Abstract

   Intracranial aneurysms (IAs) are characterized by localized dilation or
   ballooning of a cerebral artery. When IAs rupture, blood leaks into the
   space around the brain to create a subarachnoid hemorrhage. The latter
   is associated with a higher risk of disability and mortality. The aims
   of this study were to gain greater insight into the pathogenesis of
   ruptured IAs, and to clarify whether identified hub genes represent
   potential biological markers for assessing the likelihood of IA
   progression and rupture. Briefly, the [38]GSE36791 and [39]GSE73378
   datasets from the National Center of Biotechnology Information Gene
   Expression Omnibus database were reanalyzed and subjected to a weighted
   gene co-expression network analysis to test the association between
   gene sets and clinical features. The clinical significance of these
   genes as potential biomarkers was also examined, with their expression
   validated by quantitative real-time PCR. A total of 14 co-expression
   modules and 238 hub genes were identified. In particular, three modules
   (labeled turquoise, blue, and brown) were found to highly correlate
   with IA rupture events. Additionally, six potential biomarkers were
   identified (BASP1, CEBPB, ECHDC2, GZMK, KLHL3, and SLC2A3), which are
   strongly associated with the progression and rupture of IAs. Taken
   together, these findings provide novel insights into potential
   molecular mechanisms responsible for IAs and they highlight the
   potential for these particular genes to serve as biomarkers for
   monitoring IA rupture.

1. Introduction

   Intracranial aneurysms (IAs) represent a cerebrovascular disorder which
   affects between 3% and 5% of individuals. There is a potential risk
   that an IA will rupture, and this risk is higher in the posterior
   circulation [[40]1–[41]3]. With advances in intracranial imaging
   technologies, IAs have been detected more frequently. Consequently,
   clinicians are increasingly confronted with a dilemma regarding the
   choice of clinical management. Namely, whether preventive treatments
   (e.g., endovascular or surgical aneurysm repair) which are associated
   with inherent complication risks, or conservative management with or
   without follow-up imaging which leaves patients with a small, yet
   definite, risk of aneurysm rupture, should be applied [[42]3, [43]4].
   When an IA ruptures, a subarachnoid hemorrhage (SAH) develops. This
   life-threatening clinical condition has an acute mortality rate of
   approximately 50% [[44]5, [45]6]. Despite considerable advances in
   therapy for IAs, SAH remains a highly lethal condition which is
   associated with a high socioeconomic burden [[46]7–[47]9]. Thus, an
   ability to identify IAs which have a high risk of rupture and provide
   timely preventive treatment may be key to the successful management of
   IAs.

   To predict IA rupture, researchers have studied aneurysmal hemodynamics
   [[48]10, [49]11], aneurysmal morphology and location [[50]8], genetics
   [[51]12], and other factors (e.g., cigarette smoking, hypertension, and
   positive family history for SAH) [[52]13]. Only inflow angle was
   identified as a significant predictor of rupture according to
   morphological parameters [[53]8]. However, aneurysm wall inflammation
   has also been shown to play a pivotal role in aneurysm growth and
   rupture [[54]2, [55]3, [56]14]. Two scoring systems have been
   established to evaluate the risk of rupture and to guide treatment.
   These include the PHASES (population, hypertension, age, size of
   aneurysm, earlier SAH from another aneurysm, and site of aneurysm)
   system and the UIATS (unruptured IA treatment score) system [[57]8,
   [58]15, [59]16]. However, standardized management of unruptured IAs
   remains controversial, as risks of prophylactic treatment must be
   weighed against possible risk of rupture for individual aneurysms
   [[60]17].

   Despite the vast efforts made to date to prevent the rupture of IAs,
   the mechanisms mediating the pathology of IAs remains largely unknown.
   In particular, suitable biomarkers to predict IA rupture remain
   unavailable. Therefore, the aims of the present study were to gain
   greater insight into the pathogenesis of ruptured IAs and to clarify
   whether identified hub genes can be used as potential biological
   markers to assess the likelihood of IA progression and rupture ([61]Fig
   1). For these aims, the [62]GSE36791 [[63]18] and [64]GSE73378 [[65]19]
   datasets from the National Center of Biotechnology Information (NCBI)
   Gene Expression Omnibus (GEO, [66]http://www.ncbi.nlm.nih.gov/geo/)
   were reanalyzed. A weighted gene co-expression network analysis (WGCNA)
   [[67]20] was performed to test possible correlations between gene sets
   and clinical features of IAs. Clinically significant genes were
   identified and their expression levels were validated in patients with
   ruptured versus unruptured IAs by quantitative real-time (qRT)-PCR.

Fig 1. A flow chart illustrating the method used to identify six biomarkers
associated with IA rupture.

   [68]Fig 1
   [69]Open in a new tab

   The entries shown in the green box were completed in our previous
   study, while the entries shown in the red box are those that we intend
   to focus on in future studies.

2. Materials and methods

2.1 Microarray dataset

   Gene expression profile data from the [70]GSE36791 dataset were used as
   a training set to construct co-expression networks and to identify hub
   genes. This dataset was generated from peripheral blood samples
   collected from 43 patients with SAH due to ruptured IAs, and from 18
   individuals with headaches (the reference group) [[71]18]. To verify
   the results obtained, the [72]GSE73378 dataset was used as a test set.
   The [73]GSE73378 dataset was generated from peripheral blood samples
   collected from 103 patients who developed an aneurysmal SAH at least
   two years prior, and from 107 individuals used as a reference group
   [[74]19].

2.2 Data processing and WGCNA

   The [75]GSE36791 expression data were already normalized with a
   BeadArray package [[76]21] (quantile normalization). Probe sets were
   then mapped to gene symbols according to the [77]GPL10558 platform.
   After filtering the probes without a corresponding gene symbol, the
   average value of the gene symbols with multiple probes was calculated
   [[78]22]. Outlier samples were identified with hierarchical cluster
   analysis (i.e., distance matrices constructed with Pearson's
   correlation matrices and a hierarchical agglomerative method to adopt
   average linkages) by using the hclust function in WGCNA [[79]20]. Based
   on these results, three samples ([80]GSM901111, [81]GSM901112, and
   [82]GSM901161) in [83]GSE36791 were removed from subsequent analyses.

   In the present study, the co-expression analysis performed was based on
   a WGCNA, which is a systems biology method for describing correlation
   patterns among genes across microarray samples [[84]20]. Currently,
   this method is widely used in the international biomedical field. This
   method also helps identify clusters (modules) of highly correlated
   genes across samples. To identify ruptured IA-associated co-expression
   modules and their key constituents, preprocessed data of [85]GES36791
   were analyzed. Briefly, Pearson's correlation matrices were generated
   (average linkage method) for all pair-wise genes. Next, co-expression
   similarities were transformed into a weighted adjacency matrix of
   connection strengths by using a power adjacency function (cutoff height
   of 0.85). This adjacency matrix was then transformed into a topological
   overlap matrix to measure relative gene interconnectedness and
   proximity. Finally, gene co-expression modules were identified based on
   clustering (hierarchical average linkage) according to topology
   overlap.

2.3 Identification of modules with clinical significance

   In order to identify modules related to the clinical traits of IA
   (group [rupture IA/ reference] and gender), two different approaches
   were used [[86]20, [87]23]. The first approach was to determine gene
   significance (GS) from a linear regression analysis of gene expression
   data and clinical traits [[88]20]. GS was defined based on a log10
   transformation of P-values. Thus, a higher absolute value of GS[i]
   corresponds with greater biological significance of gene i [[89]20]. In
   addition, module significance (MS) is defined as the average GS value
   for all of the genes in a module, which implies that the higher the
   absolute value of MS[j], the more biologically significant is module j.
   The second approach was to define module eigengenes (MEs) as a major
   component in the principal component analysis for each module. Thus,
   MEs can be considered representative of the gene expression profiles in
   a module [[90]20]. Correlations between MEs and clinical traits were
   calculated to identify relevant modules. Modules with a P-value greater
   than 0.005 were considered to significantly correlate with certain
   clinical traits (ruptured IA).

2.4 Identification of hub genes and functional enrichment analysis

   Hub genes were defined according to module connectivity [absolute value
   of Pearson's correlation (cor.geneModuleMembership > 0.8)] and clinical
   trait relationship [absolute value of Pearson's correlation
   (cor.geneTraitSignificance > 0.5)] [[91]23]. Hub genes were highly
   interconnected with genes in the module with potential significance. To
   functionally characterize the identified hub genes, Gene Ontology (GO)
   terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway
   enrichment analysis were performed by using clusterProfiler [[92]24].
   The terms obtained from the KEGG pathway and GO analyses (including
   molecular functions, biological processes, and cellular components)
   which had false discovery rates (FDRs) < 0.05 were considered to be
   significantly enriched in the hub genes. Common hub genes were also
   screened from key genes identified in our previous study [[93]22] which
   were regarded as “candidate biomarkers” highly associated with IA
   rupture.

2.5 Analysis of the [94]GSE73378 dataset

   [95]GSE73378 expression data were already normalized (quantile
   normalization) by the surrogate variable analysis (SVA) package
   [[96]19]. Differentially expressed genes (DEGs) (ruptured IAs vs.
   references) were then screened by using the limma package [[97]25]. The