Background

Parkinson’s disease (PD) is one of the well-reported neurodegenerative disorders, only second to the Alzheimer’s disease (AD), throughout the world [1]. The primary pathology of PD is the loss of dopaminergic neurons in the substantia nigra with Lewy bodies (intracytoplasmic inclusion deposits of aggregated alpha-synuclein and ubiquitin protein, and damaged nerve cells) [2, 3].

A good number of studies have been performed to identify the causative factors and molecular markers of PD. Several previous studies have pointed out the role of different genes in this disease [4]. Gene expression profiling analysis has identified differentially expressed genes in PD [5]. Besides, differential expression of several microRNAs (miRNAs) has also been associated with the pathophysiology of several neurodegenerative diseases [6, 7] including PD [8]. Study of gene regulatory networks has emerged as an important approach for computational analyses of diseases [9]. However, limited previous studies have tried to understand the association of both of these (miRNAs and mRNAs) PD markers in the context of biological networks. In order to gain a proper understanding of this disease, one needs to study the detailed regulatory network involving genes, miRNAs and transcription factors (TFs). A thorough examination of regulatory networks can help us to identify the key genes or miRNAs as well as different network motifs associated with a disease. These network motifs, in turn, provide us several important aspects of a disease progression.

Previous studies have indicated the role of epigenetic modifications in the development of neurodegenerative diseases including Parkinson’s Disease and Alzheimer’s Disease (AD) [10, 11]. Epigenetics refers to the meiotically and mitotically heritable changes in gene expression that does not involve changes to the DNA sequence [12]. Interpretation of epigenetic profiling leads to the identification of changes in gene expression responsible for the disease progression. There are three distinct yet highly interrelated mechanisms of epigenetic regulation - DNA methylation, Histone modifications and non-coding RNA-based mechanisms [13]. Epigenetic changes can be influenced by several factors including age, environment, lifestyle and disease state [13]. Recent systematic review on neurodegenerative disease, investigated epigenetic marks in PD and identified the most consistently reported methylation genes and histone modifications associated with PD [14].

Studies have revealed that non-coding RNAs such as miRNAs (~22 nt long) and long non-coding RNAs (lncRNAs) (>200 nt long), play crucial roles in epigenetic pathways and gene silencing. The function of miRNAs involves binding to a specific sequence in the 3′ UTR of a gene and inhibiting the expression of that gene. Thus, miRNAs act as cellular post-transcriptional regulators. The miRNA profiling of PD samples offers insight into the molecular mechanism of PD progression and several miRNAs have been implicated in PD pathogenesis [8, 15, 16].

The function of lncRNAs involves diverse cellular processes, such as chromatin remodeling, cell cycle regulation and several developmental processes [17]. It can influence the post-transcriptional regulation by interfering with the miRNA pathways, by acting as competing endogenous RNAs (ceRNAs) [18]. lncRNAs possess miRNA response elements (MRE) or miRNA binding sites in them. This enables lncRNAs to act as miRNA sponges to control the availability of endogenous miRNA for binding to their target mRNAs and subsequently reducing the repression of these target mRNAs [18]. lncRNAs are implicated in neurodegenerative processes, including AD and Huntington’s disease (HD) [19, 20]. However, very little is known about the association of lncRNAs in PD [21].

Single base alteration in the gene sequence or single nucleotide polymorphism (SNP) can affect the phenotypes either by altering the amount of protein produced or by changing the type of protein produced [22]. SNPs are believed to cause differences between individuals, such as susceptibility to diseases [23]. There are numerous SNPs present in the human genome [24]. These are considered as invaluable markers and potentially powerful tools for both genetic research and applications in practice [25]. Several studies have identified SNPs associated with complex diseases, which in turn serve as a potential marker for diagnosis [26]. A recent miRNA-related SNP analysis study identified SNPs as independent prognostic markers for the survival in small cell lung cancer patients [27]. However, very few such studies have been performed for PD. A recent genome-wide association study identified significant association between bone marrow stromal cell antigen 1 SNP and increased risk of PD which is enhanced by environmental factors [28]. SNP analysis or genotyping of PD patients can be helpful to identify this disease at an earlier state. Besides single base alteration in the miRNA binding sites can give us important information about the mode of regulation of regulatory factors in this disease. It is believed that more and more genetic studies combined with machine learning and statistical methods will be required in near future to investigate the underlying molecular signature of a disease [29].

With the increase of transcriptomic data, novel system biological approaches are required that can explore the complex molecular interactions associated with a disease. In this study, we have analyzed gene co-expression network based on a PD microarray dataset. Two regulatory networks were built from the highly co-expressed genes. The mTF-miRNA-gene-gTF regulatory network involves microRNA Transcription Factor (mTF), microRNA (miRNA), gene and gene Transcription Factor (gTF) whereas, long noncoding RNA (lncRNA) mediated regulatory network involves miRNA, gene, mTF and lncRNA. mTF-miRNA-gene-gTF regulatory network identified a novel feed-forward loop. lncRNA-mediated regulatory network identified novel lncRNAs of PD and revealed the two-way regulatory pattern of PD-specific miRNAs where miRNAs can be regulated by both the TFs and lncRNAs. SNP analysis of the most significant genes of the co-expression network identified 20 SNPs. Thus, our study provides important insight into the epigenetic mechanism (lncRNA, miRNA, histone modification) associated with PD. Moreover, SNPs identified in our study can be helpful for identifying PD at an earlier stage.

Results

Figure 1 depicts the workflow of our analysis.

Fig. 1
figure 1

Workflow of the methodology used in our study

Full size image

Differentially expressed gene selection

SAM

We identified the differentially expressed (DE) genes between PD and control patients by applying the Significance Analysis of Microarray (SAM) [30]. In chip A, SAM identified 1518 DE genes at FDR value 0.19% and tail strength 44.1%. Among the 1518 DE genes 293 genes were positive (upregulated) and 1225 were negative (downregulated). In chip B, SAM identified 673 DE genes at FDR value 0.11% and tail strength 37.6%. Among the 673 differentially expressed genes, 91 genes were positive (upregulated) and 582 were negative (downregulated).

t-test

With the t-test analysis, a much higher number of DE genes were found than with SAM. Results identified 4797 and 3120 DE genes in chip A and chip B respectively at p-value 0.05 or 95% confidence level.

The common DE genes found by both SAM and t-test were considered as the most significant DE gene sets and these were used for further study. In chip A, 521 genes were found to be commonly DE in both SAM and t-test whereas, in chip B, 130 genes were found to be commonly DE in both SAM and t-test (Table 1). Out of the 521 and 130 genes from chip A and chip B only 458 genes in chip A and 105 genes in chip B were annotated.

Table 1 DE genes separately identified by SAM and t-test and DE genes commonly identified by both
Full size table

Enrichment analysis of the DE genes

The DE genes found from SAM and t-test were both annotated via EASE (Expression Analysis Systematic Explorer) [31]. The shared 458 DE genes of chip A obtained from SAM and t-test were then subjected to enrichment analysis in FatiGO (Table 2) [32]. Results of the enrichment analysis of identified several neurodegenerative disease pathways as the most significant over representative KEGG pathways such as Parkinson’s disease pathway (hsa05012), Huntington disease (hsa05016) and Alzheimer’s disease (hsa05010) (Table 3). This also signifies the importance of this gene set in the context of the PD-specific study. The 105 DE genes in chip B were not associated with any significant terms in FatiGo. Therefore DE genes of chip B were not considered for further analysis. The 458 DE genes of chip A were considered as the significant gene set for further study and were termed as common DE gene set of chip A.

Table 2 FatiGO analysis results of the common DE genes of chip A and chip B obtained from SAM and t-test
Full size table
Table 3 Highly significant KEGG pathways associated with the common 458 genes of chip A identified in FatiGO analysis
Full size table

Co-expression network construction and analysis

On the basis of the co-expression pattern, WGCNA (please refer to the methods section) divided the 458 common DE genes into six modules (turquoise, blue, brown, yellow, green and red containing 266, 56, 43, 42, 25 and 25 mRNAs). FatiGO analysis revealed that out of the six WGCNA modules Turquoise module was the most significant co-expressed module (Additional file 1: Table S1A and B).

Topological analysis of the WGCNA module and identification of hub genes

We analyzed two centrality measures - degree and betweenness centrality (BC) in tYNA [33]. We sorted the 266 genes according to their degree or connectivity. The degree represents the number of connections or edges of a particular node [34], whereas BC quantifies the flow of information through a node in the network. It specifies how a node influences the communication among other nodes [35]. In our study, the 266 genes exhibited a varied degree distribution with the highest degree of 262 and lowest degree of 1. The average degree value was found to be 217.63 with standard deviation 52.74. We found that highest BC value was 391.50 and the lowest was 0 with an average of 24.95 and standard deviation 33.71. We chose the top 8 nodes (i.e. top 3% of the total nodes) with highest degree value as High Connectivity (HC) hub nodes. AP3B2, MAGED1, NSF, STXBP1, CYB561, AF1Q, C14ORF78 and GASP were identified as HC hub genes (Additional file 2: Table S2). Interestingly these 8 HC nodes were assigned with low BC values. Surprisingly we found three nodes with high BC values but low degree value. Although these nodes have low connectivity, they might be important regarding information flow. Therefore, we identified these as High BC low connectivity (HBLC) hub nodes. HNRPC, MAN1C1 and HSPA1A were identified as HBLC hub genes (Additional file 2: Table S2). Figure 2 shows the gene co-expression network of the turquoise module with the 11 hub genes. Out of the 11 hubs, three hubs (NSF, HSPA1A and CYB561) were already found to be associated with PD. The remaining eight novel hub genes (MAGED1, AP3B2, STXBP1, AF1Q, GASP, C14ORF78, MAN1C1, HNRPC) were further studied for their association in PD.

Fig. 2
figure 2

Gene co-expression Network of the most significant co-expressed module (Turquoise module) obtained from WGCNA. Green nodes represent genes and edges represent co-expression relationship. 11 Hub genes are represented by larger node size

Full size image

Epigenetic regulation of the hub genes

In order to identify the probable epigenetic regulation of the hub genes, histone modification data for eight hub genes (Table 4) were retrieved from HHMD [36]. Table 4 shows that all the eight hub genes were associated with several histone modification sites. Further study identified the experimentally validated non-coding RNA mediated regulation of hub genes (Table 5) [37,38,39,40]. It was found that four out of the eight hub genes were associated with miRNAs already known in PD. Moreover, most of these hub gene associated miRs were in turn regulated by lncRNAs.

Table 4 Histone modification patterns (obtained from HHMD) of novel hub genes with respect to the already known histone modification sites in neurodegenerative diseases
Full size table
Table 5 Regulatory non-coding RNAs associated with the novel hub genes identified in our study
Full size table

Regulatory network construction and analysis

mTF-miRNA-gene-gTF regulatory network

In order to get a view of the regulatory pattern of the turquoise module, we built a regulatory network comprising genes of the turquoise model and the TFs associated with these genes (gTFs) [41]. 160 gTFs were found to be associated with 81 genes of the turquoise module. It was found that PSME3 and PTBP1 are the genes, which are regulated by maximum gTFs, 31 and 25 gTFs respectively. Both of these genes were found to be involved in several cancers [42, 43]. PSME3 was found to be involved in Huntington’s Disease [44]. Besides, we found gTFs for hub gene, HSPA1A, MAGED1 and NSF.

two hundred twenty-six genes of the turquoise module were found to be associated with 51 experimentally validated miRNAs. ATP6V1C1, CBFB and PSME3 are the genes, which are regulated by a maximum number of miRNAs i.e. 7, 6 and 6 miRNAs, respectively. These 51 experimentally validated miRNAs were associated with 117 mTFs as obtained from TransmiR database [45]. By combing all these regulatory information, we constructed a mTF-miRNA-gene-gTF regulatory network (Fig. 3) which represents the four layers of complex regulatory interactions taking place within the most significant WGCNA module.

Fig. 3
figure 3

The four layered mTF-miRNA-gene-gTF Regulatory Network of the turquoise module. In this network, blue rectangular nodes represent miRNAs, green circular nodes represent genes, green circular node with black border represents gene that can regulate other genes as TF, diamond shaped magenta nodes represent mTFs, diamond shaped orange nodes represent gTFs, diamond shaped pink nodes with cyan borders represent the common TFs regulating both miRNAs and genes. The Feed-Forward Loop involving hsa-miRNA-375, gene PAFAH1B1 and TF ASH1L is also shown in the network

Full size image

lncRNA-mediated gene regulatory network

Fifty-one miRNAs that were previously found to be associated with the genes of the turquoise module were further searched in the lncbase module of DIANA tools [40] to acquire information on the miRNA-lncRNA pair. Thirteen out of the 51 miRNAs were found to be PD-specific which were associated with 57 lncRNAs. These 13 miRs control 29 genes of the turquoise module and the 13 miRNAs are in turn controlled by 44 mTF. By combining all these regulatory information, we constructed a regulatory network involving the mTFs, lncRNA and genes (Table 6, Figure not Shown).

Table 6 lncRNA-mediated PD-specific regulatory network
Full size table

SNP analysis of the most significant co-expressed module

Selection of biologically significant SNPs

SNPs corresponding to the 266 genes of the turquoise module were obtained from the online database SCAN [46]. We studied for those SNPs that are present at the 3’UTR of these genes. Using MirSNP database [47] (miRNAs usually bind at the 3’UTR of the target gene and controls the expression of that gene). 1525 miRNAs were found to be associated with these SNPs. Of these 1525 miRNAs, 82 miRNAs were found to be PD related. It was found that 140 SNPs were associated with these 82 miRNAs (p-value ≤10−05) (Additional file 3: Table S3). These 140 SNPs were then searched in the dbSNP database (http://www.ncbi.nlm.nih.gov/snp/) [48] for SNP sequence, chromosome locus, and gene corresponding to each of the SNPs (data not shown). It was found that these 140 SNPs were associated with 157 genes (out of the 266) of the turquoise module.

Discussion

In our study, Co-expression network analyses revealed that out of the six WGCNA modules Turquoise module was the most significant co-expressed module. Enrichment analysis revealed that Parkinson disease pathway (hsa05012) is one of the over-representative pathways associated with this module (Additional file 1: Table S1A). Moreover, Epithelial cell signaling in Helicobacter pylori infection (hsa05120) appeared as one of the most significant KEGG pathways (Additional file 1: Table S1A). Previous studies have already reported that H. pylori infection is associated with PD [49]. Additional file 1: Table S1B depicts the highly Significant GO Biological Processes such as microtubule-based process (GO:0007017), synaptic transmission (GO:0007268), intracellular transport (GO:0046907), etc. associated with the genes of turquoise module.

Co-expression network yielded 11 hub genes based on their topological significance. Out of the 11 hubs, three hubs (NSF, HSPA1A and CYB561) were already found to be associated with PD. The remaining eight novel hub genes were further studied for their association in PD.

Differential expression pattern of the hub genes

Table 7 represents the differential expression pattern of the eight novel hub genes obtained from the co-expression network. Most of the novel hub genes (MAGED1, AP3B2, STXBP1, AF1Q, GASP, C14ORF78, and MAN1C1) were down regulated in PD with respect to control, whereas, one hub gene (HNRPC) was up-regulated in PD with respect to control.

Table 7 Differential expression pattern and fold change of the eight co-expressed hub genes
Full size table

Biological significance of hub genes

We further studied the biological significance of these eight novel hub genes. Table 8 represents the GO biological processes associated with the eight novel hub genes which shows the involvement of these genes in several PD-related processes such as protein transport, neurotransmitter release, synaptic transmission, etc. We found that a recent study has pointed out the role of MAGED1 in the central nervous system in both developmental and adult stages [50]. Studies have found reporting on the vesicle coat protein complex AP3B2 to have some neuron-specific functions such as neurotransmitter release [51, 52]. STXBP1 was found to be listed as an AD-specific marker in Genotator [53], Polysearch [54] and Pescador [55]. AF1Q is a retinoic acid target gene and reported to have an association with ovarian cancer disease [56]. GASP was found as a potential tumor marker for several cancers [57]. C14ORF78 was reported to be associated with calcium channel proteins of cardiomyocytes [58]. MAN1C1 was identified as a differentially expressed gene in PD blood samples. Transcriptome changes related to epigenetic modifications including chromatin remodeling and methylation was also studied for this gene in PD [59]. Protein products of HNRPC gene are associated with pre-mRNA processing and other aspects of mRNA metabolism and transport [60]. All these information validate our finding regarding the association of these genes as hubs in PD.

Table 8 GO Biological processes associated with the novel hub genes
Full size table

Epigenetic regulation of hub genes

The epigenetic regulations of hub genes are shown in Tables 4 and 5. We have studied the association of experimentally validated miRNAs and lncRNAs with eight hub genes. It was found that four (STXBP1, AF1Q, C14ORF78, MAN1C1) out of the eight hub genes were regulated by PD-specific miRNAs. Interestingly, AF1Q was identified to be regulated by a maximum number of PD-specific miRNAs (10). It is evident from Tables 5 and 6 that four lncRNAs namely (XIST, n406658, n338391, n410211) are regulating both PD-specific miRNAs and miRNAs not previously known in PD.

We have studied the histone modification patterns of hub genes. Histone modification refers to the posttranslational modifications of the amino-terminal tails of histone proteins which upon modification affect the downstream molecular interactions, hence regulates the gene expression. Interestingly, we found several histone modification sites those are already known to be associated with several neurodegenerative diseases [61] present within these eight hub genes (Table 4).

Identification of feed forward loop from mTF-miRNA-gene-gTF regulatory network

Analysis of regulatory network revealed the presence of an interesting FFL, where a TF regulates a miRNA and they both regulate a target gene (Fig. 3). We found such a FFL between the gene PAFAH1B1, hsa-miR-375 and TF ASH1L. TransmiR data indicated that hsa-miR-375 is activated by TF ASH1L. By combing the TransmiR and TarBase data, we found that ASH1L and hsa-miR-375 both regulate the expression of its target gene PAFAH1B1. Studies have found that ASH1L activates hsa-miR-375 and hsa-miR-375 inhibits its target PAFAH1B1. Interestingly, however, ASH1L has been found to be over-expressed in neuroblastoma cell line transfected with normal or mutated alpha-synuclein [62]. This indicates a possibility of higher expression of this TF in brain tissues of PD patients. Besides, studies have identified the association of miR-375 in gastric cancer, breast cancer, cervical cancer [63,64,65]. A recent study with AD patients has identified higher expression of this miRNA (has-miR-375) in patients than controls [66]. This information provides a link to the finding of upregulation of hsa-miR-375 by the TF ASH1L. It is possible that up-regulation of this miRNA in PD patients is responsible for the aberrant production of downstream target genes involved in the pathogenesis. Moreover, the FFL gene PAFAH1B1 has been listed in Genotator database as a responsible candidate gene in AD. PAFAH1B1 was found to be associated with epilepsy, schizophrenia, neuronal migration disorders, cerebellar hypoplasia, etc. nerve related diseases in GeneCards database (http://www.genecards.org/). Therefore, this can be considered as a validation of our findings in PD. Further study on this novel FFL can help us to understand the molecular biology of PD progression.

Significance of lncRNA-mediated gene regulatory network

This network depicts an interesting functional module where a PD-specific miRNA is being regulated by both mTF (either activation or repression) and lncRNA, and this regulatory information is then conveyed to the gene in terms of post-transcriptional repression. Modes of regulations of 44 mTFs associated with 13 PD-specific miRNAs (out of 51 miRNA of the turquoise module) indicated that most of these interactions were ‘activation’ (Table 6). The regulation of hsa-miR-103a-3p of this network is noteworthy. It is not associated with mTFs but has a maximum number of lncRNAs (14 lncRNAs) associated with it (Table 6). This miRNA represses two genes, namely, KPNA1 and NSF. NSF is known to be involved in PD [53] whereas KPNA1 is known to be involved in several neurological disorders including autism and schizophrenia [67]. In contrast to the above findings, hsa-let-7a-5p and hsa-miR-9-5p each has one identified lncRNA (Table 6). These two miRNAs are in turn repress one gene each namely HRAS and OPTN. These two genes are known to be involved in PD [68, 69]. However, both of the miRNAs are associated with 11 and 12 mTFs respectively (Table 6). Since all the 57 lncRNAs of this regulatory network are associated with known PD-specific miRNAs, they might be important epigenetic regulators in PD that were not identified by previous studies. Moreover, the conservation scores of 57 lncRNAs indicate high conservations that strengthen the association of these lncRNAs with PD (Additional file 4: Table S4).

Final screening and selection of 20 most significant SNPs associated with PD

One hundred forty SNPs were identified from the 157 co-expressed genes of the turquoise module. Out of these 157 genes, 18 genes were already known in PD. 20 SNPs are identified to be associated with 18 genes which in turn controlled by PD-specific miRNAs. This strengthens the association of these 20 SNPs in PD (Table 9). In order to find out the functional role of these 20 SNPs, we further analyzed them in F-SNP database (http://compbio.Cs.Queensu.Ca/F-SNP/) [70]. Table 10 Describes the functional category, allele and region of each SNPs. Interestingly 3 SNPs namely, rs535860, rs3814309 and rs3766286 are found to be classified as the ‘conserved’ functional category (predicted by PhastCons_8way and PhastCons_17way within F-SNP database) signifying a conserved functional role of these variations throughout the evolution. Furthermore, our study identified several SNPs associated with hsa-miR-375 involved in FFL of the regulatory network. Among them, SNP rs193223230 is present in the locus of an already known PD-related gene (YWHAZ) (Table 11). Therefore, hsa-miR-375 can be an important PD epigenetic biomarker in our study.

Table 9 20 most significant SNPs in PD with their associated PD-specific miRNAs and genes
Full size table
Table 10 Functional Categories of the 20 most significant PD-related SNPs
Full size table
Table 11 SNP associated with the FFL miRNA and PD-related gene
Full size table

Conclusion

In this study, we have analyzed gene co-expression network, gene regulatory network, and lncRNA-mediated regulatory network based on a PD microarray dataset. The co-expression network, generated through WGCNA, identified eight novel hub genes based on their topological significance in the network. The biological significance and epigenetic regulations of hub genes indicated their involvement in PD-related processes. Analysis of the gene regulatory network (mTF-miRNA-gene-gTF) resulted in the identification of a novel FFL, the regulators of which are unidentified in PD. The lncRNA-mediated regulatory network provided important insight into the lncRNA-mediated regulation of known PD miRNAs. These lncRNAs might be important epigenetic regulators in PD those were not identified by previous studies. Moreover, 57 lncRNAs obtained from lncRNA-mediated regulatory network indicate high conservations that strengthen the association of these lncRNAs with PD. Four lncRNAs (XIST, n406658, n338391, n410211) were identified to be regulating both PD-specific miRNAs and miRNAs not previously known in PD. Moreover, SNP analysis identified 20 significant SNPs along with their associated genes and regulatory miRNAs. These SNPs can be considered as potential risk factors upon further validation. Out of these 20 SNPs, 3 SNPs, namely rs535860, rs3814309 and rs3766286 have conserved functional role throughout the evolution. Thus, findings of our study will be helpful for further PD clinical research and diagnostic purposes.

Methods

Figure 1 depicts the workflow of our analysis.

Microarray data collection

Microarray data generated by Affymetrix HG_U133 array sets (A and B chips) was downloaded from GEO Dataset Browser for data set GDS3128 and series GSE 8397 (from the link http://www.ncbi.nlm.nih.gov/gds/?term=GDS3128) [71]. The microarray data contains 94 samples (47samples from chipA and 47 samples from ChipB) taken from three brain regions Frontal Cerebral Cortex (FCC), Lateral Substantia Nigra (LSN) and Medial Substantia Niagra (MSN). A total of 15 samples were taken from MSN, 9 from LSN, 5 samples from frontal cerebral cortex. 8 medial nigra control samples and 7 lateral nigra control samples and 3 frontal cerebral cortex control samples were considered. The whole dataset was normalized with GCRMA (Gene Chip Robust Multi-Array Averaging) which performs background correction, probe level intensity calculation and summarization [72].

Analysis of differential gene expression

To identify the most significant DE gene set from the microarray data, we performed both the SAM and t-test analysis. When we performed t-test and SAM, we did not get any differentially expressed genes for frontal cerebral cortex.

SAM

Significance Analysis of Microarray (SAM) [30] was used to identify the differentially expressed (DE) genes that are positively and negatively regulated genes among the control and disease samples. The test statistic of SAM is given by:

$$ {d}_i=\frac{r_i}{s_i+{s}_o} $$

Where d i is the relative difference in gene expression, r is the linear regression coefficient of gene i, s i is the standard error of r and s o is a constant chosen to minimize the coefficient of variation of d i . Thus, SAM assigns a score to each gene on the basis of change in gene expression relative to the standard deviation of repeated measurements. In chip A, SAM identified 1518 DE genes at FDR value 0.19%. In chip B, SAM identified 673 DE genes at FDR value 0.11%.

t-test

We further performed paired two sample t-test to identify differentially expressed genes in chip A and chip B. 2-tailed t-test is a measure of the statistical significance of the dataset, in terms of a test statistic t, which is given by:

$$ t=\frac{\overline{x}-\overline{y}}{\sqrt{\frac{{s_x}^2}{n}+\frac{{s_y}^2}{m}}} $$

Where \( \overline{x} \) and \( \overline{y} \) are the sample means, s x and s y are the sample standard deviations, n and m are the sample sizes for two samples, x and y. Under the null hypothesis, this test returns the probability (p-value) of observing a value as extreme or more extreme of the test statistic. Probes corresponding to a portion of the genes showed significant changes in signal intensities in disease sample groups, as compared to the control. These genes were selected as DE genes. t-test analysis identified 4797 and 3120 DE genes in chip A and chip B respectively at p-value 0.05 or 95% confidence level.

Construction of the gene co-expression network

The 458 common DE genes from the chip A were subjected to Weighted Gene co-expression Network Analysis (WGCNA) [73]. This correlation networking method deals with genes differentially expressed over two different conditions (control and disease). In this method, highly correlated nodes are placed into a single module or cluster which are thought to be regulated by the same kind of transcription factors. Therefore, identification of the hub genes of the most significant module can provide insight into the biological significance of that module [74]. Figure 2 describes the gene co-expression network of the most significant co-expressed module (Turquoise module) obtained from WGCNA.

Identification and further analysis of the hub genes

In order to find out the hub nodes, we analyzed the topological properties of the most significant turquoise module using tYNA (http://tyna.gersteinlab.org/) web interface [33]. Degree and Betweenness Centrality (BC) were selected as the criteria for hub gene selection. Nodes with high degree-low BC value (HC nodes) and nodes with high BC-low degree value (HBLC nodes) were considered as hub nodes. Histone modification data for the hub genes were retrieved from human histone modification database (HHMD, http://202.97.205.78/hhmd/index.jsp) [36]. DIANA-Tarbase [37], miRWalk database [38] and TargetScan database [39] were used to study experimentally validated non-coding miRNA-mediated regulation of hub genes. DIANA-LncBase [40] was used to study lncRNAs associated with these miRNAs. The lncRNAs, which are both experimentally validated and computationally predicted (prediction score ≥ 0.70) are considered in our study. The tissue-specific expression data of eight hub genes were collected from GNF Gene Atlas (http://biogps.org/).

Construction of regulatory networks

To get a detailed view of the regulatory pattern of the turquoise module, we built a regulatory network comprising genes of the turquoise module, TFs and miRNAs associated with these genes and TFs associated with the miRNAs. The gene-TF information was obtained from TRANSFAC [41]. Information about the miRNAs associated with the genes of the turquoise module was identified from the DIANA-TarBase database [37]. Information about the TFs regulating the transcription of these miRNAs was obtained from the TransmiR database [45]. By combing all these regulatory information, we constructed a TF-miRNA-gene-TF regulatory network (Fig. 3) which represents the four layers of complex regulatory interactions taking place within the most significant WGCNA module. The network was generated by using Cytoscape software [75].

To identify possible lncRNA-mediated regulation of the miRNAs associated with the genes of the turquoise module, we constructed a lncRNA-mediated regulatory network (Figure not shown). The PD-specific miRNAs that were previously found to be associated with the genes of the turquoise module were searched in the lncbase module of DIANA-LncBase [40] to acquire information on the miRNA-lncRNA pair. This database contains experimentally verified and computationally predicted miRNA targets on lncRNAs. The lncRNAs which are both experimentally validated and computationally predicted (prediction score ≥ 0.70) are considered in our study. In order to identify the regulation of these PD-specific miRNAs present in the turquoise module, we constructed a regulatory network involving the TFs, lncRNA and genes associated with these 13 miRNAs (Figure not shown). The TF-miR-lncRNA-gene regulatory network consisted of 44 TFs, 57 lncRNA, 13 miRNAs and 29 genes of the turquoise module (Table 6). The network was generated by using Cytoscape software [75].

SNP analysis of the highly significant WGCNA module

Figure 4 depicts the flowchart for SNP analysis performed in our study. In order to gain insight about the PD-associated SNPs, the 266 genes of the turquoise module were subjected to SNP analysis. SNPs corresponding to these genes were obtained from the online database SCAN (SNP and Copy number ANnotation database; http://www.scandb.org/) [46]. The expression data served in SCAN has been assayed in HapMap (87 CEU and 89 YRI) [76]. CEU represents the human samples of European descent from Utah and YRI represents the Yoruban samples from Ibadan Nigeria. Genes were queried to retrieve information about the relationship between SNPs and genes at user-specified p-value thresholds [77]. We chose the SNPs that predict gene expression with p-values less than ≤10−05 and frequency greater than 0.10. We obtained a huge number of SNPs corresponding to these genes. To identify only the biologically significant SNPs from this huge number of SNPs we sought the SNPs in MirSNP database (http://202.38.126.151/hmdd/mirsnp/search/) [47]. This database identifies SNPs present in the 3′ UTR of miRNA target sites. We obtained 1525 miRNAs corresponding to the SNPs of 266 genes. These 1525 miRNAs were compared with a list of 92 PD-related miRNAs that was obtained through text mining in PubMed and Human MicroRNA Disease Database (HMDD) [78]. We found 82 miRNAs, related to these 92 miRNAs already known in PD. 140 SNPs associated with these 82 miRNAs were considered as the most relevant SNPs in our study, and these were used for further screening.

Fig. 4
figure 4

Flowchart for SNP analysis performed in our study

Full size image