Identification of a Human Airway Epithelial Cell Subpopulation with Altered Biophysical, Molecular, and Metastatic Properties

The most effective therapy for early-stage lung cancer is surgical resection; however, approximately 30% of patients will eventually succumb to metastatic disease following surgery (1). Thus, micrometastatic disease may already be present at the time of surgery, but below the level currently required for clinical detection. This is consistent with the reports of circulating tumor cells in patients with stage I non–small cell lung cancer (NSCLC; refs. 2, 3). Although metastatic behavior is often considered a late event, these findings: (i) suggest that the metastatic process may be also operative in some patients with early stage disease and (ii) are consistent with laboratory-based investigations indicating that dissemination may occur during early tumor development, particularly in the context of epithelial–mesenchymal transition (EMT; refs. 4–8).

A major question is whether all cells within a tumor are capable of metastasis or whether this crucial property resides in only a subset of cells, thus representing clinically important intratumoral heterogeneity. This question also applies to premalignancy, as it has recently been theorized that cells within premalignant lesions may have metastatic capacity (6). While cells derived from a clonal population can vary widely in their gene expression profile (9), an important and often overlooked level of heterogeneity relates to cellular biophysical properties that can regulate stiffness, motility, and migration. Cell stiffness correlates with metastatic potential in a variety of malignancies (10–14), and recent attempts to isolate highly deformable cells established an association between deformability and breast cancer stem cell phenotypes (15). Attempts have been made to isolate cells based on motile behavior in vitro or in vivo (16, 17). While these studies yielded important findings, critical questions remain unanswered, including (i) is migration a heritable property? (ii) can subpopulations of cells with inherent migratory capacity be isolated in the absence of chemotactic gradients? (iii) how does enhanced migration relate to factors such as cell deformability? (iv) what genes are deregulated in migratory epithelial cells, particularly during the earliest stages of carcinogenesis? and (v) does the highly migratory subpopulation have the capacity for enhanced metastatic potential?

Unlike cancer cells, immortalized human bronchial epithelial cells (HBEC) have been shown to be a genetically stable and robust model of the pulmonary airway epithelium and its associated malignant transformation (18–22). Here, we describe phenotypic heterogeneity among a starting population of immortalized HBECs that prompted our selection (or enrichment) of a distinct subpopulation of highly migratory airway epithelial cells. Micropore selection of these cultured target cells (and controls), followed by their functional, biomechanical, and genetic interrogation, is the focus of the current investigation. Our findings provide insight into the molecular drivers of epithelial cell migration that may ultimately facilitate clinical intervention during the earliest stages of lung cancer development and dissemination.

Each parental HBEC line was derived from the large airway of an individual patient and immortalized in the absence of viral oncogenes, as previously described (21). The demographics and clinical characteristics of the patients from whom the parental HBEC lines were derived are recorded in Supplementary Table S1, along with key cell culture characteristics. These HBECs have been well characterized in 2D, 3D, and in vivo, where they maintain the capacity for differentiation (18–22). They comprise at least some of the progenitors for human lung cancer, most accurately recapitulating central airway differentiation and squamous cell carcinoma development under the defined culture conditions we utilize. Detailed information about the generation of HBEC-vector and -Snail lines was previously published, along with detailed cell culture maintenance methods (19). All cells were genotyped by STR profiling (DNA IQ System and Powerplex 1.2 system, Promega) before and after migration selection at the UCLA Genotyping and Sequencing Core. They were used within 10 passages of authentication and routinely tested to be free of mycoplasma contamination (MycoAlert, Lonza).

Western blots were performed according to standard procedures. Horseradish peroxidase–conjugated secondary antibodies (Bio-Rad) and enhanced chemiluminescence (ECL) reagent (Amersham Biosciences) were used for protein detection. Primary antibodies were from the following sources: TUBULIN and SNAIL (Cell Signaling Technologies); RHOA and RAC1 (Cytoskeleton Inc.). Densitometry was performed in ImageJ using the “analyze gels” function.

Cells were seeded into the top chamber of 12-well format Millicell inserts at 75,000 cells/well in 0.5 mL complete medium (Millipore) and incubated at 37°C. Media in both chambers were replaced after 24 hours. After 48 hours, the Transwells were swabbed, fixed in 4% Paraformaldehyde (PFA), and stained with Hoechst-33342 (Molecular Probes). Plating density was determined in unswabbed Transwells by staining with Hoechst and imaging the entire well at 40× total magnification on a Nikon Eclipse Ti inverted microscope (Nikon 4× PlanFluor PhL DL 0.13 NA). Images were captured on a 1024 × 1024 pixel Cascade II EM-CCD camera with equal exposure times using Nikon Elements AR software. Cell density was determined by fluorescence intensity using the “histogram” function in ImageJ. Cell counts described below were normalized to fluorescence intensity of the respective well. To determine the number of cells that entered the bottom chamber, cells were simultaneously stained with NucRed Live647 (stains all cells) and 1 μg/mL DAPI for 20 minutes in complete medium at 37°C (Life Technologies). After staining, the cells were washed and fixed with 4% PFA. Within two hours of staining, images of at least 25 fields in each well were captured using a 10× objective (Nikon PlanFluor Ph1 DLL 0.30 NA). The cells were enumerated using the “threshold” and “analyze particles” functions in ImageJ. Viability is reported as total cells minus dead cells, divided by total cells.

Cells (300,000 per well) were seeded in 6-well format Millicell inserts in 2 mL medium. Media in both chambers were replaced after 24 hours. After 48 hours, cells in the bottom chamber alone were harvested (trypsinization at 37°C followed by centrifugation and resuspension in medium), and the cells were added to the bottom chamber of the original plate. Cells were expanded and plated for a subsequent round of selection. Additional selection pressure was applied by selecting the cells for three rounds through 8-μm pores, then three rounds through 5-μm pores, and finally five rounds through 3-μm pores. Thus, velocity/migration was the first-order selection method applied at every passage, and deformability was the second-order selection method applied only at the two pore size steps (from 8 to 5 μm and from 5 to 3 μm).

After phenotypic validation, these highly migratory cells (and control cells) were profiled using Affymetrix GeneChip HTA 2.0 gene arrays at the UCLA Clinical Microarray Core. The raw data have been deposited at National Center for Biotechnology Information's (NCBI) gene expression omnibus (GEO; accession GSE100527). Gene-level normalization was performed using the Affymetrix Expression Console and the RMA approach. Non-annotated transcripts were then filtered out from the data before further analysis. Bioconductor limma package was used for gene expression analysis. A gene was defined as being differentially expressed if two conditions were met: (i) its fold change was greater than 1.5 and (ii) its two-tail t test P value was less than 0.1. Cluster 3.0 was used for clustering analysis. Gene set enrichment analysis (GSEA) was used for identifying deregulated pathways and biological processes associated with deregulated genes (23).

This specialized invasion assay was performed according to previously published protocols (24, 25). Assays were performed in 96-well plates (Black wells, clear bottom, PerkinElmer). Briefly, ice-cold serum-free liquid bovine collagen (PureCol) was prepared at 1 mg/mL concentration. After trypsinization, the cells were added to the collagen suspension at a final concentration of 5 × 10⁴ cells/mL, and 100 μL aliquots were dispensed into five replicate wells. The cells were subsequently spun down, then incubated in a 37°C/5% CO₂ tissue-culture incubator to allow the collagen to solidify. Finally, the collagen plug was covered by 30 μL of a mixture of the appropriate culture medium diluted 1:1 with PBS. After 48 hours, the cells were fixed with 4% PFA (final concentration) and stained with 10 μg/mL Hoechst-33342. Images of nuclei staining were obtained on an inverted microscope operated by Nikon Elements Software using a 20x air objective. Twenty-five adjacent images covering ∼85% of the well area were taken from 0 μm (bottom of the plate) to 150 μm up in the collagen plug, at 25-μm steps. Quantification of nuclei/well was obtained using the “object count” feature of Nikon Elements Software. Invasion ratio was calculated as the sum of cell counts at 50, 75, 100, 125, and 150 μm over cell counts at 0 μm.

The deformability cytometry (DC) microfluidic devices were fabricated following standard polydimethylsiloxane (PDMS) replica molding techniques (26). Documented channel design schematics were also followed for the DC chip master molds (27). Dissociated cell samples were diluted with media to achieve a concentration between 100,000–200,000 cells/mL and introduced to the DC chip at 800 μL/min using a Harvard Apparatus syringe pump. Additional details on experimental design and post-processing analysis are available in previous reports (27) and in Supplementary Methods.

Pathogen-free NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NOD scid gamma or NSG) mice were purchased from The Jackson Labs and maintained in the UCLA Biocontainment Facility. All studies were approved by the institution's animal review committee (ARC #2011-084). Cells were harvested and washed twice with sterile saline, then suspended at a concentration of 5 × 10⁵ cells in 50 μL for injection into the lateral tail vein of restrained mice. The mice were euthanized by CO₂ asphyxiation at one of two time points—early (72 hours) or late (26–74 days). The lungs from each mouse were harvested and divided into two work flows (qPCR or IHC) that allow quantitation of human HBEC cells within mouse tissues; the left lung was always utilized for qPCR and the right lungs for IHC. For qPCR analysis, genomic DNA was isolated using the DNeasy Blood & Tissue Kit (Qiagen). The percentage of human cells (HBEC) relative to the number of mouse cells (host lung) in each sample was determined using the primers and qPCR conditions previously described (28), with the exception that 100 ng of input DNA was used here. For IHC analysis, the mouse organs were fixed in 10% neutral buffered formalin for 48 hours before being rinsed thoroughly and transferred to 70% ethanol. The tissues were then processed routinely, sectioned, H&E stained, and immunostained using standard methods previously described (19). Detection of single cells or small aggregates of HBECs within the mouse lungs was confirmed by CD44 (Cell Signaling #3570) immunostaining at 1:400 in EDTA with antigen retrieval. CD44⁺ HBECs within the murine lungs were manually counted and normalized to the total lung area stained. Representative photomicrographs were obtained using an Olympus BX50 microscope, with Plan APO objective lenses, an Olympus DP21 camera, and Olympus Camedia software.

Cells were grown on 12-mm glass coverslips (Electron Microscopy Sciences) in complete medium. Cells were fixed and permeabilized using 4% PFA and 0.1% Triton X-100 using standard methods. Cells were blocked for 30 minutes in 1% BSA. For nuclear size determination, cells were labeled in blocking buffer with 1:1,000 mouse anti-Lamin B1 (AbCAM) for 30 minutes at room temperature. Cells were stained for 30 minutes at room temperature in secondary antibody (goat anti-mouse Alexafluor488, Life Technologies). For cytoskeleton fluorescence, cells were stained with phalloidin-AlexaFluor647 (Life Technologies) and counterstained with DAPI according to the manufacturer's instructions. ProLong Gold (Life Technologies) was used to mount the coverslips. Cells were imaged at room temperature using a Nikon Eclipse 90i upright fluorescent microscope and 10× or 20× objective (Nikon 10× PlanFluor DIC L/N1, 0.30 NA; Nikon 20× Plan Apo VC DIC N2, 0.75 NA). Images were captured using a Nikon Digital Sight DS-Qi1Mc camera at 1024 × 1024 or 1280 × 1024 pixel resolution. To quantify nuclear size, the images were thresholded using ImageJ. The “analyze particles” function was used to determine the nuclear area. At least 60 cells of each type were analyzed for nuclear size.

Cell lysates were harvested and utilized in a GTPase pull down assay using a commercially available kit according to the manufacturer's instructions (Activation Assay Combo Kit Cytoskeleton Inc.). Additional information can be found in Supplementary Methods.

Samples were plated and run in triplicate, unless otherwise indicated, and all in vitro experiments were performed at least three times. Data represent mean and SEM of one representative experiment, and one representative experiment/image is shown. The statistical significance of these data was determined using a nonpaired Student t test with 95% confidence interval or the Mann–Whitney test (two-tailed, 95% confidence interval) where appropriate. Data were reported significant as follows: * if P ≤ 0.05, ** if P ≤ 0.01, and *** if P ≤ 0.001.

Additional methods are included in the Supplementary Methods section.

Among unselected HBEC cell lines, our observation of a phenotypically distinct cell type that was highly migratory relative to its neighbors suggested that there is intrinsic heterogeneity in the starting population of cells that could be relevant to dissemination. Approximately 1 in 10,000 of the unselected immortalized HBEC cells was phenotypically distinct (Supplementary Fig. S1A). To isolate these rare highly migratory pulmonary epithelial cells, we devised a selection strategy using a series of microporous membranes (Fig. 1A). Cells were initially seeded on 8-μm pore Transwell membranes with basal medium in the top and bottom chambers and allowed to migrate for 48 hours. Chemotactic gradients were avoided to minimize induction of directed motility and selection based on this phenotype. To avoid inadvertent generation of a gradient due to nutrient depletion, media in both chambers were replaced after 24 hours. The cells in the bottom chamber were isolated by trypsinization and propagated in a new dish. They were replated onto a new membrane, and the process was repeated. To further increase selection pressure, the pore size was serially decreased, first to 5-μm and then 3-μm pores. It is noteworthy that an insufficient quantity of viable cells migrated through 5- or 3-μm pores within 48 hours to justify immediate selection at these sizes, thus negating the possibility of skipping a priori selection through 8-μm pores. Cells selected based on their migratory capacity are denoted HM (highly migratory), and they were compared with the unselected starting population in all subsequent studies (Fig. 1B).

The HM cells were subjected to a migration assay using 8-, 5-, and 3-μm pore membranes and stained to quantify the migration rate. Selected HBEC3 with p53 deletion and KRAS activating mutation (H3PK-HM) had a 5-fold higher migration rate through 8-μm pores, a 30-fold higher migration rate through 5-μm pores, and a 124-fold higher migration rate through 3-μm pores, compared with their unselected counterparts; similar trends were observed for H3-HM and H4-HM cell lines (Fig. 2A; Supplementary Table S2). Additional cellular processes are required for invasion, thus we sought to determine if the highly migratory cells were also highly invasive. H3-HM and H4-HM cells showed enhanced invasion in a collagen matrix compared with parental cells (Fig. 2B). H3PK-HM cells showed no difference in invasion through collagen. However, the baseline invasion rate in the H3PK cell line was already high, suggesting that migration was not rate limiting for these cells in this assay.

To determine if the migratory phenotype would be sustained through many generations, the Transwell migration assay was performed 8 weeks after selection (∼40 population doublings). Over this time, the selected cells remained highly migratory relative to their unselected counterparts (Fig. 2C), indicating that the selected migration phenotype is heritable and can be exploited to expand the cells for downstream functional analyses.

Cells with the capacity to alter their morphology by increasing their cellular protrusions may also have an enhanced ability to conform to constrictive spaces. Images of highly migratory cells were taken using phase contrast microscopy. Using ImageJ (29), outlines were made of each cell (Supplementary Fig. S1B), and the corresponding ratio of perimeter to area was calculated to determine “protrusion intensity,” which relates to the number of protrusions. H3-HM, H4-HM, and H3PK-HM cells all demonstrated increased protrusion intensity (Fig. 3A). Recently, it has been suggested that the nucleus can be a rate-limiting factor in successful migration, such that matrix metalloproteinase-independent migration was shown to decline as a linear function of pore size and with deformation of the nucleus, with arrest reached at a pore size of 10% of the nuclear cross-sectional area (30). We measured the nuclear cross-sectional area by imaging cells following nuclear lamin staining, to determine if nuclear size was a factor in the HM cell subpopulation, which was selected based on matrix metalloproteinase-independent migration. H4-HM cells have increased nuclear area, while H3-HM and H3PK-HM cells show no significant difference in nuclear size (Fig. 3B). The 3-μm pore size in the migration assay used here is much smaller than 10% of the nuclear cross section of all cell lines in the study, suggesting that nucleus size alone is not a rate limiting factor. The results indicate that our selection strategy does not bias for cells with smaller nuclei and selects cells based on migrational competence. This increased protrusion intensity suggests that HM cells may have greater control over their morphology, which would facilitate their migration through constrictive spaces.

We sought to determine if the micropore selection process coselected for increased deformability as a consequence of selecting cells capable of migrating through decreasing pore sizes that are a fraction of the cell volume. To assess cellular deformability, we utilized deformability cytometry (DC; ref. 27). This technique uses hydrodynamic forces to stretch cells uniformly in a microfluidic chip, followed by high-speed imaging to discern changes in cell shape and size as a result of applied stress. Deformability (D) is a unitless measure defined as the maximum ratio of the long to short axis of a cell stretched with a standard stress, such that a cell with a lower stiffness will have a corresponding higher deformability. There was no across-the-board increase in deformability as a result of micropore selection (Fig. 3C; Supplementary Table S3). A secondary readout of DC is cell size, measured when the epithelial cells are in suspension just prior to being deformed. As with deformability, cell line–dependent results were also observed for cell size (Fig. 3C). This suggests that the selection strategy does not bias toward biophysical properties, such as size or deformability, but rather selects cells based on migratory behavior more specifically.

While EMT is known to drive the invasion and metastasis of established tumors (31), it has also been theorized to drive dissemination during lung premalignancy (6). However, the isolated subpopulation of cells did not exhibit morphological characteristics associated with EMT (Supplementary Fig. S1B). To evaluate EMT-related phenomena in our highly migratory model, we performed selection in HBECs that have been genetically engineered to overexpress SNAIL (19). Similarly to isogenic cells, enhanced migration was observed in these SNAIL overexpressing HM cells (Supplementary Fig. S2A; Supplementary Table S2).

By dual staining for live/dead cell analysis, we found that without prior selection, migration through 3-μm pores results in a high incidence of cell death in the H3PK cells, accounting for approximately 50% viability regardless of SNAIL expression level (Supplementary Fig. S2B and S2C). This explains the need for a two-step selection process (velocity/migration the primary, and deformability the secondary) and indirectly suggests HM cells are better equipped to metastasize to distant sites via their capacity for enhanced survival during migration. Leonard Weiss first described the concept of “metastatic inefficiency,” which has been observed by others (32, 33). It refers to points in the metastatic cascade that typically result in cell death, including (i) intravasation and (ii) extravasation (34). These reports and the observations from our postmigration viability studies prompted us to test our selected cells in an in vivo model that would examine the second transition from the intravascular space to the lung parenchyma. A key technical limitation of the IV experimental metastasis model is that the first transition is not evaluable, because the cells are deposited directly into the blood stream. Here specifically, H3PKS or H3PKS-HM cells (5 × 10⁵) were injected into the lateral tail vein of NSG mice. Tissue was collected for analysis after 72 hours, matching the timeframe used in the post-migration viability study and allowing sufficient time for clearance of dead cells from the pulmonary vasculature (35). This timeframe allowed us to focus on the extravasation and survival phenotypes, rather than on tumor outgrowth. Total DNA was extracted from the left lung homogenates, and all remaining lung tissue was processed for IHC. A PCR-based method described by Malek and colleagues was used to quantify the number of human bronchial epithelial cells that arrested in the mouse lung (28). Mice injected with the micropore-selected H3PKS-HM cells showed greater than 2.5-fold more human DNA in their lungs on average compared with mice injected with unselected H3PKS cells (Fig. 4A). We next evaluated the ability of the two cell types to not only survive the initial phase of implantation in the lung, but to survive long term (26–74 days), to extravasate from the vasculature and/or lymphatics, and to successfully colonize the lung niche by expanding to form premalignant lesions or frank tumors. Among those injected with HM cells, 4 of 7 surviving mice showed evidence of human bronchial epithelial cells at the late time point, as measured by qPCR. Conversely, none of the 7 mice injected with unselected H3PKS cells showed evidence of human cells at corresponding time points (data not shown). Utilizing a human-specific CD44 antibody to delineate human cells, IHC analysis of the right lungs of corresponding mice confirmed the presence of more and larger epithelial cell aggregates in the HM mice (Fig. 4B). Furthermore, pathology review found that HPKS-HM cell aggregates could be located outside of the capillary/lymphatic beds in the lung parenchyma (Supplementary Fig. S3). Taken together with the in vitro viability studies, this suggests that the selected cells are endowed with several properties conducive to micrometastatic dissemination: enhanced extravasation from the vasculature and more robust colonization of the lung niche. Thus, enhanced migration is a biophysical trait that allows premalignant epithelial cells to overcome key aspects of metastatic inefficiency.

The actin cytoskeleton is in large part responsible for cell motility and migration through the coordination of numerous factors influencing actin polymerization and depolymerization, among others. Visualizing F-actin in cells can help determine which molecular processes are active. Thus, we grew selected and unselected cells on cover slips and stained them using fluorescent phalloidin. H3PK-HM and H4-HM cells showed no appreciable difference in F-actin structures (Supplementary Fig. S4); conversely, H3PKS-HM cells showed more F-actin bundles in the lamellipodia than their unselected counterparts (Fig. 5A). While H4S-HM cells did not show similar F-actin structures, they display striking changes in morphology and protrusions rich in F-actin (Fig. 5A). These results show changes to the cytoskeleton in the subpopulation of highly migratory cells when SNAIL is overexpressed and suggest increased Rac1 activity, a factor known to drive actin polymerization.

Rho-family GTPases are major regulators of the actin cytoskeleton, and their signaling facilitates processes such as cell polarity and motility (36). Based on our observations of F-actin, we next determined the activation status of two well-characterized GTPases, RhoA and Rac1, to determine if they are altered in the highly migratory cells. Whole-cell lysates were utilized in a pull-down assay with agarose beads conjugated to Rhotekin or p21-activated kinase binding domain (PAK), which recognize GTP-bound (activated) RhoA or Rac1, respectively. We observed no significant change in RhoA activity, but a 2-fold increase in activated Rac1 (Fig. 5B). To verify that enhanced Rac1 activity drives migration in HM cells, we pharmacologically inhibited Rac1 in these cells in a migration assay (using a concentration of 10 μmol/L inhibitor, which did not prevent cell growth; Supplementary Fig. S5). Inhibition of Rac1 activation by NSC-23766, an inhibitor of a Rac1-specific guanine nucleotide exchange factor (GEF), perturbed cell migration compared with control cells (Fig. 5C). Similar results were observed in cell line pairs without ectopic expression of Snail, suggesting that the Rac1 pathway activation is at least partially responsible for the highly migratory phenotype and is operative independent of Snail. More broadly, these results suggest that epithelial motility may be targeted pharmacologically by perturbing the Rac1 pathway.

Because we and others have shown that SNAIL leads to enhanced motility and migration, we sought to determine if highly migratory cells had greater SNAIL expression than their unselected counterparts. No differences in SNAIL protein levels were detected (Supplementary Fig. S6). To determine how gene expression was altered among HM cells we isolated total RNA from the parental line and HM subpopulation and performed microarray mRNA profiling. The raw data have been made publicly available via NCBI's GEO (accession GSE100527), and Supplementary Table S5 highlights the top gene expression changes charactering highly migratory epithelial cells. Interestingly, the H3-HM, H4-HM, and H3PK-HM subpopulations did not show significant changes in epithelial or mesenchymal genes, suggesting that EMT has not occurred in these cells, consistent with the morphological observation described earlier. However, there is a subset of genes upregulated in the HM subpopulation relative to the parental lines from which they were derived that mimic genes upregulated as a result of SNAIL expression (Fig. 6A). Next, we determined if gene expression changes in the HM subpopulation correlated with clinical datasets implicating metastatic disease. GSEA revealed genes correlated with the highly migratory phenotype were also enriched in clinical gene sets associated with poor overall survival in NSCLC, and recurrent disease in breast cancer and melanoma samples (Fig. 6B; refs. 37–39).

Among immortalized premalignant airway epithelial cell lines, we have identified a small subpopulation of cells with a highly motile and metastatic phenotype, demonstrated using in vitro and in vivo methods, and found this phenotype is heritable over 40 population doublings. The highly migratory phenotype is enhanced by Snail expression and requires Rac1 activity regardless of Snail expression levels. This system represents an important new preclinical model for understanding mechanisms and developing therapeutics. This finding also suggests that a population of premalignant airway epithelial cells are endowed with biophysical properties that enable a profound capacity for high-velocity migration.

By isolating highly migratory cells from both parental and SNAIL-overexpressing populations, we were able to study SNAIL-independent and SNAIL-dependent mechanisms of migration. This allowed us to model the SNAIL-mediated EMT program, which is known to drive migratory behavior, and is clinically relevant because SNAIL is present in lung premalignant lesions (40). As determined by deformability cytometry, we found the HM subpopulation derived from the H4 line showed increased cellular deformability. This result is in alignment with other studies showing that migration and metastasis correlate with increased cellular deformability (5, 10, 41). However, increased stiffness in H3-HM and H3PK-HM cells suggests that cellular deformability may not always be requisite for the highly motile phenotype. One explanation for the occasional disconnect between motility and deformability may be technical. Velocity/migration was the first-order selection method applied at every passage (at every pore size). Thus, motility differences between HM and control cells were most pronounced and most consistent across cell lines. Deformability was the second-order selection method applied only at the two pore size steps (8–5 μm and 5–3 μm). Thus, deformability differences between HM and control cells were not as large and not as consistent across cell lines as the motility physical characteristic. Differences in deformability among the migration-selected cells may also be explained by patient/HBEC heterogeneity. For example, Liu and colleagues have recently described different migration modes exist among cells within a single population, which were ascribed to subtle molecular differences (42). We found that Rac1 activity mediates enhanced migration, and that pharmacologic inhibition of Rac1 activation reduced migration in the selected highly migratory cells. Rac1 activity has been associated with carcinogenesis, EMT, and metastasis and is being explored as a candidate therapeutic target (43, 44). Genes whose expression were upregulated in our highly migratory cells also correlated with poor overall survival in NSCLC and increased recurrence in breast cancer and melanoma. It could be that recurrence is related to micrometastatic disease present, but not detected, at the time of resection. Prospective studies examining the migratory gene sets in pulmonary premalignancy and its relationship to survival in early-stage lung cancer patients are anticipated in future investigations.

Although most previous studies utilize fully malignant cancer cells of tumor origin for the identification and characterization of highly migratory cells, investigation of premalignant lesions may be equally important. Vaughan and colleagues have identified a pulmonary progenitor cell population capable of mobilizing to regenerate the murine lung epithelium following major injury (45). While a corresponding population in the human lung has yet to be identified, these findings support the existence of a subpopulation of highly motile pulmonary epithelial cells that may play a role in the pathogenesis of lung cancer.

Metastasis is the predominate cause of cancer death (46), and while a tumor is made up of many cancer cells, only a small percentage of these cells become metastatic (17, 47, 48). The current study provides a new approach for investigating subpopulations of premalignant human lung epithelial cells in the context of migratory and metastatic behavior. This affords the opportunity for systematic interrogation of mechanisms and potential preventive or therapeutic targets for patients at risk for lung cancer.

No potential conflicts of interest were disclosed.

Conception and design: P.C. Pagano, Z. Jing, K. Krysan, J.K. Gimzewski, J.D. Minna, T.C. Walser, S.M. Dubinett

Development of methodology: P.C. Pagano, S. O'Byrne, H.T. Tse, S. Sharma, E.L. Liclican, Z. Jing, J.K. Gimzewski, D. Di Carlo, J.D. Minna, T.C. Walser, S.M. Dubinett

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.C. Pagano, N. Bendris, S. O'Byrne, S. Sharma, J.W. Hoech, S.J. Park, E.L. Liclican, Z. Jing, R. Li, M.K. Paul, J.E. Larsen, S. Hakimi, A. Seki, J.D. Minna, T.C. Walser

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.C. Pagano, L.M. Tran, S. O'Byrne, S. Sharma, J.W. Hoech, E.L. Liclican, Z. Jing, K. Krysan, M.K. Paul, M.C. Fishbein, J.K. Gimzewski, T.C. Walser, S.M. Dubinett

Writing, review, and/or revision of the manuscript: P.C. Pagano, L.M. Tran, S. O'Byrne, S. Sharma, E.L. Liclican, Z. Jing, R. Li, K. Krysan, M.K. Paul, Y. Fontebasso, M.C. Fishbein, J.K. Gimzewski, D. Di Carlo, J.D. Minna, T.C. Walser, S.M. Dubinett

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.C. Pagano, J.W. Hoech, S.J. Park, Z. Jing, R. Li, A. Seki, T.C. Walser

Study supervision: S.M. Dubinett

Other (preparation of morphologic materials, review of microscopic slides, preparation of figures): M.C. Fishbein

Other (grant support): J.D. Minna

The authors acknowledge Ying Lin for technical assistance, the UCLA Genotyping/Sequencing Core for genotyping and microarray services, and Jessica Castillo for assistance with figure preparation. We thank the Translational Pathology Core Laboratory (TPCL) for histology services. We thank the UCLA Clinical and Translational Science Institute (CTSI), as well as the UCLA California NanoSystems Institute Advanced Light Microscopy/Spectroscopy Core for microscopy resources. We thank Lauren Winter for assistance in preparation of the manuscript.

These studies were supported by funding from the following sources: NIH/NCI#T32-CA009120-36 (S.M. Dubinett), NCI#U01CA152751 (S.M. Dubinett), NCI#1U01CA196408 (S.M. Dubinett), Department of Defense Congressionally Directed Medical Research Programs#LC130767 (S.M. Dubinett), Department of Veteran Affairs#2I01BX000359-05A1 (S.M. Dubinett), National Center for Advancing Translational Science UCLA CTSI GrantUL1TR000124 (S.M. Dubinett), University of California Tobacco-Related Disease Research Program (TRDRP) #22DT-0005 (P.C. Pagano) and #20KT-055 (T.C. Walser), NCI Lung Cancer SPORE#P50CA70907 (J.D. Minna), and the Packard Foundation Fellowship (D. Di Carlo).

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.