signature=a4d1083d28aa64b1dbf3be502f86441a,Deciphering miRNA transcription factor feed-forward loops...

CF- and CFTR-related gene and miRNA expression changes

An outline of the work flow is given in Figure 1, and a summary of CF- and CFTR-related gene and miRNA data are given in Table 1. Initially, a comprehensive list of differentially expressed genes (DEG) was compiled using diverse data sets from CF patients in addition to literature findings regarding CFTR-associated gene networks. Subsequently, common regulations of DEGs by TFs and miRNAs were investigated by means of databases searches in addition to experimental data retrieved from literature searches.

For this purpose, the publically available GEO data sets GSE2395, GSE55146, and GSE15568 were analyzed. The data informed on whole genome gene expression profiling in cystic fibrosis patients with mild and severe lung disease using either tissue samples obtained from bronchial brushings or nasal epithelium as well as rectal epithelia of CF and non-CF individuals. In all 1,042 DEGs were obtained, however there was little to no overlap among DEGs when individual studies were compared (see Figure 2A).

Figure 2

VENN diagram of CFTR- and CF-related gene and miRNA expression changes. (A) Common genes among three independent CF patient-related whole genome gene expression data sets; (B) common miRNAs identified in bronchial brushings from CF patients or CF bronchial epithelial cell lines; (C) commonality among 112 CFTR- and CF-related miRNAs derived from the study of [40]; (D) commonality among 112 CFTR- and CF-related miRNAs derived from the study [41].

Furthermore, to discriminate CF-specific and CFTR-related miRNA networks profiling data obtained from CF patient airway epithelium and CF related cell lines as well as primary human airway epithelium were considered using the findings reported by Oglesby et al. [40] and Bhattacharyya et al. [41]. As denoted for the whole genome gene expression profiling studies major discrepancies among the reported miRNA profiling studies were observed with little overlap in identified miRNAs using either bronchial brushings from CF patients or CF bronchial epithelial cell lines (see Figure 2B). In regards to the CFTR associated miRNA regulations the data reported by Ramachandran et al. [25] were used and yielded 112 differential expressed miRNAs. As depicted in Figure 2C, 31 down- and 12 upregulated miRNAs were in common when the findings of Oglesby et al. and CFTR-associated miRNAs were compared as determined in human airway epithelium. Likewise, two down- and 10 upregulated miRNAs were commonly regulated when the data reported by Bhattacharyya et al. and findings from CFTR-associated miRNAs were compared (Figure 2D). Taken collectively, a total of 93, 22, and 112 uniquely regulated miRNAs were extracted from experimental data and among the three studies seven miRNAs were in common that permitted an in-depth assessment of the miRNA-CF disease relationship.

Apart from CF-specific gene and miRNA expression changes several of the identified genes are also co-expressed or are involved in the same pathways or biological process as determined for the CFTR-associated gene network using human airway epithelium. This is consistent with our understanding of the pathogenesis of CF with most of the common regulated genes influencing folding, sorting, and degradation of proteins and included the ubiquitin mediated proteolysis, protein processing in the endoplasmic reticulum and the proteasome. It has been established that the ubiquitin-proteasome pathway controls the degradation of CFTR and therefore plays a central role in CF .

To be able to construct FFLs, different types of regulatory relationships were considered, that is, genes regulated by either miRNA (miRNA → gene) or TF (TF → gene), as well as the relationships between miRNA regulating TFs and vice versa (miRNA → TF and TF → miRNA) in addition to the gene-gene interaction as depicted in the work flow diagram (Figure 1 and Table 2). The findings entrained on the CFTR gene and miRNA networks were validated using data derived from CF patients as detailed in Table 2.

Table 2

Summary of five different kinds of regulatory relationship and constructed FFLs

miRNA gene target relationship

Initially, the miRNA targets were predicted using TargetScan (see the Method Section for further details and Table 2). There were a total of 1,615 miRNA → gene pairs, which involved 99 CFTR specific miRNAs (out of 112 miRNAs identified) and 226 CFTR-regulated genes (out of 419 genes identified from Reference [25]). Among them, the miRNAs, hsa-miR-200b, hsa-miR-200c, and hsa-miR-429 regulated the largest number of genes. The average number of targeted genes per miRNA is 16.

It is well known that the miRNAs from the same family share similar regulatory functions and mechanisms [63]. We therefore constructed a miRNA-based network using the CF gene information and investigated whether the relationship between the miRNAs from the same family was preserved as a means to verify the chosen approach. Figure 2 depicts the miRNAs network module where each node is a CF miRNA while an edge denotes the Tanimoto similarity between each of the two miRNAs. It can therefore be demonstrated that the miRNAs from the same family (for example, hsa-let-7a/b/c/e/g) were preserved with higher Tanimoto similarity. Likewise, in the constructed miRNA-gene network NEDD4L was regulated by 33 miRNAs. This gene codes for an E3 ubiquitin protein ligase and knockdown of NEDD4L in lung epithelia causes airway mucus obstruction, goblet cell hyperplasia, inflammation, fibrosis, and even death after 3 weeks of exposure in an animal disease model [64]. Such experimental data support the relevance of the constructed miRNA-gene network.

Using ChIPBase, a total of 422 miRNA → TF pairs were identified and consisted of 89 CFTR-specific miRNAs and 52 human TFs. Among the 422 miRNA → TF pairs, hsa-miR-27a was involved in the regulation of 14 TFs. Meanwhile, the genes BCL11A, SMAD2/3, and SMAD4 were regulated by the largest number (n =30) of miRNAs. Note, reduced SMAD3 protein expression and altered TGFβ1-mediated signaling in CF epithelial cells were reported [65].

TF-miRNA/gene regulatory networks

TF → miRNA circuitries were constructed using information retrieved from ChIPBase [46]. A total of 3,295 TF → miRNA combinations were computed and this involved 114 and 102 unique TFs and miRNAs, respectively (see Additional file 3: Table S3). For instance, hsa-miR-106b, hsa-miR-25, and hsa-miR-93 were regulated by 72 TFs. Similarly, a total of 16,860 TF-gene pairs were computed and involved 105 TFs and 387 gene targets (see Additional file 3: Table S3). Of the 99 TFs, c-Myc targeted the largest number of CFTR-related genes. It was earlier demonstrated that proteolysis of c-Myc in vivo is mediated by the ubiquitin-proteasome pathway [66]. Among the 387 CFTR-related genes, the gene regulated by the largest number of TFs was UBE2D3 (ubiquitin-conjugating enzyme E2 D3). We further searched for common genes among CFTR and 1,042 DEGs and found 38 genes to be mutual.

CFTR-specific feed-forward loops (FFLs)

It had been demonstrated that composite FFLs (that is, the combined miRNA and TF participating in the regulation of target genes) are more effective in unveiling disease mechanisms than single one as denoted by TF → miRNA or miRNA → TF considerations [45]. As shown in the third step of Figure 1 and as summarized in Table 2, FFLs were evaluated for their significance using a hypergeometric test with multiple testing corrections. Such analysis revealed 449 unique CFTR-entrained FFLs including 41 miRNA-FFLs, 393 TF-FFLs, and 15 composite-FFLs, as shown in Additional file 3: Table S3. The results indicated that the constructed composite-FFLs were of largest relevance followed by miRNA-FFLs and TF-FFLs. Therefore, and based on statistical significance the 15 composite FFLs were employed to search for repurposing candidates for the treatment of CF (Additional file 4: Figure S1). These FFLs contained 12 miRNAs, 11 TFs, and 104 CFTR-related genes, respectively.

We further considered the results of the Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis for the commonly targeted genes of the 15 composite-FFLs. Some composites FFLs, such as hsa-miR-192↔CTCF and hsa-miR-191↔TCF7L2, just have one gene in common; thus no enriched pathways were obtained. As depicted in Additional file 5: Figure S2, 24 different pathways belonging to 13 different functional categories were considered. Among them, two pathways, ubiquitin-mediated proteolysis and protein processing in the endoplasmic reticulum, dominated the composite FFLs, whose major function is folding, sorting, and degradation and these are key mechanism in CF [67]. Other FFLs are involved in insulin and TGF-beta signaling pathways and endocytosis. For instance, CF-related diabetes (CFRD) is a common complication of CF and insulin resistance may also affect lung function [68]. Likewise, transforming growth factor-beta (TGF-beta) plays a central role in fibrosis, contributing to the influx and activation of inflammatory cells, the epithelial to mesenchymal transdifferentiation (EMT) of cells, and the activation of fibroblasts and modulation of extracellular matrix production [69]. Downregulation of CFTR by TGF-beta limits epithelial chloride secretion, which causes mucus block [70]. It was also reported that CF is associated with a defect in apical receptor-mediated endocytosis [71].

Validation of FFLs in CF patient samples

To determine disease relevance of the FFLs and to study protein-protein-interactions (PPI) among members of the composite FFLs the following data were considered: (1) whole genome gene expression data; and (2) miRNA profiling studies using samples obtained from bronchial brushings or nasal epithelium as well as rectal epithelia of CF patients with mild and severe disease and non-CF individuals.

Initially, a total of 123 CFTR genes were retrieved from the study of Ramachandran et al. and for 80 genes a total of 135 PPIs were observed in STRING network analysis. This demonstrates that the network partners actually interact with each other. Moreover, for seven and nine disease-regulated miRNAs and TFs, respectively, a total of 97 PPIs among 66 regulated genes were observed further evidencing interactions among the predicted targets (see Figure 3).

Figure 3

Protein-protein interaction networks of CFTR-related genes. A total of 419 genes were retrieved from the study of Ramachandran et al. of which 123 could be mapped to the STRING database version 9.1. Only PPI interaction for homo sapiens were considered and a confidence score >0.4 was requested. Among them we considered those genes linked to the 15 composite FFL that were constructed. This revealed 80 genes and a total of 135 PPIs to be in common. Subsequently, for nine disease-specific composite FFLs 66 genes and 97 PPIs were observed.

Subsequently, we considered disease regulated miRNA and its directionality based on CF patient samples and therefore analyzed the data of Oglesby et al. [40] and Bhattacharyya et al. [41] with respect to the composite FFLs. This revealed a total of seven miRNAs (hsa-miR-26b, hsa-miR-29c, hsa-miR-135b, hsa-miR-155, hsa-miR-192, hsa-miR-200c, and hsa-miR-340) and nine FFLs to be CF associated. Note, in the case of miR-155 three different TFs are involved, that is, SP1, NFKB1, and EBF1, therefore giving rise to three distinct disease-relevant FFLs. We considered miRNAs whose expression was either increased or decreased in CF patient samples (see Figure 4). In order to predict targets of disease associated FFLs we employed a consensus approach using 10 different algorithms (see Additional file 6: Figure S3). The predicted gene targets were considered positive only when confirmed by at least eight different algorithms. Apart from disease specific miRNAs that were used to construct FFLs the regulation of target genes was also considered in CF patient samples. As described above we compiled a total of 1,042 DEGs derived from GEO submissions GSE2395, GSE55146, and GSE15568, and observed DEGs to be commonly regulated in CF samples and disease-specific FFLs, once again providing evidence for the clinical relevance (see Figure 5).

Figure 4

Composite FFLs of CF-regulated miRNAs. The nodes are marked as green diamonds whereas blue rectangles and gray ellipses denote transcription factors (TF), miRNAs and genes, respectively. Genes color-coded in red are among the 1,042 CF gene expression data retrieved from three independent CF gene expression data sets. The edges are t-shapes, circle-shapes, and gray solid lines, which denotes miRNAs regulating genes/TF and TFs regulating genes/miRNAs, and gene-gene interaction, respectively.

Figure 5

Protein-protein interaction networks of CF-related miRNA. A total of 7 CF regulated miRNAs were used to predict gene targets by employing a total of 10 different algorithms. This defined 263 putative targets which were mapped to the STRING database version 9.1 and revealed 247 PPI among 138 gene targets. Only PPI interaction for homo sapiens were considered and a confidence score >0.4 was requested. The predicted 138 gene targets were mapped to 1,042 DEGs identified among three independent CF patient-related whole genome gene expression data sets. This identified seven genes in common and a total of 19 PPI.

Repurposing candidates for the treatment of CF

Drug repositioning is a process of identifying alternative indications for existing drugs with acceptable safety at affordable price. To identify drugs with potential use in the treatment of CF patients, we exploited small molecules that affect the expression of miRNA which are part of the composite FFLs. We retrieved data from two databases (that is, SM2miR and Pharmaco-miR) as described in Figure 1. Notably, the SM2miR compiles a list of small molecules that interact with miRNAs from the literatures while Pharmaco-miR provides the drug-miRNA association based on the PharmGKB data [72]. We then compared the marketed drug list from DrugBank (version 3.0, [73]) with those identified by SM2miR and Pharmaco-miR as having an ability to influence the expression of miRNA which are part of the FFLs. This process identified 48 unique drugs being strongly designated as repurposing candidates for the treatment of CF patients.

To assess the validity of the CF repurposing candidates, we conducted a two-step analysis. First, we queried clinicaltrials.gov (www.clinicaltrials.gov) that archives clinical studies of human subjects conducted around the world. Collectively, Table 3 compiles all 48 repurposing candidates for CF along with their original indications and literature/clinical trial data. Note that eight out of 48 drugs were already investigated for the treatment of CF patients. For the remaining drug candidates we additionally queried PubMed using the keyword (‘drug name’ (and) ‘cystic fibrosis’) followed by reading. Here, 18 out of 43 repurposing candidates have literature citations to support their potential use in CF. Collectively, we found 54.2% of the candidates (26 drugs out of 48 repurposing candidates) to have at least one published study or clinical trial related to CF. Additional file 7: Table S4 lists the information of all 48 repurposing candidates related to drug safety and affordability that were obtained from the FDA-approved drug product labels and the DrugBank V3.0 database.

Table 3

Summary information of 48 repurposing candidates for cystic fibrosis (CF) treatment

We further assessed the therapeutic indications of the repurposing candidates and found two categories, that is, Alimentary tract and metabolism (P <0.0016) and Antineoplastic and immunomodulating agents (P <0.0009) to be significantly enriched. For the different therapeutic categories of the 48 drug repurposing candidates see Figure 6. Note that the two therapeutic categories include some drugs with boxed warning that need to be considered.

Figure 6

The distribution of repurposing candidates for CF at the first level of Anatomical Therapeutic Chemical Classification System (ATC). Each bar was divided by safety concerns including boxed warning, no boxed warning, and nutritional supplementation. The statistical significance of the therapeutic categories associated with CF are A and L based on the Fisher’s exact test with a P value cutoff of 0.01.