Identification of Gene Expression Differences between Lymphangiogenic and Non-Lymphangiogenic Non-Small Cell Lung Cancer Cell Lines

It is well established that lung tumors induce the formation of lymphatic vessels. However, the molecular mechanisms controlling tumor lymphangiogenesis in lung cancer have not been fully delineated. In the present study, we identify a panel of non-small cell lung cancer (NSCLC) cell lines that induce lymphangiogenesis and use genome-wide mRNA expression to characterize the molecular mechanisms regulating tumor lymphangiogenesis. We show that Calu-1, H1993, HCC461, HCC827, and H2122 NSCLC cell lines form tumors that induce lymphangiogenesis whereas Calu-3, H1155, H1975, and H2073 NSCLC cell lines form tumors that do not induce lymphangiogenesis. By analyzing genome-wide mRNA expression data, we identify a 17-gene expression signature that distinguishes lymphangiogenic from non-lymphangiogenic NSCLC cell lines. Importantly, VEGF-C is the only lymphatic growth factor in this expression signature and is approximately 50-fold higher in the lymphangiogenic group than in the non-lymphangiogenic group. We show that forced expression of VEGF-C by H1975 cells induces lymphangiogenesis and that knockdown of VEGF-C in H1993 cells inhibits lymphangiogenesis. Additionally, we demonstrate that the triple angiokinase inhibitor, nintedanib (small molecule that blocks all FGFRs, PDGFRs, and VEGFRs), suppresses tumor lymphangiogenesis in H1993 tumors. Together, these data suggest that VEGF-C is the dominant driver of tumor lymphangiogenesis in NSCLC and reveal a specific therapy that could potentially block tumor lymphangiogenesis in NSCLC patients.


Introduction
Lung cancer is the leading cause of cancer death among men and women in the United States [1]. Lung cancer patients typically die from the effect of metastases on distant organs. Lung cancer cells usually appear in regional lymph nodes before they are observed in distant organs. For this reason, lymph nodes are thought to function as "canaries in a coal mine" and are evaluated in order to determine whether cancer cells have spread from their primary site [2]. The presence of cancer cells in lymph nodes is associated with a poor prognosis and is one of the most important predictors of patient outcome for non-small cell lung cancer (NSCLC) and other carcinomas [2,3]. This clinical observation fueled intense research efforts to identify processes that control the lymphogenous spread of cancer and, in 2001, it was reported that lymphangiogenesis, which is the sprouting of new lymphatic vessels from pre-existing vessels, facilitates metastasis to lymph nodes [4][5][6]. This landmark finding ignited great interest in delineating the molecular mechanisms controlling tumor lymphangiogenesis.
Over the past 15 years, substantial progress has been made in the field of tumor lymphangiogenesis research. Growth factors such as Adrenomedullin, Angiopoietin-1, Angiopoietin-2, HGF, Netrin-4, PDGF-BB, VEGF-A, VEGF-C, and VEGF-D have all been reported to promote tumor lymphangiogenesis [4][5][6][7][8][9][10][11][12]. Despite this progress, the precise mechanisms governing tumor lymphangiogenesis remain incompletely understood. This is in part because many studies on tumor lymphangiogenesis use cell lines that have been genetically engineered to overexpress a lymphatic growth factor [4][5][6][7][8][9][10][11][12]. Although the evaluation of genetically modified cell lines has provided valuable information on the role lymphatic vessels serve in tumors, they have not shed light on the precise mechanisms by which cancer cells induce the formation of lymphatic vessels. A better understanding of the molecular mechanisms controlling tumor lymphangiogenesis is needed in order to develop therapies that could potentially prevent the dissemination of cancer and improve the clinical outcome of patients with early stage disease. Therefore, we set out to identify a panel of cell lines that induce lymphangiogenesis and to use genome-wide mRNA expression data to identify the molecular mechanisms governing tumor lymphangiogenesis in NSCLC.

Identification of lymphangiogenic and non-lymphangiogenic NSCLC cell lines
To identify NSCLC cell lines that induce lymphangiogenesis, we stained a panel of 13 NSCLC tumor xenograft samples from previous animal experiments with antibodies against LYVE-1 and podoplanin (Fig 1). These are two commonly assessed markers of lymphatic endothelial cells. An antibody against smooth muscle actin (SMA) was included in the podoplanin stain to help distinguish podoplanin-positive lymphatic vessels (podoplanin+;SMA-) from podoplanin-positive fibroblasts (podoplanin+;SMA+). The extent of lymphangiogenesis was quantified by counting the number of intratumoral lymphatic vessels per microscopic field and tumors were classified as being lymphangiogenic if they contained more than 5 lymphatic vessels per microscopic field or non-lymphangiogenic if they completely lacked intratumoral lymphatic vessels. Through this analysis, we were able to identify a panel of lymphangiogenic (Calu-1, H1993, HCC461, HCC827, and H2122) and non-lymphangiogenic NSCLC cell lines (Calu-3, H1155, H1975, and H2073) (Fig 1).

VEGF-C regulates lymphangiogenesis by NSCLC cells
After identifying lymphangiogenic and non-lymphangiogenic NSCLC cell lines, we set out to find differences between these two groups. We found that there was no obvious difference in the growth rate of lymphangiogenic and non-lymphangiogenic subcutaneous xenografts ( Fig  2). We also found that the ability of a cell line to induce lymphangiogenesis was not related to its subtype classification (adenocarcinoma, squamous cell carcinoma, or large cell), site of origin (primary tumor versus metastasis), or mutation status (Tables 1 and 2).
We then analyzed genome-wide mRNA expression data to identify genes differentially expressed between lymphangiogenic (Calu-1, H1993, HCC461, HCC827, and H2122) and non-lymphangiogenic (Calu-3, H1155, H1975, and H2073) cells. This analysis generated a 17-gene expression signature that distinguished lymphangiogenic from non-lymphangiogenic NSCLC cells (Fig 3). Vascular endothelial growth factor C (VEGF-C), a ligand of the receptor tyrosine kinases VEGFR2 and VEGFR3, was the only gene in this signature reported to stimulate lymphangiogenesis and was approximately 50-fold higher in the lymphangiogenic group than the non-lymphangiogenic group. We confirmed that VEGF-C was expressed at a higher level in lymphangiogenic cells than non-lymphangiogenic cells by quantitative PCR (Fig 3).
To determine whether VEGF-C expression was sufficient to induce lymphangiogenesis by NSCLC cells, we genetically engineered H1975 cells to stably express either red fluorescent protein (RFP; H1975-Ctrl) or full length human VEGF-C (H1975-VEGFC). Reverse-transcription PCR analysis showed that H1975-VEGFC cells expressed a high level of VEGF-C (Fig 4). Additionally, quantitative PCR analysis showed that the level of VEGF-C mRNA is approximately 3-fold higher in H1975-VEGFC cells than H1993 cells (data not shown). We injected these cells into the flanks of NOD/SCID mice and found that H1975-VEGFC tumors grew slightly faster than H1975-Ctrl tumors (Fig 4). The density of intratumoral blood vessels was not significantly different between H1975-VEGFC tumors (16.18 ± 1.998, n = 7) and H1975-Ctrl tumors (13.75 ± 1.263, n = 7; Fig 4). However, the density of intratumoral lymphatic vessels was significantly greater in H1975-VEGFC tumors (11.83 ± 3.125, n = 7) than H1975-Ctrl tumors (0.4286 ± 0.4286 N = 7; Fig 4). These data show that forced expression of VEGF-C is sufficient to induce lymphangiogenesis by a NSCLC cell line.

Nintedanib inhibits tumor lymphangiogenesis
Next, we sought to determine whether inhibition of the VEGF-C/VEGFR3 signaling axis with a clinically relevant compound could suppress tumor lymphangiogenesis. Nintedanib is a small molecule inhibitor that blocks all FGFRs (IC 50 = 37-108 nM), PDGFRs (IC 50 = 59-65 nM), and VEGFRs (IC 50 = 13-34 nM) by binding to the ATP-binding site in the kinase domain of the receptors [13]. Nintedanib has previously been shown to block tumor angiogenesis and growth in several mouse models [13,14]. Additionally, combination therapy of nintedanib with docetaxel has been reported to prolong the survival of stage III/IV NSCLC patients smooth muscle actin (SMA, red). (C) Graph showing intratumoral lymphatic vessel density for podoplanin and LYVE-1 positive lymphatic vessels in NSCLC xenografts grown in mice. We classified Calu-1, HCC827, HCC461, H1993, and H2122 cells as being lymphangiogenic and Calu-3, H1155, H1975, and H2073 as being non-lymphangiogenic. Graph shows mean ± SEM.   previously treated with a platinum-based therapy [15]. To determine whether nintedanib could block tumor lymphangiogenesis, we analyzed tumors from a previous study that evaluated the effect of nintedanib on the growth of H1993 tumors [14]. We found that the density of intratumoral lymphatic vessels was significantly lower in nintedanib treated H1993 tumors (5.254 ± 2.745, n = 5) than vehicle treated H1993 tumors (22.19 ± 2.536, n = 6; Fig 6). These data show that nintedanib is effective at inhibiting tumor lymphangiogenesis.

VEGF-C copy number variation influences VEGF-C expression
Cancer cells frequently undergo genomic alterations that result in the amplification and deletion of genes. These genomic alterations can impact the expression of genes. To determine whether the VEGF-C gene is amplified or deleted in lung cancer cells, we analyzed SNP array data available for 59 lung cancer cell lines. This revealed that the VEGF-C gene was amplified in 22% (13/59; range between 3-5 copies of VEGFC), present as 2 copies in 54% (32/59), and deleted in 24% (14/59) of the lung cancer cell lines that we analyzed (Fig 7). To determine whether changes in the number of copies of the VEGF-C gene influenced the expression of  VEGF-C, we evaluated log transformed microarray values for VEGF-C in this panel of 59 lung cancer cell lines. This revealed that the level of VEGF-C was significantly lower in cells that had deletion (3.582 ± 0.3033) of the VEGF-C gene compared to cells that had either 2 copies (7.029 ± 0.5023) or amplification (8.742 ± 0.6856) of the VEGF-C gene (Fig 7). These data show that VEGF-C copy number variation can affect the expression level of VEGF-C.

Discussion
The study of molecularly annotated lung cancer cell lines has increased our understanding of the pathways driving tumorigenesis and has led to the identification of novel biomarkers and therapeutic targets for lung cancer. In the present study, we use a panel of molecularly annotated NSCLC cell lines to investigate the molecular mechanisms controlling tumor lymphangiogenesis. We show that VEGF-C expression regulates tumor lymphangiogenesis by NSCLC cells and that inhibition of VEGF-C-induced signaling with nintedandinb can block tumor lymphangiogenesis by NSCLC cells.
VEGF-C has emerged as a central figure in the field of lymphangiogenesis research. VEGF-C has been shown to be sufficient to induce tumor lymphangiogenesis by melanoma [16,17], breast cancer [4,12,18], fibrosarcoma [17], and gastric carcinoma cells [19]. Additionally, inhibition of VEGF-C has been reported to suppress lymphangiogenesis by prostate [20,21], pancreatic [22], breast [23][24][25], gastric [26], and lung cancer cells [27]. VEGF-C expression has also been reported to correlate with lymphatic vessel density in many different  human tumors, including NSCLC [28][29][30][31]. We show that VEGF-C expression distinguishes NSCLC cell lines that induce lymphangiogenesis from NSCLC cell lines that do not induce lymphangiogenesis. Additionally, we show by overexpression and knockdown experiments that VEGF-C regulates tumor lymphangiogenesis by NSCLC cells. These findings further demonstrate the importance of VEGF-C in promoting tumor lymphangiogenesis and suggest that it is the dominant driver of tumor lymphangiogenesis in NSCLC.
Nintedanib is a small molecule tyrosine kinase inhibitor that blocks all FGF, PGDF, and VEGF receptors. Nintedanib was previously shown to display anti-cancer effects in a number of preclinical models of NSCLC and in NSCLC patients [14,15]. We show that nintedanib can block tumor lymphangiogenesis in a mouse model of NSCLC. Although we show that nintedanib has anti-lymphangiogenic activity, this compound was previously reported to not inhibit tumor lymphangiogenesis in a transgenic mouse model of pancreatic neuroendocrine tumor (PNET) that was genetically engineered to overexpress VEGF-C [32]. The lack of an anti-lymphangiogenic effect by nintedanib in Rip1-Tag2;Rip1-Vegfc transgenic mice could be because an extensive network of irregular lymphatic vessels might have been present in the pancreas prior to the start of therapy. It has been reported that newly formed lymphatic vessels can persist for a long period of time after the withdrawal of VEGF-C or in the face of anti-lymphangiogenic therapy [33]. Therefore, any lymphatic vessels that formed in Rip1-Tag2;Rip1-Vegfc transgenic mice prior to the start of therapy could potentially be resistant to the anti-lymphangiogenic effects of nintedanib. Alternatively, the difference between our findings and those of Bill et al., (2015) could be because we examined different tumor types or because we used an unmanipulated cell line and they used a genetically engineered model to overexpress VEGF-C.
The molecular mechanisms controlling VEGF-C mRNA levels are not well understood. We show that changes in the number of copies of the VEGF-C gene affect the expression of VEGF-C. We found that lung cancer cell lines that have lost one of their copies of the VEGF-C gene tend to express a low level of VEGF-C. However, additional mechanisms also likely control the expression of VEGF-C by NSCLC cells. The MAPK, mTOR, and NF-kB signaling pathways have all been reported to control VEGF-C expression by cancer cells [34][35][36][37]. These pathways may also play a part in controlling the expression of VEGF-C by NSCLC cells. Future studies with our panel of cell lines will help determine the potential role of these and other pathways in controlling the expression of VEGF-C by NSCLC cells.
In conclusion, the results of this study demonstrate that VEGF-C is a critical regulator of tumor lymphangiogenesis in NSCLC and show that nintedanib inhibits tumor lymphangiogenesis. These findings shed light on the molecular mechanisms driving tumor lymphangiogenesis and have the potential to influence the design of future clinical trials aimed at blocking the spread of early stage NSCLC.

Ethics Statement
The animal experiments described in this manuscript were carried out in accordance with an animal protocol (APN 2013-0121) approved by the Institutional Animal Care and Use Committee (IACUC) of UT Southwestern Medical Center. All mice in this study were purchased from an on-campus supplier. Mice were maintained in ventilated microisolater cages in a pathogen-free facility and were fed a standard irradiated diet ad libitum. Mice were provided nestlets and igloos as enrichment items. Mice were monitored for signs of distress such as lethargy and changes in fur appearance. If mice appeared severely ill or moribund, they would be euthanized by an overdose of carbon dioxide followed by cervical dislocation. No mice became severely ill or died prior to the experimental endpoint. The method of euthanasia for the experimental endpoints consisted of an inhalant overdose of carbon dioxide or isoflurane followed by cervical dislocation. These methods are consistent with the recommendations of the American Veterinary Medical Association (AVMA) Guidelines on Euthanasia.

Cell lines
The human bronchial epithelial cell line (HBEC3KT) and most of the human NSCLC cell lines used in this study (H1993, HCC461, HCC827, H2122, H1155, H1975, and H2073) were established in the laboratories of Dr. Adi Gazdar and Dr. John Minna [38][39][40]. The NSCLC cell lines Calu-1 and Calu-3 were purchased from the American Type Culture Collection (ATCC, Manassas, VA). The NSCLC cell lines were cultured in DMEM + 10% FBS under standard conditions (5% CO 2 at 37°C). The HBEC3KT cell line was cultured in keratinocyte serum-free media containing 5 ng/mL of EGF and 50 μg/mL of bovine pituitary extract under standard conditions (5% CO 2 at 37°C). All NSCLC cell lines were DNA-fingerprinted and mycoplasmatested.

Generation of stable cell lines
To overexpress VEGF-C in a non-lymphangiogenic NSCLC cell line, we infected H1975 cells with commercially available lentiviral particles that contain a plasmid that expresses full length human VEGF-C (Precision LentiORF VEGFC w/ Stop Codon; Open Biosystems, cat no: OHS5899-202618255). To generate control cells, we infected H1975 cells with lentiviral particles that express RFP (Precision LentiORF RFP Positive Control; Open Biosystems, cat no: OHS5833). Cells were grown in media containing blasticidin (30 ug/ml) for several weeks to select for stably transfected cells.
To stably knockdown VEGF-C in a lymphangiogenic NSCLC cell line, we infected H1993 cells with commercially available lentiviral particles that express an shRNA targeting VEGF-C (TRCN0000425238; Sigma, cat no: SHCLNV-NM_005429). To generate control cells, we infected H1993 cells with lentiviral particles that express GFP (MISSION1 TRC2 pLKO.5-puro-CMV-TurboGFP™ Positive Control Transduction Particles; Sigma, cat no: SHC203V). Cells were grown in media containing puromycin (1 ug/ml) for several weeks to select for stably transfected cells.

Animal experiments
NOD/SCID mice received a subcutaneous injection of 1 million H1975-Ctrl, H1975-VEGFC, H1993-Ctrl, or H1993-shVEGFC cells. Mice were weighed and tumors were measured twice a week. Tumor volumes were calculated with the formula V = (a 2 x b)/2, with a and b representing the small and large tumor diameters, respectively. Mice were euthanized before their tumors reached 1,500 mm 3 and their tissues were collected for histological analysis.
Immunofluorescence/immunohistochemistry staining Slides were de-paraffinized with xylene and rehydrated through a descending EtOH series. Antigen retrieval was performed with 0.01 M citric acid (pH 6.0) in a pressure cooker. Slides were then washed with PBS and blocked for 1 hour with TBST + 20% Aquablock. Primary antibodies diluted in TBST + 5% BSA were then added and allowed to incubate overnight at 4°C. Slides were washed with TBST then secondary antibodies diluted in TBST + 5% BSA were added and allowed to incubate for 1 hour at room temperature. Slides were then washed again with TBST and coverslips were mounted with ProLong Gold plus DAPI. Immunohistochemistry was performed using a similar protocol except endogenous peroxidase activity was blocked by incubating slides with hydrogen peroxide diluted in MeOH and signal was detected via the DAB chromogen system (Dako, cat no. K3468).

Quantitative analysis of blood and lymphatic vessels
Slides were analyzed with a Nikon Eclipse E600 microscope and images were captured using NIS-Elements imaging software. To analyze blood vessels, 4 pictures were taken of each tumor and the number of vessels was counted per microscopic field. To analyze lymphatic vessels, 3-5 pictures were taken of "hot spots" in each tumor and the number of vessels was counted per microscopic field.

Mutation status of cell lines
The mutation status of the cell lines was determined by analyzing data from published sources [38][39][40][41][42] and COSMIC (Sanger Institute, UK).

Microarray analysis
RNA quality and concentration were checked by the Bio-Rad Experion Bioanalyzer per manufacturer's protocol. 500 ng of total RNA from each sample was used to label the cRNA probes by Ambion Illumina TotalPrep RNA Amplification kit (cat no: IL1791). 1.5 ug of the amplified and labeled cRNA probes was hybridized to Illumina Human WG-6 v3.0 Expression BeadChip (cat no: BD-101-0203) overnight at 58°C, then washed, blocked and detected by streptavidin-Cy3 per manufacturer's protocol. After drying, the chips were scanned by Illumina iScan system. Bead-level data were obtained, and pre-processed using the R package mbcb for background correction and probe summarization (Ding et al, Nucl Acids Res, 36:e58, 2008). Preprocessed data were then quantile-normalized and log-transformed. Microarray results for the NSCLC cell lines were previously published [43] and archived at the Gene Expression Omnibus repository (http://www.ncbi.nlm.nih.gov/geo/; GEO accession number: GSE32036).
Differentially expressed genes between two classes of samples were determined by calculating fold change and T-test P values and using arbitrary cutoffs for selection (e.g. > 4 fold change and P < 0.01). Because of the low number of samples in each class, we did not adjust the P values with multiple testing correction.

SNP arrays
Whole genome single nucleotide polymorphism (SNP) array profiling was done with the Illumina Human1M-Duo DNA Analysis BeadChip (Illumina, Inc.) [44]. Processing was done with Illumina BeadStudio and DNA copy number was derived from the "Log R Ratio", which measures the relative probe intensity compared with normal diploid controls

Statistical analysis
Data were analyzed using GraphPad Prism statistical analysis software (Version 6.0). All results are expressed as mean ± SEM. For experiments with two groups, unpaired student's T-tests were performed to test means for significance. For experiments with more than two groups, differences were assessed by ANOVA followed by Tukey's multiple comparisons test. Data were considered significant at P < 0.05.