Abstract

   MicroRNAs (miRNAs) have been reported to contribute to the
   pathophysiology of multiple sclerosis (MS), an inflammatory disorder of
   the central nervous system. Here, we propose a new consensus-based
   strategy to analyse and integrate miRNA and gene expression data in MS
   as well as other publically available data to gain a deeper
   understanding of the role of miRNAs in MS and to overcome the
   challenges posed by studies with limited patient sample sizes. We
   processed and analysed microarray datasets, and compared the expression
   of genes and miRNAs in the blood of MS patients and controls. We then
   used our consensus and integration approach to construct two molecular
   networks dysregulated in MS: a miRNA- and a gene-based network. We
   identified 18 differentially expressed (DE) miRNAs and 128 DE genes
   that may contribute to the regulatory alterations behind MS. The miRNAs
   were linked to immunological and neurological pathways, and we exposed
   let-7b-5p and miR-345-5p as promising blood-derived disease biomarkers
   in MS. The results suggest that DE miRNAs are more informative than DE
   genes in uncovering pathways potentially involved in MS. Our findings
   provide novel insights into the regulatory mechanisms and networks
   underlying MS.
     __________________________________________________________________

   Multiple sclerosis (MS) is one of the most common neurological
   disorders in young adults and the aetiology of this chronic
   inflammatory disorder of the central nervous system (CNS) still remains
   largely unknown. Although many advances regarding MS treatments have
   been made, there is still no cure. MS is characterized by dysregulated
   immune mechanisms and seems to develop in genetically susceptible
   subjects as a result of environmental exposures[32]^1. The disease
   manifests as acute focal inflammatory demyelination with incomplete
   remyelination and axonal loss, which gradually engender multifocal
   sclerotic plaques in the CNS white matter[33]^2. These plaques in turn
   give rise to various cognitive and functional impairments. Several
   epidemiological and gene expression studies have been conducted in
   order to elucidate the underlying processes of this disease, and
   microRNAs (miRNAs), a class of non-coding RNAs, have recently been
   reported to play a role in the development and progression of MS[34]^3.

   Mature miRNAs are single-stranded endogenous RNAs approximately 22
   nucleotides in length that have the ability to posttranscriptionally
   regulate target messenger RNAs (mRNAs). They bind to the 3′untranslated
   region of their target mRNAs and translationally repress them or allow
   for their deadenylation and consequent degradation. It has been shown
   that the expression of more than 60% of mammalian protein-coding genes
   is under the control of these small RNAs and that a single miRNA may
   regulate hundreds of mRNA targets[35]^4. miRNAs partake in diverse
   biological processes such as in modulating the immune system and
   neuroinflammation[36]^5. They are present in stable form in human blood
   and plasma, and their expression profiles can be easily investigated,
   making them ideal MS biomarker candidates[37]^6. Indeed, a number of
   miRNA expression profile studies have compared peripheral blood
   constituents of MS patients to that of healthy controls (HCs),
   reporting a large number of differentially expressed (DE) miRNAs, as
   will be detailed below.

   Much effort has been devoted to integrating and analysing
   high-throughput expression and interaction data with the aim of
   understanding basic principles of human biology and disease. For
   instance, Gerstein et al. constructed a regulatory meta-network by
   hierarchically organizing the genomic binding information of 119
   transcription-related factors derived from the ENCODE project and
   merging this information with other information, including miRNA
   regulation[38]^7. This constituted the first detailed analysis of how
   regulatory information is organized in human. More specifically, Satoh
   et al. constructed molecular networks from proteomic profiling data
   derived from MS brain lesions and analysed these networks using four
   different pathway analysis tools, thereby underlining the relevance of
   extracellular matrix-mediated focal adhesion and integrin signalling in
   the development of chronic MS lesions[39]^8. Riveros et al.
   investigated whole-blood gene expression data of MS patients using a
   variety of computational methods including transcription factor binding
   motif (TFBM) overrepresentation analysis and functional profiling, and
   uncovered a network of transcription factors (TFs) that potentially
   dysregulate several genes in MS[40]^9. Similarly, Liu et al. created a
   molecular network based on differentially coexpressed TFs and genes in
   peripheral blood mononuclear cells (PBMC) of MS patients and performed
   pathway enrichment analyses to discover regulatory relationships
   between TFs and target genes[41]^10. In contrast to the three
   previously described studies, more recent studies took miRNAs into
   account when constructing MS-associated molecular networks.
   Nevertheless, non-overlapping panels of DE miRNAs resulted, possibly
   because these studies were limited in that they comprised small patient
   sample sizes, using different high-throughput technologies, or dealing
   with patients already receiving immunomodulatory treatment. Following
   microarray analysis of miRNAs and genes in PBMC of MS patients
   undergoing interferon-beta (IFN-β) treatment, Hecker et al. assembled
   an interaction network of IFN-β-responsive miRNAs and genes using
   several miRNA target databases[42]^11. Likewise, Jernås et al.
   generated an interaction network between DE miRNAs and genes in T cells
   of IFN-β treated MS patients using computationally predicted miRNA
   targets[43]^12. Another study by Angerstein et al. introduced an
   approach to construct molecular networks by integrating dysregulated
   miRNAs in MS, which were uncovered in various studies, and miRNA
   targets from target gene prediction databases[44]^13.

   Most of the aforementioned studies were conducted in small patient
   cohorts without technical replicates and independent
   validation[45]^3,[46]^14. It is thus likely that some of the findings
   are false positives. Beside small patient cohort sizes, these studies
   were performed using different samples or tissues (e.g., peripheral
   blood or monocyte), different technological microarray platforms, and
   different statistical methods to analyse the data. Consequently, little
   overlap in DE miRNAs can be observed between the various studies in
   MS[47]^3,[48]^14. Consensus methods are commonly used in medicine to
   define levels of agreement on conflicting data[49]^15. Hence, a
   consensus approach based on several expression profile studies is
   likely to reduce the finding of false positives and to improve the
   accuracy in identifying genes and miRNAs relevant in MS. In this study,
   we developed a new consensus-based method to analyse and integrate
   microarray expression data and other publically available data to gain
   a deeper understanding of the mechanistic impact of miRNAs in MS and to
   overcome the challenges posed by small studies. We created two
   regulatory networks, a miRNA- and a gene-based network, and identified
   18 DE miRNAs and 128 DE genes that may contribute to the regulatory
   alterations behind this inflammatory disease. Of the 18 miRNAs,
   let-7b-5p and miR-345-5p are the most promising biomarkers. We also
   show that DE miRNAs are more powerful than DE genes in uncovering
   pathways potentially involved in MS.

Results

miRNA-Based Network

Differential MicroRNA Expression in MS

   In order to obtain a list of miRNAs involved in MS, we preprocessed and
   analysed four miRNA microarray datasets ([50]Table 1, [51]Fig. 1). When
   comparing the miRNA expression levels in the blood of MS patients and
   HCs, we found a total of 269, 71, 398, and 83 DE miRNAs (t-test
   p-value ≤ 0.05) in the datasets [52]GSE17846[53]^16,
   [54]GSE21079[55]^17, [56]GSE31568[57]^18, and [58]GSE39643[59]^19,
   respectively, and uncovered 39 miRNAs that were significantly DE
   (p-value < 0.05) in at least 3 of the 4 datasets ([60]Supplementary
   Fig. S1). A permutation test suggested that the 39 DE miRNAs are indeed
   relevant in MS (p-value < 0.002). We next took the direction in which
   the DE miRNAs were dysregulated into consideration. We thereby
   identified 18 DE miRNAs that were significantly DE and consistently
   expressed either at higher or at lower levels in MS in at least 3 of
   the 4 datasets ([61]Table 2). A second permutation test conferred
   additional evidence supporting the implication of these 18 DE miRNAs in
   MS (p-value < 0.002). Out of these 18 candidates for the miRNA-based
   network, let-7b-5p and miR-345-5p were the only DE miRNAs
   differentially expressed in the same direction in all four datasets.
   The average fold-changes of let-7b-5p and miR-345-5p were 1.81 and 1.26
   in MS patients compared to HCs, respectively. Hence, let-7b-5p and
   miR-345-5p are promising blood-derived biomarkers of MS.

Table 1. Microarray datasets used for the differential expression analysis.

   GEO dataset Data Platform Controls MS Tissue Reference
   microRNAs
    [62]GSE17846 Normalized [63]GPL9040 21 20 Peripheral blood [64]16
    [65]GSE21079 Normalized [66]GPL8178 37 59 Peripheral blood [67]17
    [68]GSE31568 Normalized [69]GPL9040 70 23 Peripheral blood [70]18
    [71]GSE39643 Normalized [72]GPL15847 8 8 Blood-derived monocytes
   [73]19
   Genes
    [74]GSE17048 Normalized [75]GPL26947 45 99 Peripheral blood [76]9
    [77]GSE21942 Normalized [78]GPL570 15 12 PBMC [79]14
    [80]GSE41890 Raw [81]GPL6244 24 22 Peripheral blood leukocytes [82]31
    [83]GSE43591 Normalized [84]GPL570 10 10 Peripheral blood [85]12
   [86]Open in a new tab

   GEO dataset: Gene Expression Omnibus dataset (series) are represented
   by a series accession number beginning with the letters GSE; Platform:
   a platform provides the physical setup of an assay such as an array and
   is linked to a GEO platform accession number beginning with the letters
   GPL; Controls: control samples; MS: number of multiple sclerosis
   patient samples; PBMC: peripheral blood mononuclear cells.

Figure 1. Workflow and general characteristics of the networks in this study.

   [87]Figure 1
   [88]Open in a new tab

   (a) Bioinformatics workflow, illustrating the tools and databases
   employed to uncover the molecules and interactions in the multiple
   sclerosis (MS)-associated gene- and microRNA (miRNA)-based regulatory
   networks. (b) General configuration of the miRNA- (left) and gene-based
   (right) networks. The blue nodes represent transcription factors (TFs),
   the yellow node represents a miRNA, and the white nodes represent
   molecules that are neither TFs nor miRNAs. The green edges represent
   activating interactions, whereas the red one represents an inhibitory
   interaction.

Table 2. Differentially expressed microRNAs in our study and in other
multiple sclerosis studies.

   microRNA Regulation DE consensus DE in extra miRNA studies in MS
   let-7b-5p up [89]16, [90]17, [91]18, [92]19 [93]11,[94]19
   let-7g-5p up [95]16,[96]18,[97]19 [98]17,[99]46,[100]47
   miR-19b-3p up [101]16,[102]18,[103]19 [104]19,[105]49,[106]50
   miR-20b-5p down [107]16, [108]17, [109]18
   [110]16,[111]17,[112]47,[113]51,[114]52,[115]71
   miR-30a-5p up [116]16,[117]18,[118]19 [119]16,[120]51, [121]52, [122]53
   miR-125a-5p up [123]16, [124]17, [125]18
   [126]12,[127]42,[128]72,[129]73
   miR-146a-5p up [130]16,[131]18,[132]19 [133]19,[134]51,[135]74,
   [136]75, [137]76, [138]77
   miR-186-5p up [139]16,[140]18,[141]19 [142]16
   miR-221-3p up [143]16,[144]18,[145]19 [146]19,[147]47
   miR-300 down [148]16,[149]18,[150]19 —
   miR-328 up [151]16, [152]17, [153]18 [154]16,[155]51,[156]53,[157]73
   miR-345-5p up [158]16, [159]17, [160]18, [161]19 —
   miR-363-3p down [162]16, [163]17, [164]18 [165]46,[166]50,[167]73
   miR-379-5p down [168]16,[169]18,[170]19 [171]19
   miR-450b-5p down [172]16,[173]18,[174]19 —
   miR-580 down [175]16,[176]18,[177]19 —
   miR-664a-3p up [178]16,[179]18,[180]19 —
   miR-1206 down [181]16,[182]18,[183]19 [184]19
   [185]Open in a new tab

   Listed under the header “microRNA” are the 18 microRNAs (miRNAs) that
   were differentially expressed (DE) in our study and that were DE in the
   same direction in at least three of the four miRNA expression datasets
   used for this study. A brief description of these miRNA expression
   datasets can be found in [186]Table 1. “Up” regulated means that a
   miRNA is expressed at a higher level in multiple sclerosis (MS)
   patients compared to controls and vice versa for “down” regulation. In
   the third column, we provide references to the datasets in which we