Identification of key signaling pathways in cerebral small vessel disease using differential pathway network analysis

Abstract

Cerebral small vessel disease (CSVD) primarily affects the perforating cerebral arterioles and capillaries, and results in injury to subcortical grey and white matter. Despite advances in determining the genetic basis of CSVD, the molecular mechanisms underlying the development and progression of CSVD remain unclear. The present study aimed to identify significant signaling pathways associated with CSVD based on differential pathway network analysis. Combining CSVD microarray data with human protein-protein interaction data and data from the Reactome pathway database, pathway interactions were constructed using the Spearman's correlation coefficient strategy. Pathway interactions with weight values of >0.95 were selected to construct the differential pathway network, which contained 715 differential pathway interactions covering 312 nodes and was visualized using Cytoscape software. A total of 15 hub pathways with a top 5% degree distribution in the differential pathway network were identified. The top 5 hub pathways were associated with the synthesis and metabolism of fatty acids. The results of the present study indicate that the synthesis and metabolism of fatty acids is associated with the occurrence and development of CSVD, and may thus provide insights to improve the early diagnosis and treatment of CSVD.

Introduction

Cerebral small vessel disease (CSVD) predominantly affects the perforating cerebral arterioles and capillaries, and results in injury to subcortical grey and white matter (1). CSVD is associated with focal motor deficits, stroke and cognitive decline, which typically progresses to dementia (2). Genetic studies of CSVD indicate that the development and progression of the disease can be attributed to the accumulation of genomic changes (3). Gene expression profiling has been widely used to research the pathogenesis of diseases, including CSVD. However, despite advances in our knowledge of the genetic basis of CSVD, the underlying molecular mechanisms of the development and progression of CSVD remain unclear.

Pathway analysis is a useful tool for gaining insight into the biological functions of genes and proteins (4). Given the complex nature of biological systems, signaling pathways are typically required in order for systems to function in a coordinated fashion to produce the appropriate physiological responses to internal and external stimuli (5). However, previous studies have focused on identifying altered signaling pathways between normal and diseased groups, and common genes between different signaling pathways. For example, a previous study identified differential interactions between two signaling pathways across diseased and normal samples (6). Network-based methods have been used to analyze this interaction data and gain insights into the underlying molecular mechanisms by which biological systems operate (7). Sun et al (8) introduced a network-based approach, differential expression network analysis, which reflects phenotypic differences at a network level. Similarly, in the present study, a differential pathway network was constructed to conduct analysis on the pathogenesis of CSVD, in which nodes represented signaling pathways. In this way, CSVD was analyzed at a signaling pathway network level.

In the present study, this differential pathway network analysis method was applied to identify key signaling pathways associated with CSVD. CSVD data, including gene expression profiles, protein-protein interactions (PPIs) and signaling pathways, were identified and preprocessed. Differential pathway interactions in CSVD were identified using the Spearman's correlation coefficient (SCC) strategy and a differential pathway network was constructed. Topological analysis of the differential pathway network was performed to identify hub pathways, which revealed 15 hub pathways with a top 5% degree distribution. The results of the present study may aid in the identification of potential biomarkers for the early diagnosis and treatment of CSVD, and provide novel insights into the pathological mechanism underlying this disease.

Materials and methods

Gene expression data recruitment and preprocessing

Gene expression profiles from CSVD samples (accession no. E-MTAB-3408) (9) were downloaded using the ArrayExpress database (http://www.ebi.ac.uk/arrayexpress). The E-MTAB-3408 dataset was collected using the A-AFFY-14 Affymetrix GeneChip™ Human Gene 1.0 ST Array (Thermo Fisher Scientific, Inc., Waltham, MA, USA). The data were obtained from 15 CSVD samples and 15 normal controls.

The oligo package (version 1.14.0) (10) for R software was used to read the data. The robust multiarray average algorithm was applied to conduct background subtraction (11). Quantile normalization and summarization were performed using the median polish procedure, and the data were filtered using the varFilter function in R software (12). Subsequently, all of the gene annotation files were downloaded and all probes were mapped onto the genes. A total of 5,786 genes were contained in the expression profile.

PPI data recruitment and preprocessing

Global PPIs were obtained from the Search Tool of the Retrieval of Interacting Genes/Proteins database (http://string-db.org). Currently, this database covers 932,553,897 total interactions from 2,031 organisms. Here only human PPIs were retrieved, including a total of 787,896 interactions after removing self-loops and duplicated interactions. The gene expression profile was mapped onto the global PPI network and these intersections were selected to construct a novel PPI network, which was regarded as the background PPI network. In this case, a background PPI network with 89,574 interactions covering 4,971 nodes was constructed.

Signaling pathway data identification and preprocessing

All human signaling pathways were downloaded from the Reactome pathway database (http://www.reactome.org), and a total of 1,675 pathways were obtained. Pathways with too many genes may be considered too complex analysis, while pathways with too few genes may not have sufficient biological content (13). Thus, to enhance the confidence and stability of the signaling pathways identified, the number of common genes between each pathway and the background PPI network was calculated, and pathways with common gene size >100 or <5 were discarded. Following this, a total of 706 signaling pathways were obtained for further analysis.

Identifying differential interactions between signaling pathways

To evaluate interactions between signaling pathways, gene interactions were constructed randomly and then evaluated. In addition, these pathway genes were mapped onto the background PPI network. In this way, interactions between signaling pathways were constructed. In order to describe the strengths of signaling pathway interactions, the SCC method (14) was utilized to rank pairwise interactions between the CSVD samples and normal controls. The SCC absolute difference value of a pathway interaction between CSVD and normal samples was defined as the weight value of this pathway interaction.

For example, in pathway 1, genes in pathways 1 and 2 were used to construct gene interactions and these interactions were integrated with the background PPI network. All gene intersections were considered to be interactions of pathways 1 and 2. Subsequently, the pathway intersections between CSVD and normal samples were weighted by SCC, respectively. The SCC of a pair of gene interactions (a and b), was defined as:

$$\mathit{\text{SCC}}(a,b)=\frac{1}{n–1}{\sum }_{i=1}^n\left(\frac{g(a,i)–\overline{g}(a)}{\sigma (a)}\right)·\left(\frac{g(b,i)–\overline{g}(b)}{\sigma (b)}\right)$$

Where n was the number of gene interactions, g(a, i) or g(b, i) was the expression level of gene a or b in the pathway, i indicated a specific condition (CSVD or normal), g(a) and g(b) represented the mean expression level of gene a or b, respectively, and σ(a) and σ(b) represented the standard deviation of the expression level of genes a and b.

The absolute SCC difference of a gene interaction between CSVD and normal conditions was calculated, and the mean value of the absolute SCC differences of all gene interactions between pathways 1 and 2 was defined as the weight value. If there was no intersection between pathway 1 and 2, this indicated that there was no interaction between the two pathways. Similarly, gene interactions were constructed based on genes in pathways 1 and 3, and the method used to explore interactions and the weight value between the pathways was the same as described above. By such analogy, all of the interactions and weight values between any two pathways were obtained.

Differential pathway network construction

Once all pathway interactions and weight values were obtained, the weight values were set in descending order and pathway interactions with weight values >0.95 were considered as differential interactions. These differential interactions were selected to construct a differential pathway network, which was visualized using Cytoscape software (version 2.8) (15).

Centrality analysis

To further investigate the significance of signaling pathways in the differential pathway network, the biological importance of pathways was characterized using indices of centrality. Centrality analysis is a network analysis method to investigate biological networks, including gene regulatory, protein interaction and metabolic networks, in order to identify important elements of a network (16,17). Degree centrality is a simple local centrality measure, which is based on the notion of neighborhood. The degree is useful in statistical graphs to identify vertices that have the most direct connections to other vertices (18). In the present study, the nodes indicated the signaling pathways and the edges indicated the interactions between signaling pathways. The connectivity degree of a signaling pathway was quantified by considering the pathway as a node and the number of adjacent pathways as the degree value. The pathways whose degree values were at the top 5% of the degree distribution (≥95% quantile) in the network were defined as hub signaling pathways.

Network clustering

Genes and signaling pathways that are involved in a similar function are frequently coexpressed or activated, respectively, which establishes conserved transcription modules (19). These modules are groups of genes or pathways whose expression profiles or activations, respectively, are highly correlated across samples (20). In the present study, the signaling pathways in the differential pathway network were clustered using ClusterONE (version 1.0) plugin for Cytoscape, which uses a cohesiveness measurement to determine the likelihood of an overlap between a group of proteins or pathways in a complex is based on the weight of interactions within the group and with the rest of the network (21). During module searching, the three primary parameters were as follows: Module size, weighted density and overlap threshold. In the present study, pathway modules were selected under the thresholds of a module size ≥20, a weighted density ≥0.1 and an overlap threshold ≥0.5.

Results

Pathway network construction

In the present study, microarray data from CSVD samples was combined with human PPI and signaling pathway data to identify differential pathway interactions using the SCC strategy. Pathways whose gene set size was >100 or <5 were discarded, leaving a total of 706 pathways for further analysis. Once these pathways were intersected with background PPI data, the pathway interactions were screened to construct a pathway network. The results demonstrated that these 706 pathways formed 188,420 pathway interactions in total. The distribution of weight values are demonstrated in Fig. 1. It was identified that the weight values of the interactions ranged from 0.2–1.3, and that the majority of interactions ranged between 0.2 and 0.5. Because small weight values indicated small differences in pathway interactions between disease and normal conditions, pathway interactions with weight values of >0.95 were selected to construct the differential pathway network. The differential pathway network constructed included 715 differential pathway interactions and 312 pathways (Fig. 2).

Figure 1.

Weight distribution of pathway interactions. (A) Total weight distribution of pathway interactions. (B) Expanded view of weight distributions between 0.9 and 1.3.

Figure 2.

Differential pathway network in cerebral small vessel disease. Nodes represent pathways and edges indicated the interactions between pathways. Orange nodes indicate hub pathways with a top 5% degree distributions. The details of hub pathways are shown in Table I.

Centrality analysis

In order to further investigate the importance and significance of the 312 differential pathways in the differential pathway network, degree centrality analysis was applied to obtain hub pathways. Using the top 5% degree distribution, a total of 15 hub pathways were identified (Fig. 2). The detailed degrees for the 15 hub pathways are listed in Table I. Notably, signaling pathways for fatty acyl-co-enzyme A (CoA) biosynthesis (node 194), linoleic acid (LA) metabolism (node 320) and the synthesis of very long-chain fatty acyl-CoAs (node 624) produced the highest connectivity degree (degree=54). This was closely followed by α-linolenic (omega 3) and linoleic (omega 6) acid metabolism (node 32), and α-LA (ALA) metabolism (node 33), which scored a degree of 53. These signaling pathways were all associated with the synthesis of fatty acyl-CoA and LA metabolism, which suggests the involvement of these metabolic processes with the development and progression of CSVD.

Table I.

Detailed degrees of the identified hub pathways.

Node	Pathway term	Degree
194	Fatty acyl-CoA biosynthesis	54
320	Linoleic acid metabolism	54
624	Synthesis of very long-chain fatty acyl-CoAs	54
32	α-linolenic (omega 3) and linoleic (omega 6) acid metabolism	53
33	α-linolenic acid metabolism	53
683	Triglyceride biosynthesis	48
318	Laminin interactions	27
252	Glycogen breakdown (glycogenolysis)	23
613	Syndecan interactions	23
183	Ephrin signaling	21
386	Non-integrin membrane-extracellular matrix interactions	15
181	Ephrin A-mediated growth cone collapse	13
128	Cytosolic tRNA aminoacylation	11
301	Intraflagellar transport	11
684	tRNA aminoacylation	11

Acyl-CoA; acyl co-enzyme A; tRNA, transfer RNA.

Network clustering

ClusterONE was applied to identify modules from the differential pathway network that were involved in similar biological processes and thus had similar functions. After applying the thresholds for module size (≥20), weighted density (≥0.1) and the overlap threshold (≥0.5), 3 pathway modules were identified (Fig. 3). There were 6, 5 and 1 hub pathways indicated in modules 1, 2 and 3, respectively. The 6 hub pathways with the highest degree scores were all clustered in module 1, indicating that module 1 serves an important role in CSVD. As shown in Fig. 3, the 6 hub pathways in module 1 were fatty acyl-CoA biosynthesis (node 194, degree=54), LA metabolism (node 320, degree=54), synthesis of very long-chain fatty acyl-CoAs (node 624, degree=54), α-linolenic (omega3) and linoleic (omega6) acid metabolism (node 32, degree=53), ALA metabolism (node 33, degree=53), and triglyceride biosynthesis (node 683, degree=48).

Figure 3.

Modules extracted from the differential pathway network. Nodes represent pathways and edges indicate the interactions between pathways. Orange nodes indicate hub pathways with a top 5% degree distributions. The details of hub pathways are shown in Table I.

Discussion

High-throughput biological experiments that probe various genes simultaneously have generated unprecedented amounts of data. Candidate diagnostic and prognostic biomarkers are typically the most significant differentially expressed genes (DEGs) between the control and disease samples (22–24). However, the most significant DEGs obtained from different studies for a particular disease are frequently different (25). The cross validation of datasets, including via network-based methods, reduces false findings and increases the sensitivity of the identification of significant DEGs (26). Furthermore, pathway networks provide insight into the potential underlying molecular mechanisms of disease (27). Therefore, in the present study, differential pathway network analysis was used to identify hub signaling pathways in CSVD. The differential pathway network was composed of differential pathway interactions using background PPIs, the Reactome pathway database and a gene expression profile dataset. In addition, modules within the differential pathway network were identified using ClusterONE.

The results of the present study revealed that there were 15 hub pathways associated with CSVD, and that the top hits were associated with the synthesis of fatty acyl-CoA and LA-associated metabolism, which indicates that these processes contribute to the development and progression of CSVD. In addition, network clustering analysis indicated that the top 5 hub pathways were all within module 1, which suggests that these pathways exerted their role in the development and progression of CSVD in a cooperative way.

Fatty acids are a family of molecules within the lipid macronutrient class (28). A previous study indicated that saturated fatty acids and trans-fatty acids are particularly harmful to blood vessels and that omega 3 fatty acids are considered good fats (29). Long-chain acyl-CoA esters serve as important intermediates in lipid biosynthesis and fatty acid degradation. Bortz and Lynen (30) proposed that acyl-CoA esters are key regulators of fatty acid synthesis, and that long-chain acyl-CoA esters affect a large number of cellular systems and functions, including ion channels, ion pumps, enzyme activity, membrane fusion and gene regulation. Polyunsaturated fatty acids (PUFAs), particularly ALAs, have a protective function against focal and global ischemia (31,32). The ratio of membrane omega 3 to omega 6 PUFAs may be modulated by dietary intake, and this ratio influences neurotransmission and prostaglandin formation, which are vital to the maintenance of normal brain function (33). Previous studies indicate that omega 3 PUFAs may influence vascular tone by affecting membrane potential and inhibiting migration of vascular smooth muscle cells (34) and maintain vascular integrity by decreasing numerous soluble markers of endothelial hemostatic activity (35). Experimental analysis of animal models and human subjects has demonstrated that omega 3 PUFAs cause a moderate reduction in blood pressure, indicating altered vascular neuroeffector responses (36). In the present study, the top 5 hub pathways were associated with the synthesis and metabolism of fatty acids, and were clustered in module 1. These findings suggest that the synthesis and metabolism of fatty acids is associated with the occurrence and development of CSVD.

To the best of our knowledge, the present study is the first to report the analysis of CSVD using a differential pathway network method. The present study identified several pathways that were associated in the occurrence and development of CSVD. However, several limitations were associated with the present study. For example, all of the data were obtained from databases and these data themselves may be unreliable. In addition, only a small sample size was used in the present study and the results obtained using bioinformatics methods were not verified via wet lab experiments. Although disadvantages exist, the used method and the results obtained by the present study provide investigators with valuable resources for improving understanding of the underlying molecular mechanisms of CSVD, and for the identification of potential biomarkers for the early diagnosis and treatment of CSVD. Furthermore, the method proposed may be considered as a framework for optimizing further pathway analysis.