Network analysis and the impact of Aflibercept on specific mediators of angiogenesis in HUVEC cells

Abstract Angiogenesis, inflammation and endothelial cells’ migration and proliferation exert fundamental roles in different diseases. However, more studies are needed to identify key proteins and pathways involved in these processes. Aflibercept has received the approval of the US Food and Drug Administration (FDA) for the treatment of wet AMD and colorectal cancer. Moreover, the effect of Aflibercept on VEGFR2 downstream signalling pathways has not been investigated yet. Here, we integrated text mining data, protein‐protein interaction networks and multi‐experiment microarray data to specify candidate genes that are involved in VEGFA/VEGFR2 signalling pathways. Network analysis of candidate genes determined the importance of the nominated genes via different centrality parameters. Thereupon, several genes—with the highest centrality indexes—were recruited to investigate the impact of Aflibercept on their expression pattern in HUVEC cells. Real‐time PCR was performed, and relative expression of the specific genes revealed that Aflibercept modulated angiogenic process by VEGF/PI3KA/AKT/mTOR axis, invasion by MMP14/MMP9 axis and inflammation‐related angiogenesis by IL‐6‐STAT3 axis. Data showed Aflibercept simultaneously affected these processes and determined the nominated axes that had been affected by the drug. Furthermore, integrating the results of Aflibercept on expression of candidate genes with the current network analysis suggested that resistance against the Aflibercept effect is a plausible process in HUVEC cells.


| INTRODUC TI ON
Angiogenesis, process of the formation of new blood vessels from pre-existing ones, has a fundamental role in physiological conditions and different diseases. 1 VEGFA is the main key driver of angiogenesis signalling pathway and has different functions in this process.
Binding VEGFA to VEGFR2 seems to mediate several roles including induction of angiogenesis and proliferation of endothelial cells. 1 Therefore, blockade of the VEGFA-VEGFR2 signalling pathway is an important key target for developing an anti-angiogenic therapeutic system. 2 There are several mechanisms for developing VEGF (signalling pathway)-targeted agents that contain VEGF neutralizing antibodies (eg Aflibercept (Eylea)), tyrosine kinase inhibitors (eg Sorafenib) and antibodies that inhibit signalling pathway through binding to VEGFRs (Ramucirumab). 2 Binding VEGFA to its receptor (VEGFR2) leads to the initiation of different downstream signalling pathways such as PI3K-AKT-mTOR module, Ras/Raf/MEK/ERK signalling cascade, PLC-PKC, Rhoa, STAT, NF-κB, JNK, FAK and RAK modules. Moreover, there are some cytokines and matrix proteins involved in the angiogenic process. 3 Although the role of angiogenesis, inflammation and matrix-related proteins in different diseases is well established by prior studies, little attention has been paid to determine the multiple key proteins, important molecular pathways and interactions between them via system biology approaches.
Aflibercept has received the approval of the US Food and Drug Administration (FDA) for the treatment of wet AMD and colorectal cancer. 4,5 It was recruited in the study since it is a novel recombinant fusion protein that acts as a soluble decoy receptor and could bind to all isoforms of VEGFA, VEGFB and placental growth factor (PLGF). 6 Although the anti-VEGF effect of Aflibercept in pathologic angiogenesis is quietly clear, its possible role in physiologic angiogenesis state, components of VEGFR2-dependent downstream signalling pathways, cytokines and matrix protein have not been investigated yet.
Protein-protein interaction (PPI) network is one of the important tools for a comprehensive investigation of complicated biological processes in living cells. Identification of the key nodes via network analysis with topological features such as degree centrality, closeness, betweenness, centroid value, bridging, eccentricity and eigenvector centrality index enables us to determine novel therapeutic targets related to specific diseases. 7 Degree centrality is the most commonly used local parameter for identifying the regulatory importance node based on the number of edges connected to it. Highly connected nodes interact with multiple proteins and play a fundamental regulatory role in a wide range of biological activities such as signalling module coordination, amplification and gene expression pathways. [8][9][10][11] However, it is well established that the nodes with low connectivity could also mediate important functions in the proteinprotein interaction network. [12][13][14][15][16][17] This is due to the fundamental importance of these nodes in other centrality parameters which are involved in topological network analysis. Betweenness centrality is used to determine key nodes in the maintenance of the functionality and coherence of biological networks. This centrality is determined based on the shortest paths that are used to delineate the number of times that a distinct node is applied to keep distant proteins connected. Therefore, betweenness centrality identifies key nodes that are observed with a high proportion in paths, between other nodes of the network. 13,[18][19][20][21] Centroid values play a fundamental role in orchestrating the activity of distinct protein clusters. Indeed, coordination of highly connected proteins and organizing functional units are performed by nodes with highly centroid values. 22 The probability of functional association of one protein with others in the biological network evaluates by the reciprocal of the sum of the geodesic distances of a specific node to the entire network. This feature is called closeness centrality. Nodes with high closeness centrality are showing critical regulatory effect on other proteins. Moreover, any changes in the network are more likely affecting these types of proteins. 20,23 The identification of specific nodes which are easily reachable by other proteins occurs by reciprocal of the maximum of shortest path lengths. This concept is called eccentricity centrality.
As a consequence, nodes with high eccentricity are easily affected or exposed to other proteins. 20,24 Bridging centrality consists of the betweenness centrality and bridging coefficient. Nodes with high bridging coefficient have highly connected first neighbours. As a result, bridging centrality is used to determine nodes that link (due to betweenness centrality components) clusters or densely connected regions (due to bridging coefficient component). 13 Eigenvector centrality is used to distinguish central super regulatory nodes in biological networks. These nodes represent key targets in gene regulatory pathways. Nodes with high eigenvector centrality are identified by their position and the neighbouring nodes. 20,[25][26][27] Despite the distinct importance of each centrality in the biological network, for more accurate identification of the crucial proteins, all results of centralities should be integrated without any preferences. It is important to consider that high score protein in multiple centrality parameters represents great importance in the functionality of a biological network. Our ontology analysis results demonstrated that tumour necrosis factor-alpha (TNFα) signalling pathway is the most enriched pathway related to this interrelation network.
Tumour necrosis factor-alpha (TNFα) and its receptors (TNFRs) trigger several signalling pathways that regulate different cellular functions such as inflammatory gene expression, cell proliferation and programmed cell death. 28 Also, regulation of regulatory T cells (Tregs) function, endothelial cell adhesion and permeability and accumulation of immune cells including lymphocytes and monocytes to regions of inflammation was performed by TNF signalling pathway. 29 Here, we integrated text mining data, 3 angiogenesis-related protein-protein interaction networks 30,31 and multi-experiment microarray data 32-34 to find out candidate genes involved in VEGFA/ VEGFR2 signalling pathways. We then aimed to do topological analysis of candidate genes' biological networks to determine selected genes and to search for in vitro effects of Aflibercept on expression of selected genes in endothelial cells. Different kinds of endothelial cells such as SVEC4-10 (mouse), 3B-11 (mouse) and HUVEC (human) cells could be evaluated in the assessment. We selected the HUVECs (human umbilical vein endothelial cells). Anyway, the routine method for tube formation in the context of all the three cell types is the same. 35 HUVECs have been extensively used as a primary, nonimmortalized cell model to study how different manipulations and pro-and anti-angiogenic compounds affect endothelial cells' migration and proliferation, and how this regulates the formation of blood vessels. 36 Although studies in HUVECs do not represent all endothelial cell types found in an organism, HUVECs are an excellent model for the study of vascular endothelium properties and the main biological pathways that are involved in endothelium function. 37 It has been previously established that, in endothelial cells, endogenous VEGFA forms complex with VEGFR2 and leads to trigger

| Angiogenesis, inflammation and matrix protein-related genes (data sources)
A literature review from different kinds of studies, which included text mining, microarray data results and protein-protein interaction networks, was performed to obtain the information not only associated with physiological angiogenesis and VEGFA/VEGFR2 downstream signalling pathway-related proteins but also on the inflammatory and matrix proteins with important roles in angiogenesis.
This process was carried out by using the keywords: angiogenesis, VEGFA, vascular endothelial growth factor, angiogenesis signalling pathways, angiogenesis signalling network, MMP, matrix metalloproteinase, VEGFR2 and inflammation. The results were combined, and proteins were selected with the highest repetition frequencies in all kinds of studies and then defined as the seed proteins (Table 1).

| Network construction
To investigate the functional interrelation of 34 candidate genes, their binding proteins and associated signalling pathways, the protein-protein interaction network was reconstructed at Homo sapiens organism by version 3.5.1 of GeneMANIA plug-in implemented in Cytoscape software. There are different types of evidence mode as a prominent graph clustering algorithm was applied to determine densely associated regions and visualize them in biological network.

TA B L E 1 Seed proteins
This algorithm is implemented in three steps: (a) vertex weighting by the local network density which is performed based on clustering coefficient, (b) prediction of molecular complexes based on locally dense seed protein which is assigned by the highest vertex weighted and (c) adding or removing proteins due to post-processing. 41 The tendency of a graph to form different clusters (molecular complexes) is determined through the clustering coefficient parameter. 20,42 The Clustering Coefficient of i is calculated by the equation More dense complex is placed in the higher rank at the results. 41

| Enrichment analysis
To

| Topological network analysis
To analyse topological feathers of the network, centrality measurement was performed by CentiScaPe (version 2.2) and Gephi (version 0.9.2) to determine nodes with great centrality or topological importance. These nodes usually called hubs and had a fundamental role in different kinds of networks. Several types of centrality were analysed such as degree, betweenness, centroid value, closeness, bridging, eccentricity and eigenvector centrality. Then, we categorized each node according to its position in these seven states and determined important targets which is crucial for the integrity of these networks.

| HUVEC cell culture and treatment
The less than 2 and more than 10 passage number of the cells is appropriate in many functional assays such as maximum tube formation and run experiments in optimum condition. In a standard protocol, it seems that the downstream signalling pathways and molecular mechanisms that are involved in the angiogenic process and the results obtained from HUVEC cells can be generalized to

| Networks of angiogenesis, inflammation and matrix proteins
Data of the specific genes of VEGFA/VEGFR2 downstream signalling pathways, inflammatory and matrix-related proteins were obtained from text mining, microarray data results and protein-protein interaction networks, and then, the proteins with the highest repetition in all kinds of studies (=candidate genes) were utilized to reconstruct a PPI network which included angiogenesis, inflammation and matrix-related proteins simultaneously ( Figure 2). The MCODE plugin implemented in Cytoscape software was applied to determine the densely interrelated regions (clusters) in the network. As shown in

| Enrichment analysis
To determine the most relevant concepts behind the GO terms, enrichment analysis was performed on the interrelation network. which is displayed in dark red nodes in Figure 4 for more detail, see Tables S1 and S2.

| Identification of hubs
The CentiScaPe plug-in was applied to identify the most crucial nodes in a network and determine different kinds of centrality indexes for each node. Scatter plots of nodes included highest degree with betweenness centrality, closeness centrality, centroid value, bridge node and eigenvector centrality were delineated separately, Figure 5 and Figure 6 for more detail, see Table S3, Figures S1 and S2. The results obtained from CentiScaPe were categorized from the highest to the lowest scores to determine key genes in different centralities (Table S4). To facilitate the overall assessment of distinct gene status, according to Table S4 results, we assigned ranks 1-54 to the recruited genes by considering their order in different centralities (for more detail, see Table S5). We considered nodes that showed up to a rank about 25 as important nodes in the network.
Therefore, the importance of all genes in different kinds of centrality F I G U R E 2 Interrelation network which included angiogenesis, inflammation and matrix-related proteins was evaluated and the expression pattern of selected genes was investigated by qPCR in Aflibercept-treated HUVEC cells in in vitro cultures.

| D ISCUSS I ON
MTT assay demonstrated more than 95% viability in vascular endothelial cells during application of the peak serum concentration of Aflibercept (0.45 nmol/L). 47 It is well established that the Aflibercept does not affect the viability of a variety of ocular cells. 95 So, there was no need for cell viability assessment at this concentration of Aflibercept.

CO N FLI C T O F I NTE R E S T
The authors confirm that there are no conflicts of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.