Introduction
Cystoid macular edema (CME) is a retinal condition characterized by fluid accumulation in cystoid spaces within the macula, leading to visual impairment. CME often occurs as a complication of various eye diseases, including diabetic retinopathy, uveitis, retinal vein occlusion, and postoperative inflammation. The main process behind CME involves breaking down the blood-retinal barrier (BRB), increasing blood vessel permeability, and causing inflammation and changes in retinal structure.
Although treatments such as corticosteroids, anti-VEGF agents, and nonsteroidal anti-inflammatory drugs are used, CME frequently recurs and can be resistant to therapy, making it a significant clinical challenge [1,2].
While individual molecules such as vascular endothelial growth factor A (VEGFA), interleukins, and matrix metalloproteinases have been linked to CME, the disease’s pathophysiology goes beyond these isolated pathways. It involves complex molecular interactions among many proteins that promote inflammation, angiogenesis, immune cell recruitment, and BRB dysfunction. Understanding these detailed molecular relationships is essential for discovering new therapeutic targets and improving existing treatment strategies [3,4].
Protein-protein interaction (PPI) networks offer a systems-biology approach to understanding the organizational principles underlying disease-related molecular mechanisms. By depicting biological systems as networks of interacting proteins, PPI analysis provides insights into key regulatory hubs, important signal transduction mediators, and functional modules [5].
This network-centric perspective helps identify proteins of high topological significance that could serve as potential biomarkers or therapeutic targets in complex disorders such as CME.
In this study, we constructed and analyzed a PPI network associated with CME using high-confidence interaction data retrieved from the STRING database. We selected a curated panel of CME-associated proteins based on a literature review, including VEGFA, IL6, TNF, ICAM1, MMPs, chemokines, and tight junction regulators [6-8].
We evaluated the network’s topological characteristics using multiple centrality metrics, performed a modularity analysis with the Louvain algorithm, and conducted pathway enrichment analyses to infer functional relevance. This integrated approach sheds light on the molecular landscape of CME, highlighting critical proteins and biological processes that may inform future mechanistic studies and therapeutic developments.
Materials and methods
Literature-Based Identification of CME-Associated Proteins
To establish a biologically grounded foundation for network analysis, we began with a comprehensive literature review to identify proteins implicated in the pathophysiology of cystoid macular edema (CME). Searches were conducted using PubMed and Scopus databases with combinations of the following keywords: “cystoid macular edema”, “inflammation”, “angiogenesis”, “vascular permeability”, “cytokine”, “growth factor”, and “hypoxia”. Proteins were considered for inclusion if there was experimental or clinical evidence linking them to CME or related ocular diseases such as diabetic retinopathy, retinal vein occlusion, or uveitis.
Only human protein-coding genes were considered, and gene symbols were verified via the UniProt and GeneCards databases. The final curated list of 18 seed proteins spanned multiple mechanistic domains, including angiogenesis (VEGFA, ANGPT2, KDR, FLT1, PDGFB), inflammation and cytokine signaling (IL6, TNF, CXCL8, CXCL10, CCL2), vascular permeability (ICAM1, EDN1), oxidative stress and hypoxia (HIF1A, NOS2), matrix remodeling (MMP2, MMP9), and additional regulators such as TGFB1 and PTGS2 [9-13].
Retrieval of Protein-Protein Interaction Data
Protein-protein interaction (PPI) data for the seed proteins were obtained from the STRING database (version 12, https://string-db.org/), using its RESTful API and the Python requests library. The analysis was limited to Homo sapiens (taxonomy ID 9606), and only interactions with a combined STRING confidence score of 0.700 or higher were included to ensure high reliability. The returned dataset included interaction partners, preferred gene names, and combined interaction scores, integrating data from experimental results, curated databases, co-expression patterns, and computational predictions.
Network Construction and Visualization
The retrieved interaction data were used to build a weighted, undirected graph using the NetworkX library in Python. In this graph, each node represented a protein, while each edge indicated a high-confidence interaction between two proteins, weighted by the STRING combined score. Isolated proteins (with no high-confidence interactions) were excluded from the final network.
The resulting graph was visualized using spring layout algorithms to organize spatial relationships and served as the basis for topological and functional analyses.
Topological Characterization of the Network
To assess the structure and connectivity of the CME-associated protein network, we calculated several centrality and global network metrics. Degree centrality was used to identify hub proteins with the highest number of direct connections. Betweenness centrality quantified the extent to which proteins acted as bridges along the shortest paths between other nodes.
Closeness centrality was used to evaluate how quickly information could spread from a node to others. Eigenvector centrality measures a protein’s influence based on the importance of its neighbors. All centrality metrics were normalized to the range 0-1, with higher values indicating greater relative importance or connectivity within the network.
Local clustering coefficients are calculated to assess the tendency of proteins to form tightly interconnected groups. Global metrics, including graph density, average clustering coefficient, network diameter, and average shortest path length (computed over the largest connected component), provided a broader view of network architecture and navigability.
Community Detection and Functional Clustering
To uncover potential biological subsystems within the network, we employed the Louvain community detection algorithm via the community (python-louvain) package. This unsupervised method optimizes network modularity by grouping nodes into non-overlapping communities with dense internal connections. Each node was assigned to a community, and the number, size, and connectivity of the resulting modules were analyzed to detect functionally coherent clusters.
These communities potentially correspond to biological pathways or co-regulated protein complexes relevant to CME.
Functional Enrichment Analysis
The biological significance of the proteins in the extended network was evaluated using gene set enrichment analysis (GSEA) via the Enrichr platform, accessed through the GSEAPY Python package (version 1.0.5). The full set of proteins (including first-degree interactors from STRING) was submitted to three annotation databases: Gene Ontology Biological Process 2021 (GO: BP), KEGG 2021 Human, and Reactome Pathways 2022. Enrichment significance was determined using Fisher’s exact test, and multiple comparisons were adjusted using the Benjamini-Hochberg method.
Gene sets with adjusted p-values less than 0.05 were considered statistically significant. For the top enriched terms, overlap statistics and adjusted p-values were reported, and –log10-transformed adjusted p-values were visualized as bar plots to highlight the most relevant biological processes and pathways.
Computational Environment and Reproducibility
All analyses were conducted in a Python 3.10 environment using Jupyter Notebook for interactive development and documentation. The primary computational libraries included NetworkX for graph modeling and analysis, community for modularity detection, GSEAPY for enrichment analysis, and Matplotlib/Seaborn for data visualization. STRING API requests were handled using the requests package, and data wrangling was performed with Pandas and NumPy.
The entire computational pipeline was version-controlled and documented to ensure reproducibility. The full source code and instructions for replicating the analysis are available at: https://github.com/MehrdadMotamed-lab/CME_PPIN/blob/main/CME_PPIN2.ipynb.
Results
Network Construction and General Topology
The final protein-protein interaction (PPI) network, constructed using 18 seed proteins and their high-confidence first-shell interactors from the STRING database, consisted of 91 nodes and 217 edges (Fig. 1). Each node represents a protein, and edges indicate interactions with a STRING combined confidence score ≥ 0.700.
The network exhibited a graph density of 0.0505, reflecting a sparsely connected but biologically relevant interaction structure. The average clustering coefficient was 0.0554, suggesting a modest tendency for proteins to form local clusters. The diameter of the largest connected component was 4, and the average shortest-path length was 2.97, indicating efficient communication among proteins within the network (Table 1).
Centrality Metrics Identify Key Regulatory Proteins
Quantitative topological analysis was performed to identify proteins of high regulatory significance. According to degree centrality, which measures the number of direct connections a node has, the top five proteins were ANGPT2 (0.2097), FLT1 (0.1935), TNF (0.1774), KDR (0.1774), and VEGFA (0.1452). These proteins serve as network hubs, potentially controlling information flow across multiple biological modules.
Betweenness centrality, which captures the extent to which a protein lies on the shortest paths between other proteins, identified ANGPT2 (0.1899), FLT1 (0.1629), KDR (0.1456), CXCL8 (0.1399), and PTGS2 (0.1329) as major bridging proteins. These nodes are likely involved in coordinating distinct signaling pathways relevant to CME pathogenesis.
Closeness centrality, indicating the speed at which a protein can influence the entire network, was highest for ANGPT2 (0.4769), FLT1 (0.4662), KDR (0.4627), VEGFA (0.4429), and TNF (0.4397). These findings suggest that angiogenic regulators and inflammatory cytokines are optimally positioned to propagate signals throughout the CME network rapidly.
In terms of eigenvector centrality, which measures a node’s influence based on the connectivity of its neighbors, the most central proteins were ANGPT2 (0.3315), FLT1 (0.3011), KDR (0.2964), TNF (0.2920), and HIF1A (0.2605). These results further underscore the functional importance of these molecules, particularly in the context of angiogenesis, inflammation, and hypoxia-related signaling.
Modular Structure Reveals Distinct Functional Communities
Using the Louvain community detection algorithm, the network was partitioned into eight non-overlapping communities. These included clusters ranging from five to fourteen nodes, indicating modular organization within the CME interactome. The largest community (14 nodes) predominantly included cytokine receptors and inflammatory mediators, while other communities captured proteins involved in angiogenic signaling, endothelial migration, matrix remodeling, and tight junction regulation.
This modular arrangement reflects the multifactorial nature of CME pathophysiology, encompassing interlinked biological systems.
Functional Enrichment Highlights Inflammatory and Angiogenic Mechanisms
Gene ontology and pathway enrichment analysis were conducted on the full network to reveal underlying biological processes. Analysis of the GO: Biological Process (2021) dataset identified several highly significant terms. The top five enriched processes included cytokine-mediated signaling pathway (adjusted p = 1.94×10−15; 14/621 genes overlapped), cellular response to cytokine stimulus (adjusted p = 3.21×10−15; 13/482), positive regulation of MAPK cascade (adjusted p = 2.78×10−12), regulation of blood vessel endothelial cell migration (adjusted p = 6.60×10−12), and positive regulation of smooth muscle cell proliferation (adjusted p = 4.07×10−10).
These terms suggest coordinated regulation of immune activation, intracellular signaling, and vascular remodeling (Fig. 2).
In the KEGG 2021 Human pathway analysis, the most enriched pathways were the AGE-RAGE signaling pathway in diabetic complications (adjusted p = 8.28×10−15; 9/100), fluid shear stress and atherosclerosis (adjusted p = 8.78×10−14), rheumatoid arthritis (adjusted p = 2.95×10−13), TNF signaling pathway (adjusted p = 1.03×10−12), and pathways in cancer (adjusted p = 2.94×10−12). These pathways collectively highlight the intersection of systemic inflammation, vascular dysfunction, and chronic disease in CME.
Reactome Pathway enrichment analysis further reinforced the dominance of cytokine and interleukin signaling. The most significant terms were interleukin-4 and interleukin-13 signaling (adjusted p = 2.76×10−19; 11/107), signaling by interleukins (adjusted p = 1.92×10−14), interleukin-10 signaling (adjusted p = 2.76×10−13), cytokine signaling in the immune system (adjusted p = 1.80×10−12), and signal transduction (adjusted p = 1.01×10−8). These results confirm the central role of cytokine networks in orchestrating the inflammatory environment of CME.
Discussion
Cystoid macular edema (CME) is a multifactorial and vision-threatening condition resulting from the pathological accumulation of fluid within the macular region of the retina. It often arises as a complication of various retinal and systemic diseases, including diabetic retinopathy, retinal vein occlusion, uveitis, and postoperative inflammation [13]. Despite the availability of anti-VEGF therapies, corticosteroids, and NSAIDs, CME frequently presents therapeutic resistance and recurrence, underscoring the need for a deeper understanding of its molecular underpinnings [14].
In this study, we applied a systems-biology approach centered on protein-protein interaction (PPI) network analysis to investigate the molecular mechanisms and functional architecture of CME. By integrating curated literature-based protein selection, network modeling using STRING and NetworkX, topological analysis, community detection, and pathway enrichment, we delineated the complex molecular landscape that underlies CME pathophysiology.
Our network comprised 91 nodes and 217 high-confidence edges, representing both seed proteins and their interactors. The graph density (0.0505) and average clustering coefficient (0.0554) indicated a moderately sparse yet biologically meaningful structure, suggesting that CME-related proteins do not operate in isolation but instead form specialized interaction modules. This architecture reflects the inherently complex and modular nature of chronic inflammatory and vascular retinal diseases.
Centrality analysis provided a key insight into the topological prominence of several proteins, most notably ANGPT2, FLT1, KDR, TNF, VEGFA, and IL6. ANGPT2, or angiopoietin-2, emerged as the most influential protein across all major centrality metrics, including degree, betweenness, closeness, and eigenvector centrality. ANGPT2 is a context-dependent antagonist or agonist of the TIE2 receptor and plays a pivotal role in vascular destabilization and endothelial permeability, both hallmarks of CME [15].
Its high degree and betweenness indicate its broad connectivity and regulatory potential, acting as a molecular bridge between angiogenesis and inflammatory modules [16]. These findings align with reports showing elevated vitreous ANGPT2 levels in patients with diabetic macular edema and other neovascular retinal diseases [17,18].
Closely trailing ANGPT2 in centrality were FLT1 (VEGFR-1) and KDR (VEGFR-2)—two key receptors for VEGFA. The inclusion of both receptors among the top central nodes reinforces the centrality of the VEGF signaling axis in CME, particularly in its vascular leakage phenotype. Notably, VEGFA itself showed a high degree and closeness centrality, further supporting the hypothesis that this growth factor operates at the center of the CME molecular network [19,20].
These observations support the continued therapeutic relevance of anti-VEGF agents in CME treatment and suggest that VEGF-independent nodes, such as ANGPT2, may serve as alternative or adjunct therapeutic targets, particularly in cases refractory to anti-VEGF therapy.
Interestingly, TNF and IL6—two well-established inflammatory cytokines—also ranked highly in network centrality. TNF showed high betweenness and eigenvector centrality, suggesting its role as a critical hub linking inflammatory and vascular dysfunction modules [21,22]. Similarly, IL6, though slightly lower in the overall centrality rankings, remains biologically significant due to its documented involvement in uveitis-associated CME and response to IL6 inhibitors such as tocilizumab [23].
Together, these findings reaffirm the dual contribution of angiogenic and inflammatory signaling axes in CME, consistent with its variable clinical presentation and response to therapy across different etiologies.
Community detection using the Louvain algorithm revealed eight distinct functional modules, each potentially representing discrete biological programs contributing to CME. The largest community contained key cytokines, interleukins, and signaling intermediates, suggesting an inflammatory core within the network. Other communities were composed of angiogenic mediators, tight junction proteins, and hypoxia response factors, suggesting that CME arises from the convergence of multiple biological subsystems rather than a single linear pathway.
This modular structure echoes findings from retinal transcriptomic studies, which have identified simultaneous activation of immune, oxidative, and angiogenic signatures in eyes with edema [24-26].
Functional enrichment analysis further corroborated these insights. Among the top Gene Ontology biological processes, cytokine-mediated signaling and cellular responses to cytokines were the most significantly enriched, followed by MAPK cascade regulation and endothelial cell migration. These results indicate that inflammation, cytokine signaling, and endothelial remodeling form a triad of key biological processes driving CME.
Enrichment of the AGE-RAGE signaling pathway in KEGG, especially in the context of diabetic complications, underscores the relevance of advanced glycation and oxidative stress pathways in diabetes-associated CME. This is particularly relevant given the clinical burden of diabetic macular edema and its often-suboptimal response to standard therapies.
The enrichment of fluid shear stress and atherosclerosis, TNF signaling, and rheumatoid arthritis pathways further supports a systemic inflammatory influence on the retinal microenvironment. These findings highlight potential shared mechanisms between systemic autoimmune diseases and retinal inflammation, suggesting that CME may act as a sentinel marker of systemic immune dysregulation in certain patients.
Reactome analysis provided additional granularity, highlighting interleukin-4, interleukin-10, and interleukin-13 signaling pathways as among the most significantly enriched. These interleukins are central to Th2-mediated immune responses and vascular homeostasis [27]. For instance, IL-4 and IL-13 are known to disrupt endothelial barrier integrity and modulate macrophage polarization, both of which may contribute to CME formation [28].
The prominence of IL-10 signaling, an anti-inflammatory cytokine, suggests an endogenous counter-regulatory response to ongoing inflammation in the retinal microenvironment [29]. The Reactome category “cytokine signaling in the immune system” was also strongly enriched, again reinforcing the idea that CME exists within a tightly regulated immune signaling landscape.
Collectively, these findings present a compelling case for considering CME not merely as a localized ocular phenomenon but as the product of integrated and systemic molecular dysregulation. The strong network connectivity and pathway enrichment of proteins involved in inflammation, angiogenesis, vascular remodeling, and oxidative stress support a multifactorial pathophysiology that varies with the underlying disease etiology. This has therapeutic implications.
While anti-VEGF therapy remains a cornerstone for many CME cases, our network analysis identifies several alternative targets—most notably ANGPT2, IL6, and TNF—which may warrant therapeutic consideration, especially in refractory or recurrent CME.
Moreover, this systems-level approach provides a framework for identifying biomarker candidates. Proteins with high centrality metrics may serve as accessible indicators of disease activity or treatment response, particularly if their expression correlates with clinical severity. For instance, aqueous or serum levels of ANGPT2 or TNF could be evaluated longitudinally in patients undergoing anti-VEGF or steroid therapy to assess predictive value.
However, this study is not without limitations. The network was constructed using high-confidence interactions from the STRING database, which, while comprehensive, is inherently static and does not account for tissue-specific expression, temporal dynamics, or post-translational regulation. Moreover, the PPI model reflects potential physical or functional interactions, not necessarily active pathways in CME-affected tissue.
Thus, future work should incorporate multi-omics data—including single-cell RNA-sequencing and phosphoproteomics—from retinal tissues to refine network relevance. Experimental validation of hub protein expression and interaction dynamics in CME models will also be essential to translate these findings into therapeutic applications.
Conclusion
In summary, our integrative PPI network analysis reveals a complex but coherent molecular architecture underlying cystoid macular edema. The prominence of ANGPT2, FLT1, VEGFA, TNF, and IL6 underscores the importance of both angiogenic and inflammatory signaling pathways. Modular and enrichment analyses point to distinct yet interrelated processes that drive disease, including cytokine-mediated inflammation, endothelial permeability, and tissue remodeling.
These insights may inform the development of more personalized and effective therapeutic strategies for CME, especially for patients unresponsive to conventional treatments. Our findings highlight the value of systems biology in ophthalmic disease research and open new avenues for translational investigation into CME.