Neuronal microRNA regulation in Experimental Autoimmune Encephalomyelitis

Multiple sclerosis (MS) is an autoimmune, neurodegenerative disease but the molecular mechanisms underlying neurodegenerative aspects of the disease are poorly understood. microRNAs (miRNAs) are powerful regulators of gene expression that regulate numerous mRNAs simultaneously and can thus regulate programs of gene expression. Here, we describe miRNA expression in neurons captured from mice subjected to experimental autoimmune encephalomyelitis (EAE), a model of central nervous system (CNS) inflammation. Lumbar motor neurons and retinal neurons were laser captured from EAE mice and miRNA expression was assessed by next-generation sequencing and validated by qPCR. We describe 14 miRNAs that are differentially regulated in both neuronal subtypes and determine putative mRNA targets though in silico analysis. Several upregulated neuronal miRNAs are predicted to target pathways that could mediate repair and regeneration during EAE. This work identifies miRNAs that are affected by inflammation and suggests novel candidates that may be targeted to improve neuroprotection in the context of pathological inflammation.

Multiple sclerosis (MS) is an autoimmune neurodegenerative disease characterized by the infiltration of peripherally activated immune cells into the central nervous system (CNS) 1 . Inflammation results in multiple foci of oligodendrocyte damage and demyelination throughout the CNS including the cerebral cortex, spinal cord, and optic nerve 2,3 . Active lesions also contain axonal ovoids, indicative of newly damaged axons, and extensive accumulation of amyloid precursor protein (APP) resulting from impaired axonal transport 4 . Axonal transection is apparent early in the disease and is thought to be a major cause of persistent neurological disability 2,4 . Gray matter atrophy and cortical thinning, as a result of neuronal death, also increase with MS-related disability, and are characteristic of the progressive forms of MS [5][6][7] .
In experimental autoimmune encephalomyelitis (EAE), an animal model of CNS inflammation induced by active sensitization with myelin antigens, foci of demyelination and axon damage are also apparent 8 . Infiltration begins at the lumbar region of the spinal cord and progresses anteriorly resulting in ascending paralysis. Evidence of sub-lethal neuronal damage includes dendrite shortening, thinning and fragmentation while axons display accumulations of APP and transections [9][10][11] . Significant loss of αand γ-motor neurons in lumbar spinal cord has also been reported over the course of the disease, beginning as early as 14 days post immunization (dpi) 11 . Similarly, the visual system is affected in EAE with loss of retinal ganglion cells (RGCs) and optic nerve pathology including immune cell infiltration, demyelination and glial activation [12][13][14] . The pathology is delayed in the optic nerve relative to the spinal cord, with loss of RGCs occurring after the peak stage, around 35 dpi 13 .
The molecular changes occurring within neurons in EAE and MS that mediate pathological responses to inflammation have not been fully characterized. Understanding the molecular response to damage will be critical to devising neuroprotective strategies to mitigate sub-lethal axonal damage and neuronal cell death. We sought to identify conserved molecular networks that regulate the response to injury in all neuronal subtypes by profiling neuronal microRNAs (miRNAs) in affected neuronal subtypes. miRNAs are short RNA molecules approximately To validate the miR-Seq data, miRNA expression was analyzed using Taqman MicroRNA Assays. For the novel miRNAs and mmu-miR-92b-3p, custom Taqman MicroRNA Assays were designed (Supplementary  Table S2). Probes for miR-92b-3p and miR-novel-chr2_10423 failed quality control testing thus qPCR validation was performed for 41 of the 43 regulated miRNAs. The fold change miRNA expression was relative to naïve levels, after normalization to endogenous control snoRNA202, using the 2 −ΔΔCt method 25 . Of the 41 miRNAs significantly regulated by miR-Seq, 24 were significantly regulated similarly between the miR-Seq and qPCR analyses ( Fig. 3 and Table 1). Three miRNAs, miR-1969, miR-7056-5p, and miR-novel-chr7_31864, were regulated in the opposite direction compared to the miR-Seq analysis. These miRNAs carry GC rich areas which create an alignment issue during NGS, making quality reads in those regions more difficult to produce 26 . Low quality reads for a miRNA are discarded during the NGS process; this alignment issue is reduced with increased amounts of total RNA. As our total RNA was collected from LCM material, we were able to submit approximately 300 ng per Heat Map summary of significantly regulated miRNA as identified by miR-Seq in EAE. miRNA and animals (N = naive, O = onset, P = peak) are hierarchically clustered by Euclidean distance using rlog transformed counts. Blue data represents low expression of that microRNA within its own row, and red indicates high expression. 997 miRNAs were identified by miR-Seq, 43 of these were identified as significantly regulated (p < 0.05); 6 of which are novel miRNAs.

miR-Seq identified and validated miRNAs in EAE neurons are found regulated in other EAE
In silico assessment of putative targets identifies novel pathways predicted to be affected in the neurons of EAE mice. To predict programs of gene expression that may be regulated by miRNAs that were commonly regulated in motor neurons and RGCs, we performed a non-biased in silico analysis to identify putative mRNA targets of regulated miRNAs (Fig. 5).
We mined predicted targets of miRNAs using Diana-microT 51,52 , microRNA.org 53 , miRDP 54 , miRTarBase 55 , RNA22 56 , TargetScan 57 , and TarBase 58 . mRNAs that were predicted targets in at least 3 of 7 databases were coined 'filtered targets' . To determine conserved pathways between the neuronal subtypes in EAE, we overlapped the filtered targets of regulated miRNAs in lumbar motor neurons and the RGC layer and performed an overrepresentation test using Protein ANalysis THrough Evolutionary Relationships (PANTHER) tools 59,60 . We identified several enriched PANTHER Pathways, Reactome Pathways and gene ontology (GO) cellular components that are targeted by upregulated miRNAs (Table 2). Specific gene targets identified in the PANTHER, Reactome and GO cellular component pathways are recorded in Supplementary Tables S3-5, respectively. The upregulated miRNAs in neurons harvested from EAE mice are predicted to target and downregulate genes in PANTHER Pathways including hypoxia response via hypoxia-inducible factors (HIF) activation, and axon guidance mediated by Slit/ Robo. Predicted Reactome Pathways include: KIT signaling; signaling by bone morphogenic protein (BMP); netrin-1 signaling; CD28 co-stimulation; protein lysine methyltransferases (PKMTs) methylate histone lysines; and synthesis of phosphatidylinositol phosphates (PIPs) at the plasma membrane. PANTHER and Reactome analyses differ in how they classify terms 59 . The GO cellular components targeted by the upregulated miRNAs are the CCR4-NOT complex, and cytoplasmic stress granules (SG). The predicted targets suggest that upregulated miRNAs target pathways converging on cell survival and growth, cytoskeleton rearrangement, and the stress response. These pathways and how they may act in concert are summarized in Fig. 6.
To validate if these in silico identified pathways are regulated, we evaluated the expression of key genes described in Supplementary Tables 3-5, focusing on representative genes regulating cell survival and growth, cytoskeleton rearrangement, and the stress response; as well as genes targeted by multiple miRNAs. Gene expression was assessed in the RGC layer over the EAE disease course, and compared to the miRNA expression of the respective targeting miRNAs (Fig. 7). Genes Bmpr1a, Cdc42, Hif1a, Pik3r1, Eif4e, and Mbnl1 showed a significant downregulation at peak disease, corresponding with increased expression of their targeting miRNAs at peak disease. Genes Rictor and Smad4 showed the same trend, but were not significant. Larp1 was uniquely downregulated at chronic disease stage, and accordingly its targeting miRNAs were upregulated at the chronic stage. Genes Pik3c2a and Pum1 were downregulated throughout EAE relative to presymptomatic levels. Pum1 is targeted by the most miRNAs, and showed the strongest downregulation throughout EAE, emphasizing the validity of our in silico approach for determining the (1) filtered targets of our differentially regulated miRNAs; and (2) PANTHER pathways and cellular components assessment of the filtered targets. The overrepresentation test for filtered targets of miRNAs that were downregulated in both neuronal subtypes generated a small list of 143 genes, and this was insufficient to identify overrepresented pathways or cellular components. Therefore, we performed an overrepresentation analysis for pathways that were predicted to be downregulated in lumbar motor neurons Scientific REPORtS | (2018) 8:13437 | DOI:10.1038/s41598-018-31542-y alone. We identified several enriched PANTHER Pathways, Reactome Pathways, and GO cellular components that were targeted by downregulated miRNAs ( Table 3). The PANTHER Pathways included signaling mediated by the following: histamine H 1 receptor; oxytocin receptor; phosphoinositide 3-kinase (PI3K); thyrotropin-releasing hormone receptor; Alzheimer's disease (AD)-amyloid secretase; epidermal growth factor receptor (EGFR); 5-HT 2 type receptor; and fibroblast growth factor (FGF). The Reactome Pathways included: post-transcriptional silencing by small RNAs; signaling by TGF-β Receptor Complex; and several events activating PI3K/Akt via GAB1 signalsome, EGFR, and FGF receptor (FGFR). Finally, the cellular components enriched in targeted genes of downregulated motor neuron miRNAs included protein phosphatase type 2A (PP2A) complex, and endocytic vesicles. Specific gene targets identified in the PANTHER Pathways, Reactome Pathways, and GO cellular components are recorded in Supplementary Tables S6-8, respectively. As these pathways and cellular components were targeted by the downregulated miRNAs, it is suggested that they are now activated or relieved of inhibition.

Discussion
In this study, we took an unbiased approach to determine the changes in neuronal miRNA expression in a model of CNS inflammation. We compared the miR-Seq validated profile of lumbar motor neurons to RGCs in EAE, and identified potentially regulated pathways and cellular components during neuroinflammation, using an in silico bioinformatics approach. This provides the first analysis of miRNA expression in neurons over the course  17 or NAWM 21,46 have been isolated to profile differential miRNA expression in MS.
The time course assessment of miRNA expression revealed massive upregulation at peak disease. The loss of motor neurons and axons in the ventrolateral and dorsal columns begins early at onset of EAE disease; thus, we hypothesized that lumbar motor neurons would exhibit miRNA regulation at disease onset 11,62 . Rather, almost all miRNAs were uniquely upregulated at peak disease alone, in some cases trending towards upregulation at onset (Fig. 3). Animals may have slightly different kinetics of disease and have more variability at onset.
There are several potential explanations for exaggerated miRNA upregulation at peak disease. Specifically, miRNAs may be trapped in the cell body compartment as result of axonal transection and compromised axonal transport early in EAE, leading to their accumulation over the course of disease 9,62 . Alternatively, it is predicted that genes involved in post-transcriptional silencing by small RNAs would be dramatically enriched 13.5 fold as a consequence of downregulation of select miRNAs in motor neurons ( Table 3). The predicted increase in neuronal RISC pathway components contrasts with a report of decreased miR-mediated transcriptional silencing in the T cells and oligodendrocytes of EAE mice 23 . Lastly, it is also possible that negative feedback loops required to limit miRNA expression may be compromised in inflamed neurons. For example, miR-223-3p is part of an autoregulatory negative feedback loop with transcription factor E2F1, and disruption of this loop may contribute to the progression of acute myeloid leukemia 63 . Dysregulation of such feedback loops may thus be affected in the neurons of EAE mice.
Next, our comparison of miRNA expression across cell types in EAE and/or MS revealed that miRNA expression is dependent on the cell type even when compared within the same disease model or group of patients (Table 1). Of the differentially expressed miRNAs in motor neurons, half were regulated in the same direction in the RGC layer. This identifies some similarities in neuronal responses to inflammation irrespective of neuronal function. Differences in miRNA regulation in the two neuronal populations could represent physiologic diversity in responses. Alternatively, these divergences could reflect differences in the time course of inflammation in the motor and visual system. Motor neuron loss begins as soon as 14 dpi in the EAE model 11 , whereas RGC loss becomes significant between 25 and 40 dpi 11,13 . Another consideration is that motor neurons and their axons can be exposed to a common inflammatory milieu whereas RGCs are compartmentalized from their axons within the optic nerve. Similar proportions of T cell, B cell, macrophage and microglia populations have been described within the spinal cord and optic nerve 64 . However, there is a lack of immune infiltrates within the choroid or retina surrounding the RGCs, with CD3+ infiltrates exclusively occurring in the optic nerve whereas activated  12,65 . Alternatively, in the spinal cord, CD3+ infiltrates are described in close proximity to motor neurons 11 . We cannot rule out the possibility that some differences result from comparing pure motor neurons to the RGC layer, which contains amacrine cells and retinal astrocytes in addition to approximately 50% RGCs 28 .
When comparing our analysis of neurons to prior studies examining EAE or MS tissue, we find that most of the neuronally-regulated miRNAs were not similarly regulated in other EAE/MS tissue. Several miRNAs including miR-127-3p, miR-223-3p, miR-7a-5p, miR-203-3p and miR-340-5p are differentially regulated depending on the tissue source (Table 1). For example, miR-340-5p was reportedly upregulated in CD4+ T cells in MS but downregulated in MS lesions 17,41 . The only prior study to investigate neuronal miRNAs compared miRNA expression by fluorescence intensity between myelinated and demyelinated hippocampal human post-mortem MS tissue 44 . They investigated a list of candidate miRNAs based on their specific associations with mRNAs that are changed in demyelinated MS hippocampi. Regulated miRNAs in that study were not differentially regulated in our screen.
Our results highlight the importance of performing cell type specific analyses of miRNAs because target genes can have distinct effects in individual cell types. For example, miR-183-5p is elevated in the neurons of a model of spinal muscular atrophy (SMA) and promotes neurodegeneration 66 . However, miR-183 elevation is also important for blocking the cytotoxic effects of natural killer (NK) cells by targeting required receptors for NK cell  Table 2. Overrepresented pathways and cellular components by the upregulated miRNAs in EAE neurons.  67 . This could have potential protective roles in the inflammatory context of MS 68 . In both models miR-183-5p is upregulated, yet its elevation produces drastically different biologies in the two cell types. Finally, the putative pathways targeted by the differentially regulated neuronal miRNAs suggest that these miRNAs contribute to axonal pathology and cell death. Using a bioinformatics approach, we determined putative pathways targeted by upregulated miRNAs during neuroinflammation. Many target pathways such as CD28 co-stimulation are known to affect the biology of non-neuronal cells (Supplementary Tables S3 and 4). While it is possible that neuronal miRNAs may be released in exosomes to affect the biology of other cells 33 , we will limit our discussion to potential roles in the neuronal response.
Many of the pathways and cellular components described in Table 2 converge on cell survival and growth, cytoskeleton dynamics, and stress response. The predicted direction of regulation suggest that regulated miRNAs could contribute to neuronal degeneration, cell death and an aberrant stress response.
Many genes of the pathways described in Table 2 converge on the promotion of cell survival and growth via the PI3K/Akt/mTOR cascade 69,70 . Genes Src, Mtor, Rictor, Rptor, Akt1, Akt2, Akt3, Pik3r1, and Pik3c2a are directly involved in positive mTOR signaling 70,71 , and are predicted to be targeted by several of our upregulated miRNAs (Supplementary Table S4). Such a regulation would result in stunted growth and cell death. HIF signaling, downstream of mTOR, promotes survival during hypoxia 70,72 . HIF signaling genes HIF1a and Arnt are also targeted by our upregulated miRNAs 72 . In vivo knockdown or inhibition of HIF activity limits axon regeneration in axotomized sensory neurons 73,74 , and exacerbates cerebral ischemia-induced tissue damage 75 ; whilst the pharmacological upregulation of HIF in animal models of cerebral ischemia 76 , Parkinson's disease 77 , and AD 78 alleviates disease. Targeting of the PI3K/Akt/mTOR cascade and its downstream pathways suggests our upregulated miRNAs promote cell death of EAE neurons. This is emphasized by the significant downregulation of genes HIF1a, Pik3c2a and Pik3r1 at peak disease in the RGC layer of EAE mice (Fig. 7a). Another pattern that emerged was that stress granule (SG) formation may be disrupted in response to miRNA upregulation (Tables 2 and S5). Many SG initiating genes, including Tia1, Tial1, G3bp1, Zfp36, Fmr1, Pum1, and Pum2, are targeted by the upregulated miRNAs [79][80][81][82][83] , along with other essential components of SGs listed in Supplementary Table S5. Some of these genes were downregulated throughout EAE, with Pum1 significantly downregulated at both peak and chronic EAE in the RGC layer; where Pum1 was one of the most targeted genes by the upregulated miRNAs (Fig. 7b). SGs are non-membrane bound cytoplasmic aggregates of RNA and protein that form in response to environmental stress such as hypoxia, endoplasmic reticulum (ER) stress, reactive oxygen species, and nutrient deprivation 84 . ER stress is a hallmark of MS and chronic stress in neurons that leads to failed SG assembly [85][86][87] . SGs are transient and can act to protect untranslated mRNA from stress-dependent damage 88 . Use of the SG-stabilizing drug guanabenz in EAE alleviates clinical symptoms 89,90 . The implication from our study is that upregulated miRNAs may block adaptive SG formation, limiting the ability of neurons to recover from chronic inflammatory stress associated with EAE.
Interestingly, we also identified genes ascribed to axon guidance pathways as targets of upregulated miR-NAs. Mena, Rac1, Cdc42, DCC, Smad4, and Bmpr1a are some predicted gene targets of the upregulated miRNAs (Supplementary Tables S3 and 4). Several of these targets can positively influence cytoskeleton rearrangement, such as Bmpr1a and Cdc42, which we identified as significantly downregulated at peak disease in the RGC layer (Fig. 7c). This raises the possibility that miRNA-dependent downregulation contributes to the formation of retraction bulbs in MS lesions and failure to reform a new growth cone and mount a regenerative response [91][92][93][94][95] .
The current understanding of how neurons are affected during EAE and MS has been unclear. Our assessment of the affected miRNAs and their targeted pathways over the course of EAE suggests that the upregulated miRNAs themselves target pathways related to cell survival and growth, cytoskeletal rearrangement, and stress response. We also identified the downregulation of representative targets in the same tissue, validating our in silico approach and emphasizing the contribution of these pathways to EAE. This novel information concerning the neuronal response lends information on what we may be able to target therapeutically to promote neuroprotection or repair for a disease largely discussed in an inflammatory context alone.

Methods
Active EAE Disease Induction and Scoring. EAE was induced in 6-9 week old female C57BL/6 mice as previously published 96 . Animals were immunized subcutaneously with 200 μg of myelin oligodendrocyte glycoprotein 35-55 (MOG 35-55 ) (MEVGWYRSPFSRVVHLYRNGK; Alpha Diagnostic International) in a 100 μl emulsion of Complete Freund's Adjuvant (4 mg/ml Mycobacterium tuberculosis; Fisher Scientific). On day 0 and day 2, Pertussis toxin (500 ng PTX, Sigma-Aldrich) was injected intra-peritoneally (i.p.). Animals were scored as follows: 0 = normal; 1 = limp tail; 2 = slow righting-reflex; 2.5 = difficulty walking/ataxia; 3 = paralysis of one hindlimb (monoparalysis); 3.5 = hindlimb monoparalysis and severe weakness in the other hindlimb; 4 = paralysis of both hindlimbs (paraparalysis); 4.5 = hindlimbs paraparalysis and forelimbs weakness; 5 = moribund (requires sacrifice). All animal procedures were approved by the Centre de Recherche du Centre Hospitalier de l'Université de Montréal Animal Care Committee (N11023APs) and followed guidelines of the Canadian Council on Animal Care.  Animals showing no sign of pedal and palpebral reflex were perfused intracardially using cold saline. The spinal cord and retina were isolated and frozen at −80 °C in Tissue-Tek ® O.C.T. compound. Spinal cords were collected at naïve (score 0), onset (score 0.5-1), and peak (score 3-3.5) clinical points; and retinae were collected at presymptomatic (score 0, 8 dpi), peak (score 3.5-4), and chronic endpoints (score 2, endpoint at 35 dpi). Frozen sections were obtained using a Leica ® cryostat CM3050S. Sections were mounted onto RNase-free MembraneSlide , and Perl 10 script to remove the reads that contained ambiguous bases "N", poly(A/T); and the read lengths <14. Q-score fols were 25 for the 3′-end, 10 for <4 bases or 13 for <6 bases in the first 30 bases, and 20 for >75% of all bases. The adaptor sequences were trimmed with cutadapt called from Python environment with-overlap set to 2. Trimmed reads were aligned to a combined mouse pre-miRNA in miRBase 21 and the predicted novel pre-miRNAs with Novoalign v2.07.11, allowing no more than 1 mismatch. miRDeep2 was used to predict novel miRNAs with the trimmed reads. miRNA read counts were processed and normalized with the DeSeq2 (version 3.5) Bioconductor package and used to detect differentially expressed miRNAs in lumbar motor neurons 97 . Significantly differentially expressed miRNAs (adjusted p < 0.05) were visualized as heatmaps created with the Morpheus tool (http://software.broadinstitute.org/morpheus/). Quantitative RT-PCR (qRT-PCR) -For qRT-PCR validation of miRNA expression in lumbar motor neurons, samples from deep sequencing and additional naïve and peak time points samples were analyzed (n = 3-6). The expression of 41 differentially regulated miRNAs was analyzed using multiplex qRT-PCR. Forty-one miRNAs were profiled using a Taqman MicroRNA Assay, of which 5 were novel miRNAs that required custom designs (Supplementary Table S2). Multiplexed RT reactions were performed using a mix of miRNA-specific RT primers and a TaqMan miRNA RT kit, with 10 specific primers per RT reaction and 10 ng of RNA per RT reaction. A pre-amplification reaction was run with 5 μl of the RT-generated cDNA using again a pooled set of TaqMan miRNA-specific probes to increase the number of target copies, as previously described 98 . Individual miRNA expression assays were performed using specific miRNA TaqMan probes on the pre-amplified cDNA material. Small nuclear RNA (snRNA) snoRNA202 was used as an endogenous control. Fold change calculations for miRNA expression were performed using the 2 −ΔΔCt method 25 , with normalized miRNA expression compared to the naïve mouse controls (score 0). miRNAs identified as significantly regulated in the lumbar motor neurons were assessed in the RGC layer of EAE mice using the multiplex qRT-PCR system with miRNA expression compared to presymptomatic mouse controls (score 0). For the RGC layer, 3-8 animals were assessed per condition. Statistical analysis was done using one-way ANOVA, and a Dunnett's multiple comparisons test (GraphPad Prism 6). For qRT-PCR of mRNA, total RNA from cell culture or LCM tissue was transcribed using Superscript Vilo or IV Vilo cDNA Synthesis Kit (ThermoFisher), respectively, using manufacturer's instructions. Individual gene expression was determined using FAM-labeled Taqman probes, and as endogenous controls FAM-labeled GapDH probe for cell culture tissue and VIC-labeled GapDH probe for LCM tissue. For determination of neuronal purity in LCM tissue, RNA markers for other cell types were relative to neuronal Tubb3 expression (n = 3) after GapDH normalization. For assessment of in silico identified genes in LCM RGC layer tissue, mRNA expression was compared to presymptomatic levels. Statistical analysis was done using one-way ANOVA, and a Tukey's multiple comparisons test with n = 3-6 per disease stage (GraphPad Prism 6).
In silico assessment of predicated targets. For significantly regulated miRNAs, a bioinformatics assessment of putative target genes was performed based on a comparative analysis by seven prediction programs. These include: Diana-microT 51,52 , microRNA.org 53 , miRDP 54 , miRTarBase 55 , RNA22 56 , TargetScan 57 , and TarBase 58 . Only mRNAs that were identified as putative targets across 3 of the 7 prediction programs were analyzed further. These were termed as "filtered targets". To determine conserved pathways between the neuronal subtypes in EAE, we overlapped the filtered targets of the upregulated miRNAs in lumbar motor neurons and the RGC layer (3969 genes). The functional classification of the overlapping filtered targets was performed using an overrepresentation test by PANTHER classification system (http://www.pantherdb.org/) 59,60 . The p-values were determined by PANTHER using the binomial statistic with a Bonferroni correction for multiple testing. From the overrepresentation assessment, we focused on PANTHER Pathways, Reactome Pathways and GO cellular components with a Fold Enrichment above 3. An overrepresentation assessment was repeated for the filtered targets of miRNAs downregulated in both neuronal subtypes (143 genes); however, this list was not comprehensive enough to determine any overrepresentation. An overrepresentation test of the filtered targets of the upregulated miRNAs in the motor neurons alone (4346 genes); upregulated miRNAs in the RGC layer alone (4226 genes); downregulated miRNAs in the motor neurons alone (1422 genes) were also performed. Again, the list of filtered targets of downregulated miRNAs in the RGC layer alone (143 genes) was not comprehensive enough to determine any overrepresentation.