Inferring immune‐associated signatures based on a co‐expression network in Guillain‐Barré syndrome

Abstract Objectives Guillain‐Barré syndrome (GBS) is a type of acute autoimmune disease, which occurs in peripheral nerves and their roots. There is extensive evidence that suggests many immune‐associated genes have essential roles in GBS. However, the associations between immune genes and GBS have not been sufficiently examined as most previous studies have only focused on individual genes rather than their entire interaction networks. Materials and methods In this study, multiple levels of data including immune‐associated genes, GBS‐associated genes, protein‐protein interaction (PPI) networks and gene expression profiles were integrated, and an immune or GBS‐directed neighbour co‐expressed network (IOGDNC network) and a GBS‐directed neighbour co‐expressed network (GDNC network) were constructed. Results Our analysis shows the immune‐associated genes are strongly related to GBS‐associated genes whether at the interaction level or gene expression level. Five immune‐associated modules were also identified which could distinguish between GBS and normal samples. In addition, functional analysis indicated that immune‐associated genes are essential in GBS. Conclusions Overall, these results highlight a strong relationship between immune‐associated genes and GBS existed and provide a potential role for immune‐associated genes as novel diagnostic and therapeutic biomarkers for GBS.

interact with gangliosides. 6 Supportive medical care and immunotherapy are two main treatment modes for GBS. Although most patients should recover by nonspecific immunotherapy, there is still a need for optimized treatment and patient care throughout the course of disease. 7 It is also essential to optimize supportive medical care to prevent or treat disease-related complications. Therefore, there is an urgent need to identify novel biomarkers and treatment methods to improve the clinical features of GBS.
Guillain-Barré syndrome has been proven to be related to immune dysfunction. A previous study reported that humoral immune and autoimmune responses could be triggered by pathogen infection (eg, Campylobacter jejuni) leading to nerve dysfunction for GBS. 8 It has also been demonstrated that the activation of prominent T lymphocytes (T cells) occurs in GBS and is essential in the pathogenesis of a rat model of the disease. The integrins on T cells and adhesion molecules are up-regulated on the endoneurial endothelium in GBS. 9 Although some studies reported a relationship between the immune system and GBS, the roles of immune factors in the pathogenesis of GBS have not been explored in detail and systematically investigated.
Genetic factors are also important contributors for the pathogenesis of GBS. Others have reported associations between the development of GBS and CD1 SNPs in European populations, 10 a finding also reported in the Chinese population. 11 Yuan et.al suggested that CCR2 is responsive to pharmacologic blockade CCR2, making a candidate drug target for GBS which could provide novel insight into exploring target-specific anti-inflammatory treatment methods for peripheral neuroinflammation. 12 Genome-wide gene expression analyses of GBS suggest there are gender differences based on GBS patients. 13 Chang et.al identified networks and key pathways associated with GBS using gene expression profiles. 14 These studies implicate certain genes in the development of GBS, and gene expression profiles may help identify GBS-associated pathways. However, most of these studies concentrated on individual genes and either used cell lines or limited patient samples. Specially, the global and systematic investigation focus on the association between immune and GBS using genome-wide expression profiling is absent.
In the current study, an immune-or GBS-directed neighbour co-expressed network (IOGDNC network) including immuneand GBS-associated genes, immune-associated genes only, GBS-associated genes only, and other genes was constructed.
We also extracted the first neighbours of GBS-associated genes from IOGDNC networks to construct a GBS-directed neighbour co-expressed network (GDNC network). We use the two networks to illustrate the roles of immune-associated genes in GBS.
We identified immune-associated genes in the IOGDNC network with hub topological characteristics related to GBS. The immune-and GBS-associated genes have the highest co-expression level are also found. We further identify five clusters from GDNC network which include some key immune-associated and GBS-associated genes, suggesting specific functions of the immune-associated genes in GBS. Specially, some of these clusters could distinguish the GBS from control samples when used as a ebiomarker. A functional analysis reveals that the genes in these clusters are co-related to immune-related processes and others in GBS. Our results highlight the specific contribution of an immune-directed network for GBS and could be as an effective resources and candidate for further research and treatment of GBS.

| Expression profile of genes and PPI network data set
High-throughput microarray gene expression data were obtained from Gene expression Omnibus (GEO, accession number: GSE31014). This data set comprises 7 Guillain-Barré syndrome samples and 7 normal controls. We downloaded high-confidence protein-protein interaction (PPI) data from the Human Protein Reference Database (HPRD), 15 which is an integrated protein database that records various protein features. PPI data from HPRD have been applied to thousands of network studies, especially in the field of disease gene discovery. 16

| Human immune-related gene data sets
All immune-related genes in Homo sapiens were obtained from AmiGo, and 3068 immune genes were obtained based on 651 records. 17

| Guillain-Barré syndrome (GBS)associated genes
We downloaded all GBS-associated genes from the DisGeNET database, which stores data on human disease-related genes and variants. We obtained 561 119 gene-disease associations comprising 20 370 diseases or phenotypes and 17 074 genes.

| Immune-or GBS-directed neighbour coexpressed network construction (IOGDNC)
Firstly, we calculated the Pearson correlation for gene expression between any two gene pairs. Initial gene co-expression networks were obtained by limiting the expression correlation coefficient (absolute coefficient value > 0.3) and false discovery rate (FDR < 0.05). For the second step, we mapped the PPI network pairs to our co-expression network and only retained the gene pairs that were common to the PPI network. The final network is a Guillain-Barré-specific co-expression network. Network visualization was performed using Cytoscape. Interaction levels were distinguished by high and modest Pearson correlation coefficients. Most biological networks are scalefree networks. We therefore checked the power law distribution of our co-expression network in MATLAB, using the degree distribution data from our network.

| Dissecting Guillain-Barré syndrome and immune-associated gene features in network
We classified the genes into five groups: GBS (Guillain-Barré syndrome)-associated genes, immune genes, GBS-and immune-associated genes, GBS-and non-immune-associated genes, and other genes. In order to construct the GBS-directed neighbour co-expression network (GDNC network), we extracted the first neighbours of GBS genes and GBS-and non-immune-associated genes from IOGDNC network to get two GBS-associated networks. The onestep neighbour network for the GBS and GBS-and non-immuneassociated genes was visualized using Cytoscape, with different node colours representing different gene types. For each gene set, we compared the number of their first neighbours in the network. Next, to analyse the level of interaction with neighbours among different gene sets, we used a cumulative distribution function (CDF) to estimate the degree of the expression correlation for each gene category. Wilcoxon rank-sum tests were used to compare the co-expression correlation coefficients between gene set pairs.

| Network cluster mining and validation of its classification power
We used GraphWeb tool mine important network clusters that are associated with GBS, 18 using our constructed co-expression networks as the input file. Each of the output clusters was plotted using the Cytoscape program. For each cluster, the gene expression data were then used to classify the 14 samples in our study using a consensus clustering method. 19 This was performed using the ConsensusClusterPlus package in R. We chose the optimum category number determined by the point at which the increase in the area under the cumulative distribution function curve is small. Combining the classification results of the consensus clustering and the real category (disease and control) of the samples, we used a chi-squared test to investigate the association between the two classification methods. We considered the two type of class to be associated when the test result was significant (P < 0.05).

| Identifying gene expression patterns in network modules
To explore the genes' inter-relationship in each module, we computed the Pearson correlation coefficient S between each gene pair in the module. Correlation heatmaps were plotted in R using the pheatmap package. We also compared the gene expression patterns between GBS and normal control samples in each module.
The differentially expressed genes were identified using a t test in R, with a significance threshold (P value) of 0.05. Finally, we used a hypergeometric test to validate the enrichment between all differentially expressed genes and the differentially expressed genes in all modules.

| KEGG pathway enrichment analysis
We performed functional enrichment analysis using the GSEApy package in Python. Briefly, the genes in each module and all modules were tested against each KEGG pathway, respectively. Significant enrichment results (adjusted P value < 0.05) were retained for the following analysis.

| RE SULTS
3.1 | Immune-associated genes are essential in the process of GBS There were 58 genes common to the immune-related and GBSrelated gene sets such as FCRL3, CXCL1 and HPRT1. Twenty-four GBS-specific genes and 2859 immune-specific genes were also identified. We constructed an immune-or GBS-directed neighbour co-expressed network (IOGDNC network), which is a subnetwork extracted from the PPI network. The IOGDNC network included 2497 nodes and 3442 edges ( Figure 1A). First, the degree distribution of all genes was analysed, with the degree showing the scalefree distribution ( Figure 1B). The R-squared was 0.9946. There were 23 immune-and GBS-associated genes, 10 GBS-associated genes and 834 immune-associated genes in the network ( Figure 1C). Next, we compared the degree of different gene types and found GBS-associated and non-immune-associated genes, and immune-associated genes showed the highest degree ( Figure 1D). The top five genes with highest degree were CASP3, PRKCD, AKT1, EGFR and SUMO4, and three of these were immune-associated genes. This suggests an essential role for immune-related genes in the IOGDNC network.
Collectively, these results indicate that immune-related genes could act as essential contributors for GBS.

| Immune-associated genes interact with GBS genes and show stronger co-expression patterns
We also constructed a network termed GBS-directed neighbour co-expressed network (GDNC network) to further consider the association between immune-associated and GBS-associated genes ( Figure 2A). The GDNC network is a subnetwork of the IOGDNC network based on extracting GBS-associated genes and their directly interacting genes. There were 20 GBS-and immune-associated genes, 10 GBS-specific genes and 63 immune-specific genes.
We further analysed the co-expression (Pearson correlation) levels of the interactions between different gene types ( Figure 2B). The GBM-e and immune-associated genes showed the highest co-expression level with their interacting genes. This result suggests there are strong relationships between immune-associated genes and GBS-associated genes based on topological structure and expression patterns. Additionally, we discovered the correlation of expression was significantly distinct among diverse kinds of genes ( Figure 2C). A subnetwork only comprising GBS-associated genes and their directly interacting genes was extracted from the GDNC network ( Figure 2D). This network included nine GBS-associated genes, all of which interacted with immune-associated genes. These GBS-associated genes also showed strong correlations with immune-associated genes ( Figure 2E). For example, GBS-associated gene NR3C1 was strongly co-expressed with immune-associated genes including STAT3, RAF1 and TRIM28. These results indicate that there are complex features including topological interactions and expression patterns among the associations between GBS-associated genes and immune-associated genes.

| Immune-associated network clusters are special classifiers in GBS
We used a public web server, GraphWeb, to identify the vital module clusters from our IOGDNC network. A total of five clusters were identified, comprising several important GBS-or immune-associated genes ( Figure 3A). The number of genes in each module was variably distributed ( Figure 3B)

| Immune-and GBS-associated genes are related on expression level in clusters
The similarity of the modules based on their co-expression status was dissected. We found these modules all had a very high co-expression pattern for any gene pair. The fraction of the co-expression value >0.5 against all gene pairs reached 90% for all module clusters.
In particular, the proportion of the co-expression values more than 0.7 in each module was 73.6% on average, with the minimum value being 58.2% in the second module and the maximum value being 86.1% in the fourth module. The third and fifth module's co-expression patterns were very significant ( Figure 4A,C, for other modules, see Figure S2).
We also analysed gene expression in the five modules by comparing the expression values for all genes between GBS and normal control samples. Interestingly, in all modules, the gene expression value differed between disease and control samples. For modules 1-3 and 5, the gene expression in GBS was higher than controls; on the contrary, the module 4 gene expression in normal control samples was higher than GBS samples ( Figure 4B,D, and Figure S2). We could infer that these genes have a consistent impact in GBS. In addition, we discovered significantly differentially expressed genes by a t test ( Figure 4B,D). Although the total number of differentially expressed genes in our module was small, it may be related to the low gene numbers in each module. We assessed whether genes with differential expression patterns between groups were enriched among our module gene sets (Materials and methods). Unsurprisingly, this enrichment was significant (P = 0.021).

| The functional analyses show the potential significance of immune-associated clusters in GBS
KEGG pathway enrichment analysis was performed for our module clusters. Because of the low number of module genes, KEGG analysis was also performed for all genes in each module ( Figure 5, the whole result pathways are available in Table S1). We obtained several important pathways that are related to the immune system, such as antigen processing and presentation, leucocyte transendothelial migration, chemokine signalling pathway, Fc gamma R-mediated phagocytosis and T-cell receptor signalling pathway. When using all module genes to perform pathway enrichment, we find there is a common pathway, chemokine signalling pathway, which is implicated in several biological reactions including immune responses. All module genes except for those in the third module were enriched in this pathway. Early studies identified a role for chemokines in recruiting immune cells to the site of inflammation. 24 After years of study, new roles of chemokines have been discovered. Chemokines also regulate other cells to impact embryonic development and metastasis. CXCL12 is a well-known chemokine, and it can bind with CXCR4 ( Figure 6A, module 5) to activate downstream signalling reactions. 25 Moreover, we also identified a significant immune-related pathway, TNF (tumour necrosis factor) signalling pathway ( Figure 6B). The most famous member of the TNF family is TNF-α ( Figure 6B), and its initial function comprises regulating the immune cells. Many studies have demonstrated that the abnormal expression of TNF-α is implicated in multiple human diseases. [26][27][28] Researchers had shown that the crosstalk between TNF-α and TNFR1 have a vital effect in the tumour microenvironment. They also have an important role in tumour progression, and impact metastasis, apoptosis and survival of tumour cells. Dysregulation of module cluster genes might therefore impact GBS status through acting on these immune-associated pathways, which are potential signatures for GBS.

| D ISCUSS I ON
The current study integrates interaction network and expression Guillain-Barré syndrome is an acute immune disease which occurs at nervous system, with a serious prognosis. It has been reported that GBS could be induced by immunization with peripheral-nerve F I G U R E 4 Co-expression and gene expression feature in the clusters. A,C, The co-expression heatmap of genes in the third and fourth module cluster. B,D, Gene expression pattern of the third and fourth module, significantly differentially expressed genes have been labelled (* and ** indicate the significance level of 0.05 and 0.01, respectively) proteins. 29 In the current analysis, we also concentrated on the association between immune-and GBS-related genes on gene expression patterns. The immune-and GBS-related gene interaction network and expression profile data were both integrated to study the role of immune genes for GBS. Collectively, all results suggest there are strong associations between immune-and GBS-related genes not only on network structure but also on expression patterns. Some immuneassociated genes may have potential as diagnostic and treatment biomarkers of GBS based on differential expression and functional analyses.
In summary, interaction networks and expression profiles were integrated to reveal novel roles and insights for immune-related F I G U R E 5 KEGG pathway enrichment analysis in genes. The KEGG enrichment results for the five module genes and all module genes, respectively F I G U R E 6 KEGG pathway enrichment analysis in signalling pathways. A, The map of the chemokine signalling pathway. B, The map of the TNF signalling pathway genes in GBS. Immune-and GBS-related genes showed close relations both at the level of network interaction structure and expression patterns. Some immune-associated clusters were also identified for GBS, and these core clusters comprised special expression features. Specially, gene expression of these modules could distinguish the GBS and normal samples. The functional analyses highlighted the molecular mechanisms of immune-associated genes in GBS.

| CON CLUS IONS
We constructed a GBS-directed neighbour co-expressed network (IOGDNC network) and a GBS-directed neighbour co-expressed network (GDNC network) and found an important role of immune and GBS gene in GBS. We identified five clusters from the GDNC network, some of which can serve as biomarkers to distinguish between GBS and control samples. In addition, the genes in these clusters displayed different expression characteristics between GBS and control samples. Furthermore, pathway enrichment analyses clearly illustrate the functions for these genes in each cluster. Our results provide support for these immune-associated genes as potential therapeutic targets for GBS.

CO N FLI C T S O F I NTE R E S T
The authors declare that they have no competing interests.