Background
The process by which new blood vessels sprout from pre-existing blood vessels is called angiogenesis. This step is normal and important in wound healing, cell growth and development [
1]. However, angiogenesis is also a prerequisite for tumour progression and metastasis. Tumour angiogenesis provides oxygen and nutrients to cancer cells and also provides a route for cancer cell metastasis [
2‐
4]. High tumour angiogenesis activity is associated with advanced tumour growth and metastasis in human cancers [
5]. Moreover, the non-small-cell lung cancer (NSCLC) patients with higher microvessel densities in tumour tissues have a worse survival [
6,
7].
The tumour microenvironment is composed of tumour cells, blood vessels, immune cells, fibroblasts and extracellular matrix (ECM) [
8]. Many investigations have shown that the microenvironment plays a crucial role in tumour growth, metastasis and angiogenesis. Particularly, the proliferation and motility of endothelial cells are essential for microvessel sprout formation and angiogenesis, which correlates with cancer metastasis. The interaction between cancer, endothelial cells and other components of the microenvironment may have a reciprocal effect on angiogenesis, cancer cell proliferation and dissemination [
9]. Vascular endothelial growth factor (VEGF) is a key regulator of angiogenesis. A previous meta-analysis showed that VEGF overexpression in tumour indicates an unfavourable prognosis in NSCLC patients [
10]. Collagen VI, a major ECM protein in the microenvironment, can induce angiogenesis and promote tumour progression [
11]. Cancer-associated fibroblasts promote angiogenesis, invasion and metastasis in many cancers [
12,
13]. Tumour-associated macrophages stimulate tumour progression by inducing angiogenesis and suppressing adaptive immunity [
14]. Our group also reported that M1-type macrophages could reduce tumour growth and angiogenesis in vitro and in vivo [
15].
Dozens of studies have illustrated that the PI3K/Akt signalling pathway is critical in tumour growth, proliferation and survival. This makes PI3K/Akt and downstream signalling components suitable therapeutic targets [
16,
17]. Moreover, several reports also show that PI3K and Akt play crucial roles in endothelial cell survival and angiogenesis and activated by VEGF, fibroblast growth factor (FGF) stimulation [
18,
19]. However, the actual mechanisms that regulate tumour angiogenesis and cancer progression within the tumour microenvironment remain unclear. Previous data showed that glioma conditioned medium increased the proliferation, migration and tube formation of human brain endothelial cells via Roundabout4 (Robo4) down-regulation [
20]. However, the mechanism behind the association between endothelial cells and lung cancer has not been defined. Lung cancer is one of the top ten causes of death in human malignancy and shows a poorer prognosis than other human cancers [
21]. Therefore, the clarification of the molecular mechanism is essential for the discovery of new therapeutic reagents in the treatment of NSCLC.
In this study, we identified the differentially expressed genes of human endothelial cells after interaction with a human lung cancer cell line through an indirect co-culture system by microarray. A panel of genes involved in migration, tube formation and apoptosis were identified and validated. Inhibitors of the key activated genes (such as PI3K and COX-2) were applied to confirm the effect of these activated genes on the angiogenesis of endothelial cells after co-culture with lung cancer cells. These results may provide new therapeutic targets for anti-angiogenesis therapy for NSCLC in the future.
Methods
Cells
Human umbilical vein endothelial cells (HUVECs, pooled donors) were obtained from Lonza (Walkersville, MA) and maintained in Endothelial Basel Medium-2 (EBM-2) with supplements (Lonza). Replicated cultures were obtained by trypsinization and were used at passages < 6. The human lung adenocarcinoma cell line CL1-5 was established previously [
22]. The cells were grown in RPMI 1640 medium (Thermo Fisher Scientific, Rockford, IL) supplemented with 10% FBS. All cell lines were kept at 37 °C in a humidified atmosphere containing 5% CO
2
. The Transwell Permeable Supports (Corning, Tewksbury, MA) with a 0.4 μm polycarbonate membrane were used in the co-culture model system to separate HUVECs and CL1-5 cells into the different compartments. One hour prior to co-culture, 5 × 10
4 HUVECs in 500 μl EBM-2 were grown in 24-well plate with or without Matrigel coating (BD Biosciences, San Jose, CA). An equivalent number of CL1-5 cells in 200 μl EBM-2 were seeded into the top chamber of a transwell insert, which was then placed directly on top of the 24-well plate containing the HUVECs. When performing tube formation and TUNEL assay, HUVECs were seeded on Matrigel in 24-well plate. After incubation of overnight (mRNA and protein expression) or the indicated time (tube formation and TUNEL assay), HUVECs were harvested for further analysis as described below. For the motility assay, the 8 μm polycarbonate membrane insert was used in the co-culture system where HUVECs were in the upper and CL1-5 cells were in the lower compartment as described in the migration assay section.
F-actin staining
For F-actin staining, cells were plated on 12-mm-diameter coverslips for 24 h, washed twice with phosphate-buffered saline (PBS) and fixed in 4% paraformaldehyde for 10 min at room temperature. After two additional washes with PBS, cells were permeabilized with 0.1% Triton X-100 for 2–5 min and washed again with PBS. Rhodamine conjugates of Phalloidin (Thermo Fisher Scientific) were used to stain F-actin. The cells were stained for 40 min at room temperature. The nuclei were stained with 4',6-diamidino-2-phenylindole (DAPI). The fluorescence images of F-actin and nuclei were visualized by confocal microscopy (ZEISS, Germany).
Migration assay
The Transwell Permeable Support (Corning) with an 8-μm polycarbonate membrane insert was used in the cell migration model where HUVECs were allowed to migrate. Approximately 2 × 104 HUVECs in 200 μL EBM-2 medium were loaded into each 24-well insert in triplicate with or without CL1-5 cells in the lower chamber at 37 °C in a 5% CO2 incubator. After approximately 20 h, the migrated cells were fixed with methanol, stained with Giemsa solution (Sigma, St. Louis, MO) and counted at 200x magnification under a light microscope.
TUNEL assay
Detection and quantification of apoptosis were performed by the TUNEL reaction, using the In Situ Cell Death Detection Kit, Fluorescein (Roche Diagnostics, Indianapolis, IN). Cells were recovered from Matrigel by Cell Recovery Solution (Corning) after culture for 6, 12, 24 and 30 h, seeded onto slides by cytospin and stained following the standard protocol to label DNA strand breaks with fluorescein-dUTP. Propidium iodide (PI) was used to label all nuclei. The image data were analysed under a fluorescence microscope. Experiments were evaluated in triplicate, and 10 fields of view were quantified for each sample.
Matrigel Basement Membrane Matrix (BD Biosciences) was diluted with EBM-2 medium and coated in 24-well plates at 37 °C for 1 h. Then, 5 × 104 HUVECs were seeded alone or co-cultured with an equivalent number of CL1-5 cells in the EBM-2 medium on Matrigel. Co-cultured CL1-5 cells were seeded in transwells and incubated in the same well with HUVECs. The tube formation ability of HUVECs was measured at 1, 2, 6, 12 and 24 h with or without CL1-5 cells. In inhibitor experiments, HUVECs were treated with the PI3K inhibitor LY294002 (5 μM) and the COX-2 inhibitor celecoxib (10 μM) (Sigma) for 12 h and co-cultured with CL1-5 cells. After incubation, the number of tubes and nodes of the tubular structures was quantified.
Real-time quantitative PCR
Total RNA was extracted from HUVECs, which were co-cultured with or without CL1-5 cells. First-strand cDNA for real-time quantitative PCR (QPCR) analysis was obtained from 5 μg of total RNA using a random primer and SuperScript III Reverse Transcriptase kit (Thermo Fisher Scientific) according to the manufacturer’s instructions. Reactions were detected by the SYBR Green approach (Thermo Fisher Scientific). Ten nanograms of cDNAs served as templates to detect gene expression. Experiments were performed three times in triplicate. Details of the specific primers designed for QPCR to determine relative levels of gene expression are shown in Table
1.
Table 1
Primer sequences used in real-time PCR experiments
ACTN1 | Forward: AACTGTCACTTGGCGGGCAGGG |
Reverse: AAGGGCATCAGCCAGGAGCAGAT |
AKT3 | Forward: CCTTCCAGACAAAAGACCGTTT |
Reverse: ATGTAGATAGTCCAAGGCAGAGACAA |
CTNNB1 | Forward: AGCTAAAATGGCAGTGCGTTTAG |
Reverse: ACTAGCCAGTATGATGAGCTTGCTT |
CXCL8 | Forward: TTGGCAGCCTTCCTGATTTC |
Reverse: AACTTCTCCACAACCCTCTGCA |
ICAM1 | Forward: CGATGACCATCTACAGCTTTCCGG |
Reverse: GCTGCTACCACAGTGATGATGACAA |
ITGAV | Forward: CTTCCAATTGAGGA ATCACCAACT |
Reverse: CAATCCTGCTAGAACTGCTAAAATGA |
ITGB3 | Forward: CGACCGAAAAGAATTCGCTAAA |
Reverse: GGTACGTGATATTGGTGAAGGTAGAC |
PIK3CA | Forward: AACACTCAAAGAGTACCTTGTTCCAA |
Reverse: TAGCACCCTTTCGGCCTTTA |
PIK3R1 | Forward: GCGAGATGGCACTTTTCTTGT |
Reverse: TACTTCGCCGTCCACCACTAC |
PIK3R3 | Forward: GATGCCCTATTCGACAGAACTGA |
Reverse: TTGGAACTGCTGAAGTCATTGG |
PTGS2 | Forward: CCCTTGGGTGTCAAAGGTAA |
Reverse: GCCCTCGCTTATGATCTGTC |
RAC1 | Forward: AAGCTGACTCCCATCACCTATCCG |
Reverse: CGAGGGGCTGAGACATTTACAACA |
VCAM1 | Forward: GGGAAGATGGTCGTGATCCTT |
Reverse: TCTGGGGTGGTCTCGATTTTA |
Western blot
All experiments were performed as previously described [
23]. After transfer to nitrocellulose membranes, the following primary antibodies were used: α-actinin (Merck Millipore, Billerica, MA), β-catenin (SANTA CRUZ BIOTECHNOLOGY, Dallas, Texas), Akt (Cell Signaling Technology, Beverly, MA), phospho-Akt (Ser473) (Cell Signaling Technology), PI3K (SANTA CRUZ BIOTECHNOLOGY), phospho-PI3K p85 (Tyr458)/p55 (Tyr199) (Cell Signaling Technology), PARP (Cell Signaling Technology) and Caspase 3 (Cell Signaling Technology). Chemiluminescent signals were detected by the Fujifilm LAS-3000 system (Fujifilm, Tokyo, Japan), and β-actin and α-tubulin (Sigma) (Merck Millipore) were used as the loading control. To determine the Rac-1 activity, the Active Rac1 Pull-Down and Detection Kit was used, according to the manufacturer’s protocol (Thermo Fisher Scientific).
Microarray analysis
The mRNA profiles of HUVECs co-cultured with or without CL1-5 cells were analysed using the Affymetrix Human Genome U133 Plus 2.0 GeneChip according to the manufacturer’s protocols (Santa Clara, CA) by the National Taiwan University Microarray Core Facility for Genomic Medicine. The raw data were analysed by GeneChip Operating software (GCOS). Pathway analyses of the differentially expressed genes were performed using the DAVID programme [
24].
Statistical analysis
The data were presented as the means ± standard deviations, and the significance of differences was analysed using Student’s t-test. All experiments were performed in triplicate. To evaluate the prognostic ability of the selected candidate genes, we examined their association with clinical data using a published microarray dataset GSE30219 [
25]. The intensity values of the probes from expression profile of GSE30219 were first rescaled using a quantile normalization method, and then log-transformed to a base-2 scale. The prognostic test was performed as previously described [
15]. Briefly, the differentially expressed genes were employed to test its association with overall survival and disease-free survival by univariate Cox proportional hazard regression analysis. For those genes with a significant Cox regression coefficient, a risk score method was used to calculate the signature and construct the risk score function. The risk score function was a linear combination of gene expression weighted by the regression coefficient from Cox regression. The median risk score was used as the cut-off point for patient classification. Kaplan-Meier method was used to estimate survival curve and difference between curves was evaluated by log-rank test. Multivariate Cox proportional hazards regression analysis was employed to evaluate independent prognostic factors, and age, gender and stage were used as covariates. All statistical tests were two-tailed and
P < 0.05 was considered statistically significant.
Discussion
The tumour microenvironment is a complex network composed of ECM, signalling molecules and different types of stromal cells, including infiltrating immune cells, cancer-associated fibroblasts, endothelial cells and pericytes [
31]. They all play their own role to support tumour growth and restrict drugs from targeting the tumour centre. During tumour progression, malignant cells could also affect the microenvironment in many ways to promote immune tolerance and angiogenesis and sustain proliferative signalling. This influence makes for favourable surroundings and benefits tumour survival, proliferation and metastasis in cancer development [
32]. In the present study, we found that the interaction between endothelial and lung cancer cells changed the HUVECs into a mesenchymal-like morphology, decreased the apoptotic percentage and increased the migration and tube forming ability of HUVECs. Using microarrays and the DAVID bioinformatics database, we identified several genes and signal transduction pathways affected in HUVECs after co-culture with CL1-5 cells. These differentially expressed genes were in accordance with the biological phenotype changes of HUVECs. Manipulating PI3K and COX-2 activity by inhibitors could reverse the tube formation ability and apoptotic resistance of HUVECs. These results could help to identify the feasible therapeutic target candidates for lung cancer anti-angiogenesis therapy in the future.
Angiogenesis is crucial in normal physiology. However, the imbalance of tumour angiogenesis activity is correlated with malignant tumour growth and metastasis in human cancers. To target angiogenesis in cancer therapy, the characterization of tumour-derived endothelial cells (TEC) from normal endothelial cells (NEC) is important. It has been shown that human hepatocellular carcinoma-derived endothelial cells showed increased proliferation, apoptosis resistance, migration and tube formation ability compared with NEC [
33]. A previous study also showed that A549 conditioned medium (CM) increased HUVECs cell survival and wound healing migration ability and decreased apoptosis via Akt activation. Knockdown of Akt or treatment with the PI3K inhibitor wortmannin blocked A549 CM-induced cell survival and the migration ability of HUVECs [
34]. Although cancer cells do not contact with endothelial cells in the Transwell or CM co-culture system, cancer cells can produce many growth factors and cytokines, such as VEGF, basic fibroblast growth factor (bFGF) and IL-8, in the medium to promote angiogenesis. VEGF is the most potent angiogenic factor that can stimulate endothelial cell proliferation and migration and decrease apoptosis via VEGF-VEGF receptor 2 signalling pathway. It also induces the adhesion molecules, ICAM1 and VCAM1, expression. In addition, bFGF not only modulates the expression of integrins but also stimulates VEGF secretion [
35‐
37]. Interestingly, our data showed that some of the above molecules are also upregulated in HUVECs co-cultured with lung cancer cells.
In the present study, we also found that interaction with lung cancer cells changed the HUVECs into a longer, mesenchymal-like morphology and increased the migration, anti-apoptosis and tube formation ability (Fig.
1). However, the proliferation rate did not change in our system (data not shown). Furthermore, we identified that not only Akt but PI3K and Rac-1 were activated in HUVECs after co-culture with CL1-5 cells (Fig.
3a, c). In the previous report, the authors only investigated the apoptosis resistance of NEC or A549 CM-treated HUVECs by inoculating endothelial cells in culture plates in serum starvation. Our data provided further evidence that the interaction with cancer cells decreased the apoptosis percentage of HUVECs on Matrigel, which is more similar to ECM than the culture dish only. Furthermore, the tube formation assay is closer to the process of angiogenesis in physiological conditions than the wound healing assay to mimic vessel formation.
PI3K/Akt signalling elicits many downstream signalling pathways, which are related to cell growth, cell survival, cell cycle, apoptosis, cell motility, glucose metabolism and angiogenesis [
38]. Our study revealed that the interaction with cancer cells increased the PI3K and Akt mRNA levels of HUVECs, as well as their activities. This activation turned on the downstream expression of genes such as β–catenin and stimulated tube formation and apoptosis induction (Figs.
1c, d and
2a). Previous studies reported that β3 integrin induces calpain-dependent integrin cluster formation, triggers Rac-1 activation and ultimately leads to the formation of Rac-1-induced focal complexes. Rac-1 is an important regulator of actin polymerization at the cell’s leading edge. Knockdown of α-actinin impairs Rac-1 induced dorsal stress fibre formation [
39,
40]. Here, we showed that the mRNA of integrin alpha V, integrin beta 3, α-actinin and Rac-1, as well as the protein level of α-actinin and active Rac-1 of HUVECs, were elevated after interaction with cancer cells (Figs.
2b and
3b, c). These changes resulted in filopodia and lamellipodia formation and the increase of migration (Figs.
1b and
3d, e). COX-2, IL-8, ICAM1 and VCAM1 are angiogenesis-related genes [
37]. IL-8 induces ICAM1, VCAM1 and COX-2 expression [
36]. Celecoxib decreases ICAM1 and VCAM1 expression in HUVECs and inhibits HT29 cells’ adhesion to endothelial cells [
41]. Our data showed that the interaction of cancer and endothelial cells increases COX-2, IL-8, ICAM1 and VCAM1 mRNA levels of HUVECs (Fig.
2c). The inhibition of COX-2 by celecoxib prevented tube formation when co-cultured with cancer cells (Fig.
4a).
COX-2 is overexpressed in many cancers and is associated with angiogenesis through the Rac/Cdc42 and PI3K-Ras signalling pathway [
42‐
44]. It has been shown that COX-2 cross-talks with PI3K/Akt in epithelial ovarian cancer. The inhibition of COX-2 by aspirin and COX-2 siRNA decreased Akt phosphorylation and the downstream signalling pathway [
45]. In angiogenesis, PI3K/Akt activation induces VEGF production in cancer cells. Once VEGF binds to the receptor, this binding also turns on the PI3K signalling pathway in endothelial cells and causes actin reorganization and cell migration and finally increases angiogenesis. Here, we also showed that the mRNA level of COX-2 and PI3K and the quantity of active PI3K protein was increased in HUVECs after co-culture with cancer cells (Figs.
2 and
3a). When we treated HUVECs with PI3K and COX-2 inhibitors in the co-culture system, the angiogenesis ability of HUVECs was reduced (Fig.
4). This indicated that cancer cells could influence the endothelial cell phenotype via the PI3K and COX-2 pathways, at least in our experimental conditions. Taken together, COX-2 and PI3K could not only be activated in cancer cells but also in endothelial cells to induce angiogenesis through the interaction of cancer and endothelial cells. This cross-talk creates a microenvironment beneficial for tumour growth and finally makes tumour cells malignant.
In clinical outcome analysis, tumour samples are typically collected to determine the gene expression profiles and survival information. However, the tumour mass was composed not only of cancer cells but also of other surrounding cell types, including endothelial cells, immune cells and stroma cells, which exist in the tumour microenvironment [
8]. Thus, we tried to use the differentially expressed genes in HUVECs after co-culture with cancer cells to predict the clinical outcome. Our data indicated that these cancer cell-stimulated genes could estimate patients’ overall survival and disease-free survival (Fig.
5). This result suggested that these genes were also important in cancer cells. The cause may be the cross-talk between cancer and the microenvironment as previously described. Furthermore, we also compared the predictive power between our HUVECs derived gene signatures and the published prognostic gene signatures, 8-gene [
46] and 7-gene [
47], using the same dataset GSE30219 as ours. To perform the comparisons appropriately, the 8- and 7-gene signatures were employed to construct the risk score functions according to our method and then subjected to overall survival analyses. Interestingly, the re-calculated log-rank tests were more significant (
P < 0.0001 and
P = 0.0045, respectively; Additional file
1: Figure S1a and S1b) compared to the original studies (
P = 0.000513 and
P = 0.0071, respectively) in Kaplan-Meier estimates. The results also showed that both our gene signatures could predict patient prognosis as well as the published 8-gene signature did and better than the 7-gene signature in early-stage lung adenocarcinoma (
P < 0.0001 and
P = 0.0029, respectively; log rank test; Additional file
1: Figure S1c and S1d). It is worth mentioning that our 5-gene signature has the fewest genes among the signatures compared here. Therefore, taking the gene expression profile of endothelial cells into account may improve the prognostic prediction.