

Metastasis is responsible for nearly all solid tumor-related deaths (1). Metastasis proceeds through a multistep process wherein tumor cells invade the surrounding normal tissue adjacent to the primary tumor, enter the bloodstream (for tumors that spread via hematogenous route), colonize, and ultimately grow within a distant tissue (2). Preclinical and patient-level data suggest that therapies designed to disrupt the early steps of the metastatic cascade, such as invasion through adjacent tissues and intravasation into the bloodstream, are less likely to prevent clinically relevant metastatic lesions, as disseminated malignant cells have likely colonized distant tissues prior to the detection of a primary tumor (3–5). Development of therapies that target the later elements of the metastatic cascade is more likely to have a clinical impact but has remained largely theoretical because of our crude understanding of the mechanisms that drive the progression of tumor growth in distant tissues following colonization. Understanding how disseminated tumor cells transform the foreign and otherwise hostile microenvironments of metastatic sites into accommodating environments that promote tumor growth may expose novel metastatic vulnerabilities (6).

The lung is a common site of metastasis for tumors that originate in diverse developmental origins (7). Physiologically, the lung not only facilitates gas exchange but also creates a barrier to the invasion of microbes, toxins, pollutants, and allergens. Accordingly, the lung has evolved complex mechanisms for recognizing and repairing injury without causing excessive tissue damage or inflammation (8). Certain pathologic conditions such as idiopathic pulmonary fibrosis (IPF) arise when the mechanisms of lung repair become altered. In these altered states, dysregulation of these wound-healing mechanisms leads to progressive dysfunction (9). At the cellular level, IPF is characterized by epithelial hyperplasia, accumulation of fibroblasts, and the excessive deposition of extracellular matrix (ECM) proteins such as fibronectin and collagen, all confined within an immunosuppressive microenvironment (10). How these structural elements of the lung respond to metastatic cancer cells remains poorly understood. The profoundly inefficient nature of the metastatic process and the dramatic differences evident in the comparison of normal lung to metastatic lesions suggest that the soil of the lung must be shaped by cancer cells in order for tumors to successfully colonize and grow in the organ (11). Therefore, disrupting the tumor education of the lung microenvironment is an attractive therapeutic approach to targeting metastatic disease.

Osteosarcoma is the most common malignant bone tumor in young people and ostensibly metastasizes to the lung, leading to poor patient outcomes (12–14). Given the organotropism that osteosarcoma displays for the lung, and the critical clinical problem lung metastasis poses to patients with osteosarcoma, we leveraged several murine and human xenograft models to elucidate how the lung responds to metastatic osteosarcoma cells. We uncovered that osteosarcoma cells arrest within the distal alveolar epithelial niche, damage alveolar epithelial cells, and trigger an abnormal alveolar wound-healing program that promotes metastatic progression. Utilizing single-cell RNA sequencing (scRNA-seq) and spatial transcriptomic analysis, we found that the stroma of established metastases resembles a fibrotic, nonhealing wound. Translationally, disrupting the lung fibrosis response with nintedanib, a tyrosine kinase inhibitor (TKI) utilized clinically for treating patients with IPF that has also previously been shown to reduce osteosarcoma metastases, is used as a mechanistic tool to reprogram the osteosarcoma metastatic microenvironment through the inhibition of FGFR transcriptional responses. Collectively, our study sheds light on how tumors co-opt the lung microenvironment during metastasis and provides proof of concept that targeting the metastatic niche is a viable therapeutic strategy.

All cell lines utilized herein were routinely authenticated with short tandem repeat (STR) and Mycoplasma testing through the Laboratory Corporation of America. Cells utilized for experiments were within 20 passage doublings after reviving from liquid nitrogen storage. OS-17 cells were derived from an early passage of the OS-17 patient-derived xenograft created from a primary femur biopsy performed at St. Jude’s Children’s Research Hospital in Memphis and were a gift from Peter Houghton. K7M2 cells are a highly metastatic derivative cell line derived from a spontaneous osteosarcoma in a BALB/c mouse and were obtained from the ATCC (CRL2836). F420 cells are a murine osteosarcoma derived from spontaneous lung metastases arising in the genetically engineered osteosarcoma mouse model Col2a-Cre mutant Tp53 and were a gift from Jason Yustine at Emory University, Atlanta, Georgia. HBEC3-KT (human bronchial epithelial cells [HBEC]) is an immortalized nontumorigenic human lung epithelial cell line and was obtained from the ATCC (CRL-4051). F420 and K7M2 cells were propagated in culture in DMEM (Corning, 10-013-CV) supplemented with 10% FBS (R&D Systems, S11150H), whereas OS-17 cells were cultured in RPMI (Corning, 10-040-CV) supplemented with 10% FBS. HBEC cells were propagated in human airway epithelial media (ATCC, PCS-300-030) with the Bronchial Epithelial Cell Growth Kit (ATCC, PCS-300-040). OS-17 and HBEC cells were plated 1:1 on to glass coverslips in DMEM/F12 (Corning, 10-092-CV) supplemented with 1% FBS. HBEC-conditioned media were obtained after 72 hours of incubation (DMEM/F12 with 1% FBS) and were cleared with centrifugation at 1,000 rpm prior to treating osteosarcoma cells.

All animal studies carried out in this study were done with the approval of the Institutional Animal Care and Use Committee of the Research Institute at Nationwide Children’s Hospital (IACUC protocol AR15-00022) prior to conducting experiments. Six- to eight-week-old female C57BL/6 (for F420; RRID: MGI:2159769), BALB/c (for K7M2; RRID: IMSR_APB:4790), and CB17-SCID [for human OS-17 xenograft- CB17SC scid−/− (RRID: IMSR_TAC:cb17sc)] mice were used for all studies. For generation of primary tumors, single-cell suspensions of 5 × 10⁵ cells were injected intratibially. Primary tumors were excised once they grew to 800 mm³. For generation of experimental metastasis, F420 (5 × 10⁵), K7M2 (5 × 10⁵), or OS-17 (1 × 10⁶) cells were inoculated intravenously via tail vein. Cell number utilized for experimental metastasis is titrated by a model to ensure survival of mice after injection and to ensure metastasis rates of at least 80% (15). Mice were followed until designated endpoint of experiment or until arriving at humane endpoint. For nintedanib (MedChemExpress, HY-50904) treatment, experimental metastasis was induced as described above. To determine the effect of nintedanib on metastatic progression, 1 day after injection, mice were randomized to receive vehicle (0.5% hydroxyethylcellulose, Eisen-Golden Laboratories, EG-HEC-3) or nintedanib 50 mg/kg twice per day. Mice were monitored, and once an ill mouse (which was a vehicle treated mouse in all three models) was confirmed to have lung metastasis, all vehicle and nintedanib treatment mice were sacrificed. All mice were euthanized by IACUC-approved CO₂ inhalation method prior to processing as described below. To define the effects of nintedanib on established metastases using single-nucleus transcriptomics, mice were injected with F420 cells expressing luciferase. The development of metastatic disease was monitored weekly by intraperitoneal injection of luciferin followed by bioluminescence imaging. On day 28 postinjection, asymptomatic mice with identifiable metastasis were randomized to receive vehicle or nintedanib 50 mg/kg twice daily for 14 days and processed for single-nucleus RNA-seq or histology as described below.

Lungs harvested from mice were processed using the human tumor dissociation kit (Miltenyi Biotec, 130-095-929) with a gentleMACS Octo Dissociator with Heaters (Miltenyi Biotec, 130-096-427) for 3′ single-cell analysis. For single-nucleus analysis, cryopreserved metastasis-bearing lungs were isolated and permeabilized for single-cell gene expression sequencing following the Nuclei Isolation from Complex Tissues for Single Cell Multiome ATAC + Gene Expression Sequencing protocol (10x Genomics, CG000375 Rev C). Following dissociation into single cells or nuclei, 3′ RNA-seq libraries were prepared using Chromium Single Cell 3′ RNA-seq system (10x Genomics) with the Reagent Kit v3.1 (10x Genomics, PN-1000121) according to the manufacturer’s instructions. cDNA library quality and quantity were determined using an Agilent 2100 bioanalyzer using the High Sensitivity DNA Reagents kit (Agilent, 5607-4626) and then sequenced on an Illumina NovaSeq 6000. Raw data and initial processing were performed using Cell Ranger version 7 (10x Genomics) with reads mapped to mm10 and hg19 versions of the mouse and human genomes.

A histologic section was generated from a mouse with lung metastases from injected F420 tumor cells as outlined above. This section was stained with hematoxylin and eosin (H&E) and imaged using NIS-Elements Antigen retrieval (AR) software (v. 5.42) using a Nikon Eclipse Ti2-E microscope with a Nikon DS-Ri2 color camera and a 10× Plan Apochromat Lambda D objective. The section was then mounted into a CytAssist instrument (10x Genomics), and spatial libraries were prepared according to the manufacturer’s recommendations. The resulting library was sequenced on an Illumina NextSeq 2000. The SpaceRanger software (10x Genomics, version 2.1.1) was used to align the data to the mm10 mouse genome, assigning reads to both genes and spots on the Visium slide and aligning the data to H&E images.

10× single-cell and single nuclei sequencing data were loaded into Seurat toolkit version V5 in R for further analysis (16). Cells were filtered by minimum and maximum unique molecular identifier (UMI) counts appropriate to each sample (available in the GitHub repository [https://github.com/kidcancerlab/epithelial/]), and any cell with mitochondrial DNA >20% was removed. RNA counts were normalized with regularized negative binomial regression via SCTransform. After dimensionality reduction with principal component analysis and Uniform Manifold Approximation and Projection (UMAP) embedding, unsupervised clustering was achieved using “FindNeighbors” and “FindClusters” function in Seurat using PCA dimensions 1:30. Clustering resolution was determined using the clustree R package (RRID: SCR_016293). Differential gene expression between clusters was determined using FindMarkers (Seurat) and the Wilcox method. For single-cell data, cell types were approximated with SingleR using the Human Lung Cell Atlas and the Human Primary Cell Atlas as references (17, 18). For single-nucleus data, we used publicly available single-nucleus reference data as a reference (19). Clusters for both types of data were then curated manually based on the automated assignments and review of canonical marker expression.

Human normal lung single-cell data were obtained from the Human Lung Cell Atlas, with samples filtered using the accompanying metadata to isolate cells from healthy lungs, nonsmokers, and lung parenchyma, and then downsampled in a weighted fashion to include similar proportions of whole-cell and nuclear libraries as the lung metastasis samples (17).

Differentially expressed genes were evaluated by gene set enrichment analysis (GSEA) against Kyoto Encyclopedia of Genes and Genomes (KEGG), GO:Molecular Function, GO:Biological Processes, and transcription factor activity (all from MSigDB) using fgsea (RRID: SCR_020938). Results were plotted in R using the ggplot suite.

The nichenetr R package was used to identify cell–cell interactions within the tumor microenvironment in our single-cell RNA-seq data that might be disrupted by nintedanib treatment (20). To identify all potential interactions, each cell type was, in turn, considered a “receiver” cell type, whereas the other cell types were “sender” cells producing ligands. For each receiver cell type, gene expression fold change values were calculated for genes upregulated in tumor-bearing lung compared with healthy lung. Potential ligands and receptors for each cell–cell interaction were selected by removing those not expressed in sender and receiver cells, respectively. The output from nichenetr was filtered for receptors potentially inhibited by nintedanib to generate a list of ligand–receptor interactions to assess in samples treated by nintedanib (21, 22). The nintedanib-treated samples were single nuclei and so were suboptimal to perform nichenetr analyses in directly because of a lack of steady-state cytoplasmic RNAs. However, we were able use these samples to assess the downstream targets of the ligand–receptor interactions identified in single-cell data as outlined above. Briefly, the databases provided with the nichenetr package were used to identify all genes predicted to be downstream of the ligand–receptor interactions identified in the single-cell data as potentially inhibited by nintedanib. These gene lists were then used as input for the AUCell R package to generate activation scores for each nucleus in the dataset (23). The distribution of these activation scores was plotted as violin plots for the untreated F420-bearing lung as well as the F420-bearing lungs treated with nintedanib at either 3- or 28-day post tumor tail vein injection. To assess the differences between groups, we calculated FDR-adjusted P values using t tests and Cohen’s d statistics as an estimate of effect size.

Visium spatial RNA-seq data were loaded into Seurat V5 in R for further analysis. Data outside of the tissue were filtered, and the remaining data were normalized using SCTransform, and then principal component analysis, cluster assignment, and further dimensional reduction by Uniform Manifold Approximation and Projection were performed. The single-cell data generated by this study were used as a reference to deconvolute each spot in the spatial dataset using the spacexr R package (24).

To assess the activity of pathways potentially inhibited by nintedanib (as identified in single-cell data as outlined above), we used the AUCell R package to generate activation scores for each spot in the Visium data. These activation scores were then plotted to show spatial patterns of potential pathway activation in tumor-bearing lung.

Sternotomy was performed, and the lungs were flushed via slow instillation of PBS(Corning, 21-031-CV) through the right ventricle until the lungs were cleared of blood. The lungs were dissected from mouse en bloc along with the trachea and heart. The lungs were slowly instilled with 10% neutral-buffered formalin (NBF; Sigma-Aldrich, HT5012-1CS) until they fully expanded, and the trachea was then tied off with suture and placed in NBF for 24 hours at 4°C. After repeated washing in PBS, lung lobes were carefully dissected from each other and processed for formalin-fixed, paraffin embedding (FFPE) or immunohistochemistry frozen (IHC-F). For FFPE, 4-μm sections were deparaffinized with xylene and rehydrated through ethanol series. For cryosections, lung lobes were cryoprotected through a sucrose gradient (10%, 15%, 30%), mounted in Optimum Cutting Temperature compound (Thermo Fisher Scientific, 23-730-571), frozen in liquid nitrogen, and then sectioned to 10 μm. To quantify metastasis burden, FFPE sections were counterstained with hematoxylin (Sigma-Aldrich, MHS16-500ML) and eosin (Abcam, AB246824) via conventional methods. Heat-mediated AR methods utilized citrate buffer (CB; pH 6.9) or Tris-EDTA buffer (TE; pH 9) with Tween 20 (Fisher Bioreagents, BP337-100) at 0.2% v/v. AR buffer used was antibody dependent (see below).

For thick tissue analysis, Ce3D-mediated tissue clearing was performed as described previously (25, 26).

FFPE sections were rehydrated in PBS for 30 minutes. Human tissue sections utilized in this study were obtained from the Nationwide Children’s Pathology department under institutional review board–approved study (IRB1100478). Cells growing on #1.5 glass coverslips (Electron Microscopy Sciences, 72204-04) were fixed in NBF for 10 minutes at room temperature. Following three rinses in PBS, coverslip or histologic sections were then incubated for at least 1 hour in blocking solution consisting of PBS, 0.2% Triton 100 (v/v; Sigma, X-100-100mL), and 3% bovine serum albumin (w/v; Sigma, A-7888) at room temperature. For Ce3D specimens, blocking and permeabilization were performed overnight at room temperature. Primary antibodies used were rabbit anti-fibronectin (Abcam, ab2413, AR = TE), rabbit anti-mouse Sftpc (Abcam, ab211326, AR = CB or TE), rabbit anti-mouse Ly6G (Cell Signaling Technology, 87048, AR = CB), hamster anti-podoplanin [Developmental Studies Hybridoma Bank, 8.1.1, AR = TE], rat anti-KRT8 (Developmental Studies Hybridoma Bank, TROMA-I, AR = TE), rat anti-Ki67 (Invitrogen, 14-5698-82, AR = CB), rabbit anti–human Sftpc (Millipore, ABC99, AR = TE), mouse anti–human fibronectin (R&D, MAB1918, AR = TE), and rabbit anti–mouse Cd68 (R&D, MAB101141, AR = CB or TE). Primary antibodies were diluted in blocking solution, and sections were incubated in primary antibody solution overnight at 4°C. For Ce3D, primary antibody incubation was performed for 72 hours 4°C. Following three rinses in PBS (for Ce3D, 3 washes × 2 hours/wash), sections were incubated with appropriate Alexa Fluor–labeled secondary antibodies (Invitrogen), DAPI (Invitrogen, D1306), and, where indicated, Phalloidin Alexa Fluor 647 (Invitrogen, A12379) or wheat germ agglutinin Alexa Fluor 647 (Invitrogen, W11261) diluted in blocking solution for 1 hour at room temperature. For Ce3D, tissue was incubated for 24 hours at room temperature. Following PBS rinses, coverslips and tissue sections were mounted in Fluoromount-G (Invitrogen, 00495802).

Fluorescent microscopy images were obtained on an LSM800 confocal microscope using 20× air or 63× oil objectives. Tile scanning was used for high-resolution imaging of large areas with images stitched together with 10% overlap using ZenBlue software. CZI files were then uploaded and processed using ImageJ (RRID: SCR_003070). All manipulations of images (brightness/contrast and smoothing) were done uniformly for each image. For quantifying Krt8 and fibronectin fluorescence, samples were processed and imaged simultaneously under the same microscope settings, and then an identical manual threshold was applied to all images in ImageJ. Fluorescence was measured for each area with “measure” analysis function. Krt8 expression was normalized to Krt8 fluorescence in areas without visible metastases. The cell counter plugin was used to manually quantify Ki67⁺/ Sftpc⁺ cells, Ly6G⁺ neutrophils, or CD68⁺ macrophages. Metastasis burden was quantified from H&E sections by a blinded examiner using Aperio ImageScope software (V12.4.3). Whole lung lobe images were obtained with Aperio FL ScanScope digital slide scanner with 20× objective.

Statistical analysis for animal and bench studies was performed using GraphPad Prism V9. The number of samples and mice is explicitly stated in text with appropriate statistical test and P values. Data reported with error bars represent mean ± SEM. Statistical analyses of the sequencing data are described in the relevant sections.

The code used to analyze scRNA-seq data can be found on GitHub at https://github.com/kidcancerlab/epithelial. Previously unpublished raw sequencing data and the associated count tables with metadata are available through GEO accession GSE252703. Human data are available on dbGaP through accession phs003569.v1.p1 and GEO accession GSE270231. The cell lines utilized in this study are available from commercially available resources or can be obtained upon request to corresponding author.

We first sought to determine the spatial localization of osteosarcoma cells as they begin to colonize the lung, which would provide evidence for the specific lung environment most relevant to these earliest steps in metastatic colonization. To control the timing of dissemination, we inoculated immunocompetent C57BL/6 mice intravenously (tail vein) with F420 murine osteosarcoma cells. We then collected the lungs from mice euthanized at 1, 7, and 14 days after injection and processed the lungs onto slides to evaluate the development of the earliest metastatic niche. F420 cells demonstrated a stereotyped pattern of lung colonization characterized by extravasation and invasion through the alveolar epithelial layer, inducing injury to podoplanin-positive type 1 alveolar epithelial cells (AEC1). We identified a compensatory increase of surfactant protein C–positive type 2 alveolar epithelial cells (AEC2) over areas of injured AEC1 (Fig. 1A), reminiscent of the process triggered by other types of alveolar injury (27). Similar results were seen in the K7M2 murine osteosarcoma model (Supplementary Fig. S1A).

In other nonmalignant models of alveolar injury, the disruption of alveolar integrity promotes neutrophil followed by macrophage infiltration into the lung (28). We likewise observed a short burst of neutrophil infiltration shortly after colonization that diminished over time and was followed by a significant and sustained accumulation of macrophages (Supplementary Fig. S1B), mirroring the kinetics of immune infiltration observed in other alveolar injury models (28). Taken together, these data suggested that disseminated osteosarcoma cells colonize the distal alveolar niche and induce an acute lung injury response during early metastatic colonization.

To understand how these changes affect the behavior of the surviving osteosarcoma cells and the surrounding lung tissues, we isolated early F420 and K7M2 metastases using a stereomicroscope and processed them for scRNA-seq (Fig. 1B). Cell cluster annotation was done using a combination of automated and manual approaches using established markers (Supplementary Fig. S2A and S2B; “Materials and Methods”). Multiple cell populations demonstrated enrichment in metastasis when compared with non–tumor-bearing control lung tissues (Fig. 1C; Supplementary Fig. S2C). We divided cells computationally into infiltrating/immune and native/stromal compartments to quantify the relative frequencies of each distinct cell population. We stress that this quantification has inherent bias in the capture of certain cell populations by droplet-based scRNA-seq methods (29). With this caveat in mind, we found that metastasis-bearing lungs exhibit a marked enrichment in infiltrating myeloid cells (particularly macrophages) and distal airway epithelial cells (Fig. 1D and E). Immunofluorescence imaging confirmed that developing metastases become highly enriched in macrophages and distal airway epithelial cells when compared with an uninvolved lung (Supplementary Fig. S2D and S2E). We further confirmed the enrichment of epithelial cells in metastasis compared with unaffected lung areas using the thick-tissue Ce3D tissue-clearing method (Supplementary Videos S1 and S2). We next isolated the alveolar epithelial cells and macrophages to evaluate the phenotypic changes in the two most clearly altered stromal and immune cell populations, hypothesizing that the invasion of the osteosarcoma cells influences these changes and that these constitute important elements of tumor education within the developing metastatic niche.

Distal epithelial cell clusters from metastases and healthy (non–metastasis bearing) C57BL/6 (control for F420) or BALB/c (control for K7M2) mice were merged, integrated, and reanalyzed through our Seurat-based pipeline. Several populations of epithelial cells were enriched in metastasis-bearing lungs compared with control (Fig. 2A). Five distal epithelial clusters were identified in F420 and K7M2 metastases, each with distinct gene expression signatures (Fig. 2B).

AEC1 and AEC2 populations were discernible based on the enrichment of genes such as Cav1, Hopx, and Sftpc. Two additional AEC2-like populations were found within the metastatic samples, one of which expressed genes such as Lcn2 and Lrg1, whereas the other demonstrated upregulation of genes involved in cell cycle such as Mki67 and Cdk1. This finding was consistent with the physiologic reaction to alveolar injury, as AEC2 cells are facultative stem cells within the alveolus, functioning to reestablish epithelial barrier integrity after AEC1 injury (27). In the canonical wound-healing process, AEC2 is activated into a “primed” pAEC2 state by AEC1 injury (30). pAEC2 then enters the cell cycle (cAEC2) before differentiating into AEC1 cells to reestablish alveolar epithelial integrity. Given the phenotypic resemblance of these previously defined subtypes, we assigned the Lcn2⁺ population a label of “primed” AEC2 (pAEC2) and the Mki67⁺ cluster a label of “cycling” AEC2 (cAEC2), consistent with the established literature for murine lung injury models (30).

The fifth epithelial cluster demonstrated upregulation of atypical cytokeratins, such as Krt8 and Krt18/19, and genes such as Cdkn1a, Sfn, and Nupr1. This was an interesting finding, as Krt8⁺ is a primary marker of a recently described fibrogenic intermediate produced as activated AEC2 cells differentiate into AEC1 cells in response to injury (30–32). Under normal conditions, these intermediates are short lived and do not persist beyond the acute phase of the injury. However, under certain chronic inflammatory conditions, these damage-associated transient progenitors (DATP; also referred to as alveolar differentiation intermediates and prealveolar type 1 transitional cell state) become arrested in the Krt8⁺ transitional state and can promote pathologic fibrosis. Indeed, a similar process has been shown to occur in human patients with chronic IPF (30–32).

To ensure that these findings reflect the biology that is relevant to the human disease, we collected tissue samples from patients undergoing metastatectomy at our local institution and performed single-cell sequencing using the same protocol. To enhance the power of this analysis and ensure the broad applicability of these findings, we combined these with previously published datasets containing both normal lung tissues and metastatic osteosarcoma lesions (17, 18, 33–35). This assessment of human tissues revealed changes in both epithelial content and phenotype, very similar to those observed in our murine models, with several epithelial cell clusters enriched in metastasis relative to a healthy lung (Fig. 2C and D).

Of note, in human lung pathologic conditions, AEC2 has been shown to differentiate through a disease-specific, aberrant KRT17⁺ basaloid cell state. These KRT17⁺ cells stain positive for KRT8 by IHC and cluster with murine Krt8⁺ DATP in cross-species single-cell transcriptomic analyses (32, 36). Thus, they are likely to be an orthologous AEC intermediate. Our human osteosarcoma metastasis datasets replicated the finding in the murine models, with strong enrichment of KRT17⁺ basaloid cells (Fig. 2C and D).

We next performed spatial transcriptomics to localize epithelial subsets relative to metastatic lesions in the F420 immunocompetent murine model (Fig. 2E). Epithelial subsets like cAEC2 and DATP were only found in proximity to metastatic lesions, supporting that they are metastasis-induced cell types. pAEC2 and cAEC2 localized in the metastasis/lung border, whereas DATP cells infiltrated into metastatic lesions. We confirmed these sequencing and spatial transcriptomic findings in tissue slices by showing that AEC2 within the developing metastatic niches was hyperproliferative (Ki67⁺) relative to AEC2 in uninvolved areas of the lung across multiple models of osteosarcoma metastasis (Fig. 2F and G). Additionally, accumulation of metastasis-associated Krt8⁺ epithelial cells occurs across multiple osteosarcoma models (Fig. 2H and I). Lastly, KRT8⁺ AEC intermediates were prevalent within human lung metastasis samples by immunofluorescence (Supplementary Fig. S3). These findings demonstrate that basaloid/DATP accumulation is a conserved phenomenon in osteosarcoma metastasis and not unique to mouse models.

We conclude that partially differentiated epithelial intermediates, which have been associated with aberrant nonhealing lung wounds in other pathologic processes, consistently accumulate within the developing metastatic niche in both mice and humans. Functionally, this finding suggested that impaired epithelial wound healing is central to osteosarcoma lung metastasis. As lung alveolar wound repair is a highly stereotyped and regulated process that requires coordination with other cell types [such as fibroblasts and macrophages (37)], we wondered whether the monocyte-derived macrophages that accumulate around developing metastatic lesions might evidence changes characteristic of either physiologic or pathologic wound healing. We thus leveraged our single-cell datasets to define the macrophage phenotypes associated with osteosarcoma metastasis.

Macrophages can play paradoxical roles during distinct phases of wound healing (38). In the early wound-healing response, macrophages are predominantly inflammatory in nature (39, 40), marked by a high expression of antigen presentation proteins and inflammatory cytokines in the early phases of wound healing (41). Once the inciting noxious stimulus is controlled, anti-inflammatory/resolution macrophages promote healing by reorganizing the ECM, phagocytosing apoptotic cells, and dampening inflammation, though their uncontrolled accumulation can promote chronic, nonhealing wounds (42).

To ascertain the macrophage phenotypes associated with developing osteosarcoma metastases, we isolated myeloid cells from our murine and human scRNA-seq datasets. In this analysis, we also included primary murine tibial tumors to define how the surrounding microenvironment impacts macrophage phenotypes in the primary versus metastatic microenvironment. Murine osteosarcoma metastases demonstrated marked accumulation of several macrophage populations that were transcriptionally distinct from populations present in healthy lung and tibial tumors (Fig. 3A and B). Tumor-associated macrophage annotations were made using recently published criteria (43). Notably, a population of macrophages expressing Trem2, Cd9, and Spp1 was highly enriched in lung metastasis compared with healthy lung and primary tibial tumors. This macrophage population shares transcriptional profile with scar-like macrophages (44), a macrophage phenotype characterized by expression of genes such as Cd9, Trem2, Spp1, Gpnmb, and Fabp5, which have recently been identified as a crucial driver of pathologic fibrosis within the lung and during liver metastasis (45–47). Their enrichment in osteosarcoma lung metastasis, but not in primary tibial tumors, suggests a microenvironmental influence on macrophage phenotypes, with chronic injury/fibrosis being promoted specifically in the lung, likely secondary to osteosarcoma-induced aberrant wound-healing process. Similar macrophage subpopulations were enriched in human osteosarcoma metastasis relative to a healthy human lung (Fig. 3C and D).

Given the complexity of macrophage phenotypes and overlapping marker expression, we utilized spatial transcriptomics to define the localization of fibrotic macrophage subsets relative to established metastases (Fig. 3E). Scar macrophages were only localized to the metastatic compartment of the lung, such as epithelial fibrotic cell types, and confined to the lung–metastasis border. Spatial analysis also demonstrated marked enrichment of actively proliferating (cycling) macrophages and inflammatory macrophages as well as recruitment of classic monocytes to metastatic lesions relative to an unaffected lung. Based on our epithelial and macrophage data, we propose a model in which the activation of epithelial wound healing, triggered by the infiltrating osteosarcoma cells at the initial stages of metastasis and continuing along the leading edge during lesional growth, coincides with the enrichment of inflammatory macrophages. Subsequently, the tumor-educated inability to resolve the wound causes macrophages with chronic wound phenotypes to accumulate within the metastatic lesion. In other disease states, this chronic wound-healing process ultimately leads to pulmonary fibrosis, characterized by the pathologic deposition of ECM proteins such as fibronectin (48). Therefore, we next investigated if osteosarcoma metastases are fibrotic.

Fibronectin is an ECM protein that is necessary for proper wound healing but can promote progressive organ dysfunction if not tightly regulated (49). In all three of our osteosarcoma models tested, fibronectin markedly accumulated within developing metastatic lesions, relative to an adjacent unaffected lung. Given the findings above, this is consistent with a metastasis-induced fibrotic transformation of the osteosarcoma-associated lung (Fig. 4A). To elucidate whether osteosarcoma-associated fibronectin deposition is metastasis specific, we compared primary F420 and K7M2 tibial tumors with lung metastases. Primary tibial tumors expressed low levels of intracellular fibronectin, whereas metastases were characterized by extensive extracellular, fibrillar fibronectin matrices at the lung–metastasis interface, suggesting that fibronectin matrix formation is metastasis specific and not an inherent trait of osteosarcoma tumors (Fig. 4B).

In nonmalignant lung fibrosis diseases, activated myofibroblast secretes and organizes fibronectin into dense matrices (50). Additionally, fibronectin deposition by cancer-associated fibroblasts is also noted in other tumor types (51). However, our scRNA-seq data did not demonstrate enrichment of myofibroblasts in osteosarcoma metastases, suggesting an alternative source of fibrotic matrix in osteosarcoma metastases. Given the mesenchymal origin of osteosarcoma, it is developmentally related to fibroblasts and demonstrates strong matrix-forming tendencies. We leveraged the xenogeneic nature of the OS-17 metastasis model to help determine whether metastasis-associated fibronectin is tumor cell derived. To do this, we stained metastatic lesions with a human-specific fibronectin antibody. Human (tumor)-derived fibronectin was highly expressed throughout metastatic lesions. However, ECM fibrils were enriched at the lung–metastasis interface, suggesting that local microenvironment interactions favor ECM deposition at this location.

In lung fibrotic disorders, epithelial damage triggers fibronectin secretion and fibrosis by stimulating mesenchymal cells (52). Based on our data demonstrating osteosarcoma cells induce epithelial injury, we hypothesized that epithelial cells may induce osteosarcoma fibronectin deposition. Indeed, in our OS-17 model, fibronectin matrix deposition correlated with dense collections of Krt8⁺ DATP cells (Fig. 4C). To test this hypothesis within a reductionist system, OS-17 cells were cocultured with human immortalized lung epithelial cells (HBEC) or stimulated with epithelial-conditioned media. OS-17 cells generated a complex fibronectin matrix in the presence of lung epithelial cells (Fig. 4D). This reaction was replicated by stimulation with epithelial-conditioned media (Fig. 4E), suggesting that the reaction is triggered by soluble signals produced by the epithelial cells.

In summary, chronic unresolved injury induced by osteosarcoma lung colonization culminated in microenvironmental fibrosis. This fibronectin-rich matrix was synthesized by osteosarcoma cells in a metastasis-dependent manner because of microenvironment-specific epithelial interactions. We thus hypothesized that osteosarcoma-induced chronic injury and subsequent fibrotic transformation of the lung niche are key drivers of osteosarcoma metastasis progression and that inhibiting this metastasis-associated fibrosis would prevent osteosarcoma metastasis growth.

Fibrotic lung conditions are progressive, often fatal diseases. Nintedanib is an orally available TKI with a broad target profile that has demonstrated clinical efficacy in slowing the progression of fibrotic lung diseases (22, 53, 54). The pathophysiologic overlap with osteosarcoma metastases and lung fibrotic conditions suggested that antifibrotic agents such as nintedanib could disrupt metastasis progression. To test the effect of nintedanib on osteosarcoma metastasis progression, tumor cells were injected intravenously via tail vein to promote experimental metastasis formation. Nintedanib or vehicle was initiated at 24 hours after injection to focus on the impact of nintedanib on metastasis progression, not extravasation into lung (Fig. 5A). Study mice were monitored daily for signs of metastasis. When metastasis was confirmed in one mouse, all mice were humanely euthanized and lung metastases quantified (Fig. 5B–G). We found that nintedanib significantly hindered metastatic progression in both immunocompetent (F420 and K7M2) and immunodeficient (OS-17) models.

Interestingly, nintedanib did not significantly affect F420 or OS-17 cell viability in vitro (Supplementary Fig. S4A). Additionally, nintedanib did not affect primary tibial tumor growth in an F420 model while causing slowing but not regression of OS-17 tibial growth (Supplementary Fig. S4B–S4D). Collectively, the conserved activity of nintedanib in disrupting metastasis progression across models and within three distinct mouse strains strongly supports the antimetastatic effect in osteosarcoma. Given nintedanib’s clinical activity in fibrotic diseases, and its marked effect on metastatic osteosarcoma compared with primary tibial tumors or cells growing in vitro, we surmised that nintedanib likely inhibited metastatic progression by disrupting the osteosarcoma-induced, pathologically aberrant wound-healing process within the lung.

Nintedanib is a promiscuous TKI with documented activity on multiple receptor tyrosine kinase receptors such as FGFR1-4, VEGFR, and PDGFRα/β (21, 22). We next utilized a combination of spatial and single-cell transcriptomics to gain insights into the mechanisms underlying the potent effects of nintedanib on osteosarcoma lung metastasis. We devised an experimental approach using NicheNet, a bioinformatic tool that identifies transcriptomic changes induced by ligand–receptor interactions, to identify transcriptional activation of nintedanib target pathways in the metastatic lung relative to an unaffected lung. We then performed single-nucleus RNA-seq on mice treated with nintedanib to identify cell types and pathways impacted by nintedanib during metastasis progression (immediate treatment) or on established metastases (delayed treatment; Fig. 6A). First, we projected nintedanib target pathways identified by NicheNet analysis (Supplementary Fig. S5A–S5G) onto our spatial dataset to identify where nintedanib targets are active in metastatic lesions. Strikingly, although activation of pathways such as FGFR and PDGFR was markedly upregulated in metastatic lesions relative to an unaffected lung, FGFR activation was spatially restricted to the lung–metastasis border, whereas PDGFR was preferentially activated in the metastasis core (Fig. 6B).

To understand how nintedanib affects the cell types that constitute the metastatic niche and the pathways impacted by nintedanib, we treated mice with nintedanib shortly after metastatic colonization (3 days after injection) or once metastases were established and confirmed by bioluminescence imaging (28 days after injection, Fig. 6C demonstrates treatment schematic and representative H&E images for each treatment group; see Supplementary Fig. S6 for bioluminescence imaging). Single-nucleus transcriptomic analysis demonstrated that nintedanib prevented the accumulation of fibrotic DATP and scar-macs in the immediate treatment group, consistent with nintedanib inhibiting metastasis-induced fibrosis (Fig. 6D). Remarkably, nintedanib also reversed metastasis-induced fibrotic phenotypes, as evidenced by a decrease in DATP and scar-macs in the delayed treatment group (Fig. 6D). Lastly, we validated the effect of nintedanib at the protein level in vitro and in vivo, demonstrating that nintedanib inhibited epithelial-induced tumor fibronectin deposition as well as decreased fibronectin and DATP cells in established metastases in vivo (Supplementary Fig. S7A–S7D). We concluded that nintedanib fundamentally alters the cellular and matrix composition of osteosarcoma metastases, impacting both tumor cells and surrounding stromal cells.

We next utilized our NicheNet results to ask what nintedanib-targeted pathways are downregulated by nintedanib in key cell types: tumor cells, infiltrating myeloid cells, and epithelial cells (Fig. 6E). FGFR activation was downregulated by nintedanib in all three cell types, whereas the impact on PDGFR was limited to tumor cells and infiltrating myeloid cells. VEGFR signaling was not impacted in these three cell types, which is consistent with VEGFR signaling being primarily restricted to endothelial cells (Supplementary Fig. S8). Collectively, the results of our preclinical nintedanib trial, coupled with this molecularly guided pharmacodynamic analysis of nintedanib-treated metastases, suggest a tight linkage between the antimetastatic activity and the anti-fibrogenic activity of nintedanib. These results also suggest a requirement for fibrotic reprogramming of the lung microenvironment during osteosarcoma metastasis progression. Inhibition of this fibrogenic, aberrant wound-healing program represents a novel therapeutic vulnerability with the potential for combating osteosarcoma lung metastasis.

How disseminated cancer cells colonize, integrate with, and coerce foreign organs to promote metastatic growth remains a poorly understood phenomenon. With respect to osteosarcoma, specifically, our data support a model in which the tumor cells trigger and then perpetuate the lung’s intrinsic wound-healing program, beginning with dissemination and resulting in a chronic, nonhealing fibrotic wound. The characterization of a tumor as a nonhealing wound is not new. It was first postulated by Virchow and then further conceptualized by Dvorak. However, this paradigm largely focused on tumorigenesis of primary tumors in adult epithelial carcinomas (55). Our results demonstrate that osteosarcoma, a mesenchymal pediatric tumor, induces alveolar epithelial injury upon colonization of the lung, leading to a chronic nonhealing wound and fibrotic transformation of the surrounding lung. In many ways, osteosarcoma–epithelial interactions mimic, albeit in a highly pathologic sense, epithelial–lung fibroblast interactions that occur during nonmalignant lung wounds (52, 56). This observation may suggest that aberrant epithelial–mesenchymal interactions are central to lung metastasis across diverse histologic subtypes.

Indeed, an analogous scenario has been demonstrated in breast cancer lung metastasis, in which epithelial tumor cells co-opt and activate normal lung mesenchymal fibroblasts to drive metastasis, with fibrogenic wound-like processes playing a central role in the mechanisms for colonization (57). It may be that in the case of breast cancer metastasis, the tumor cells play the role of the epithelial cells and the fibroblasts the role of the mesenchymal cell in a process similar to that observed here. If true, this aberrant wound-healing process and the resulting fibrogenic changes may represent mechanisms broadly conserved across tumor types, though this requires rigorous and direct exploration across tumor models. Notably, organs with facultative repair capacity that are susceptible to fibrosis, such as the lung, liver, and pancreas, are organs commonly affected by high metastatic rates (or, in the case of the pancreas, characterized by fibrotic/desmoplastic primary tumors; refs. 8, 58, 59). It will be critical from a therapeutic perspective to elucidate whether initiation of abnormal wound healing and progression to organ fibrosis is specific to a handful of histologic tumor types or a generalizable mechanism.

Our work significantly advances the mechanistic understanding of osteosarcoma lung metastasis. Our findings converge on previous findings by Zhang and colleagues (60) but add several critical new observations. Notably, our work demonstrates that fibrogenic reprogramming is a niche-wide event initiated by osteosarcoma–epithelial interactions and provides critical pharmacodynamic analysis of nintedanib-treated metastases in an immunocompetent host. Our mechanisms also unify and advance concepts emerging from other recent publications, suggesting that cancer cells disrupt the physiology of the lung epithelia during colonization (61, 62). Long considered inert bystanders in the processes that facilitate malignant progression, stromal cells such as alveolar epithelial cells have received relatively little attention as potential drivers of metastasis biology. However, our work establishes the key role that epithelial cells play in metastatic colonization. Indeed, the clear physical proximity and interactions of early disseminated osteosarcoma cells with alveolar epithelial cells, together with the functional data demonstrating that epithelial cells initiate fibrogenic reprogramming of osteosarcoma cells, strongly suggest active two-way paracrine signaling between osteosarcoma cells and epithelial cells as a key event driving osteosarcoma colonization of the lung. Future work is required to clarify the mediators of these osteosarcoma–epithelial interactions.

The use of TKIs in patients with sarcoma is well established, especially in the relapse setting (63, 64). Indeed, many have been evaluated, several of which have partially overlapping targets with nintedanib. For instance, cabozantinib is an orally available TKI that inhibits VEGFR2, c-MET, and RET and has activity in the relapsed/metastatic setting (65). Nearly 40% to 50% of osteosarcomas demonstrate amplification of VEGFA (66), and MET signaling can transform osteoblasts (67), the likely cell of origin for osteosarcoma. Accordingly, cabozantinib could work on both the microenvironment (inhibiting angiogenesis) and tumor cells (inhibiting MET signaling) but likely would not inhibit tumor or stromal FGFR and PDGFR signaling as implicated by our single-cell and spatial transcriptomic analysis. To our knowledge, the tumor cell-intrinsic and cell-extrinsic mechanisms mediating the antimetastatic effects of cabozantinib and other clinically validated TKIs remain poorly understood. Our data suggest that nintedanib exerts its effect in a context-dependent manner (lung metastasis) and provide evidence that nintedanib’s antimetastatic activity likely involves the suppression of fibrogenic signaling pathways involving both tumor and host cells, particularly those involving FGFR-induced transcriptional responses. Even in the treatment of pulmonary fibrosis, the most clinically important target(s) of nintedanib remain unclear. Nintedanib’s spectrum of activity includes inhibition of VEGFR, PDGFR, and FGFR signaling, as well as several intracellular kinases such as SRC, which is critical for the signal transduction of many RTKs inhibited by nintedanib (21). This target promiscuity is common among the TKIs, including those already used to treat osteosarcoma (68). Although several of these targets are shared, this work provides an initial suggestion that the unique targets may account for the clinical effects of individual agents and that the on-target, off-tumor effects may be just as important as those intrinsic to the tumor cells. Here, we demonstrated how single-cell and spatial transcriptomics can be leveraged to further our mechanistic understanding of the holistic impact that therapeutic interventions have on the metastatic microenvironment. Given that TKIs often demonstrate activity against unanticipated targets (69), a more comprehensive evaluation using such approaches may point to novel mechanisms and suggest strategic combinations of agents that would otherwise not be pursued.

Elucidating why fibronectin/fibrosis is required for osteosarcoma metastatic progression remains an important area of ongoing investigation. Potential, nonmutually exclusive mechanisms include directly providing survival signals to tumor cells during early lung colonization (70), resistance to chemotherapy (71, 72), protection against innate/adaptive antitumor immunity (73), and reprogramming of niche cells such as macrophages. We note that the effectiveness of nintedanib was similar in mice with or without intact adaptive immunity, which does not rule out an important role for fibrosis in controlling antitumor immunity but does support the idea that fibronectin/fibrosis could function through both tumor cell–intrinsic and non–tumor cell–intrinsic mechanisms to promote metastatic progression.

In conclusion, we note that metastases are complex multicellular structures composed of tumor, immune, and altered resident stromal cells. Utilizing scRNA-seq and spatial transcriptomics, we illuminate how osteosarcoma cells manipulate the lung microenvironment into a chronic state of wound healing. From a translational perspective, we demonstrate that targeting the microenvironment is a viable therapeutic option across multiple osteosarcoma models, which is an important consideration due to the substantial intertumoral heterogeneity inherent to osteosarcoma.

J.B. Reinecke reports grants from Alex’s Lemonade Stand Young Investigator, Alex’s Lemonade Stand Crazy 8, NIH R01CA260178, NIH CTSA Grant UL1TR002733, Translational Studies Grant from the Osteosarcoma Institute, CancerFree KIDS Foundation, Steps for Sarcoma Foundation, and Sarcoma Foundation of America during the conduct of the study. L. Jimenez Garcia reports grants from Alex’s Lemonade Stand Crazy 8 Award, Alex’s Lemonade Stand Foundation Young Investigator Award, NIH R01CA260178, NIH CTSA Grant UL1TR002733, Translational Studies Grant from the Osteosarcoma Institute, CancerFree KIDS Foundation, Steps for Sarcoma Foundation, and Sarcoma Foundation of America during the conduct of the study. A.C. Gross reports grants from Alex’s Lemonade Stand Crazy 8 Award, Alex’s Lemonade Stand Foundation Young Investigator Award, the NIH, Osteosarcoma Institute, CancerFree KIDS Foundation, Steps for Sarcoma Foundation, and Sarcoma Foundation of America during the conduct of the study. M. Cam reports grants from Alex’s Lemonade Stand Crazy 8 Award, Alex’s Lemonade Stand Foundation Young Investigator Award, NIH R01CA260178, NIH CTSA Grant UL1TR002733, Translational Studies Grant from the Osteosarcoma Institute, CancerFree KIDS Foundation, Steps for Sarcoma Foundation, and Sarcoma Foundation of America during the conduct of the study. M.V. Cannon reports grants from Alex’s Lemonade Stand Crazy 8 Award, Alex’s Lemonade Stand Foundation Young Investigator Award, NIH R01CA260178, NIH CTSA Grant UL1TR002733, Translational Studies Grant from the Osteosarcoma Institute, CancerFree KIDS Foundation, Steps for Sarcoma Foundation, and Sarcoma Foundation of America during the conduct of the study. R.D. Roberts reports grants from Alex’s Lemonade Stand Foundation, Osteosarcoma Institute, the NIH, CancerFree KIDS Foundation, Steps for Sarcoma Foundation, and Sarcoma Foundation of America during the conduct of the study, as well as a patent for CA3079781A1, IL-8 inhibitors for Use in the Treatment of Some Sarcomas issued. No disclosures were reported by the other authors.

J.B. Reinecke: Conceptualization, data curation, formal analysis, funding acquisition, investigation, methodology, writing–original draft, writing–review and editing. L. Jimenez Garcia: Data curation, formal analysis, methodology, writing–review and editing. A.C. Gross: Data curation, formal analysis, methodology, writing–review and editing. M. Cam: Data curation, methodology, writing–review and editing. M.V. Cannon: Data curation, formal analysis, methodology, writing–review and editing. M.J. Gust: Formal analysis. J.P. Sheridan: Formal analysis, methodology. B.E. Gryder: Funding acquisition, writing–review and editing. R. Dries: Funding acquisition, writing–review and editing. R.D. Roberts: Conceptualization, data curation, formal analysis, supervision, funding acquisition, methodology, writing–original draft, writing–review and editing.

We would like to thank the patients and their families for donation of tissue. We would like to thank the Histopathology, Microscopy, Biopathology Center Processing and Banking and Institute for Genomic Medicine sequencing cores at Nationwide Children’s Hospital for their technical assistance. This work is supported by Alex’s Lemonade Stand Crazy 8 Award (R.D. Roberts, R. Dries, B.E. Gryder), Alex’s Lemonade Stand Foundation Young Investigator Award (J.B. Reinecke), NIH R01CA260178 (R.D. Roberts), NIH CTSA Grant UL1TR002733 (R.D. Roberts), Translational Studies Grant from the Osteosarcoma Institute (R.D. Roberts), CancerFree KIDS Foundation (R.D. Roberts, J.B. Reinecke), Steps for Sarcoma Foundation (R.D. Roberts, J.B. Reinecke), Sarcoma Foundation of America (R.D. Roberts), and Nationwide Children’s Director Strategic Development Fund (R.D. Roberts).