Vimentin Is Required for Lung Adenocarcinoma Metastasis via Heterotypic Tumor Cell–Cancer-Associated Fibroblast Interactions during Collective Invasion

This article is featured in Highlights of This Issue, p. 251

Epithelial-to-mesenchymal transition (EMT) has served as a guiding principle underlying the initiation and progression of metastatic disease (1). During EMT, epithelial cells from the primary site undergo genetic (2) and epigenetic (3) alterations that trigger cancer cell invasion. A hallmark of EMT is the loss of epithelial proteins, such as E-cadherin (4), and gain of mesenchymal proteins, such as vimentin (5). These events can be transient as cells transition through partial and reversible EMT states (6–9), and it is debated whether a full EMT is necessary for metastasis (10, 11).

Over half of all lung adenocarcinoma cases present with advanced metastatic disease, which is associated with a 5-year survival rate of 4% (12). As the driving genomic aberrations of lung adenocarcinoma have become clearer, comutations in KRAS and LKB1 are recognized as major drivers of the disease (13, 14). The KRAS/LKB1 comutation has previously been modeled in a Cre-dependent LSL-Kras^G12D, Lkb1^fl/fl genetically engineered mouse model (GEMM; ref. 15) to investigate this comutation in the context of metabolism (16), therapeutic response (17, 18), and microenvironment remodeling (19). In this GEMM, primary lung tumors develop at 8 to 10 weeks with 52% of mice having metastatic disease to draining lymph nodes (15, 18). When Cre is delivered intranasally via lentiviral infection, 100% of primary tumors are lung adenocarcinomas that reproduce the stages of tumor progression from adenocarcinoma in situ to invasive disease (18).

To assess EMT in lung adenocarcinoma, and more specifically the role of the canonical mesenchymal marker vimentin, we generated an LSL-Kras^G12D/Lkb1^fl/fl/Vim^−/− (KLV^−/−) metastatic lung adenocarcinoma GEMM model, which contains a whole-body vimentin knockout (20). Vimentin is an intermediate filament protein, which is associated with increased metastatic potential (21), high nuclear grade (22), and poor overall survival (23) across most solid tumor types including lung, prostate, and breast cancers (24–26). Because vimentin is integral for the structural integrity of the cell (27), lamellipodia formation (28), and adhesion signaling (29, 30), downregulating vimentin expression is sufficient to alter cell morphology in vitro (31) as well as inhibit cell motility and invasion (32).

The data presented here from GEMMs, human lung adenocarcinoma samples, and in vitro 3D models show that vimentin is required for lung adenocarcinoma to metastasize at an early stage by acting in cancer-associated fibroblasts (CAF) that surround epithelial-like collective invasion packs (CIPs). These data support the concept that vimentin is critical for lung adenocarcinoma metastasis and could be a potential target for antimetastatic therapies.

KLV^−/− mouse was generated by crossing the LSL-Kras^G12D/LKB1^fl/fl mouse (15) with the Vimentin^−/− (20) mouse until a homozygous KLV^−/− mouse was generated. KLV^−/− mouse took approximately four generations over 7 months to create. KLV^+/− mice were generated by crossing KLV^+/+ mice with KLV^−/− mice. All mice were housed and treated according to Institutional Animal Care and Use Committee protocol #2003525 by the Emory University Division of Animal Resources. Kras- and Lkb1-mutant mice were purchased from The Jackson Laboratory. Vimentin-null mice were obtained from the EMMA repository (20). Lentiviral Cre (1.2 × 10⁷ i.u.) was administered intranasally to mice 8 to 12 weeks old that were sedated with Avertin (20 mg/mL working dilution). Infected mice were then monitored until tumorigenic symptoms (i.e., weight loss greater than 10%, respiratory distress, inactivity) presented. All mice infected with lentiviral Cre were monitored to a 25-week postinfection endpoint and were sacrificed prior if they presented with clinical symptoms including weight loss, respiratory distress, and/or hunching. Upon sacrifice, lungs and mediastinal lymph nodes were harvested in either 10% formalin or flash frozen in OCT pending use.

Deidentified human lung adenocarcinoma tissue is a mixed cohort of samples from Emory and Wellstar tumor banks and Emory clinical formalin-fixed paraffin-embedded (FFPE) tumor specimens (IRB00009857). All Emory samples were sequenced using a workflow described (18). Lung adenocarcinoma staging was performed by a board-certified thoracic pathologist according to the AJCC Cancer Staging Manual 7^th Edition. Vimentin immunohistochemistry (IHC) was scored by a board-certified thoracic pathologist with a 40X objective on a scale from 0 to 3 based on stain intensity and was stratified by stromal or tumor compartment. Q score was determined by the multiplication of stain intensity by percentage of cells staining at that intensity for a given cell compartment (i.e., 30% of stromal cells at level 3 stain has Q score of 90).

FFPE tissues were hydrated in a series of washes from xylenes to ethanol dilutions to water. Heat-mediated antigen retrieval was conducted with citrate buffer when necessary. Endogenous peroxidase activity was blocked with 3% hydrogen peroxide. Samples were blocked with 2.5% normal goat serum (Vector Labs) and incubated in primary antibody overnight in 1:1 NGS:PBS solution (Vimentin R28 Cell Signaling Technology 1:300, alpha SMA Invitrogen 1:1,000, Pro-SPC Santa Cruz FL-197 1:150, CD31 ab28364 1:50 CD3 ab16669 1:100). Samples were incubated in secondary antibody (Rabbit 1:200), biotin, and streptavidin (Vector Labs) according to kit protocol. Staining was developed using Peroxidase Substrate Kit DAB (Vector Labs). Proliferation was measured by staining for Ki67 (ab16667, 1:200) using IMMpact Kit (Vector Labs).

Tissue sections frozen in OCT Compound were sliced in 10 μm sections and fixed in 100% acetone. Sections were stored at −80°C until used. Sections were blocked for 1 hour in 10% normal goat serum in PBS. Primary antibodies were incubated in PBS overnight for the following antibodies at indicated concentrations: vimentin R28 Cell Signaling Technology 1:50, F4/80 Abcam 1:100, FSP1 ab27957 1:200. FFPE tissues were hydrated as described above. Heat-mediated antigen retrieval was also conducted as above. A blocking buffer of 5% BSA in PBS was used. Primary antibodies were incubated at 4°C overnight in 1% BSA in PBS (FSP1 ab27957 1:200; E-cadherin BD Biosciences 610181 1:1,000). Secondary antibodies were incubated on tissues for 1 hour (1:1,000).

CAFs were fixed and stained as described (30). Primary antibodies were diluted in 5% normal goat serum (Vimentin R28 Cell Signaling Technology 1:50) and added to cells on coverslips overnight at 4°C. The coverslips were mounted on glass slides using Prolong diamond mounting medium (Invitrogen). The slides were imaged using Leica SP8 confocal microscope.

Spheroids embedded in matrigel were fixed with 4% paraformaldehyde for 20 minutes. Glycine rinse (TBS/glycine 130 mmol/L NaCl, 7 mmol/L Na₂HPO₄, 3.5 mmol/L NaH2PO4, 100 mmol/L glycine) performed three times for 10 minutes at room temperature. The immunofluorescence (IF) buffer used for protocol consisted of 130 mmol/L NaCl, 7 mmol/L Na₂HPO₄, 3.5 mmol/L NaH₂PO₄, 0.5% Triton X-100, and 0.5%, Tween-20. Samples were blocked in IF buffer supplemented with 3% BSA, 5% goat serum, and 0.3% Triton X-100 for 2 hours at room temperature. Primary antibodies incubated overnight in blocking buffer (FSP1 1:250) at room temperature. IF buffer was used for three 20-minute washes. Secondary antibody was diluted 1:500 in blocking buffer (Phalloidin 488 1:40 and DAPI added) and incubated at room temperature overnight. Fluorescence was imaged on Confocal Leica SP8 microscope.

Second harmonic generation (SHG) images of hematoxylin and eosin (H&E) sections were taken using a Zeiss Axio Examiner Z1 microscope with 20X water immersion objective (1.0 NA DIC [UV] VIS-IR) as described (33). The SHG signal was obtained using a band pass cube of 380 to 430 nm. Images were taken with a Coherent Chameleon Verdi laser at 820 nm wavelength. Z-stack images were taken with a 1-μm interval.

Lung tissue was harvested and flash frozen in liquid nitrogen. Frozen tissue was cut into 2-mm pieces and ground in glass mortar and pestle in 1 mL of lysis buffer TNES buffer (50 mmol/L Tris, pH 7.5, 100 mmol/L NaCl, 2 mmol/L EDTA, 1% Nonidet P-40, 1X Roche Complete Protease Inhibitors, 10 mmol/L NaF, 1 mmol/L NaVO4, 2 mmol/L sodium pyrophosphate, and 2 mmol/L β-glycerophosphate). Mixture was further homogenized with a sonicator. Lysate was centrifuged, and supernatant was collected. Protein assay and Western blot were performed as described (30). Primary antibody, diluted in either 5% BSA in TBST (Vimentin R28 Cell Signaling Technology 1:1,000) or 5% milk in TBST (Tubulin Millipore MAB1864 1:20,000, GAPDH CST 1:30,000), was added to membrane overnight. Secondary antibodies were incubated at a concentration of 1:1,000 in 5% milk in TBST for 1 hour.

H460 lung cancer cell line was purchased from the ATCC and subcultured in RPMI 1640 media supplemented with 10% FBS, 1% penicillin/streptomycin, and 1% kanamycin. Human lung adenocarcinoma CAFs were purchased from Vitro Biopharma and subcultured in MSC Gro VitroPlus III, Low Serum media supplemented with 1% penicillin/streptomycin and 1% kanamycin. All cells were cultured in a humidified chamber at 37°C with 5% CO₂. Experiments were performed with cells below passage 10. All cells were verified as mycoplasma-free on October 4, 2017, by DAPI staining.

For the production of stable vimentin knockdown CAFs, pLKO.1 vector and vimentin shRNA virus described previously (30) were added to cells 1:2 in media with 8 μg/mL polybrene three times. Cells expressing the constructs were selected using puromycin 2 μg/mL treatment.

Spheroids of H460s, CAFs, and H460/CAF 50/50 cocultures were generated as described (18). Spheroids were collected and resuspended in 250 μL of 2.0 mg/mL MatriGel (VWR) in a 35-mm Dish with a 14-mm bottom well (Cellvis), and a coverslip was attached with Vaseline. After 45-minute incubation at 37°C, 2 mL of media was added. Images were taken at 0 and 24 hours on Olympus IX51 microscope. Spheroid invasion was quantified using ImageJ (NIH). For 3D IF, spheroids were formed as described above. Each well of a 96-well flat-bottomed plate (Corning) was coated with 8 mg/mL MatriGel. Spheroids were collected and resuspended in 50 μL of 2.0 mg/mL MatriGel. After a 45-minute incubation at 37°C, 100 μL of media was added. Spheroids were allowed to invade for 48 hours.

Tissue was collected from candidate pups prior to weaning. Genomic DNA was extracted using Alkaline Lysis Buffer at 100°C for 45 minutes. A 40 mmol/L Tris-HCL Neutralization Buffer was then added, and samples were stored at −20°C. PCR was performed using a GoTaq Flexi system from Promega. Samples were run on Biorad iCycler at recommended annealing temperatures.

H&E and IHC-stained tissue samples were imaged on the Zeiss Axioplan 2 widefield microscope, or whole slides were scanned into ImageScope software. IF images were taken on the Confocal Leica SP8. All images were analyzed with ImageJ and Photoshop Elements. CellProfiler was used for fluorescent image quantification. Imaris software was used for vim/FSP1 colocalization analysis where a mask was generated for each marker and then a colocalization mask was generated from this. The 2D histogram was generated in Imaris by removing background levels of each marker. Pearson's coefficient was generated from the histogram. Imaris was also used for CAF cell size analysis.

Stromal and tumor compartments in KLV mouse were identified by a board-certified thoracic pathologist. KLV model tumor grading was performed as described (34).

P values of significance were obtained using a Fischer exact test, χ² test, or unpaired t test when comparing two groups (i.e., KLV^+/+ vs. KLV^−/− or metastatic vs. nonmetastatic). Data with more than two groups were analyzed by one-way ANOVA followed by Tukey's multiple comparisons test. Analysis of vimentin IHC in patient samples was performed using the Wilcoxon Rank Sum test. Kaplan–Meier curves were analyzed by Mantel–Cox and Gehan–Breslow–Wilcoxon tests.

Metastasis is well characterized in the LSL-Kras^G12D/LKB1^fl/fl (KLV^+/+) model (15, 18), where approximately 52% of mice that develop primary lung tumors at 8 to 10 weeks also develop metastasis to mediastinal lymph nodes at 12 to 16 weeks (18). A whole-body vimentin^−/− mouse (20) was crossed to this KLV^+/+ mouse to test the hypothesis that vimentin is necessary for metastatic disease. Through this cross, a novel LSL-Kras^G12D,LKB1^fl/fl,Vim^−/− (KLV^−/−) mouse was generated (Supplementary Fig. S1A) that lacks vimentin expression throughout and has the LSL-Kras^G12D,LKB1^fl/fl conditionally mutated in the lungs via intranasal delivery of lentiviral Cre recombinase. Disruption of the vimentin allele and lack of protein expression in the KLV^−/− mouse were validated by RT-PCR, Western blotting, and IF (Supplementary Fig. S1B–S1D). Disruption of one vimentin allele in the heterozygous KLV^+/− mouse did not greatly reduce vimentin protein expression (Supplementary Fig. S1C and S1D). These results demonstrate the development of a novel KLV^−/− GEMM to probe the role of vimentin in lung adenocarcinoma metastasis.

KLV^+/+ and KLV^−/− were infected intranasally with Cre lentivirus, and a comparison of tumor burden between the two genotypes (either at 25 weeks or before due to morbidity) revealed no significant differences in primary tumor initiation or growth (Table 1 and Fig. 1A). In both genotypes, 100% of primary tumors were lung adenocarcinomas with no significant difference in tumor incidence (Table 1). Both KLV^+/+ and KLV^−/− mice generated primary lung tumors (Fig. 1A) without a significant difference in lung weight (Fig. 1B) or tumor multiplicity (number of tumor foci per H&E slice; Fig. 1C). Tumor histology was similar across genotypes with both KLV^+/+ and KLV^−/− mice developing tumors with and without invasive fronts (Fig. 1D). Proliferative index using Ki-67 staining was also not significantly different between the two genotypes (Fig. 1E and F). Based on these data, we conclude that vimentin loss does not affect primary lung tumor initiation or growth. A Kaplan–Meier curve to clinical symptoms shows that KLV^−/− mice have decreased survival compared with KLV^+/+ (Supplementary Fig. S2).

The KLV^+/+ model develops robust primary lung tumors and metastasis to the mediastinal lymph node (15, 18). We leveraged this phenotype to investigate how the vimentin loss affects the metastatic potential of Kras/Lkb1 comutated lung adenocarcinomas. Vimentin loss in KLV^−/− mice results in 54% less mediastinal lymph nodes metastasis compared with KLV^+/+mice (Fig. 2A and B). In mice with clinical symptoms (respiratory distress, weight loss, hunching), vimentin loss was associated with significantly less metastatic disease to the draining lymph nodes compared with KLV^+/+ mice (Fig. 2C). In addition to decreased incidence of lymph node metastasis, vimentin loss in the KLV^−/− is associated with significantly less invasive foci than KLV^+/+ mice tumors as measured by the number of invasive fields (defined as a field with focal invasion) per mouse (Fig. 2D).

To assess the impact of vimentin loss on tumor progression, primary tumors of KLV^+/+ and KLV^−/− were graded on a scale of 1 to 4 with grade 1 tumors classified as atypical adenomatous hyperplasia and grade 4 tumors demonstrating stromal reaction in addition to irregular mitoses and invasive borders along lymphatic vessels. Higher tumor grade is also associated with more invasive and metastatic tumors (34). Based on this analysis, the KLV^+/+ mice developed significantly more high-grade tumors than KLV^−/− mice (Fig. 2E), further supporting the claim that vimentin loss reduces metastatic disease and tumor progression. Taken together, these data show that vimentin is required for efficient metastasis in Kras/Lkb1 comutated lung adenocarcinoma.

To determine vimentin localization within the primary tumor, vimentin IHC was performed on KLV^+/+ primary tumor samples (Fig. 3A). The vimentin antibody was validated for specificity by IHC in vimentin-null tissue and by Western blot (Supplementary Fig. S3). To our surprise, in KLV^+/+ samples, vimentin expression was absent in tumor cells but present in the stromal niche often surrounding tumor cells that appear to have budded off from the primary tumor. We previously termed these tumor cell histologies CIPs (18). CIPs are surrounded by vimentin-positive, elongated stromal cells that resemble fibroblasts in morphology (Fig. 3A).

CIPs are positive for the lung adenocarcinoma marker pro-SPC (ref. 35; Fig. 3B), indicating their type II pneumocyte origin, retain epithelial morphology (nonelongated), and express E-cadherin (18), while lacking vimentin protein expression (Fig. 3A). This pattern suggests that tumor cells within CIPs retain homotypic cell contacts and are likely not undergoing a traditional EMT. CIPs range in size from 5 to 25 cells (Fig. 3C) and are found within this range in KLV^+/+ and KLV^−/− primary tumor samples. CIP density in each genotype was assessed, and these data show that KLV^−/− mice have significantly less CIPs per invasive field than KLV^+/+ mice overall (Fig. 3D). When CIP density is stratified by tumor grade (Fig. 3E), high-grade KLV^−/− tumors have significantly less CIPs than high-grade KLV^+/+ tumors, indicating that vimentin is required to maintain CIP density in invading tumors. Further, these data indicate that within the KLV models, the predominant role of vimentin lies within stromal cells rather than tumor cells.

To determine the role of CIPs in tumor progression and metastasis, CIP density and invasive area were stratified by extent of disease or tumor stage. These data demonstrate that CIPs are found in both early (nonmetastatic) and late (metastatic) stage KLV tumors (Fig. 3F and G) with significantly higher CIP density and invasion in late-stage tumors. To probe CIP biology, IF of extracellular matrix (ECM) components was performed. Fibronectin staining reveals that cancer cells themselves are negative for fibronectin, but fibronectin is expressed in the regions surrounding emerging CIPs at the invasive front of the primary tumor in both KLV^+/+ and KLV^−/− samples (Fig. 3H; Supplementary Fig. S4A). Interestingly, fibronectin expression is lost in CIPs farther away from the primary tumor, suggesting that fibronectin may only be important for early invasion events (Fig. 3H).

Second harmonic generation imaging was performed to assess the collagen ECM which showed high levels of aligned collagen at the invasive front of the primary tumor and surrounding CIPs in the KLV models (Fig. 3I; Supplementary Fig. S4B). This aligned collagen persisted at metastatic lymph nodes, indicating that collagen alignment and secretion are maintained at the distant metastatic site (Fig. 3I). To probe cancer invasion and EMT, IF of TGFβ1 ligand was performed in primary tumor samples (Supplementary Fig. S4C). High expression was observed in areas of focal invasion at or near the primary tumor indicating TGFβ1 is promoting cancer invasion and progression in the KLV models without inducing some canonical EMT markers.

To characterize vimentin-positive cells surrounding CIPs, FSP1 (Fibroblast Specific Protein 1) and alpha-Smooth Muscle Actin (α-SMA) were used to mark CAFs (36). Dual labeling of vimentin and FSP1 in KLV^+/+ primary tumor samples demonstrates that vim⁺ stromal cells are also FSP1⁺ (Fig. 4A). FSP1⁺/vim⁺ cells are randomly localized within the primary tumor (Fig. 4A, left) but surround CIPs at sites of focal invasion (Fig. 4A, right). Quantitative analysis of FSP1⁺/vim⁺ cells surrounding CIPs shows colocalization between vimentin and FSP1 (Pearson coefficient = 0.59; Fig. 4B). IHC of serial sections shows that vim⁺ cells are also α-SMA⁺ (Supplementary Fig. S5A). These data strongly suggest that vim⁺ cells are CAFs. FSP1⁺ CAFs are found at the invasive fronts of both KLV^+/+ and KLV^−/− primary tumors (Supplementary Fig. S5B); however, the significantly greater extent of focal invasion in KLV^+/+ tumors (Fig. 3D) indicates increased CAF recruitment in this model.

Because many tumor-promoting components of the tumor microenvironment express vimentin, several members of the microenvironment were probed within the KLV models, including macrophages, inflammatory cells, and vascular cells. IF of tumor-associated macrophages with F4/80 in KLV^+/+ and KLV^−/− primary tumor samples shows no significant difference in macrophage recruitment (Supplementary Fig. S6). IHC of a Pan T-cell marker (CD3) and endothelial marker (CD31) also demonstrated no difference in recruitment of these factors (Supplementary Fig. S7A and S7B). Taken together, these data indicate that vimentin loss does not impede the recruitment of vascular or inflammatory cells to the primary tumor microenvironment.

To determine the relationship between CAF recruitment and metastatic disease, the presence or absence of CAFs was stratified by genotype and metastatic phenotype. These results show that 100% of metastatic mice contained CAFs at the primary tumor site (Fig. 4C), but less than 50% of nonmetastatic mice had CAFs recruited to the primary tumor in both KLV^+/+ and KLV^−/− genotypes. Vimentin IHC identified that CAFs are also found surrounding CIPs at the metastatic lymph nodes, demonstrating that CAFs are important components of the tumor microenvironment at the secondary site (Fig. 4D) and perhaps travel with tumor cells to the lymph node. These data suggest that CAF recruitment contributes to tumor metastasis by promoting CIP formation and facilitating metastasis to the secondary site.

To probe the mechanisms of the heterotypic tumor cell–CAF interactions and CAF recruitment found in the KLV models, in vitro studies were performed. Stable knockdown of vimentin was achieved in human lung adenocarcinoma CAFs via shRNA (Fig. 5A and B). CAFs lacking vimentin (shVIM19 and shVIM22) were significantly smaller than pLKO.1 vector control (Fig. 5C and D). To test the hypothesis that vimentin depletion inhibits CAF invasion in the KLV models, 3D spheroid invasion assays were performed with pLKO.1 and shVIM CAFs embedded in MatriGel (Fig. 5E). Results show that shVIM CAFs invaded significantly less than pLKO.1 CAFs (Fig. 5F), supporting the hypothesis that CAF motility is impaired with vimentin loss in the KLV models.

Coculture 3D spheroid invasion assays were then performed to assess heterotypic cancer cell–CAF interactions (Fig. 5G). Invasion analysis revealed that the addition of CAFs to spheroids of H460s (lung adenocarcinoma cell line) decreases circularity, a surrogate for collective invasion branching (i.e., increased circularity means less CIPs; Supplementary Fig. S8). Coculture spheroids with shVIM CAFs exhibit significantly higher circularity (decreased branching) compared with those with pLKO.1 CAFs (Fig. 5H). These same coculture spheroids show significantly less invasive chains (37) per spheroid with shVIM CAFs compared with pLKO.1 CAFs (Fig. 5I). These data suggest impairment in shVIM CAFs to drive collective invasion. IF of FSP1 and actin in coculture spheroids (Fig. 5J) shows that CAFs facilitate collective invasion by leading chains of H460s into the surrounding ECM. CAFs lacking vimentin expression lose the ability to lead cancer cells out of the spheroid as evidenced by the decrease in invasive branches with CAFs leading the chain (Fig. 5H). Together, these data indicate that heterotypic in vitro 3D invasion is vimentin-dependent and consistent with the in vivo GEMM results.

To analyze the clinical relevance of the CIP phenotype found in the KLV models, vimentin staining in human lung adenocarcinoma patients with varying genetic subtypes was performed to assess vimentin localization, invasive histology, and EMT (Fig. 6A). A similar pattern to the KLV^+/+ mouse samples was observed in human tissues, where the vast majority of patient samples (25 of 26) expressed vimentin in the mesenchymal stroma and not within the collectively invading cancer cells (Fig. 6A). Vimentin stain intensity was scored and was significantly higher in stromal cells compared with tumor cells across the majority of samples, independent of driver mutation (Fig. 6B). Furthermore, there were no observable differences in vimentin stromal Q score (product of stain intensity and percentage of vimentin-positive cells in stromal compartment) based upon genetic subtype (Supplementary Fig. S9).

Because tumor cells lacked vimentin staining, the epithelial cell marker, E-cadherin, was used to further probe for evidence of EMT. Tumor cells within patient CIPs were E-cadherin–positive (ref. 18; Fig. 6C) and pro–SPC-positive (Fig. 6D), which is also consistent with the data in the KLV models. IF of FSP1 was performed to determine if patient CIPs are surrounded by CAFs similar to the KLV model. These data show an expression pattern consistent with the KLV model in which FSP1⁺ CAFs surround CIPs of tumor cells (Fig. 6E). Taken together, these data suggest that in human lung adenocarcinoma samples tumor cells are not undergoing EMT but instead contain invasive CIPs surrounded by vim⁺/FSP⁺ CAFs, indicating that the predominant role of vimentin in lung adenocarcinoma lies within CAFs and not tumor cells.

To determine the clinical relevance of CIPs, CIP density was stratified in human lung adenocarcinoma patient samples (Fig. 6F and G). Stage 1 patients exhibit significantly higher CIP density than higher stage patients indicating an early dissemination of CIP and CAF invasion, though there are still CIPs present in late-stage patients (P < 0.05). Taken together, the formation of CIPs in early-stage tumors supports an early dissemination model (38) in which metastatic seeds such as CIPs disseminate early during tumor progression.

Vimentin expression has historically correlated with increased metastatic potential across numerous solid tumor types (24, 39); however, a functional role for vimentin in cancer progression has yet to be elucidated. Data from the KLV^−/− GEMM show that vimentin is required for metastasis and progression but not primary tumor formation, whereby KLV^−/− mice had less mediastinal lymph nodes metastasis, less invasive foci, and lower grade tumors compared with their KLV^+/+ counterparts (Figs. 1 and 2); therefore, this supports a model that vimentin is more than an EMT biomarker and plays a functional role in lung adenocarcinoma invasion and metastasis. Despite decreased metastatic rates in the KLV^−/− model, there is no survival advantage with vimentin depletion (Supplementary Fig. S2), which is consistent with previous reports demonstrating increased morbidity in vimentin-null mice (40). In the KLV^−/− model, primary tumor burden could be exacerbating vimentin-dependent deficiencies in arterial remodeling (41) and wound healing (42).

EMT has been the canonical mechanism (9) for how cancer cells lose their epithelial morphology (4), invade through the basement membrane, and navigate the surrounding microenvironment (1, 43); however, this model is challenged in part by studies showing a more epithelial and collective-based migration by tumor cell clusters or packs (18, 44). The data presented here support the concept that a classical EMT does not occur in the majority of lung adenocarcinoma patients or in the lung adenocarcinoma KLV GEMM, but rather cancer cells undergo epithelial-like collective invasion and are surrounded by vim⁺/FSP1⁺ CAFs (Figs. 3 and 4). Across 96% of lung adenocarcinoma patient samples tested (n = 26), independent of driver mutation, vimentin was expressed in CAFs surrounding epithelial CIPs and only rarely in the tumor cells (Fig. 6). Similarly, vimentin was rarely expressed within the tumor cells of the KLV^+/+ GEMM, and only in the CAFs surrounding CIPs at the primary tumor site and at secondary metastatic lymph nodes (Figs. 3 and 4). This supports the concept that malignant cancer cells are not undergoing EMT but partnering with vim⁺/FSP1⁺ CAFs to potentially cometastasize to a secondary site (45). This finding is consistent with studies demonstrating that circulating tumor cell clusters have metastatic potential up to 50 times greater than single circulating tumor cells (46), suggesting that tumor cell cooperativity can lead to greater metastatic success. Furthermore, these heterotypic CAF-CIP clusters are enriched in early-stage lung adenocarcinoma patient samples (Fig. 6), indicating potential early dissemination (38, 47) of metastatic seeding.

Studies investigating vimentin within CAFs in homotypic (CAFs only) and heterotypic (CAFs + cancer cells) in vitro cell cultures demonstrate that vimentin drives not only homotypic CAF invasion, but also the formation and maintenance of heterotypic collective invasion chains (Fig. 5). We show that heterotypic invasive chains in vimentin-depleted CAFs are impaired compared with their wild-type counterparts. Nevertheless, some collective invasion remains suggesting compensatory or redundant mechanisms within vimentin-null CAFs. Because actin, focal adhesion kinase, and other components of the motility machinery (48) are not directly targeted, it is possible that some degree of vimentin-independent cell motility is maintained within the vimentin-depleted CAFs. This is also consistent with the in vivo studies showing that although significantly reduced, CIPs still form in the KLV^−/− mice.

Taken together, the in vivo and in vitro data here support a model where vimentin is required for collective lung cancer invasion and metastasis by facilitating epithelial tumor cell CIP formation via CAF recruitment to the CIP. When vimentin function is impaired, heterotypic interactions are significantly reduced likely due to the inability of vimentin-depleted CAFs to move efficiently. These data support the claim that there are early-stage, epithelial, and collective modes of lung cancer metastasis that drive cellular escape from the primary tumor and require vimentin-positive CAFs. We speculate that vimentin could be a target for antimetastatic therapies that are given at an early stage and in conjunction with traditional cytotoxics (49, 50).

No potential conflicts of interest were disclosed.

Conception and design: A.M. Richardson, L.S. Havel, A.E. Koyen, W.D. Martin, A.I. Marcus

Development of methodology: A.M. Richardson, L.S. Havel, A.E. Koyen, W.D. Martin, M. Gilbert-Ross, A.I. Marcus

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): A.M. Richardson, A.E. Koyen, J.M. Konen, J. Shupe, W.G. Wiles IV, W.D. Martin, G. Sica, M. Gilbert-Ross

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): A.M. Richardson, A.E. Koyen, J.M. Konen, H.E. Grossniklaus, A.I. Marcus

Writing, review, and/or revision of the manuscript: A.M. Richardson, L.S. Havel, A.E. Koyen, J.M. Konen, J. Shupe, H.E. Grossniklaus, M. Gilbert-Ross, A.I. Marcus

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): A.M. Richardson, J. Shupe, W.G. Wiles IV

Study supervision: A.M. Richardson, J. Shupe, A.I. Marcus

This work was supported through funds from the NCI (1R01CA194027, 1RO1CA201340, and 1R01CA142858; awarded to A.I. Marcus and M. Gilbert-Ross) and through a Ruth L. Kirschstein National Research Service Award awarded (to A.M. Richardson). Grant 1U54 CA209992 was awarded to A.I. Marcus. The research in this publication was supported in part by the Winship Cancer Institute NCI/NIH funding (P30CA138292) and several Winship Core Facilities—the Emory Integrated Cellular Imaging, the Winship Core Pathology Laboratory, the Winship Biostatistics and Bioinformatics Shared Resource, and the Winship Cancer Animal Models Core. Special thanks to Jamie Arnst, Rachel Commander, Jamie King, Junghui Koo, Janna Mouw, Brian Pedro, Katelyn Ponder, Emily Summerbell, Carol Tucker-Burden, and Paula Vertino for helpful discussions and technical advice.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.