A Unique TGFB 1-Driven Genomic Program Links Astrocytosis , Low-Grade Inflammation and Partial Demyelination in Periplaques of Spinal Cords from Progressive Multiple Sclerosis Patients

We previously reported that in multiple sclerosis (MS) patients with a progressive form of the disease, spinal cord periplaques extend distance away from plaque borders and are characterized by the co-occurrence of partial demyelination, astrocytosis and low-grade inflammation. However, transcriptomic analyses comparing periplaques to adjacent normal-appearing white matter (NAWM) areas did not allow providing a comprehensive view of molecular events in astrocytes vs oligodendrocytes. Here, we re-assessed our transcriptomic data with the aim of identifying functionally-relevant co-expression networks that would reflect astrocyte vs oligodendrocyte molecular signatures in periplaques. We identified an astrocytosis-related gene module comprising GFAP, the hub gene CX43/GJA1 and a set of transcripts forming a TGFB/SMAD1/SMAD2 genomic signature. Partial demyelination was characterized by a co-expression network which, besides myelin genes, comprised a highly significant number of translation/elongation-related genes. Interestingly, the main oligodendrocyte-related hub we identified was NDRG1, a gene previously shown to be specifically silenced in the NAWM of MS patients. This result indicated that NDRG1 down-regulation could be an important event in the process of periplaque partial demyelination. To establish a putative link between NDRG1 down-regulation and a cytokine/chemokine signature, we then sought to identify cytokine/chemokine genes whose mRNA levels inversely correlated with those of NDRG1. Following this approach we found 5 candidate immune-related genes whose up-regulation associated with NDRG1 down-regulation: TGFB1, PDGFC, IL-17D, IL-33, and IL-12A. From these results we propose that in the spinal cord of MS patients with progressive forms of the disease, TGFB1 may Preprints (www.preprints.org) | NOT PEER-REVIEWED | Posted: 23 August 2017 doi:10.20944/preprints201708.0080.v1 © 2017 by the author(s). Distributed under a Creative Commons CC BY license.


Introduction
Multiple sclerosis (MS) is a neuroinflammatory disorder of the central nervous system (CNS) and the leading cause of permanent neurological disability in young adults [1].While autoimmune mechanisms targeting the myelin sheath have been extensively demonstrated, there are still uncertainties regarding the pathophysiology of MS progressive forms [2][3][4][5].Importantly, patients suffering from secondary or primary progressive multiple sclerosis (SPMS and PPMS respectively) frequently exhibit slowly evolving spinal cord-related symptoms and studies based on magnetic resonance imaging (MRI) or neuropathological analyses demonstrated a high occurrence of spinal cord lesions as compared to patients with a relapsing-remitting form of the disease (RRMS) [6,7].In this context, we recently described a particular type of tissue alteration in the spinal cord of MS patients with SPMS or PPMS [8].We found that periplaques i.e. areas that surround the borders of fully-demyelinated lesions, are characterized by a partial loss of myelin, a low-grade inflammatory process and an astrocytosis signed by a constant upregulation of the astrocyte markers Cx43/GJA1 and AQP4.Spinal cord periplaques extend distance away from the plaque borders and appear to evolve, at least in part, independently from plaque activity.However, when performing paired comparisons between periplaques and areas of adjacent normal-appearing white matter (NAWM), whole genome transcriptomic analyses did not allow providing a comprehensive picture of the molecular events driving astrocytosis, inflammation and partial demyelination in periplaques.Indeed, only 34 genes were found to be constantly up-regulated in periplaques and 57 constantly down-regulated.A potential caveat of the differential expression approach we opted for, relies on the varied levels of molecular alterations that may occur in areas of NAWM [9,10].In particular, previous studies demonstrated that NAWM in MS brains or spinal cords exhibit; i) signs of low-grade inflammation [9,10] and astrocytosis [11][12][13] ii) an abnormal lipid and protein composition of myelin sheaths [14][15][16] and iii) an epigenetic silencing of the oligodendrocyte gene NDRG1 [17].Thus, when performing paired comparisons between periplaques and NAWM, the genomic alterations occurring in NAWM might have masked functionally-relevant molecular events taking place in both periplaques and NAWM.
In the present paper, we re-assessed our transcriptomic data with the aim of identifying gene co-expression networks that would link astrocytosis, inflammation and myelin alterations in spinal cord periplaques.In a first step, we identified Cx43/GJA1 as the only astrocyte-specific hub gene among genes that constantly co-upregulated in periplaques.To identify a molecular signature of astrocytosis in periplaques, we then used Cx43/GJA1 as a "bait" and determined the top-200 genes that more closely co-upregulated with Cx43/GJA1.Interestingly, besides GFAP, this Cx43/GJA1 genomic module comprised the two cytokines IL17D and IL33 and a set of genes forming a TGFB1/SMAD1/SMAD2 progliotic signature.In parallel, we identified NDRG1 as the only oligodendrocyte-specific hub gene among genes that constantly co-downregulated in spinal cord periplaques.To identify a molecular signature of oligodendrocytes in periplaques and NAWM, we then used NDRG1 as a "bait" and determined the top-200 genes that more closely co-dowregulated with NDRG1.Interestingly, NDRG1 co-expression module comprised not only the myelin genes PLP1, MBP and MOBP but a highly significant number of transcripts involved in the process of mRNA translation and elongation.Finally, to unravel molecular mechanisms potentially linking partial demyelination to inflammation, we established a list of 193 candidate genes coding for cytokines, chemokines or growth factors from which we identified the top-5 genes whose mRNA levels inversely correlated with those of NDRG1 in periplaques.We found that NDRG1 mRNA levels inversely correlated with TGFB1, IL33, IL17D, IL12A and PDGFC.

Workflow and results summary
Our workflow and main results are summarized in Figures 1 and 2.

Identification of Cx43/GJA1 as an upregulated astrocyte-related hub gene in spinal cord periplaques
Based on the data we previously obtained by paired comparisons of periplaques vs NAWM transcriptomic profiles, we assessed whether the 34 coding genes identified as constantly upregulated in periplaques were indeed co-upregulated in periplaques.We identified a set of 21 co-upregulated mRNA species among which Cx43/GJA1 (Gap junction protein alpha 1 also named Connexin 43) was the only astrocyte-related hub gene (Figure 3).Importantly, Cx43 on astrocytes was previously reported to be up-regulated in the NAWM of MS patients as compared to controls [18].
Figure 3: Network of co-upregulated genes in spinal cord periplaques.A co-expression network was generated with the set of genes that are constantly upregulated in periplaques as compared to adjacent NAWM.In this co-expression module, hubs were defined as genes (nodes) being connected to at least half of the total number of genes (nodes) forming the network.Two hub genes (black-filled nodes) were identified, namely CPE (Carboxypeptidase E) and the astrocyte-specific gene marker Cx43/GJA1 (Gap junction protein alpha 1 also named Connexin 43).Genes are designated by their gene symbols.

Identification of NDRG1 as a downregulated oligodendrocyte-related hub gene in spinal cord periplaques
We then assessed whether the 57 coding genes identified as constantly downregulated in paired comparisons of periplaques vs NAWM were co-downregulated in periplaques.We identified a set of 44 co-downregulated genes (Figure 4 and Table 1) among which NDRG1 was the only hub gene previously recognized as being specifically expressed by oligodendrocytes in the central nervous system [19][20][21].A co-expression network was generated with the set of genes that are constantly downregulated in periplaques as compared to adjacent NAWM.In this co-expression module, hubs were defined as genes (nodes) being connected to at least half of the total number of genes (nodes) forming the network.Five hub genes (black-filled nodes) were identified, namely NDRG1 (N-myc downstream regulated 1), PEA15 (phosphoprotein enriched in astrocytes 15), RASSF2 (Ras association domain family member 2), PAQR6 (progestin and adipoQ receptor family member 6) and LHPP (phospholysine phosphohistidine inorganic pyrophosphate phosphatase).Genes are designated by their gene symbols.Full names of functionally-relevant genes are provided in Table 1.Legend: A co-expression network was generated with the set of genes that constantly downregulated in periplaques as compared to adjacent NAWM.In this co-expression module, hubs were defined as genes (nodes) being connected to at least half of the total number of genes (nodes) forming the network.

A unique set of genes co-upregulate with CX43/GJA1 in periplaques of MS spinal cords
To identify a putative astrocyte molecular signature in periplaque areas of MS spinal cords, we used the astrocyte-specific hub gene Cx43/GJA1 as a "bait" and retrieved the top-200 mRNA species that more closely co-expressed with Cx43/GJA1 in our whole set of data.Functionally-relevant genes are shown in Table 2 and Figure 5.The full list of Cx43/GJA1 co-expressed genes is provided in data supplement 2. To identify an astrocyte molecular signature in spinal cord periplaques, we used the astrocyte-specific hub gene Cx43/GJA1 as a "bait" and retrieved the top-200 mRNA species that most closely co-upregulated with Cx43/GJA1.The co-expression network formed by functionally-relevant genes is shown.Genes are designated by their gene symbols.Gene full names are provided in Table 2. Legend: A co-expression analysis was performed in which the astrocyte-specific gene CX43/GJA1 was used as a "bait" to identify a putative molecular signature of reactive astrocytes in periplaque areas of MS spinal cords.The top-200 genes that more closely co-expressed with CX43/GJA1 were identified.Listed are genes considered as functionally-relevant in the context of astrocytosis and inflammation.
Besides GFAP, such a genomic network comprised 3 groups of genes that we found of particular relevance in the context of astrocytosis, inflammation and tissue remodeling: i) transcription factors (TFs), ii) cytokines and iii) genes coding for extra-cellular matrix (ECM) molecules (Figure 3 and Table 2).Interestingly, the TFs that co-expressed with Cx43/GJA1 comprised SMAD1 and SMAD2, two major signaling components of the TGF-beta and Bone Morphogenetic Proteins (BMPs) pro-gliotic pathways [22][23][24][25].Moreover, TFs that co-expressed with Cx43/GJA1 also comprised SOX2 a stem cell-related gene expressed by proliferating astrocytes [26,27] and NFAI, a positive regulator of GFAP transcription [28].
On the other hand, the androgen receptor (AR), which was also found to co-express with Cx43/GJA1, was previously shown to inhibit reactive astrocytosis under varied CNS conditions [29][30][31][32][33][34].Of note, AR co-expressed also with Cx43/GJA1 when excluding the 2 male patients from our analysis (data supplement 3).To establish a potential link between Cx43/GJA1 co-expressed TFs and other genes of the CX43/GJA1 module, we performed an enrichment analysis using ChEA2016, a Chip-Seq and Chip-Chip database and webtool [35].We found that the GJA1/Cx43 module was significantly enriched (adjusted p-value < 0.01) in previously identified gene targets of AR or SOX2 in human cells (data supplement 4).However, the GJA1/Cx43 module was not enriched in known functional pathways as assessed with the Reactome 2016 database and webtool [36].Surprisingly, we found that only 2 cytokines, namely IL-17D and IL33, co-upregulated with Cx43/GJA1 in spinal cord periplaque areas.Other inflammation-related genes comprised notably CD44 and CD200 that were both previously demonstrated on reactive astrocytes [37,38].Finally, the GJA1/Cx43 module comprised several ECM-related genes that are expressed by

A unique set of genes co-express with NDRG1 in periplaque areas of MS spinal cords
To unravel a molecular signature of oligodendrocytes in periplaque areas of MS spinal cords, we used the oligodendrocyte-related hub gene NDRG1 as a "bait" and retrieved the top-200 mRNA species that more closely co-expressed with NDRG1.The full list of NDRG1 co-expressed genes is provided in data supplement 5.As expected, such a module comprised genes coding for myelin proteins, namely MBP, MOBP and PLP1.Similarly to the Cx43/GJA1 module, the NDRG1 module also comprised several TFs.However, the network of NDRG1 co-expressed genes was not enriched in known targets of these TFs, as assessed with the ChEA2016 database and webtool.A survey of the "TargetScan microRNA" [39] database showed that the NDRG1 co-expression module was not significantly enriched in known targets of miRNAs.However, pathway analysis with the webtool "Reactome 2016" showed a highly significantly enrichment in genes involved in the functional pathway "Eukaryotic Translation Elongation" (Reactome 2016, adjusted p-value=1.37E-9)(Figure6 and Table 3).3.  Legend: A co-expression analysis was performed in which the oligodendrocyte-specific gene NDRG1 was used as a "bait" to identify a putative molecular signature of oligodendrocytes in spinal cord periplaques.The top-200 genes that more closely co-express with NDRG1 in spinal cord periplaque areas were identified.Listed are genes considered as functionally-relevant in the context of myelin loss.
This finding appears of particular interest since, in oligodendrocytes , the translation of MBP and MOBP was previously shown to rely on specific mechanisms requiring a transport of the protein translation machinery (including ribosomal proteins) along the cell processes of oligodendrocytes [40][41][42][43][44][45].To explore the existence of potential links between NDRG1 and the processes of translation/elongation, we performed a meta-analysis of NDRG1 protein interactants.While the STRING database allowed to recover only 20 direct protein interactants with no significant enrichment relating with translation/elongation, the Wiki-Pi database, gathering data from high throughput technologies, listed 64 NDRG1 interactants (data supplement 6) that showed a highly significant enrichment in the "Translation" pathway (Reactome 2016, adjusted P-value = 4.80E-11).Similar results were obtained when assessing NDRG1 interactants in the protein interactome database TissueNet2 [46](data not shown).This result points to a potential link between NDRG1 down-regulation in oligodendrocytes and a subsequent altered translation/elongation of myelin genes leading to partial demyelination.

Identification of candidate soluble factors that may trigger NDRG1 silencing in periplaque areas of MS spinal cords
While a previous study showed that NDRG1 is specifically silenced in the NAWM of MS patients [19][20][21], we found that NDRG1 was constantly downregulated in spinal cord periplaques as compared to adjacent NAWM.We thus sought to identify candidate soluble factors (in particular immune molecules) that would be potentially involved in a repression of NDRG1 expression in periplaque areas of MS spinal cords.To this aim, we first established a list of 193 candidate genes coding for cytokines, chemokines or growth factors.We then used the GeneMANIA software to identify the top-5 candidate genes whose mRNA levels harbored the more closely inverse correlation with NDRG1 mRNA levels in our whole set of data.This top-5 genes was established as follows: IL17D, IL33, IL12A, PDGF-C and TGFB1 (Figure 7).

Identification of a molecular pathway linking MYCN to PDGFC and TGFB1
NDRG1 was originally identified as a major target gene repressed by the transcription factor MYCN (also named N-Myc) in neuroblastoma cells [47,48].Since then, in multiple cell types, MYCN was shown to silence NDRG1 gene expression via a process involving a hypermethylation of the NDRG1 promoter [47][48][49][50][51][52] .On this basis, we performed a meta-analysis of the human interactome to determine whether signaling pathways could link MYCN to the identified candidate molecules (i.e IL17D, IL33, IL12A, PDGF-C and TGFB1).We found that PDFGRA (a PDGFC receptor) and TGFBR1 (a TGFB1 receptor) are second shell interactants of MAX (Myc associated factor X) and SP1 (Sp1 transcription factor)(Figure 8), two TFs that are recruited by MYCN during the transcriptional repression of specific target genes [49].

Discussion
In this paper, gene co-expression analyses were performed in order to infer astrocytes vs oligodendrocytes molecular signatures in spinal cord periplaques of MS patients.Regarding astrocytes, genes that were found to co-upregulate with GJA1/Cx43 in periplaques formed a pro-gliotic signature which composition is compatible with a combined TGFB/SAMD1/SMAD2-and SOX2-driven program.While TGFB and SOX2 were both previously identified as astrocytosis-promoting factors [22][23][24][25], TGFB was further shown to induce an astrocyte-specific developmental program that, in turn, prevents effective remyelination in MS plaques and periplaques [53].Superimposing to the TGFB/SMAD1/SMAD2 and SOX2 progliotic signatures, we also found that the GJA1/Cx43 module was characterized by a highly significant enrichment in genes that are regulated by the AR.Interestingly, even when excluding males from our study, the AR was still co-expressed with GJA1/Cx43.As androgens are known to prevent astrocytosis [29][30][31][32][33][34], our results raise the possibility that in males and females suffering from progressive forms of MS, activation of the AR is somehow defective and fails to downregulate a large set of progliotic genes induced by the TGFB/SMAD1/SMAD2 pathway.This could be due to an age-related decrease of circulating testosterone or of the circulating androgen precursors (Dehydroepiandrosterone and Androstendiol) that can be intracellularly metabolized into testosterone [54][55][56][57].Another explanation, not exclusive from the former one, could be that the AR signaling pathway might be hampered or disrupted in periplaque astrocytes despite a normal level of circulating or intracrine testosterone.Supporting this view, a meta-analysis of the human AR interactome shows that AR interactants are significantly enriched in SMAD-binding proteins (data not shown).Moreover, mechanisms of mutual exclusion between AR and TGFB signaling pathways were previously demonstrated in different cell types [58][59][60][61][62][63].
With regard to inflammation, we found that in spinal cord periplaques, the astrocytois signature formed by GJA1/Cx43 gene module comprised only two cytokines, namely IL17D and IL33.IL17D is a pro-inflammatory cytokine that binds to a yet unknown receptor and directs in vivo the intra-tissular recruitment of NK T-cells [64,65].Of note, IL17D, as all the members of the IL17 family, is also a potent inducer of TGFB secretion in many cell types [66,67].Interestingly also, the astrocytic expression of IL-33 was recently demonstrated in MS plaques and IL33 was shown to inhibit the de novo myelination of rat axons in vitro [68].However, other works reported that IL-33 promotes remyelination [69] and exerts potent anti-inflammatory and neuroprotective functions [70].Finally, besides TFs and immune genes, we observed that GJA1/Cx43 module included several key extra-cellular matrix (ECM)-related genes involved in tissue remodeling.Among these, Procollagen Lysyl Hydroxylase 2 (PLOD2) is a TGFB-induced enzyme [71][72][73] that promotes fibrosis via the crosslinking of collagen molecules [74][75][76].
The second aim of our work was to identify a molecular signature of oligodendrocytes in the spinal cord periplaques of MS patients.Among the network of genes that were constantly down-regulated in periplaques, NDRG1 was identified as the only oligodendrocyte-specific hub gene.Such a finding has to be interpreted in light of a recent work demonstrating that NDRG1 is also the only oligodendrocyte-specific gene whose silencing was demonstrated in the NAWM of MS patients.These data, while further supporting the role of NDRG1 as a master regulator of oligodendrocyte differentiation and myelin maintenance [19][20][21], suggest also that NDRG1 may be a major target of a yet uncharacterized process leading to diffuse myelin alterations in periplaques.When using NDRG1 as a bait gene to identify a large co-expression module in oligodendrocytes, we found that only 3 genes coding for myelin proteins co-downregulated with NDRG1 in periplaques: MBP, MOBP and PLP1.Interestingly, as opposed to the GJA1/Cx43 module, no specific enrichment in known targets of co-expressed TF could be evidenced in the NDRG1 module.Moreover, we were neither able to identify enrichment in apoptosis-related pathways nor in miRNA targets, which would have provided molecular schemes explaining oligodendrocyte cell loss or altered myelin gene expression.In contrast, the NDRG1 co-expression module was highly significantly enriched in genes involved in translation/elongation of mRNAs.Although, at this stage, one may only extrapolate on the significance of such an enrichment, it is important to underscore that the translation of both MBP and MOBP mRNAs in oligodendrocytes rely on a unique mechanism during which various components of the protein translation machinery (including ribosomal proteins) are transported along the cell processes of oligodendrocytes [40][41][42][43][44][45] .Strikingly, when performing a meta-analysis of human NDRG1 protein interactants, we found a highly significant enrichment in proteins involved in the "Translation" pathway (Reactome 2016, adjusted p-value = 1.80E-11)(data supplement 6).
On this basis, one may thus propose that NDRG1 silencing in oligodendrocytes may hamper myelination via the coordinated downregulation of a specific subset of ribosomal genes supporting the translation/elongation of MBP and MOBP mRNAs.Finally, we attempted to identify candidate soluble molecules that could be responsible for NDRG1 down-regulation in spinal cord periplaques.Our analysis pointed to a network of 5 genes whose expression levels inversely correlated with those of NDRG1 with the highest levels of confidence: i) IL-17D and Il-33, the two cytokines found to co-upregulate with GJA1/Cx43 in periplaques, ii) IL12-A, a potent inducer of TH1 (T-Helper 1) polarization [77], iii) PDGFC (Platelet-derived growth factor C), a profibrotic gene [78,79] and a regulator of the neurovascular unit [80] and iv) TGFB1 (Transforming growth factor-1).As a matter of fact, these correlations do not allow establishing a functional causative link between an up-regulation of these cytokines and the down-regulation of NDRG1 expression.Again, we used bioinformatics tools to explore such putative links and performed a meta-analysis of the protein interactants potentially linking the identified candidate cytokines with N-Myc.Indeed, a well identified mechanism of NDRG1 repression relies on the binding of N-Myc to the promoter region of NDRG1 and the subsequent recruitment of histone deacetylases leading to promoter hypermethylation and an epigenetic silencing of NDRG1 [47][48][49][50][51][52].Our meta-analysis showed that only TGFBR1 (an oligodendrocyte-expressed receptor to TGFB1) and PDGRA (an oligodendrocyte-expressed receptor to PDGFC) formed a close (second shell) protein network with N-MyC.
Overall, since the NDRG1 promoter was shown to be hypermethylated in the NAWM of MS patients [20], our results indicate that TGFB1 and/or PDGFC released in MS periplaques could be responsible for a silencing of NDRG1 in periplaques and, to a lesser extent, in the NAWM.
Overall, our work points to a major role of TGF-beta1 in the development of astrocytosis and diffuse myelin alterations in the spinal cord periplaques of MS patients.In particular, we propose a new pathophysiological scheme where TGF-beta1 would be central to a self-perpetuated process during which: i) IL17-D, synthesized by reactive astrocytes, triggers the production of TGFB1 and ii) astrocyte-derived TGFB1 fuels chronic gliosis and hamper myelination via a transcriptional repression of NDRG1 .While further studies are needed to test this hypothesis, it is important to notice that the monoclonal antibody Fresolimumab directed against all TGFB isoforms is available for clinical trials [81], cross the blood brain barrier [82] and is currently under evaluation in patients suffering from systemic sclerosis [83].

Transcriptomic data
Co-expression networks were generated from a previously published set of transcriptomic data that our group obtained by whole genome analyses of paired NAWM and periplaque samples derived from 8 SPMS patients [8].Data are available at the public gene array repository bank GEO DataSets (Ref: GSE52139).To ensure the highest level of confidence to co-expression analyses, we first submitted transcriptomic data to a stringent quality control using the arrayQualityMetrics Bioconductor package (v3.28.2) [84] in R (v3.3.0).We removed from further analyses the transcriptomic profile obtained from 1 out of 16 samples as it was detected as an outlier by both boxplots and MA plots criteria.

Generation of a global co-expression network.
An expression file was generated that comprised an "OBJECT" column in which gene symbols were provided for each row and 15 "SAMPLE" columns compiling normalized data values for each sample and each transcript.This expression file was then imported in the GeneMANIA Cytoscape plugin (version 3.3.4data version 2014-08-02-core) [85,86] allowing the generation of a co-expression network that, based on the Pearson correlation test, gathered 735 240 interactions.

Connectivity analysis of gene co-expression networks
Starting from the global co-expression network generated from periplaque and NAWM samples, we investigated the correlation links that could be established between genes that were previously found to be differentially regulated in paired comparisons between periplaques and NAWM samples.Using the GeneMANIA plugin in Cytoscape software, 2 sets of genes were entered separately as independent queries: i) the 34 genes found to be constantly upregulated in periplaques [8] and ii) the 57 genes found to be constantly downregulated in periplaques [8].For each analysis, the number of additional connected genes to be identified was set to zero in order to limit the search for co-expression links to query genes only.Then, to identify hub genes among these modules, we used the "Network analysis" tool in Cytoscape software.A hub gene was defined as a gene (node) being connected to at least half of the total number of genes (nodes) forming the network.

Identification and analysis of the Cx43/GJA1 co-upregulated gene network
Starting from the global expression network loaded on Cytoscape/GeneMANIA, CX43/GJA1 (GJA1) was entered as a query gene, and the number of connected genes to be identified was set to 200.The generated list was then assessed with regard to the presence of genes considered as functionally relevant in the context of astrocytosis, inflammation and tissue remodeling.These notably included genes coding for: i) cytokines, ii) extracellular matrix (ECM) or iii) transcription factors (TF).Identification of functionally-relevant genes was performed by a survey of the literature, on the basis of GO terms annotation, and using the TF database "AnimalTFDB" [87].Finally, the 200 genes forming the Cx43/GJA1 co-expression network were also analyzed with regard to their enrichment in: i) genes involved in specific pathways, ii) genes identified as known targets of human transcription factors (TF), iii) genes identified as known targets of miRNA.These enrichment analyses were performed using the EnrichR webtools [88]: "ChEA2016" [35] (allowing to survey a total of 622 ChIP-ChIP or ChIP-seq sets of data that were manually curated), "TargetScan microRNA" [39] and "Reactome 2016" [36] .

Identification and analysis of the NDRG1 co-downregulated gene network
Starting from the global expression network loaded on Cytoscape/GeneMANIA, NDRG1 was entered as a query gene, and the number of connected genes to be identified was set to 200.The generated list was then assessed with regard to the presence of genes coding for myelin or oligodendrocyte-related molecules as determined by GO terms annotation and a survey of the literature.The complete list of genes forming the NDRG1 co-expression network was also analyzed with regard to its enrichment in: i) genes involved in specific pathways, ii) genes identified as known targets of human transcription factors (TF), iii) genes identified as known targets of miRNA.These enrichments analysis were performed using the open source website EnrichR [88] and the same webtools as described above.

Identification of a network of NDRG1 inversely correlated genes coding for cytokines/chemokines or growth factors
As GeneMANIA recognizes only positive correlation links, its use for the identification of mRNA species that inversely correlate with a given query gene requires the prior replacement of the corresponding mRNA values by opposite values.To identify a network of cytokines, chemokines or growth factors that inversely correlated with NDRG1, we first established a list of 193 candidate genes (data supplement 1) that included notably 38 interleukins, 17 interferons, 36 chemokines, 35 growth factors, 22 metalloproteases, 6 angiogenic factors and 5 granzymes.GeneMANIA was then loaded with an expression file that comprised: i) the original values obtained for each of the probes corresponding to the above-described list of cytokines/chemokines or growth factors, ii) the values opposite to the original ones obtained with the NDRG1 probe.Again, the number of additional connected genes to be identified was set to zero.NDRG1 was entered as a query gene, and the number of connected genes to be identified was set to 10.As GeneMANIA was conceived to process large sets of data, its algorithms include a sparsification process during which only the 50 strongest correlation links for each object are retained [85].Therefore, for expression files comprising a low number of objects, correlation links being ranked in the top 50 might still not reach significance.To circumvent this possible drawback, for each of the 10 genes that were initially identified as inversely correlated with NDRG1 with the GeneMANIA software, the statistical significance of the inverse correlation was checked using the Pearson correlation test.Only genes for which p values were ≤0.01 were taken into account.

Meta-analysis of MYCN protein interactants
When needed, the open source websites "STRING" [89] and Wiki-Pi [90] were used to perform meta-analyses of protein-protein interactions.STRING allows protein interaction networks to be generated via a survey of distinct databases that list known protein interactions from published and manually-curated biochemical studies [89].Here, we used STRING to identify interactants that would link MYCN (also named N-Myc) to the receptors IL12RB1, IL12RB2, IL1RL1 (IL33 receptor), PDGFRA (PDGFC receptor), PDGFRB (PDGFC receptor), TGFBR1, TGFBR2 and TGFBR3.Several round of analyses were performed in which two query proteins were entered for each analysis: MYCN and one of the above-mentioned receptors.The number of first-shell protein interactants to be identified was set to 500 and only high confidence interactions (score ≥ 0.7 according to the classification provided by the STRING website) retrieved from published and manually-curated biochemical studies were taken into account.Selected interaction networks were then loaded in Cytoscape in order to facilitate the visualization of identified pathways.In another set of analyses, we used the Wiki-Pi database to retrieve NDRG1 protein interactants.Wiki-Pi is a resource that is not manually-curated but allows a large survey of both Biogrid [91] and HPRD [92] databases obtained by high throughput technologies.Wiki-Pi compiles 48419 unique binary biophysical interactions among 10492 human proteins.

Figure 4 :
Figure 4: Network of co-downregulated genes in spinal cord periplaques.A co-expression network was generated with the set of genes that are constantly downregulated in periplaques as compared to adjacent NAWM.In this co-expression module, hubs were defined as genes (nodes) being connected to at least half of the total number of genes (nodes) forming the network.Five hub genes (black-filled nodes) were identified, namely NDRG1 (N-myc downstream regulated 1), PEA15 (phosphoprotein enriched in astrocytes 15), RASSF2 (Ras association domain family member 2), PAQR6 (progestin and adipoQ receptor family member 6) and LHPP (phospholysine phosphohistidine inorganic

Figure 5 :
Figure 5: Astrocyte signature inferred from the GJA1/Cx43 co-expression module in spinal cord periplaques.To identify an astrocyte molecular signature in spinal cord periplaques, we used the astrocyte-specific hub gene Cx43/GJA1 as a "bait" and retrieved the top-200 mRNA species that most closely co-upregulated with Cx43/GJA1.The co-expression network formed by functionally-relevant genes is shown.Genes are designated by their gene symbols.Gene full names are provided in Table2.

Figure 6 :
Figure 6: Oligodendrocyte signature inferred from the NDRG1 co-expression module in spinal cord periplaques.To identify an oligodendrocyte molecular signature in spinal cord periplaques, we used the oligodendrocyte-specific hub gene NDRG1 as a "bait" and retrieved the top-200 mRNA species that most closely co-downregulated with NDRG1.The co-expression network formed by functionally-relevant genes is shown.Genes are designated by their gene symbols.Gene full names are provided in Table3.

Figure 7 :
Figure 7: Identification of candidate soluble factors that putatively repress NDRG1 expression.Starting from a list of 193 candidate genes coding for cytokines, chemokines or growth factors, we used the GeneMANIA software to identify mRNA species whose levels inversely correlated with those of NDRG1 in spinal cord periplaques.The top 5 genes whose expression levels inversely correlated with NDRG1 were: IL17D, IL33, IL12A, PDGF-C and TGFB1.

Figure 8 :
Figure 8: Identification of molecular pathways linking PDGFC and TGFB1 to MYCN.A meta-analysis of the human interactome was performed to determine whether signaling pathways could link MYCN (a known repressor of NDRG1 expression) to the identified candidate IL17D, IL33, IL12A, PDGF-C and TGFB1.Data show that PDFGRA (a PDGFC receptor) and TGFBR1 (a TGFB1 receptor) are second shell interactants of MAX (Myc associated factor X) and SP1 (Sp1 transcription factor), two transcription factors that are recruited by MYCN during the transcriptional repression of specific target genes.

Table 3 :
List of functionally-relevant genes that co-express with NDRG1 in spinal cord periplaque areas