Vimentin is required for tumor progression and metastasis in a mouse model of non–small cell lung cancer

Vimentin is highly expressed in metastatic cancers, and its expression correlates with poor patient prognoses. However, no causal in vivo studies linking vimentin and non–small cell lung cancer (NSCLC) progression existed until now. We use three complementary in vivo models to show that vimentin is required for the progression of NSCLC. First, we crossed LSL-KrasG12D; Tp53fl/fl mice (KPV+/+) with vimentin knockout mice (KPV−/−) to demonstrate that KPV−/− mice have attenuated tumor growth and improved survival compared with KPV+/+ mice. Next, we therapeutically treated KPV+/+ mice with withaferin A (WFA), an agent that disrupts vimentin intermediate filaments (IFs). We show that WFA suppresses tumor growth and reduces tumor burden in the lung. Finally, luciferase-expressing KPV+/+, KPV−/−, or KPVY117L cells were implanted into the flanks of athymic mice to track cancer metastasis to the lung. In KPVY117L cells, vimentin forms oligomers called unit-length filaments but cannot assemble into mature vimentin IFs. KPV–/– and KPVY117L cells fail to metastasize, suggesting that cell-autonomous metastasis requires mature vimentin IFs. Integrative metabolomic and transcriptomic analysis reveals that KPV–/– cells upregulate genes associated with ferroptosis, an iron-dependent form of regulated cell death. KPV–/– cells have reduced glutathione peroxidase 4 (GPX4) levels, resulting in the accumulation of toxic lipid peroxides and increased ferroptosis. Together, our results demonstrate that vimentin is required for rapid tumor growth, metastasis, and protection from ferroptosis in NSCLC.


INTRODUCTION
Lung adenocarcinoma (LUAD) is a type of non-small cell lung cancer (NSCLC) and is the most common subtype of lung cancer, representing 40% of all lung cancer diagnoses [1]. Vimentin, a type III intermediate filament, is highly expressed in lung cancer and is associated with metastatic tissue and lower survival rates in patients with NSCLC [2][3][4]. Despite this association, a causal, in vivo link between vimentin and disease progression has not been established.
Vimentin's role in the metastatic process has been extensively investigated in vitro. During epithelial-to-mesenchymal transition (EMT), epithelial cells lose their apicobasal polarity and intercellular adhesion properties by downregulating epithelial cellassociated genes, including E-cadherin and cytokeratins, and progressively acquire migratory and invasive capabilities associated with the mesenchymal cell phenotype by upregulating genes, such as N-cadherin and vimentin [5,6]. EMT-mediated tumor metastasis is stimulated by transcription factors, such as TWIST1, SNAI1, and ZEB1, known to upregulate vimentin [7][8][9][10]. Formation and elongation of the invadopodia, a proteolytically active plasma membrane projection that facilitates cancer cell invasion across the basement membrane and migration through the collagen-rich interstitial space, requires vimentin [11]. Vimentin is crucial for establishing front-rear polarity, which is necessary for the efficient migration of tumor cells [12]. Previous studies identified these vimentin-dependent mechanisms in cancer cell invasive migration, metastasis, and poor prognosis in patients with lung, breast, head and neck, and bone cancer cells [9,[13][14][15][16][17][18][19][20][21].
Tumors must maintain redox homeostasis to survive and grow. An emerging mechanism by which tumor cells manage oxidative stress is ferroptosis. Ferroptosis is a form of iron-dependent regulated cell death (RCD) initiated by the peroxidization of phospholipids (PLs) containing polyunsaturated fatty acid chains (PUFA-PLs). Extrinsic and intrinsic pathways regulate ferroptosis [22]. The extrinsic pathway regulates ferroptosis by inhibiting cell membrane transporters, such as the system Xc− cystine/glutamate transporter (consisting of SLC7A11 and SLC3A2 subunits) or through iron uptake via transferrin and lactotransferrin transporters. The intrinsic pathway is activated by blocking intracellular antioxidant enzymes, such as glutathione peroxidase 4 (GPX4). Glutathione is a cofactor for GPX4, which neutralizes lipid hydroperoxides and protects the cell from ferroptosis. KRASmutant LUAD cells express high levels of SLC7A11 and GPX4; pharmacological or genetic inhibition of either SLC7A11 and GPX4 attenuates tumor growth in vivo [23,24]. Of note, the regulation of ferroptosis is not entirely understood.
Transcriptional profiling reveals a less differentiated cancer phenotype in vimentin-null cancer cells To better understand how vimentin is involved in the molecular pathways of LUAD, we used RNA sequencing (RNA-seq) to identify genes that have altered expression in the absence of vimentin. Epithelial-derived cancer cells (CD45-negative, EpCAM-positive) were isolated from KPV +/+ and KPV −/− lungs at 6 w.p.i. with Ad-Cre (hereafter referred to as KPV +/+ and KPV −/− cells). Western blot and immunofluorescence staining confirmed the absence of vimentin in KPV −/− cells ( Supplementary Fig. S2A S2E). K-means clustering identified two clusters: Cluster 1 contained 316 upregulated genes while Cluster 2 had 588 downregulated genes in KPV +/+ cells compared to KPV −/− cells ( Fig. 2A). Cluster 1 was enriched for genes associated with Gene Ontology (GO) processes mesenchymal cell proliferation and cell migration. In contrast, Cluster 2 was enriched for genes related to metabolic processes, epithelial cell differentiation, and cell adhesion (Fig. 2B). By exploring the DEGs that contribute to mesenchymal cell proliferation and epithelial cell differentiation, we found several EMT-associated genes upregulated in KPV +/+ cells, including Vimentin, Twist1, and Cdh2, the gene that codes for N-cadherin (Fig. 2C). This finding was confirmed by IHC of tumors and Western blots on isolated tumor cells, which showed an increase in N-cadherin in KPV +/+ cells compared to KPV −/− cells ( Supplementary Fig. S2A, F). In contrast, genes associated with epithelial cell phenotype, including claudins Cldn2, Cldn8, and Cldn18, and the epithelial marker Epcam and the surfactant protein Sftpd, were upregulated in KPV −/− cells. These data suggest that KPV −/− cells, but not KPV +/+ cells, fail to upregulate essential mesenchymal genes that confer metastatic potential.
Withaferin A (WFA) is a steroidal lactone used as an anticancer agent in various murine tumor models, including breast, prostate, and ovarian cancer [34][35][36]. WFA treatment leads to the phosphorylation of vimentin at serine 38 (Ser38) and serine 56 (Ser56), resulting in the aggregation of vimentin Messenger RNA collected from KPV +/+ and KPV −/− cells was subjected to RNA sequencing. A Differentially expressed genes (DEGs) between KPV +/+ and KPV −/− cells were clustered using K-means clustering. B Genes enriched in Cluster 1 (316 genes) and Cluster 2 (588 genes) were subjected to GO enrichment analysis. GO Processes with FDR < 0.05 are shown. GeneRatio is the number of genes in the cluster associated with the GO process divided by the total number of genes in that GO process. C Expression values (counts per million; CPM) of select genes are shown. N = 3 for each group. Data in panel C are presented as the mean ± standard deviation. All gene comparisons (KPV +/+ vs. KPV −/− ) have FDR < 0.05 after adjusting for multiple comparisons; therefore, all gene differences shown between KPV +/+ vs. KPV −/− cells are statistically significant.
intermediate filaments (IFs) [37]. To test the hypothesis that WFA disrupts vimentin IFs and impairs migration and invasion in vitro, we treated KPV +/+ cells and human lung adenocarcinoma (A549) cells with WFA ( Fig. 3D and Supplementary Fig.  S3A). In untreated cells, vimentin IFs were extended to the plasma membrane from the nucleus. In WFA-treated cells, vimentin IFs were retracted from the plasma membrane and collapsed around the nucleus ( Fig. 3D and Supplementary Fig.  S3A). There was no change in vimentin protein levels in cells treated with WFA ( Supplementary Fig. S3B) [38]. WFA treatment decreased KPV +/+ cell migration by~46% (Fig. 3E) and completely suppressed cell invasion (Fig. 3F). We observed a similar trend in human-derived A549 cells, which exhibited a dose-dependent decrease in cell migration following treatment with WFA ( Supplementary Fig. S3C). These data suggest that an intact vimentin intermediate filament network is required for migration and invasion in an in vitro model of NSCLC. Vimentin is required for in vitro cancer cell migration and invasion. A A scratch wound assay was used to evaluate cell migration. Representative images are shown at 0 and 6 h following scratch formation. Wound area closure was compared to the starting value and quantified for KPV +/+ (n = 11) and KPV −/− (n = 17) cells; each point represents a separate scratch wound. B Cell invasion through a Matrigel-coated transwell was measured over 48 h. Invasive index is the mean number of cells invaded per 20× magnification imaging field. KPV +/+ (n = 9) and KPV −/− (n = 11) cell invasion data are plotted so that each point represents data from a single transwell assay. C KPV +/+ and KPV −/− spheroids were suspended in type I collagen and spheroid growth was tracked over 48 h. Spheroid area was quantified relative to the initial area of each spheroid (n = 4 independent experiments). Scale bar: 200 µM. D KPV +/+ cells were treated with withaferin A (WFA; 5 or 10 µM) or DMSO vehicle control for 6 h. Cells were stained for vimentin (white; dotted line represents cell outline). Scale bar: 10 µm; inset: 5 µm. E KPV +/+ cells were treated with vehicle (n = 6) or 5 µM WFA (n = 4) and were subjected to a scratch wound assay. Wound area was quantified at 6 h. F KPV +/+ cells were plated atop a Matrigel-coated transwell and were treated with vehicle control (n = 6) or 5 µM WFA (n = 9); invasion was quantified at 48 h via invasive index as described above. Data are presented as the mean ± standard deviation. The pvalues were calculated using an unpaired, two-tailed t-test, except for panel C, in which data were compared using a repeated-measure twoway ANOVA with multiple comparisons. ( * p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001).
Vimentin confers cancer cell protection from ferroptosis Cystine/glutamate transporter is an antiporter encoded by the Slc7a11 and Slc3a2 genes. This transporter brings cystine into the cell, where it is immediately converted to cysteine, the ratelimiting precursor of the antioxidant glutathione. KRAS-mutant lung adenocarcinoma cells are susceptible to SLC7A11 inhibitorinduced cell death, resulting in attenuated tumor growth in vivo [23]. KPV −/− cells express higher levels of Slc7a11 and Slc3a2 than KPV +/+ cells. We reasoned that increased activity of this transporter might lead to higher levels of glutathione in KPV −/− cells. Indeed, we identified genes involved in glutathione production, including Slc1a5, a glutamine transporter; Glul, the enzyme involved in the conversion of glutamine to glutamate; Shmt1, an enzyme required for the metabolism of serine to glycine; and Gclm and Gclc, enzymes involved in the conversion of glutamate to glutathione ( Supplementary Fig. S5A).
Despite expressing similar levels of SLC7A11 protein (Supplementary Fig. S5A), KPV −/− cells had much higher levels of glutathione and other metabolites involved in its production than KPV +/+ cells (Fig. 5A, B, Supplementary Fig. S5C). To test whether KPV −/− cells rely on glutathione for survival, we treated the cells with BSO, a well-characterized glutathione-depleting agent [40]. As expected, BSO treatment significantly reduced the glutathione levels compared to cells at baseline and abrogated the difference between KPV +/+ and KPV −/− cells (Fig. 5B). Interestingly, BSO treatment caused elevated cell death in KPV −/− cells but did not affect KPV +/+ cells, suggesting that KPV −/− cells rely more on glutathione for survival than KPV +/+ cells (Fig. 5C).
Extrinsic factors and intrinsic factors regulate ferroptosis. Typically, the cystine-glutathione axis is the main extrinsic factor regulating ferroptosis, but because of the seemingly contradictory evidence of increased ferroptosis despite higher levels of the protective antioxidant glutathione in KPV −/− cells, we explored other pathways of ferroptosis regulation. Specifically, we assessed whether the intrinsic factor, GPX4 or the extrinsic factor, iron transport, regulates ferroptosis in KPV −/− cells. Recently, WFA, which disrupts the vimentin intermediate filament network (Figs. 3  and 4), was shown to induce lipid peroxidation and ferroptosis by suppressing GPX4 activity [44]. GPX4, a key upstream regulator of ferroptosis, is responsible for both converting GSH to glutathione disulfide (GSSG) and reducing phospholipid hydroperoxides (PL-OOHs) to their corresponding alcohol (PL-OHs), thereby preventing the accumulation of lipid peroxides and leading to suppression of ferroptosis. We found that KPV −/− cells had reduced Gpx4 A Differentially produced metabolites are plotted with each row representing z-scores for each metabolite. Metabolite data were log-transformed and then subjected to an unpaired two-tailed t-test; p-values were corrected for multiple comparisons (all comparisons shown have adjusted p-value < 0.05). B Total glutathione and GSSG were quantified in KPV +/+ and KPV −/− cells treated with a vehicle control or BSO for 24 h using a luminescence-based assay (RLU = relative light units). GSH levels were calculated by subtracting GSSG from total glutathione. C Cells were treated with indicated reagents for 48 h. Cell death was measured through flow cytometry (n = 6-8). D Z-scores representing gene expression data from RNA-sequencing. All gene comparisons (KPV +/+ vs. KPV −/− ) have FDR < 0.05 after adjusting for multiple comparisons; therefore, all gene differences shown between KPV +/+ vs. KPV −/− cells are statistically significant. E BODIPY was measured by flow cytometry and normalized to the average vehicle values per experiment (n = 6-8). F GPX4 levels were measured by Western blot and normalized to an actin loading control and the KPV +/+ vehicle control. Quantification is from cells treated with 0.1 µM ML162; data were collected from three independent blots (n = 3-6). In B, C, E, and F groups were compared using a two-way ANOVA with multiple comparisons. All data are presented as the mean ± standard deviation. ( * p < 0.05; ** p < 0.01; *** p < 0.001; **** p < 0.0001).

DISCUSSION
Clinically, vimentin expression correlates with increased metastatic potential [18] (Supplementary Fig. S2C), high nuclear grade [53], and poor overall survival across most solid tumor types, including lung, prostate, and breast cancers [2][3][4]. Vimentin has also been implicated in many aspects of cancer initiation and progression, including tumorigenesis, EMT, and the metastatic spread of cancer [5]. These reports often relied on in vitro experiments comparing cancer cell lines and suppressing vimentin expression with siRNA, shRNA, and pharmacological agents or with cells derived from Vim −/− mice but lacking oncogene expression. Our data provide causal evidence that vimentin is required for the metastasis of Kras-mutant, Tp53-null lung cancer cells in vivo and in vitro. Data from the KPV −/− GEMM show that vimentin is required for metastasis and tumor progression (Figs. 1 and 6), as KPV −/− mice had decreased lung tumor burden (Fig. 1C, Supplementary Fig.  S1H), lower-grade tumors ( Supplementary Fig. S1K), and no metastasis from primary tumors in the flank to the lung ( Fig. 6A-C). Consistent with the decreased metastatic rates, we observed a survival advantage in the KPV −/− mice (Fig. 1B). These results were recapitulated in KPV +/+ mice by disrupting vimentin filaments with WFA treatment two weeks after tumor initiation (Fig. 4). Collectively, these data provide evidence that vimentin is integral in the progression and metastasis of lung cancer.
EMT is the canonical mechanism by which cancer cells lose their epithelial morphology, form invadopodia, and degrade the surrounding basement membrane to promote the invasive spread of cancer [5,11]. Vimentin expression is upregulated in epithelial cells undergoing EMT in malignant tumors, and several studies support the notion that vimentin functions as a positive regulator of EMT [5,7,[54][55][56]. In this respect, it has been proposed that vimentin intermediate filaments provide a scaffold for the recruitment of transcription factors, such as Twist1 and Slug [7]. Specifically, Twist1 upregulates Cullin2 circular RNA, which absorbs vimentin-targeting miRNAs, and thus increases vimentin expression [10]. Accordingly, we found that KPV −/− cells lack expression of Twist1 (Fig. 2). Similarly, when transforming growth factor β (TGF-β) is used to activate Smad-mediated EMT in primary alveolar epithelial cells, the cells undergo a rapid induction of vimentin expression regulated by a Smad-binding-element located in the 5′ promotor region of the Vim gene [57]. We used RNA-seq to show that KPV −/− cells derived from primary lung tumors display a distinct transcriptional phenotype, characterized by the suppression of genes directly involved in EMT (Fig. 2). Collectively, these data suggest that vimentin should no longer be considered a biomarker of EMT but rather a required protein for transcriptional activation and interaction with transcription factors such Twist and Slug, which serve to induce EMT in epithelial cells.
After a tumor cell has undergone EMT, it will invade the surrounding tissue. We previously reported that the transient expression of vimentin in epithelial cells, which typically express type I and type II keratin intermediate filaments, causes epithelial cells to be transformed into mesenchymal cells, which is accompanied by changes in cell shape, increased cell motility, and focal adhesion dynamics [9,57]. We use four different methods to show that vimentin is required for cancer cell invasion across the basement membrane and through the interstitial matrix. Direct evidence supporting the role of vimentin in the migration of Kras-mutant, Tp53-null lung cancer cells was demonstrated by disrupting vimentin expression genetically (Fig. 3A) and pharmacologically in mouse- (Fig. 3E) and humanderived ( Supplementary Fig. S3C) cancer cells resulting in impaired migration. Furthermore, loss of vimentin through genetic manipulation (Fig. 3B) and pharmacological interference with WFA ( Fig. 3F) confers an inability of cells to invade a layer of the basement membrane. KPV −/− cells also fail to invade through a collagen-rich matrix in a three-dimensional experimental model (Fig. 3C). Importantly, we provide evidence that vimentin is respectively) were injected subcutaneously into the right flank of nude mice. At 3 weeks post-injection, primary tumors were removed and lung metastases were tracked for an additional 1 week. Shown are representative IVIS images of mice (n = 9-11 per group). The coronal views shown were acquired after masking the flank tumor to minimize bleed-through of the signal. Intensity overlay shows the accumulation of luciferaselabeled cells. B Luciferin signal was quantified from the lungs. An ordinary one-way ANOVA with multiple comparisons was used to compare groups at week 4 ( * p < 0.05). C Lungs at week 4 and D excised flank tumors from week 3 were fixed, sectioned, and subjected to H&E staining and vimentin immunohistochemical staining. Positive vimentin staining is brown, and nuclei are blue. Scale bars: 1 mm (whole tumor/lung, left), 100 μm (inset, right). required for in vivo metastasis. In the KPV +/+ mice, we observed metastases in the liver at 12 weeks after tumor induction; KPV −/− mice had no evidence of metastatic lesions (Supplementary Fig.  S1D). To confirm this observation, we implanted cells in the flanks of mice. KPV +/+ cells, but not KPV −/− cells nor KPV Y117L cells, were able to invade and migrate away from the primary tumor and seed metastatic lesions in the lung (Fig. 6). Together, these results provide direct evidence that an intact vimentin intermediate filament network is required for the cell motility that leads to metastasis.
To our knowledge, this is the first report to show that vimentin protects cancer cells from ferroptosis (Fig. 5). Specifically, we show that KPV −/− cells are more susceptible to BSO-induced death than KPV +/+ cells, and that this effect can be rescued with Fer-1. Importantly, we show that several convergent mechanisms of ferroptosis are activated in KPV −/− , but not KPV +/+ , cells ( Supplementary Fig. S5F). These include loss of the intrinsic mechanism of GPX4-mediated protection from lipid peroxidation and elevation of the extrinsic mechanism of iron transport. Interestingly, previous studies support our findings that suggest vimentin may interact with GPX4, be involved in iron metabolism, and play a role in lipid droplet (LD) maintenance. A report from Hassannia et al. recently showed that WFA treatment directly targets vimentin also binds and inactivates GPX4, thus inducing ferroptosis [44]. In line with this finding, we show that GPX4 levels are lower in KPV −/− cells compared to KPV +/+ cells (Fig. 5D, F). Additionally, suppressors of the redox pathway upstream of GPX4 (glutathione, SLC7A11) are increased or unchanged in KPV −/− cells (Fig. 5A, B, D, Supplementary Fig. S5A-C). In contrast, downstream readouts such as lipid peroxidation and cell death are decreased (Fig. 5C-E), suggesting that there may be a direct relationship between GPX4 and vimentin. Several iron-related genes, including the transferrin receptor (Trfc), Ltf, Slc39a14, Slc25a37, Ncoa4, and Bach1, are upregulated in KPV −/− cells (Fig. 5D). In contrast, genes associated with negative regulation of the cytosolic iron pool, such as Ftl, Fth1, and Cisd1, are downregulated in KPV −/− cells. However, there is no difference in iron levels between KPV +/+ and KPV −/− cells (Supplementary Fig. S5E) and no difference between groups after treatment with the iron chelator DFO alone or with ML162 ( Supplementary Fig. S5F). Interestingly, vimentin knockdown regulatory T cells (Tregs) express higher transferrin receptor levels and take up more fatty acids than wildtype Tregs [58]. However, the precise role of vimentin in iron metabolism remains to be explored. Degradation of LDs promotes ferroptosis [59]. Interestingly, vimentin forms a cage around LDs and physically anchors LDs via the linker protein perilipin [60]. Vimentin-null cells have smaller LDs, impaired lipid motility, and impaired lipolysis [61,62]. In support of the hypothesis that differences in lipolysis might lead to susceptibility to ferroptosis in KPV −/− cells, we observed that Gch1, which is the rate-limiting enzyme of BH 4 -mediated lipid remodeling, and Tpd52l1, which promotes lipid storage, are upregulated in KPV +/+ cells compared to KPV −/− cells [59,63] (Fig. 5D). These data suggest that the interaction between vimentin and lipid droplets may mediate ferroptosis. Of note, many of the studies mentioned in this section were conducted in non-cancer cell types, and the precise role of vimentin in regulating ferroptosis in KRAS-mutant cancer remains to be identified.
Global vimentin depletion confers a survival advantage in the Kras-mutant, Tp53-null lung cancer model. Additionally, by treating KPV +/+ mice with WFA following tumor development, we show that the delayed tumor growth and metastasis observed when vimentin is suppressed is not due solely to delayed onset of tumor growth, but rather to attenuated growth kinetics in tumor cells that lack an intact vimentin network. Although vimentin-null mice were first reported to display no obvious phenotype, these data along with previous reports suggest that loss of vimentin is protective against a range of disease states including lung cancer, acute lung injury, acute respiratory distress syndrome, idiopathic pulmonary fibrosis, bacterial meningitis, cerebral ischemia, and acute colitis [26,[64][65][66][67][68]. These global vimentin knockout mice lack vimentin expression in tumor cells and tumor microenvironment (TME). The TME, which includes immune and stromal cells, can promote metastasis of primary tumor cells. The cytokine interleukin 1β (IL-1β), produced by macrophages in the lung, is a mediator of cancer growth, metastasis, and tumor-associated macrophage infiltration [65,69]. Our group has reported that vimentin is required for mature IL-1β release [65]. Because of the potential contribution of the TME in vimentin-null mice, we set out to ensure that vimentin is sufficient in cancer cells to promote metastasis. We injected KPV +/+ or KPV −/− cells into wildtype nude mice; these mice lack T cells but retain innate immune cells. We observed recruited mesenchymal and immune cells in subcutaneous flank tumors (Fig. 6D), suggesting that the vimentin-positive TME in this model participates in growth of the primary tumor. Despite being in the presence of vimentin-expressing stromal and immune cells, KPV −/− cells failed to metastasize. Therefore, while other groups have found that vimentin deficiency impairs function in cancer-associated fibroblasts and immune cells such as macrophages and T cells [56,58,65,70], a vimentin-expressing microenvironment is not sufficient to promote metastasis in the time frame evaluated. To better understand how vimentin participates in different compartments of the TME, we recognize that animal models with immune-, mesenchymal-, and epithelialspecific deletion of vimentin will need to be created.
This study fills a major gap in the literature by providing novel causal data that vimentin is required for the in vivo progression of NSCLC at several steps of the metastatic cascade. Importantly, our in vivo data validated our in vitro data and in vitro data from previously published reports [11,12,[71][72][73][74]. Additionally, this work gives causal context to clinical data associating vimentin expression with tumor progression in patients [2,18,75]. By using in vivo models and disrupting vimentin both genetically and pharmacologically, we have presented physiological context to these reports. Through this study, we have identified vimentin as a new clinical target for metastatic lung cancer.

Magnetic resonance imaging
Scheduled magnetic resonance imaging (MRI) was performed at Northwestern University Center for Translational Imaging (Chicago, IL) via a 7-tesla system (Clinscan, Bruker) using a four-channel mouse body coil at set time points (2, 6, and 10 weeks after Ad-Cre administration). Mice were anesthetized with isoflurane (2% isoflurane in oxygen for induction, followed by 1.5-2% via nose cone for maintenance during imaging) to permit tolerance to imaging. Pulse oximetry and respiration were recorded and used to trigger the MRI to avoid motion artifacts. Turbo Multi Spin Echo imaging sequence was used in conjunction with respiratory triggering to acquire in vivo MRI coronal images covering all the lung area and portions of abdomen, including liver and kidneys (ST = 0.5 mm, In plane = 120 μm, TR = 1000 msec, TE = 20 msec). Gradient Echo sequence was used with cardiac triggering (pulse oximeter rate) covering the lung area transversally (ST = 0.5 mm, In plane = 120 μm, TR~20 msec, TE~2 msec). Jim software was used to quantify tumor burden (Xinapse).

Immunohistochemistry
Mice were anesthetized and lungs were perfused via the right ventricle with 4% paraformaldehyde in phosphate-buffered saline (PBS). A 20-gauge angiocatheter was sutured into the trachea, heart and lungs were removed en bloc, and then lungs were inflated with 0.8 mL of 4% paraformaldehyde at a pressure not exceeding 16 cm H 2 O. Tissue was fixed in 4% paraformaldehyde in PBS overnight at 4°C, then processed, embedded in paraffin, and sectioned (4-5 μm). Sections (5 μm) were cut from three distinct zones (top, middle, bottom) of each paraffin block and were stained with hematoxylin and eosin (H&E) or used for immunohistochemistry. H&E-stained samples were assessed for tumor type (hyperplasia, adenoma, adenocarcinoma), numbers, and areas using AxioVision digital image processing software per manufacturer's protocol. Briefly, the desired quantification region was imaged, total region area measured (μm 2 ), pixels matching tumor cells selected, and a binary mask created to compute tumor numbers and area. The measurements were independently confirmed by manual grading of tumor lesions. Tumor lesions were graded by a pathologist as described previously [76].
For immunohistochemistry, tissue was rehydrated and subjected to antigen retrieval in 10 mM sodium citrate (pH = 6.0) with 0.05% Tween-20 for 20 min at 96-98°C, followed by 20 min of cooling. Tissue sections were blocked in 3% hydrogen peroxide for 5 min. A Vector Laboratories avidin/ biotin blocking kit (SP-2001), Vectastain ABC kit (PK-4001), and 3,3′diaminobenzidine (DAB) peroxidase substrate kit (SK-4100) were used according to the manufacturer's protocols. Nuclei were counterstained with hematoxylin (Thermo Scientific 72604) and treated with bluing solution (Thermo Scientific 7301), and then coverslips were mounted with Cytoseal 60 (Thermo Scientific 8310-4). A TissueGnostics automated slide imaging system was used to acquire whole-tissue images and measure area. The semi-quantitative immunohistochemistry (IHC) method was performed using ImageJ Fiji software to conduct deconvolution and downstream analysis as described [77]. Human tissue was purchased from Biomax (LC2083) and stained as described above.
Cell isolation and culture KPV +/+ and KPV −/− mice were treated with Ad-Cre as described above; after 6 weeks, mice were sacrificed and lung tumors were excised. The tissue was dissociated into a single-cell suspension in 0.2 mg/mL DNase and 2 mg/mL collagenase D and filtered through a 40 μm filter. Cells then underwent two rounds of selection. First, cells were treated with anti-CD45 magnetic beads (Miltenyi Biotec, 130-052-301) and were passed through a magnetic column. CD45-negative cells were then subjected to anti-EpCAM magnetic beads (Miltenyi Biotec, 130-105-958) and underwent positive selection. CD45negative, EpCAM-positive cells were expanded in vitro and were used in experiments between passages 1 and 10. Cells derived from a human lung adenocarcinoma (A549) were obtained from the American Type Culture Collection (ATCC, Manassas, VA). All cells were maintained in Dulbecco's modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum, 100 U/mL penicillin, 100 µg/mL streptomycin, and HEPES buffer. All cells were grown in a humidified incubator of 5% CO 2 /95% air at 37°C.

Polymerase chain reaction
Mice were infected with Ad-Null or Ad-Cre; mice were sacrificed at 2, 8, and 12 w.p.i. and lungs were harvested. Lungs were lysed, and DNA was extracted and amplified by polymerase chain reaction (PCR) using the following primers: Kras forward, GGC CTG CTG AAA ATG ACT GAG TAT A; Kras reverse, CTG TAT CGT CAA GGC GCT CTT; Kras-G12D forward, CTTGTGGTGGTTGGAGCTGA; and Kras-G12D reverse, TCCAAGAGACAGGT TTCTCCA. DNA products were run on an agarose gel and imaged with the Li-Cor Odyssey imaging system.

Western blotting
Western blot analysis was utilized to quantify protein levels in cell lysates. The protein was separated using 12% sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) and transferred onto nitrocellulose membranes. Membranes were then blocked with Odyssey blocking buffer (Li-Cor Biosciences) and incubated with the appropriate primary antibodies overnight at 4°C. IRDye secondary antibodies were then used (Li-Cor Biosciences, 1:10,000) for 2 h at room temperature. Images of blots were acquired using the Li-Cor Odyssey Fc Imaging System.

RNA-sequencing
Tumor cells were isolated from KPV +/+ and KPV −/− mice at 6 w.p.i. and underwent CD45-negative, EpCAM-positive magnetic-activated cell sorting (MACS) selection as described above. Cells were cultured for one passage, then lysed using RLT lysis buffer (Qiagen), and total RNA was isolated with the RNeasy Plus Mini Kit (Qiagen). RNA quality was confirmed with a TapeStation 4200 (Agilent); all samples had an RNA integrity number (RIN) score equal to or greater than 9.8. Next, mRNA was isolated via poly(A) enrichment (NEBNext). Libraries were prepared using NEBNext RNA Ultra chemistry (New England Biolabs). Sequencing was performed on an Illumina NextSeq 500 using a 75-cycle single-end high-output sequencing kit. Reads were demultiplexed (bcl2fastq), and fastq files were aligned to the mm10 mouse reference genome with TopHat2. Htseq was used to obtain counts. The resulting data were filtered, and differentially expressed genes (DEGs) were identified using the edgeR package. DEGs were selected using a false discovery rate (FDR) cutoff of <0.05, with a 1.0-fold change cutoff for pairwise comparison. K-means clustering and heat map visualization was performed using the Morpheus web tool (https:// software.broadinstitute.org/morpheus). Enrichment analysis was performed using Gorilla [78,79]. The RNA-seq data have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE225083.

Withaferin A treatments
Withaferin A (WFA) was purchased from Enzo Life Sciences and dissolved in dimethyl sulfoxide (DMSO; Sigma-Aldrich) to a final concentration of 5 μM unless noted otherwise. For in vivo experiments, jelly pellets were utilized to provide an oral, voluntary method of drug delivery. Using a 24well flat-bottom tissue culture plate as the jelly mold, WFA (4 mg/kg in DMSO) or vehicle control (DMSO only) were combined with gelatin and Splenda for flavoring as described elsewhere [80]. Tumor development was induced with Ad-Cre as described above. Two weeks following Ad-Cre administration, mice were fed jelly pellets every other day for 4 weeks. Survival was tracked daily and weight was measured weekly.

Scratch wound assay
Cells were grown to 100% confluence in 6-well plates. A pipette tip was used to make a single scratch in the monolayer. The cells were washed with 1× PBS to remove debris and imaged at 0 and 6 h. For WFA conditions, WFA or DMSO was added at 0 h (when the scratch was created). Rate of cell migration was calculated using ImageJ software. Results were normalized to the initial wound area at 0 h.

Matrigel invasion assay
Transwell inserts with 8 μm pores were coated with Matrigel (200 μg/mL), and 5 × 10 4 KPV +/+ or KPV −/− cells were seeded atop each transwell in serum-free media. For WFA experiments, cells were resuspended in WFA or DMSO containing media directly before being seeded in transwells. Media containing 10% fetal bovine serum was added to the bottom well to serve as a chemoattractant. Cells were placed at 37°C for 48 h. Following incubation, the Matrigel with the cells remaining on the upper surface of the transwell was removed with a cotton swab. The cells remaining on the bottom of the membrane were fixed in 2% paraformaldehyde and incubated with Hoechst nuclear dye (Invitrogen; 1:10,000 in 1× PBS). Five random 10× magnification fields were imaged, and the average number of cells per field was quantified; this average is reported as the "invasive index."

Spheroid culture
Spheroids were generated as described by Gilbert-Ross et al. [81]. Briefly, cells were grown in Nunclon Sphera 96-well plates (Thermo-Fisher Scientific) at a concentration of 3000 cells per well. After 3 days in culture, cells were transferred using a wide-bore pipette tip to 2 mg/mL collagen (Corning) in 4-well LabTek plates (Nunc). Collagen was allowed to gel at 37°C for 1 h; then, complete media was added to the spheroids. Gels were imaged using a Ti2 widefield microscope (Nikon) at 0, 24, and 48 h. Spheroid area was quantified using Fiji software. Reported spheroid area values are normalized to 0-h spheroid area of the same spheroid.

Metabolomics
KPV +/+ and KPV −/− cells were grown in 6-well plates. High-performance liquid chromatography (HPLC) grade methanol (80% in water) was added to cells, and plates were incubated at −80°C for 20 min. Lysates were collected and centrifuged, and the supernatant was collected and analyzed by high-performance liquid chromatography and high-resolution mass spectrometry and tandem mass spectrometry (HPLC-MS/MS). Specifically, the system consisted of a Thermo Q-Exactive in line with an electrospray source and an Ultimate3000 (Thermo) series HPLC consisting of a binary pump, degasser, and auto-sampler outfitted with a Xbridge Amide column (Waters; dimensions of 4.6 mm × 100 mm and a 3.5 µm particle size). The mobile phase A contained 95% (vol/vol) water, 5% (vol/vol) acetonitrile, 20 mM ammonium hydroxide, 20 mM ammonium acetate, pH = 9.0; B was 100% acetonitrile. The gradient was as follows: 0 min, 15% A; 2.5 min, 30% A; 7 min, 43% A; 16 min, 62% A; 16.1-18 min, 75% A; 18-25 min, 15% A with a flow rate of 400 μL/min. The capillary of the ESI source was set to 275°C, with sheath gas at 45 arbitrary units, auxiliary gas at 5 arbitrary units, and the spray voltage at 4.0 kV. In positive/negative polarity switching mode, an m/z scan range from 70 to 850 was chosen and MS1 data were collected at a resolution of 70,000. The automatic gain control (AGC) target was set at 1 × 10 6 and the maximum injection time was 200 ms. The top 5 precursor ions were subsequently fragmented, in a data-dependent manner, using the higher energy collisional dissociation (HCD) cell set to 30% normalized collision energy in MS2 at a resolution power of 17,500. Besides matching m/z, metabolites were identified by matching retention time with analytical standards and/or MS2 fragmentation pattern. Data acquisition and analysis were performed using Xcalibur 4.1 software and Tracefinder 4.1 software, respectively (both from Thermo Fisher Scientific). For each sample, peak area of each metabolite was normalized to total ion count per sample. Data were log-transformed and compared with a two-tailed, unpaired t-test. Data were analyzed with MetaboAnalyst software [82].

Tracking of tumor growth in subcutaneous flank injection model
Male nude (NU/J) mice were purchased from Jackson Laboratories; 8-12week-old mice were anesthetized with 2% isoflurane in oxygen and were given a subcutaneous injection of cells (1 × 10 6 cells in 100 μL of 1× PBS) on their right flanks. Weight and tumor volume were monitored weekly. For IVIS imaging, mice were injected with 150 mg of D-luciferin per kilogram of body weight (PerkinElmer 770504). After 10 min, IVIS images were captured. At week 3 post-injection, tumors were removed. Briefly, mice were anesthetized with ketamine (100 mg/kg body weight) and xylazine (10 mg/kg body weight). Tumor area was disinfected with 70% ethanol and iodide solution. Tumors were excised and placed in 4% paraformaldehyde for immunohistochemistry. Wounds were closed with simple interrupted nylon sutures (Ethilon). Mice were monitored until they recovered from anesthesia; they were housed singly and treated with Meloxicam as an analgesic. The following week, mice underwent a final IVIS imaging session and were sacrificed.

Immunofluorescence confocal microscopy
Cells were grown on no. 1 glass coverslips for all immunofluorescent immunocytochemistry experiments. Following treatment, KPV +/+ and KPV −/− cells were fixed in methanol for 3-5 min. A549 cells were fixed with 2% paraformaldehyde for 7-10 min. KPV +/+ and KPV −/− cells were blocked in 5% normal goat serum (NGS) for 1 h at room temperature. A549 cells were blocked with 1.5% NGS for 30 min at 37°C. Cells were then treated with the indicated primary antibodies overnight at 4°C. Cells were washed twice in PBS with 0.10% Tween-20 for 3 min each and treated with secondary antibodies conjugated with Alexa Fluor 488 (Invitrogen A-11039, 1:200) and/or Alexa Fluor 568 (Invitrogen A-11004, 1:200), as well as Hoechst nuclear dye (Invitrogen H3570, 1:10,000). Coverslips were mounted and sealed. A Nikon A1R+ laser scanning confocal microscope equipped with ×60 and ×100 objective lenses was used to acquire images. For experiments with A549 cells, a Zeiss LSM 510 laser scanning confocal microscope equipped with a ×63 objective lens was used to acquire images. Nikon NIS-Elements software and ImageJ were used for image processing.

Flow cytometry
Lipid peroxidation was assessed by flow cytometry with BODIPY™ 581/591 C11 (Thermo Fisher Scientific, United States) following the manufacturer's instructions. In brief, cells were pretreated with 1 µM BSO or 20 µM Fer-1. Cells were collected at 48 h for lipid peroxidation and to measure cell viability. Cells were resuspended in 400 μl of serum-free medium that contained BODIPY™ 581/591 C11 (5 µM) at 37°C in the dark for 30 min, to detect lipid peroxidation. Finally, cells were stained with Ghost Dye Red 780 (Tonbo Biosciences) and analyzed by flow cytometry immediately. FeRhoNox (Gorya Chemical) was used following manufacturer's instructions. Briefly, cells were plated overnight and incubated with FeRhoNox in HBSS for 1 h then immediately analyzed by flow cytometry. Data were collected using a BD FACSymphony cell analyzer, and a minimum of 10,000 cells were analyzed per condition. Cells were gated to exclude debris and doublets using FlowJo software and statistical analysis was conducted as described below.

Statistical analysis
Statistical analyses were performed in GraphPad Prism v.9 software using statistical tests indicated in the figure legends. Data are presented as mean ± s.d. with a minimum of n = 3 independent experiments unless otherwise indicated. Specific numbers of replicates are indicated in scatter plots. Experiments were neither randomized nor blinded, excluding histological analysis. Differences between two groups were assessed by using a student's t-test. Differences between three or more groups were assessed using one-way analysis of variance (ANOVA) with a Bonferroni multiple comparisons test. Values of p < 0.05 were significant. The log rank test, which calculates the chi-square (χ 2 ) for each event time for each group and sums the results, was used in the Kaplan-Meier curve analysis. Minimum sample group sizes were determined through power analysis. Specific tests are indicated in figure legends. Plated cells were allocated randomly to each treatment group; mice were assigned randomly to each treatment group.