Research Paper

High-resolution transcriptome analysis revealsneuropathic pain gene-expression signatures inspinal microglia after nerve injuryHeejin Jeonga,b, Young-Ji Nac, Kihwan Leea,b, Yong Ho Kima,b, Yunsin Leea,b, Minho Kangd, Bao-Chun Jiange,Young Il Yeomd, Long-Jun Wue, Yong-Jing Gaof, Junhyong Kimc,g, Seog Bae Oha,b,h,*

AbstractMicroglial cells, the resident immune cells of the spinal cord, become activated in response to peripheral nerve injury. Microgliaactivation contributes to the development of neuropathic pain. Here we employed microarray analysis of individually collected poolsof 10 spinal microglia cells to identify changes of levels and cell-to-cell expression variance of microglial genes during their activationafter peripheral nerve injury. The analysis of microglia on postoperative day 1 (POD1) identified miR-29c as a critical factor formicroglial activation and the development of neuropathic pain. Early POD1 microglia exhibited a very distinct expression profilecompared to late POD7 microglia, possibly leading to the transition from initiation to maintenance of neuropathic pain. We foundsample variance patterns that were consistent with the hypothesis that microglia were highly heterogeneous at the level of individualcells, and variation analysis identified 56 microglial genes potentially linked to the maintenance of neuropathic pain which includedGria1. This study provides insights into spinal microglial biology and reveals novel microglial targets for the treatment of neuropathicpain.

Keywords: Spinal microglia, High-resolution transcriptome, Neuropathic pain, miR-29c, Grai1

1. Introduction

Chronic pain is a main public medical issue that undermines thequality of life, especially because of the current lack of effectivetherapies controlling pain in patients.12,15 Neuropathic pain is themost difficult form of chronic pain and sciatic nerve injury inrodents leads to persistent tactile allodynia in which innocuousstimuli elicit abnormal pain hypersensitivity and spontaneouspain, 2 main symptoms of neuropathic pain.7 Interestingly,neuropathic pain is regulated by immune reactions in the spinal

cord, which are mediated by resident microglia, astrocytes, andinfiltrating immune cells.20,42,47

Microglia are resident immune cells of the central nervoussystem (CNS) which participate both in maintaining homeostasisand in sensing pathological changes.48 In response to peripheralnerve injury, spinal microglia become activated through a pro-gressive series of cellular and molecular changes.6,23 Activatedmicroglia in spinal dorsal horns contribute to central sensitization,the key mechanism for the development of neuropathicpain.18,24,57 Despite recent progress in our understanding ofmicroglial biology5,9,26 and identification of several microglialfactors which contribute to neuropathic pain,10,54,55 the molec-ular mechanisms by which resting microglia are switched toactivated forms after nerve injury have not yet fully understood.Given the poor efficacy of current drugs targeting neuronal activityin neuropathic pain,20 identifying microglial factors could beimportant for unraveling the molecular mechanisms and findingnovel targets.

Cellular phenotypes are determined by multiple processesinvolving gene and protein expression.58 Large-scale screeningfor pain-related genes on spinal cord or dorsal root ganglia (DRG)has revealed substantial expressional changes in hundreds ofgenes.11,36,51 However, most of these genes are neuronal notmicroglial, indicating that whole tissue analysis may havelimitations in finding targets from the spinal microglia whosetranscriptome is expected to be a minor fraction of whole tissuetranscriptome. Single-cell transcriptome analysis has revealedcell-to-cell variations in the same tissuewhichweremasked at thewhole tissue level.16,49 These studies have suggested that analternative approach for identifying microglial targets is to performexpression profiling at higher resolutionwhich allows us to identifyrare transcripts and screen for effects on cell-to-cell variability.

Sponsorships or competing interests that may be relevant to content are disclosed

at the end of this article.

a Pain Cognitive Function Research Center, Seoul National University, Seoul,

Republic of Korea, b Dental Research Institute and Department of Neurobiology and

Physiology, School of Dentistry, Seoul National University, Seoul, Republic of Korea,c Department of Biology, University of Pennsylvania, Philadelphia, PA, USA,d Medical Genomics Research Center, Korea Research Institute of Bioscience &

Biotechnology, Daejeon, Korea, e Department of Cell Biology and Neuroscience,

Rutgers University, Piscataway, NJ, USA, f Pain Research Laboratory, Institute of

Nautical Medicine, Nantong University, Nantong, Jiangsu, China, g Department of

Computer and Information Science, University of Pennsylvania, Philadelphia, PA,

USA, h Department of Brain and Cognitive Sciences, College of Natural Sciences,

Seoul National University, Seoul, Republic of Korea

*Corresponding author. Address: Pain Cognitive Function Research Center,

Department of Brain and Cognitive Sciences, College of Natural Sciences, Seoul

National University, 101 Daehakro, Jongno-Ku, Seoul 110-744, Korea. Tel.: (82)

2-740-8656; fax: (82) 2-762-5107. E-mail address: odolbae@snu.ac.kr (S. B. Oh).

Supplemental digital content is available for this article. Direct URL citations appear

in the printed text and are provided in the HTML and PDF versions of this article on

the journal’s Web site (www.painjournalonline.com).

PAIN 157 (2016) 964–976

© 2016 International Association for the Study of Pain

http://dx.doi.org/10.1097/j.pain.0000000000000470

964 H. Jeong et al.·157 (2016) 964–976 PAIN®

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

In this study, we employed near single-cell microarray analysisof microglia from spinal cord slices after L4 spinal nervetransection (SNT). Recent researchers have achieved single-cellresolution for the whole transcriptome35 but given the size of themicroglia, we collected pools of 10 cells. Although the collectionof 10 cells is not at the level of single cells, mechanical selection ofindividual cells helped to generate microglia-enriched samples.Using this approach, we investigated a murine microglialtranscriptome between nerve-injured vs sham-operated micro-glia at POD1 and POD7. By comparing gene profiles betweenthese 2 time points, we aimed to identify novel genes exclusivelyinvolved in microglia activation after nerve injury, which may leadto initiation of neuropathic pain and to find novel microglial genesspecifically involved in the transition to neuropathic pain de-velopment and/or maintenance.

2. Materials and methods

2.1. Mice

Adult C57BL/6 and heterozygous Cx3cr11/GFP male mice (8-10weeks old) produced by breeding Cx3cr1GFP/GFP mice30 withwild type C57BL/6 were used for this work. Five mice per cagewere kept. All animals were housed on a 12 hour/12 hour light/dark cycle with food and water provided ad libitum. Mice wereacclimatized for at least one week before the experiments.

2.2. Animal models

All surgical and experimental procedures were reviewed andapproved by the Institutional Animal Care and Use Committee ofthe School of Dentistry, Seoul National University before theexperiments. Animal treatments were performed according to theGuidelines of the International Association for the Study of Pain. Forthe spinal nerve transection model, the surgery was performed asdescribed previously.13,32 Briefly, under sodium pentobarbitalanesthesia, a 1 cm longitudinal incision overlying the L3-L6 sectionwas made. The right paraspinal muscles were separated from thesuperior articular processes and the transverse processes toexpose the L3 and L4 spinal nerves. The L4 spinal nerve wastransected without damaging the L3 spinal nerve. The wound wasclosed by suturing the muscle and skin layers. The sham groupreceived a sham surgery without nerve transection.

2.3. Behavioral test

Mechanical sensitivity was assessed at 1-day and 7-day afternerve injury during daytime by the up–down method usingcalibrated von Frey filaments as described previously.8 Six micefor each group were tested. Mice were acclimatized on a metalmesh floor in transparent plastic boxes (603 1003 60 mm) for 1hour. The mechanical sensitivity was evaluated using a set of 8calibrated von Frey filaments (0.02, 0.04, 0.07, 0.16, 0.4, 0.6, 1.0,and 1.4 g; Stoelting, Wood Dale, IL) that were applied to theplantar surface of the hind paw until the filament bent slightly fora few seconds. A withdrawal reflex of the hind paw duringstimulation or immediately after stimulus removal was considereda positive response. The first stimulus was always the 0.4 gfilament. When there was a positive response, the next lowerfilament was applied, and when there was no response, the nexthigher filament was used. After the first change in responses, 4additional responses were observed and the 50% paw with-drawal threshold value was calculated.14 Sensory motor co-ordination was examined by using the rotarod test. All behavioraltests were performed blindly.

2.4. Microglial cell collection

Mice were deeply anesthetized with excess isoflurane anddecapitated. The spinal cord including the lumbosacral enlarge-ment (L3-L5) was exposed by a dorsal laminectomy. Dissectedtissue blocks were placed into ice-cold cutting solution contain-ing (in millimolars) 245 sucrose, 3 KCl, 6.0 MgCl2, 0.5 CaCl2, 26NaHCO3, 1.25 NaH2PO4, 11 glucose, 5 N-[2-hydroxyethyl]piperazine-N9-[2-ethanesulfonic acid] (HEPES), 1.0 Kynurenicacid (pH 7.4, when bubbled with 95% O2/5% CO2). Transverseslices (300 mm) were prepared using vibroslicer (Lieca VT1000Plus; Leica Microsystems GmbH, Wetzlar, Germany) andcollected in a slice chamber containing recording artificialcerebrospinal fluid (aCSF) composed of (in millimolars) 126 NaCl,3 KCl, 1.3 MgCl2, 2.5 CaCl2, 26 NaHCO3, 1.25 NaH2PO4, 11glucose, and 5 HEPES (pH 7.4, when bubbled with 95% O2 and5% CO2 and had osmolarity of 305-310 mOsmol). The sliceswere initially maintained at 32˚C for 45 minutes to recover andtransferred to the slice chamber kept at room temperature (2561˚C) until used. Slices were continuously superfused withdiethylpyrocarbonate (DEPC)-treated aCSF (;5 mL per minute).Individual cell collection was performed as previously de-scribed.33 Briefly, CX3CR1-enhanced green fluorescentprotein-positive microglia located approximately 30 to 60 mmunderneath the superficial dorsal horn (laminae I and II) wasvisualized under fluorescence microscopy. Individual spinalmicroglia from 3 different spinal cord slices were randomlycollected with approximately 10-mmglass pipettes gently put intoa reaction tube containing reverse transcription reagents. In-dividually identified microglia forming samples of 10 cells wereused for each microarray transcriptome assay. Each samplerepresents an average of 10 cells and although the measurementis not at the single-cell level, the sample-to-sample variation isexpected to be proportional to the single cell variation. If thevariance among single cells is v, then by assuming a random poolof 10 cells, the variance of the 10-cell averages is expected to beproportional to v/10. The cDNA was synthesized from 10 pooledmicroglia collected from three spinal cord slices obtained fromone animal, and it was used for each microarray chip. Tenmicroglia were selected 6 times for each experimental groupwhich consists of 6 mice. A total of 24 mice were tested.

2.5. RNA isolation and microarray expression profiling

Microarray samples were prepared as described previously.17

Briefly, the total RNA from 10 pooled microglia was reverse-transcribed into cDNA using a SMARTer Pico PCR cDNASynthesis Kit (Takara Clontech, Kyoto, Japan). Advantage2Polymerase Mix (Takara Clontech) was used to amplify the wholecDNA according to the manufacturer’s instructions. The reactionprofile was 1 minute at 95˚C; 20 cycles of 5 seconds at 95˚C, 5seconds at 65˚C, or 6 minutes at 68˚C. We then added 2 mL of 12mM modified Primer IIA with the T7 promoter and 2 mL of 503Advantage2 Polymerase Mix again and performed another thermalcycling for 1 minute at 95˚C, 10 cycles of 5 seconds at 95˚C, 5seconds at 65˚C, 6minutes at 68˚C, or 6minutes at 68˚C. To assessthe quality of cDNA prepared from 10 pooled cells for microarrayexperiment, a portion of the cDNA (1mL) was used as a template fornested polymerase chain reaction (PCR) to detect b-actin andCD11b expression. Aspirated bath solution was used as thenegative control. All PCRamplificationswereperformedwith nestedprimers. The sequence used are as follows: CD11b, 59-ACATGT-GAG CCCCATAAAGC-39 and 59-TCAGGGCTTCAAAGTTGTCC-39 in the first PCR (326 bp), 59-ACATGTGAGCCCCATAAAGC-39

April 2016·Volume 157·Number 4 www.painjournalonline.com 965

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

and 59-AATGACCCCTGCTCTGTCTG-39 in the second nestedPCR (205 bp); b-actin, 59-CATCACTATTGGCAACGAGCG-39 and59-ACATCTGCTGGAAGGTGGACAG-39 in the first PCR (325 bp),59-GGCTCTTTTCCAGCCTTCCTT-39 and 59-CCACCGATCCA-CACAGAGTACT-39 in the second nested PCR (251 bp).

With samples verified with reverse transcription polymerasechain reaction (RT-PCR), 3.4 mg per cDNA sample was used asinput into the Affymetrix procedure as recommended by protocol.Amplified RNA (cRNA) was generated from the double-strandedcDNA template through an in vitro transcription (IVT) reaction andpurified with the Affymetrix sample cleanup module. The cDNAwas regenerated through a random-primed reverse transcriptionusing a dNTP mix containing dUTP. The cDNA was thenfragmented by UDG and APE 1 restriction endonucleases andend-labeled by a terminal transferase reaction incorporatinga biotinylated dideoxynucleotide. Fragmented end-labeled cDNAwas hybridized to the Affymetrix GeneChip Mouse Gene 1.0 STarrays for 16 hours at 45˚C and 60 rpm as described in the GeneChip Whole Transcript Sense Target Labeling Assay Manual(Affymetrix, Santa Clara, CA). After hybridization, the chips werestained and washed in a Genechip Fluidics Station 450(Affymetrix) and scanned by using a Genechip Array scanner3000 7G (Affymetrix). Six biologically independent hybridizationswere performed for each condition. The raw data for individualplates (libraries) are accessible in Gene Expression Omnibusunder the accession code GSE60670.

2.6. Microarray data processing

The image data were extracted through Affymetrix CommandConsole Software 1.1. The raw .cel file generated through theabovementioned procedure meant expression intensity data andwas used for the next step. Expression data were generated byAffymetrix Expression Console Software version 1.1. Global geneexpression profiles and single cell variation analysis wereperformed. Six biologically independent hybridizations wereperformed for each experimental group excluding the shamPOD1group (n 5 5 for shamPOD1 group). We excluded one of“shamPOD1” sample from the 6 shamPOD1 samples because itturned out that the shamPOD1 sample was an outlier accordingto the Spearman’s correlation coefficients analysis.

Wenormalized the arrays usingRobustMultichip Average (RMA)normalization.28 In order to obtain robust expression values from10-cell samples, we summarized perfect match probes on thearray by specifically taking the log of the second highest perfectmatch probe intensity. In addition, we removed batch effects usinga 2-stage least squaresmethod for estimation of location and scaleparameters because nonbiological variability, such as differencesin sample preparation and hybridization protocols, can bepotentially confounding biological differences.19 For differentialexpression analysis, a t test was performed. To compare thedistribution of coefficient of variation (CV) between conditions,Wilcoxon rank-sum test was conducted. For heat maps, we usedfollowing web site; http://www.tm4.org/mev.html.

Metric multidimensional scaling (MDS) was performed tovisualize the heterogeneity of gene expression profile betweenany pair of samples. Briefly, metric MDS finds an orthogonalprojection to a set of axes ordered by the amount of variance. Inaddition, we applied a linear discriminant analysis (LDA) analysisto investigate whether different classes have distinct patternswhere the axes were rescaled to make the 2 sets of projectscomparable. Linear discriminant analysis finds a linear function ofthe variables that maximizes the ratio of between-class varianceand minimizes the ratio of within-class variance. Metric

multidimensional scaling and LDA were performed using animplementation in the R Statistical Software package down-loaded from www.r-project.org.

To assess within-group similarity of gene expression, we usedthe symmetric Jensen-Shannon divergence (JSD) index, whichranges from 0 (identical) to 1 (“distant” distributions) such thatlower JSD scores reflect higher similarity between transcrip-tomes. Tomeasure cell-to-cell variability captured in 10-cell units,we used the CV defined by the ratio of the SD to the mean (CV5SD/mean). The variances between groups were assessed usingan equality of variances F test. A 2-sample F test was performedto compare 2 population variances.

2.7. Functional and pathway analysis

To understand molecular functions involved in the early and latemicroglia, each of the POD1-specific, POD7-specific, andcommon differentially expressed genes (DEGs) were input toPathway Studio (version 9.0; Ariadne Genomics, Rockville, MD)43

and Gene Ontology (GO) classification was performed. Gene setenrichment analysis was applied to search for groups of genesinvolved in the same processes (gene sets) that were alteredsignificantly by nerve injury. Cutoff parameters for significancewereP value,0.01 and overlap (the number of overapped genes) .2.We analyzed disease pathways using Pathway Studio. Diseaseentities were identified using the “common targets” option.

2.8. Primary microglia culture

Mixed glial cultures were prepared from cerebral cortex of 1 to 2-day-old postnatal mice in accordance with themethod describedpreviously.2 Briefly, mixed glial cells were cultured andmaintainedfor 14 to 16 days in high-glucose DMEM medium including 10%fetal bovine serum. Microglia were harvested from the mixed gliaby shaking T75 flasks. The media including floating cells werecollected and microglia were plated onto 6-well plates in thedensity of 1.0 3 106 cells per milliliter for subsequent ATPtreatments or used for transfection. Microglia purity was greaterthan 95%, as confirmed by Iba1 staining. Two days after plating,microglia were treated with 50 mMATP (Sigma, St. Louis, MO) for1 hour which was previously shown to activate microglia.54

2.9. Quantitative real-time reverse transcription polymerasechain reaction

Total RNA or specifically the small RNA fraction was extracted byusinganRNeasyMini Kit ormiRNeasyMini kit (Qiagen, Valencia,CA)respectively, and 0.5 to 1.0 mg was processed for cDNA synthesisusing M-MLV Reverse Transcriptase (Invitrogen, Carlsbad, CA) ormiScript PCRStarter Kit (Qiagen, Hilden, Germany) according to themanufacturer’s instructions. The primers used formiRNAweremiR-29b-3p, 59-UAGCACCAUUUGAAAUCAGUGUU-39 (cat no.MS00005936, miScript Primer Assays); miR-29c-3p, 59-UAGCAC-CAUUUGAAAUCGGUUA-39 (cat no. MS00001379, miScriptPrimer Assays); miR-137-3p, and 59-UUAUUGCUUAA-GAAUACGCGUAG-39 (cat no. MS00001589, miScript PrimerAssays); snRNA RNU6B (RNU6-2), which were provided ina miScript PCR Starter Kit (cat no. MS00033740; miScript PrimerAssays). Real-time PCR was performed with cDNA prepared formicroarray experiments or obtained from primary cultured microgliaas a template using a 7500 Real-Time PCR System (AppliedBiosystems, Foster City, CA). The threshold cycle (Ct) of theGAPDHor snRNA RNU6B was used as a reference control to normalize theexpression level of the target gene (DCt) to correct for experimental

966 H. Jeong et al.·157 (2016) 964–976 PAIN®

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

variation. Relative mRNA levels were calculated according tothe 2-DDCt method.39 Real-time RT-PCR experimentswere performed at least 3 times. Values are presented as mean6 SEM unless otherwise noted. The PCR primer sequencesused in this study are as follows; GAPDH, 59-TCCATGA-CAACTTTGGCATTG-39 and 59-CAGTCTTCTGGGTGG-CAGTGA-39; Il-1b, 59-ACCTGCTGGTGTGTGACGTTC-39and 59-CAGCACGAGGCTTTTTTGTTGT-39; Tnfa, 59-AGCAAA ECA CCA AGT GGA GGA-39 and 59- GCT GGCACCACTAGTTGGTTGT-39; Cx3cr1, 59-CAGCATCGACCGG-TACCTT-39 and 59-GCTGCACTGTCCGGTTGTT-39; P2rx4,59-TGGCCGACTATGTGGTCCCA-39 and 59-GGTTCACGGTGAC-GATCATG-39; C5ar1, 59-GTCCTGTTCACGACCGTTTT-39 and 59-ACGGTCGGCACTAATGGTAG-39; Hexb, 59-AGCGCTGTTGGT-GAGAGACT-39 and 59-CTATTCCACGGCTGACCATT-39;Gria1, 59-TCAATGAAGCCATACGGACA-39 and 59-GCTGAC-CACTCTGCCATTCT-39; P2ry12, 59-TTTGTTCCCTTCCACTTTGC-39 and 59-AGGGTGCTCTCCTTCACGTA-39; Glul, 59-CTGCCA-TACCAACTTCAGCA-39 and 59-TGTGGTACTGGTGCCTCTTG-39; Trf, 59-CCGGGTTAAGGCTGTACTGA-39 and 59-ACA-GAAGGTCCTTGGTGGTG-39; Spg21, 59-GAGTGCGCTGTCAA-CAGAAG-39 and 59-CTCCTGCACAGGTAGGGAAA-39; Cdk5, 59-GTCCATCGACATGTGGTCAG-39 and 59-GCTGGTCATCCA-CATCATTG-39; Bank1, 59-TTCAGCAGGAAAAGCTACGG-39and 59-TGGTGCACAATGGTCAGTTT-39; Arf6, 59-CAGGGTCT-GATCTTCGTGGT-39 and 59-CTCATCTCCCGGTCATTGAT-39.

2.10. Western blotting

Proteins were prepared using Pro-Prep solution (Intron bio-technology, Seoul, Korea). Protein concentration was deter-mined using a BCA assay kit (Pierce, Rockford, IL). Thirtymicrograms of proteins were separated by 10% SDS-PAGEelectrophoresis and transferred onto a PVDF membrane. Afterblocking, the membrane was probed with anti-pp38 (CellSignaling Technology, Beverly, MA, cat no. 4631s, 1:1000) andp38 (Cell Signaling Technology, cat no. 9212, 1:1000) overnightfollowed by horseradish peroxidase-conjugated antirabbit (SantaCruz Biotechnology, Santa Cruz, CA, cat no. sc2004, 1:5000) for1 hour at room temperature. Blots were detected and quantifiedusing a Gel-Doc XRS1 system (BioRad, Hercules, CA) by usinga chemiluminescent reagent (SuperSignal West Pico; ThermoFisher Scientific, Waltham, MA). Independent experiments wereconducted at least 3 times.

2.11. siRNA/miRNA inhibitor and miRNA mimic transfection

AccuTarget Gria1 siRNAs (1365968; Bioneer, Daejeon, Korea)and negative control siRNA were purchased from Bioneer. Theantisense 29-O-methyl (29-O-Me) oligonucleotide for miR-29cinhibitor (29-O-Me-29c-AS) and NC5 negative control wereobtained from Integrated DNA Technologies (San Diego, CA).The sequences of 29-O-Me-29c-AS and NC5 negative controlare 59-mU/ZEN/mAmAmCmCmGmAmUmUmUmCmAmAmA-mUmGmGmUmGmCmU/3ZEN-39 (r, RNA base; m, 29O-methyl base) and 59-mG/ZEN/mCmGmAmCmUmAmU-mAmCmGmCmGmCmAmAmUmAmUmGmG/3ZEN-39 (r, RNAbase; m, 29 O-methyl base). To increase miR-29c levels, weadministered a chemically modified double-stranded miRNAmimic. Primary cultured microglia were transfected with Gria1siRNA (10 nM) or miR-29c inhibitor (10 nM) or miR-29c mimic (10nM) by electroporation using a microporator (MP-100; Digital BioTechnology Co., Seoul, Korea) following the manufacturer’sinstructions. Transfected microglia was seeded at 1 3 106 cells

per well of 6-well plates. Real-time PCR was used to determineknockdown efficacy of siRNA, miRNA inhibitor, and miR-29cmimic. After 48 hours for Gria1 and miR-29c mimic and 60 hoursformiR-29c inhibitor, we harvested the transfectedmicroglia and,after centrifugation, transferred only microglial cell resuspendedin PBS, without culture media, into animals.

2.12. Intrathecal injection of microglia and drug injection

Gria1 knockdown microglia or miR-29c inhibited microglia (1 3106 cells, 10 mL) were injected intrathecally in normal mice asdescribed previously.27 Briefly, under isoflurane anesthesia, micewere injected using a 50 mL Hamilton syringe fitted with a 31gauge needle by direct lumbar puncture between the L5 and L6levels. Control animals were injected with microglia transfectedwith the NC5 negative control (miRNA inhibitor control, NC) or thenegative control siRNA (siScram). Paw withdrawal threshold wastested 1, 3, and 5 hours after injection.

The TNFa antagonist etanercept (25 mg per vial; Enbrel,Wyeth, Taplow, Maidenhead, United Kingdom) was adminis-trated by systemic intraperitoneal injection 1 hour after microglialinjection. Dilution before administrations was made in PBS fortreatment of mice.

2.13. Preparation of DRG neurons

Dorsal root ganglia neurons obtained from 4 to 6-week-old micewere prepared. Animals were anesthetized with overdose ofisoflurane, decapitated, and DRG were rapidly removed underaseptic conditions and placed in HBSS (Gibco, Gaithersburg,MD). Dorsal root ganglia were digested in 1mg/mL collagenase A(Roche, Mannheim, Germany) and in 2.4 U/mL dispase II (Roche)in HBSS for 60 minutes, respectively, followed by 8 minutes in0.25% trypsin (Sigma), all at 37˚C. The DRG were then washed inneurobasal medium (Gibco) 3 times and resuspended in DMEMmedium supplemented with 10% FBS (Invitrogen, Carlsbad, CA)and 1%penicillin/streptomycin (Sigma). They were thenmechan-ically dissociated using fire-polished glass pipettes, centrifuged(800 rpm, 5 minutes), and resuspended in DMEM mediumsupplemented with 10% FBS (Invitrogen), 13 B-27 supplement(Invitrogen), 13 N-2 supplement (Invitrogen), and 1% penicillin/streptomycin (Invitrogen), and plated on 0.5mg/mL poly-D-lysine(Sigma)-coated glass coverslips. Cells were maintained at 37˚C ina 5% CO2 incubator.

2.14. Ca21 imaging

We performed fura-2 AM-based (Molecular Probes, Eugene,OR) Ca21 imaging experiments. The DRG neurons were loadedwith fura-2 AM (2 mM) for 40 minutes at 37˚C in a balanced saltsolution containing (in millimolars) 140 NaCl, 5 KCl, 2 CaCl2, 1MgCl2, 10 HEPES, 10 glucose, adjusted to pH 7.3 with NaOH.The cells were then rinsed with the Neurobasal medium andincubated in it for an additional 20 minutes to deesterify the dye.Cells on slides were placed onto an inverted microscope andilluminated with a 175 W xenon arc lamp; excitation wave-lengths (340/380 nm) were selected by a monochromatorwavelength changer. Intracellular calcium concentrations([Ca21]i) were measured after every 1-second interval by digitalvideo microfluorometry with an intensified charge-coupled-device (CCD) camera (CasCade; Roper Scientific, Duluth, GA)coupled to the microscope and a computer with MetaFluorsoftware (Universal Imaging Corporation, West Chester, PA).Independent experiments were conducted at least 3 times.

April 2016·Volume 157·Number 4 www.painjournalonline.com 967

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

2.15. Electrophysiology

Electrophysiological recordings of postsynaptic field potentials(fPSPs) evoked by dorsal root stimulation were taken usinga whole spinal cord preparation. Adult male C57BL/6 micewere anesthetized with excess isoflurane and the lumbarspinal column was removed and placed into an ice-coldcutting solution containing (in millimolars) 245 sucrose, 3 KCl,6.0 MgCl2, 0.5 CaCl2, 26 NaHCO3, 1.25 NaH2PO4, 11glucose, 5 HEPES, and 1.0 kynurenic acid (pH 7.4, whenbubbled with 95% O2/5% CO2). After that the lumbar spinalcord was quickly removed by means of laminectomy andplaced in aCSF composed of (in millimolars) 126 NaCl, 3 KCl,1.3 MgCl2, 2.5 CaCl2, 26 NaHCO3, 1.25 NaH2PO4, 11glucose, and 5 HEPES (pH 7.4, when bubbled with 95% O2

and 5% CO2 and had osmolarity of 305-310 mOsmol) for 1hour before recordings. The fPSPs data were recorded andacquired using an EPC-10 amplifier (HEKA, Lambrecht,Germany) and Pulse 8.30 software (HEKA). The fPSP record-ings were performed at room temperature (236 1˚C) using;3MV borosilicate capillaries (World Precision Instruments,Sarasota, FL) filled with aCSF. The electrodes were insertedinto the dorsal side of the spinal cord in Lissauer’s tract.Electrical stimuli (0.01 millisecond, 0.0167 Hz) were deliveredthrough a theta glass suction electrode, which was filled withaCSF and placed in the dorsal root. After a baseline recording(10 minutes), the microglial cells (1 3 106 in 500 mL) werefocally perfused into spinal dorsal horn through a borosilicatecapillary placed toward the recording area. The amplitude ofevoked fPSPs was measured, 2 consecutive fPSP wereaveraged, and these measurements were expressed relativeto the normalized baseline.

2.16. Immunofluorescence

After fixing microglia with 4% paraformaldehyde for 10 minutes atroom temperature, microglia were incubated for 1 hour ina blocking solution containing 5% normal donkey serum(Jackson ImmunoResearch Laboratories, Bar Harbor, ME) and5% FBS (Gibco) in PBS-T (0.1% Triton X-100). The cells wereincubated with primary antibody for rabbit anti-GluA1 (1: 500;Millipore, cat no. AB1504) and rabbit anti-Iba1 (1:1000; Wako,Tokyo, Japan, cat no. 019-1974) at 4˚C overnight. After rinsingwith PBS, the cells were incubated with Alexa fluor 488 goatantirabbit (1:200; Jackson ImmunoResearch Laboratories) andcy3 fluor goat antirabbit (1:200; Jackson ImmunoResearchLaboratories) at room temperature for 1 hour. Following washingwith PBS, cells were mounted with the Vectashield mountingmedium (Vector Laboratories, Burlingame, CA). Fluorescentimages were obtained using a confocal microscope (LSM700;Carl Zeiss, Germany). Independent experiments were performedat least 3 times.

2.17. Enzyme-linked immunosorbent assay (ELISA)

After transfection, primary cultured microglia were plated 1.53106 cells per well in 6-well plates and cell supernatants werecollected 48 hours after transfection. The levels of TNFa insupernatants were quantified using the mouse TNFa DuosetELISA kit (R&D Systems, Minneapolis, MN, cat no. DY410-05).For each reaction in a 96-well plate, 100 mL of supernatantswere used, and ELISA was performed according to themanufacturer’s instructions. Independent experiments wereperformed 3 times.

2.18. In situ hybridization of miR-29c

In situ hybridizations were performed in 14-mmcryosections fromspinal cord. Sections were fixed in 4% paraformaldehyde-0.1 MPBS for 30 minutes followed by washes in diethylpyrocarbonate-treated ultrapure water. After treatment with a mixture of 30%H2O2 andmethanol (v/v5 1:50) for 30minutes, the sections weretreated with proteinase K (Boster, China) for 2 minutes at roomtemperature. The sections were prehybridized in a hybridizationsolution (50% deionised formamide, 0.3 M NaCl, 5 mM EDTA, 10mM NaPO4, 0.5 mg/mL yeast tRNA, 10% dextran Sulfate, and13 denhardt’s solution) at 25˚C below the predicted Tm value ofthe LNA probe for 2 hours. Next, sectionswere incubatedwith 59-DIG–labeled and 39-DIG–labeledmaturemiR-29cmiRCURY LNAdetection probe (/5DigN/TAA CCG ATT TCA AAT GGT GCT A/3Dig_N/) or microRNA detection control probes for ISH (/5DigN/GTG TAA CAC GTC TAT ACG CCC A/3Dig_N/) (Exiqon,Vedbaek, Denmark) at 53˚C overnight. Sections were thenincubated in blocking solution at 37˚C for 30 minutes and inmouse anti-DIG–biotin for 60 minutes, washed, and incubatedusing an SABC-FITC regent (Boster) for 30 minutes. To identifythe cell types expressing miR-29c, the above sections under ISHwere incubated overnight at 4˚C with primary antibodies againstGFAP (mouse, 1:6000; Millipore), NeuN (mouse, 1:1000; Milli-pore), or IBA-1 (rabbit, 1:3000;Wako). On the following day, Cy3-conjugated secondary antibody was added and incubated for 2hours. The signal was detected with a Leica fluorescencemicroscope, and images were captured with a CCD Spotcamera.

2.19. Statistics

All data were checked for normality (Shapiro Wilk) beforestatistical testing. For the data showing nonnormal distribution,we performed Mann-Whitney U test. Data were presented asmean 6 SEM and were analyzed using SPSS and GraphPadPrism version 5.0 (GraphPad Software, San Diego, CA). Forsample sizes, no statistical methods were used to predeterminesample sizes, but our sample sizes are similar to those generallyemployed in this field. Statistical analyses of the 50% withdrawalthresholds were performed using the Mann–Whitney U test atindividual time points or 2-way repeated ANOVA. The unpaired 2-tailed Student t test was used for real-time PCR analysis, westernblotting, and calcium imaging analysis. The value of P, 0.05 wasconsidered statistically significant.

3. Results

3.1. Validation of 10-cell microarray data

L4 spinal nerve transection produced significant mechanicalallodynia from POD1 and peaked at POD7 (Fig. 1A). Microgliain the ipsilateral dorsal horn largely consisted of activatedmicroglia, which was identified by CX3CR1-enhanced greenfluorescent protein–positive expression (Fig. 1B). We selectedthese 2 time points, POD1 and POD7 to compare microglialgene profiles during neuropathic pain initiation and develop-ment or maintenance against respective sham-operatedcontrols (Fig. 1A). Reverse transcription polymerase chainreaction revealed the expression of CD11b (Fig. 1D) inmicroglia randomly collected with glass pipettes from super-ficial laminae of L3 to L5 spinal dorsal horn (Fig. 1C), but not innegative controls consisting of aspirates of the bath solution(Fig. 1D), indicating there were no detectable levels of RNAcontamination.

968 H. Jeong et al.·157 (2016) 964–976 PAIN®

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

To check the consistency of 10-cell microarray data, 14microglial genes previously known to increase their expressionafter nerve injury and closely associated with neuropathicpain21,52,55,59 were examined (Fig. 2A). Most of the 14 geneswere upregulated after injury. In contrast, housekeeping genesand most marker genes for other CNS cell types showedinsignificant changes (Figs. 2A, E). We further validated the datawith real-time PCR by confirming increased expression ofrandomly selected 5 genes among the 14 microglial genes (Fig.2D). We also examined the 29 spinal cord microglia-specificgenes recently reported byChiu et al. (Fig. 2B).9We found that 28genes (except one gene missing on our microarray chip) werehighly expressed both in sham and injured microglia (Fig. 2B andFig. S1A; http://links.lww.com/PAIN/A202). Butovsky et al.reported 40 microglial genes that were uniquely expressed incentralmicroglia but not in peripheralmonocytes.5 These 40 geneswere also expressed both in sham and injuredmicroglia (Fig. S1B),

and upregulated after nerve injury compared with their respectivesham (Figs. 2C, E). Taken together, these results demonstrate thatthe 14 previously known genes in 10-cell microarray results wereconsistent with expression profiles from previous studies andsuitable for screening neuropathic pain-linked genes with minimalcontamination from other CNS cell types.

3.2. Time-dependent differential changes in microglial geneexpression after nerve injury

To find early and late responsive microglia genes for peripheral nerveinjury,wedeterminedDEGsbycomparing the results fromPOD1andPOD7 with their respective sham-operated controls. A total of 559geneswhoseexpression significantly changed inmicroglia after nerveinjury were identified. Of these, 312 geneswere upregulated and 247

Figure 1. Experimental design for microarray with 10 pooled spinal microgliaafter L4 SNT. (A) Left, L4 SNT microglial cells were compared to shammicroglia after nerve injury on day 1 and day 7, respectively. Right, mechanicalthreshold was measured in sham and L4 SNT mice on the first, third, andseventh day after surgery (n 5 12 for each experimental group, total n 5 24mice). Mechanical allodynia was fully developed on POD7. P5 0.0432 for day1, P5 0.006 for day 3, and P5 0.0070 for day 7, as compared with those ofrespective sham controls. Data are shown as means 6 SEM. (B)Representative images of GFP-labeled microglia (green) in spinal cord ofmouse on POD1 and POD7 after L4 SNT. Scale bar: 100 mm (C) Imagesshowing individual collection of spinal microglia into a glass pipette. Scale bar:40 mm. (D) A representative image of 10 pooled-cell RT-PCR analysis showingthe expression of CD11b, a well-known microglia marker, in 10 pooledmicroglia (MG) and the PC but not in NC. b-actin was used as a endogenouspositive control. cDNA from primary cultured microglia was used as a PC. M,a marker for DNA size; PC, positive control; NC, negative control; SNT, spinalnerve transection; RT-PCR, reverse transcription polymerase chain reaction.

Figure 2. Validation of single-cell microarray data (A) A heat map of expressionprofile with 14 known microglial genes associated with neuropathic pain,comparing the injury group with the sham-operated group at POD1 andPOD7, respectively. No significant changes were detected in the expressionlevel of housekeeping genes between sham and SNT microglia. Marker genesfor other CNS cell types were used as the negative control. (B) A heat map ofthe expression profile of 28 spinal microglia-specific genes. (C) A heat mapof expression profile of 40 brain-derived microglial unique genes. (D) Validationof single-cell microarray data using single-cell microarray samples by real-timePCR, which revealed upregulation of Il-1b, Tnfa, C5ar1, Cx3cr1, and P2rx4.Data are shown asmean6SEM. (E) A summarizing graph showing expressionchanges of the different sets of genes. Previously knownmicroglial genes wereupregulated. Other CNS cell types and housekeeping genes were used asnegative control (CNS, central nervous system; POD, postoperative day).

April 2016·Volume 157·Number 4 www.painjournalonline.com 969

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

genes were downregulated (Fig. 3A). Of the 312 upregulated genes,123 were significantly upregulated only at POD1 (POD1-specificDEG) and 167were upregulated only at POD7 (POD7-specific DEG).The overlap between that obtained at POD1 and POD7 (commonDEG) yielded 22 genes that were upregulated throughout the courseof pain (Fig. 3B and Table. S1; http://links.lww.com/PAIN/A203).Interestingly, thePOD1-specific DEGs included3miRNAgenes suchas miR-137, miR-29b-1, and miR-29c.

Next, we analyzed DEGs using GO terms to assess whethercertain functions or molecular pathways were significantlyenriched within and across each time point. A number of genesrelated to chemotaxis, C5a signaling, succinate metabolicprocess, cytokine-mediated signaling, mitotic cell cycle, andcell shape were significantly enriched within the POD1 DEGs.For POD7, genes associated with Type I interferon mediatedsignaling, cellular response to calcium ion, cytokine-mediatedsignaling, interferon-g, chemotaxis, and phospholipid meta-bolic process were significantly regulated (Fig. 3C). Genefunctions involved in interleukin-6, chemokine, kinase, NF-kB,and toll-like receptor (TLR) signaling pathway were enriched inDEGs common to both time points (Fig. 3C). Interestingly,early-phase and late-phase microglia were associated with

distinct cellular compartments (Fig. 3D). Genes associatedwith membrane, mitochondrial matrix, and nucleus changed atPOD1, whereas genes located in the trans-Golgi transportvesicle, cytosol, and endoplasmic reticulum were largelyaltered at POD7. Plasma membrane-related genes wereupregulated throughout the course of pain (Fig. 3D). AlthoughGO analysis is often weak, these results suggest that genesrelated to sensing function dominate at POD1 whereasexpressions of genes induced by signaling pathways are moreprevalent at POD7.

3.3. Identification of candidate pain-maintenanceneuropathic microglial genes by variation analysis

We next investigated the sample-to-sample dispersion patternsto gain insights into potential single-cell variability of the microglia(we expected the 10-cell samples to reflect individual variabilitywith a scaling factor of sqrt(n), n5 10; see Methods). To visualizethe dispersal patterns, metric MDS was performed. Visually,POD7 and shamPOD7 were more dispersed than POD1 andshamPOD1, respectively (Fig. 4A). We also examined 2-dimensional projections of a subset of housekeeping genes

Figure 3. Time-dependent differential expression changes of spinal microglial genes after peripheral nerve injury (A) A Venn diagram showing the number ofupregulated and downregulated genes meeting fold change (log 2) .1.5 and statistical thresholds (P . 0.05 by t test) in each time point. (B) Gene expressionprofile of DEGs grouped into 3 clusters (POD1-specific, POD7-specific, and common genes) in the pseudocolor scheme. (C and D) Strongly regulated microglialgenes according to biological process (C) and cellular component (D) in each time point. Red, POD1; blue, POD7; and yellow, common DEGs (DEG, differentiallyexpressed genes; POD1, postoperative day 1; POD7, postoperative day 7).

970 H. Jeong et al.·157 (2016) 964–976 PAIN®

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

(Fig. 4A, left) and all genes (Fig. 4A, right). Housekeeping genesshowed less dispersal than all expressed genes in the spinalmicroglia (Fig. 4A), suggesting that not all genes show a broadvariation (see also Fig. S2C; http://links.lww.com/PAIN/A204).The JSD index was used to numerically quantify the visual pattern(Fig 4B). As seen in Figure 4B for both sham and injuryconditions, we observed a general increase in variability over time(Fig. S2A). We additionally compared the ratio of variances ofeach gene (POD7/POD1 and shamPOD7/shamPOD1) by an Ftest. In the injury condition, a smaller number of genes increasedin variation over time than in the sham condition (Fig. S2B). Wenote that unlike tissue samples with millions of cells that mightaverage across any hidden subtypes, our 10-cell samplecollection might be disproportionately affected by hiddensubtypes with the result that replicate samples cannot beguaranteed to match their cell collection. If some samples have

hidden subtypes while others do not, it may affect estimates ofvariation. However, we also note that the same hidden subtypefactor exists for all our treatment conditions and we are makinginferences based on differential variation across the treatmentconditions.

We hypothesized that the cells in an organismal environmentshow natural single-cell variability. When these cells are stressedor stimulated, such as by injury signals from primary afferentneurons, a subset of the transcriptome may respond withactivation of specific pathways. We also hypothesized that suchactivation will have the effect of synchronizing or resetting thegenes in those pathways to a coordinated expression pattern andthereby reduce variability across the samples for that subset ofgenes. In a temporal sequence, the transcriptome may showbroad coordinated expression from the initial stress response toboth injury and the collection protocol and then subsequently

Figure 4. Identification of candidate pain maintenance neuropathic microglial genes by variation analysis (A) MDS plots of the spinal microglia transcriptome. Leftpanel shows the dispersion using only 6 housekeeping genes (Tuba1a, Tuba1b, Hsp90ab1, Hsp90aa1, Hsp90b1, and Actb) as negative control. Right panelshows the dispersion using all genes (20,694 genes including the housekeeping genes). Each dot represents 10 pooled microglia. (B) Box plot of the JSD of thetranscriptome amongst samples of the same cell type, which captures the information-theoretic difference between 2 distributions (ie, transcriptomes). (C) A Venndiagram showing the overlap of gene identities of the entire transcriptome with time-dependent changes in their variation. (D) LDA plots showing classified genesinto 4 groups according to variation change from day 1 to day 7 after nerve injury. In particular, the group C genes (56 genes) show clearly distinct clusteringpatterns within the groups. In group A, the variation increased for both injury and sham and in group B, variation increased for injury but decreased for sham; groupC, variation decreased for injury but increased for sham; group D, variation decreased for both injury and sham (MSD, multidimensional scaling; JSD, Jensen-Shannon divergence; LDA, linear discriminant analysis).

April 2016·Volume 157·Number 4 www.painjournalonline.com 971

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

relax back to its natural heterogeneous state except for thespecifically activated pathways. Therefore, we reasoned that keygene classes linked to themaintenance of neuropathic painmightbe found among genes that show relatively low variation in the latephase of POD7. To explore this hypothesis, we first examined thepatterns of time-dependent changes in cell-to-cell heterogeneityfor a collection of 172 genes hypothesized to be important in paincategorized into 6 functional groups relevant to microgliaactivation31: cytoskeleton/ECM, enzyme, GPCR, receptor, se-creted molecule, and transcription regulation (Table. S2; http://links.lww.com/PAIN/A205). Among these, some categories suchas GPCRs and transcription regulation showed decreased cell-to-cell variation with time for the injury condition (Fig. S2C). Inparticular, we saw a significantly lower variation in secretedmolecules between POD1 and POD7 but not in the shamconditions (Fig. S2C), which might be consistent with co-ordinated expression of pain maintenance factors becauseconstant release of proinflammatory mediators contributes tocentral sensitization in spinal cord and chronic pain.57

Given this observation, we conducted a screening strategy forthe entire transcriptome to identify key genes by their pattern oftemporal cell-to-cell variation. We examined the overlap of geneidentities for the genes with time-dependent changes in theirvariation (Fig. 4C) and classified genes into 4 groups according tovariation changeof gene expression fromday 1 to day 7: groupA5estimated variation increased for both injury and sham, group B5estimated variation increased for injury but decreased for sham,group C 5 estimated variation decreased for injury but increasedfor sham, and group D 5 estimated variation decreased for bothinjury and sham. Among these groups, we expected group Cgenes to be the most plausible candidates for microglia activationand pain.We examined the clustering pattern of the samples usingonly the A, B, C, and D subset of genes by performing LDAprojection (Fig. 4D). In this figure, each dot is a 10-cell sample andthe colors indicate the experimental treatments. Group C genes (56genes; Table S3; http://links.lww.com/PAIN/A206) were particularlynoticeable in inducing tight groupings of the 4 experimentalconditionsconsistentwith the idea that thesegeneswere informativefor distinguishing the microglia treatment states. Interestingly, groupC contains P2ry12, which has been known to regulate microgliaactivation in neuropathic pain.25,53 These observations led us tofunctionally test genes selected from this group.

3.4. Validation of candidate pain genes

As mentioned above, we identified a gene set of interest using 2different analytic approaches: DEGs and variation analysis.Among the POD1-specific DEGs, we focused on 3 miRNAs(miR-137, miR-29b-1, and miR-29c) because miRNAs areknown to regulate transcription and have considerable effectson cellular functions.50 Expression of miR-137 was increased atPOD1, whereas expression of miR-29b-1 and miR-29c weresignificantly lower at POD1 (Fig. 5A). We validated miR-29cexpression using real-time PCR with ipsilateral spinal cord afterL4 SNT (Fig. 5B). Because P2X and P2Y signals are known to beimportant for pain hypersensitivity following peripheral nerveinjury,1,34 we adopted in vitro ATP stimulation to activatemicrogliaand assess the expression of these 3 miRNAs. Consistent withour 10-cell microarray data, real-time PCR showed significantdownregulation of miR-29c (Fig. 5C) and miR-29b-1, but noupregulation of miR-137 (data not shown), in ATP-stimulatedmicroglia. AlthoughmiR-137might still have expression changes,these results suggested a focus on miR-29b-1 and miR-29c astargets for in vivo functional assays. Activated microglial cells are

known to increase the expression of proinflammatory cytokines,such as TNFa and Il-1b, and phosphorylated p38 MAP kinase(pp38).29 We measured the expression of cytokines in miR-29cinhibitor-transfected microglia. MiR-29c inhibitor increased TNFaexpression compared to inhibitor control group, whereas wefound no difference in Il-1bmRNA expression in primary culturedmicroglia (Fig. 5C). We also found a significant increase of pp38expression in the miR-29c-inhibitor–transfected microglia(Fig. 5D). Together, these results were consistent with thehypothesis that miR-29c downregulation recapitulates thecytokine responses found in activated microglia. We nextexamined physiological consequences of miR-29c downregula-tion in microglia. Application of miR-29c-inhibitor–transfectedmicroglia elicited calcium responses in DRG neurons (Fig. 5E).With ex vivo whole spinal cord preparation, we also foundsignificant long-term potentiation (LTP) of evoked postsynapticfield potentials (fPSPs) in spinal dorsal horn following localapplication of miR-29C-inhibitor–transfected microglia (Fig. 5F).Intrathecal application of activated microglia by ATP producesmechanical allodynia in rats.54 To dissect their role in pain, weinvestigated whether miR-29c inhibitor-transfected microglia aresufficient to produce tactile allodynia when injected into naiveanimals. The paw withdrawal threshold decreased markedly innormal mice after the intrathecal injection of miR-29c-inhibitor–transfected microglia, but not in the inhibitor control group (Fig.5G), suggesting that downregulation of microglial miR-29c inspinal cord is sufficient for developing mechanical allodynia. Inaddition, we tested whether miR-29c overexpressing microgliacan resolve mechanical allodynia induced by ATP-stimulatedmicroglia. Application of miR-29c-mimic–transfected microgliasignificantly reversed mechanical allodynia, compared to control-miRNA–transfected microglia at 5 hours after injection; miceinjected with control miRNA were comparable to mice injectedwith ATP-stimulated microglia (Fig. 5H). Motor behavior,assessed by rotarod test, was not changed bymicroglial injectionper se (data not shown). Collectively, these results suggest thatdownregulation of miR-29c in microglia may play an importantrole in the development of neuropathic pain.

Next, we focused on 56 genes found through temporalvariation analysis (group C; see above) to assess whether anyof these genes may play a role in the development/maintenanceof neuropathic pain. Of the 56 identified genes, we usedinformation from previous literature to focus on changes in theexpression level of 9 genes (Hexb, Gria1, Glul, P2ry12, Trf,Spg21, Cdk5, Bank1, and Arf6) in the ipsilateral spinal cord atPOD7. Of the 9 genes, only the P2ry12 expression showeda significant difference between that of sham and POD7, butwe found no significant difference in other 8 genes (Fig. S4A;http://links.lww.com/PAIN/A208). Because microglial changescould be masked in the whole spinal cord tissue, we additionallytested expression changes under in vitro ATP stimulationconditions. Under ATP simulation, we only found a significantchange in the expression of Gria1, the gene encoding AMPAreceptor subunit GluA1, both at mRNA and protein levels (Figs.6A, B), but not for the other 8 genes (Fig. S4B). Animmunofluorescence analysis revealed that microglia stainedwith Iba-1 expressed GluA1 in the resting state, whereas itsexpression significantly decreased when microglia were treatedwith ATP (Fig. 6B). A previous study suggested that activation ofAMPA receptor by glutamate promotes release of TNFawhereasthe potentiation ofmicroglial AMPA receptors by AMPA receptorsagonist inhibits TNFa release.22 Microglial Gria1 knockdowninduced TNFa release and increased mRNA expression (Figs.6C, D). We tested live animal response to Gria1 expression

972 H. Jeong et al.·157 (2016) 964–976 PAIN®

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

changes by intrathecal injection of siGria1microglia. Similar to ourresults for miR-29c, we found that microglial Gria1 knockdownproduced mechanical allodynia. Paw withdrawal thresholddecreased only after the injection of siGria1 microglia, but not ofsiScram microglia (Fig. 6E). With administration of etanercept(1 mg, i.p.), a TNFa inhibitor, 1 hour after injection of siGria1microglia, siGria1 microglia induced pain was significantlyreversed at 5 hour after microglial injection (Fig. 6F). Takentogether, these results suggest that Gria1 downregulation maycontribute to mechanical allodynia through TNFa production.

4. Discussion

In this study, we adopted a high resolution, individually collected,transcriptome assay to screen for microglia-specific factors forthe first time in a rodent neuropathic pain model. Mechanicalcollection of marked microglia allowed us to assess pure cellpopulations for expression changes resulting in an unbiased (upto probes on themicroarray) discovery process. Furthermore, ourapproach allowed us to further examine the diversity of geneprofiles within multiple small pools of cells coupled with a time-dependent variation analysis resulting in a novel approach forcandidate gene discovery. Many microglial changes could bemasked in the analysis with the whole spinal cord tissue (Fig.S4A), and analyses based on cell variation would not have beenpossible with the whole tissue level data. From our analysis, wefound downregulated miR-29c in early spinal microglia to bea possible factor for the microglia activation and the initiation ordevelopment of neuropathic pain and the possible role of Gria1 intime-dependent injury and pain response. The GO analysisfurther suggested that early and late microglia could havefunctionally distinct phenotypes following peripheral nerve injuryalthough they are difficult to distinguish by morphology alone.

4.1. Validity of transcriptome analysis

Despite being obtained from a very small number of cells, thegene profile of previously known genes from this transcriptomedata was largely consistent with general expectations fromprevious studies. Our 10-cell microglia data contained manyspinal microglia-specific genes that were recently reported byChiu et al.9 We also minimized the possibility of contamination ofinfiltrating macrophage, given that brain microglia signaturegenes but not the peripheral monocyte signature genes wereenriched in our microarray data.5 Indeed, recent studies havesuggested that infiltratingmacrophagesmay not contribute to theneuropathic pain after peripheral nerve injury.20

4.2. Distinctive transcriptome profiles of early-phase andlate-phase microglia

We found early microglia genes that are actively responding toperipheral nerve injury and leading to upregulation of receptorsand channels at POD1 (and in some case at POD7 as well).Notably, in addition to TLR4 previously shown to be upregulatedafter nerve injury,3 we also identified other toll-like receptors suchas TLR7 and TLR13, which could be also important for the

Figure 5. Spinal administration of miR-29c knockdown microglia in normalmice produces mechanical allodynia by means of microglia activation (A) Foldchanges of microRNAs, miR-137, miR-29b-1, and miR-29c, which weresignificantly changed exclusively at POD1 in spinal microglia after nerve injury.(B) Relative expression of miR-29c in ipsilateral spinal cord 1 day after L4 SNT.(C) Validation ofmiR-29c expression using primary culturedmicroglia activatedby ATP application (50 mM, for 1 hour) by real-time qPCR. Data are normalizedagainst RNU6B. (D) Changes in mRNA expression of cytokines, TNFa and Il-1b, in miR-29c inhibitor-transfected microglia using real-time qPCR. (E)Expression of pp38 revealed by western blot analysis in each group. Therelative amount of pp38 protein was normalized to expression level of p38 ininhibitor control and miR-29c inhibitor-transfected microglia. (F) Increasedintracellular calcium in primary DRG neurons by miR-29c inhibitor-transfectedmicroglia. Top, the average (mean6SEM) traces of [Ca21]i changes in all DRGneurons with microglia treatment from one particular imaging experiment.Bottom, summary of Ca21 transient responses in DRG neurons to themicroglia application in an unbiased manner. P , 0.0001. (G) Long-termpotentiation of fPSPs induced by miR-29c inhibitor-transfected microgliaapplication. Insets show representative traces fromone experiment at baseline(1) and after 120minutes (2). Two-way repeatedmeasures ANOVA values. P50.0015. (H) Paw withdrawal threshold after intrathecal injection of miR-29cinhibitor-transfected microglia into naı̈ve mice. Microglia transfected withinhibitor control was used as the control. n 5 6 for inhibitor control, n 5 7 formiR-29c inhibitor. *P, 0.05, ***P, 0.001. (I) Paw withdrawal threshold afterintrathecal injection of miR-29c mimic-transfected microglia after ATP

stimulation. ATP-treatedmicroglia was used as positive control. n5 6 for ATP-treated microglia, n5 6 for control miRNA, and ATP microglia, n5 8 for miR-29cmimic, and ATPmciroglia. *P, 0.05 compared to control miRNA and ATPmicroglia at 5 hour. All data in this figure are shown as mean 6 SEM, unlessotherwise stated.

April 2016·Volume 157·Number 4 www.painjournalonline.com 973

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

microglia activation and neuropathic pain. Interestingly, a recentstudy showed that extracellular miRNAs can activate nociceptorneurons to elicit pain through TLR7.45 It is also possible thatextracellular miRNAs released from central primary afferents mayactivate microglia through TLR7. It has been reported that let-7binduces TNFa release from brain microglia through TLR7.38

Furthermore, our transcriptome profiles revealed the distinctivedifference in their functional phenotypes between the early-phaseand late-phase microglia although they are difficult to bedistinguished by morphology alone (Fig. 1B).

4.3. Role of microglial miR-29c in initiation/development ofneuropathic pain

We observed that downregulation of miR-29c might be involvedin initial pain response. From in vitro studies with cytokine profilingand pp38 analysis, we verified that miR-29c inhibition recapit-ulates the profile of activated microglia and TNFa could bedownstream in the pathway targeted by miR-29c. Behavioral

studies revealed that downregulation of miR-29c in early spinalmicroglia after nerve injury modulates pain response, whichmay recapitulate initiation/development of neuropathic pain.Growing evidence indicates that miRNAs have a substantialrole in modulating complicated pain pathways.4,40,44,45,56

Furthermore, a recent study also demonstrated that the levelof miR-29c is decreased in the serum of complex regional painsyndrome patients,41 suggesting potential role of miR-29c inchronic pain. Given that Gria1 might not be a downstreamtarget of miR-29c (Fig. S4D), further work is required toidentify target genes regulated by miR-29c and how thosegenes have a functional significance in the neuropathic painconditions.

4.4. Individual heterogeneity of spinal microgliain transcriptome

Like individual cells found in other tissues,37 spinal microgliadisplayed a substantial estimated individual variation in theirtranscriptome. Our measurement of variation is not as sensitiveas single-cell measurements because of the pooling of the 10cells. Under random sampling, the variance of the mean of thesamples is expected to be proportional to the variance of theindividual cells. Nevertheless, we note that pooled cellmeasurements are indirect estimates of individual variation.We hypothesized that individual microglia in their normal statemight exhibit heterogeneous gene expression to dynamicallyreflect microenvironmental changes. However, external stimuli,such as skin lesions, may lead to transient coordinatedexpression of a subset of genes, which may later relax backinto natural cell-to-cell heterogeneity. In the late phase of injury,under the continuous stimuli from injured nerves, a certainspecific set of genes may remain coordinately expressed,leading to reduced variability in that gene set which arefunctionally associated with the pathophysiological neuro-pathic pain outcome. Thus, examination of single-cell levelvariability may be a powerful tool in identifying novel key genesunder specific stimulus.

4.5. Role of microglial Gria1 in maintenance ofneuropathic pain

Since some previous studies have mostly focused on theupregulated genes after nerve injury,21,51 most of the 56 genesidentified with variation analysis in this study were rarely detectedfrom such whole tissue expression profiling. From these genes,we focused on Gria1 as a strong candidate gene for functionalrole and knockdown of microglial Gria1 produced mechanicalallodynia in naive mice, which was reversed by TNFa inhibitor(Fig. 6F). Previous studies demonstrated that expression of Gria1is significantly decreased in activated brain microglia andunexpectedly, potentiating microglial AMPA receptors by AMPAreceptor agonist inhibits TNFa release, whereas activation ofAMPA receptor by glutamate promotes release of TNF by ratmicroglia.22 In this study, we found that the downregulation ofmicroglial GluR1 may lead to increased TNFa release46 contrib-uting to central sensitization in the spinal cord that might lead tothe maintenance of neuropathic pain. It was found that both miR-29c and Gria1 were also highly expressed in neurons in the spinalcord (Fig. S3B; http://links.lww.com/PAIN/A207) such that thestrong signals from the neurons made it difficult to determinetheir levels in microglia. However, we believe this does notaffect the hypothesis that miR-29c and Gria 1 are involved inmicroglial regulation as the functional role of the genes may be

Figure 6. Spinal administration of Gria1 knockdown microglia in normal miceproduces mechanical allodynia reversed by TNFa inhibitor (A) Validation ofGria1 expression using primary culturedmicroglia activated by ATP application(50 mM for 1 hour) by real-time qPCR. (B) Immunofluorescence of GluR1 (red)in primary cultured microglia following ATP (50 mM for 1 hour) application.Scale bar: 20 mm. (C) Increased TNFa mRNA in siGria1 microglia comparedwith that of siScram revealed by real-time qPCR. (D) Increased TNFa release insiGria1 microglia was measured by ELISA. (E) Paw withdrawal threshold afterintrathecal injection of siGria1 microglia into naive mice. siScram was used asnegative control. n 5 6 for siScram, n 5 7 for siGria1. ***P , 0.001. (F)Etanercept reversed the paw withdrawal threshold which was produced by anintrathecal injection of siGria1 microglia. The n 5 7 for siGria1, n 5 6 forsiScram and etanercept, n 5 10 for siGria1 and etanercept. **P , 0.01compared to siGria1 group. siGria1, siGria1 transfected microglia; siScram,Scrambled siRNA transfected microglia (see Figs. S3C, D and S4C for efficacyof miR-29c inhibitor, miRNA mimic, and Gria1 siRNA). All data in this figure areshown as mean 6 SEM.

974 H. Jeong et al.·157 (2016) 964–976 PAIN®

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

cell-type–specific. Indeed, it is notable that we clearly observeddownregulation of miR-29c in the spinal dorsal horn after nerveinjury (Fig. S3A) and in the original transcriptome data.

In summary, this study demonstrates that transcriptomeanalysis with individual collections of spinal microglia can bea relevant tool to reveal a microglial gene-expression signaturefollowing peripheral nerve injury. This high-resolution analysisreveals significant differences in the gene profile between early(POD1) and late (POD7) microglia and it is likely that microgliaexhibit functionally distinctive phenotypic changes over timefollowing nerve injury. Of note, downregulation of miR-29c in earlyspinal microglia following peripheral nerve injury might beimportant for the initiation or development of neuropathic painand Gria1 for the maintenance of neuropathic pain. Our workprovides insight into spinal microglial biology and potential newmicroglial targets for the treatment of chronic pain such asneuropathic pain.

Conflict of interest statement

The authors have no conflicts of interest to declare.This workwas supported by the National Research Foundation

(NRF) grant (2012R1A3A2048834) funded by the Koreangovernment (MEST). This work was also funded in part by NIHU01MH98953 grant to J. Kim. The funding agencies had noassociation with the results of this study.

Acknowledgements

The authors thank Dr Ji-Young Moon for her technical support tointrathecal injection.Author contributions: H. Jeong and S. B. Oh conceived the ideaand designed the experiments. H. Jeong performed the single-cell microarray experiments, transfection, real-time PCR, westernblotting, surgery and behavioral experiments. Y. H. Kimperformed slice preparation, single-cell collection, calciumimaging, and electrophysiological recordings. K. Lee carried outmiRNA validation. M. Kang and Y. I. Yeom performed bioinfor-matics analyses; Y. -J. Na and J. Kim analyzed the single-celldata, Y. Lee did primarymicroglia cultures andGria1 staining, andY. -J. Gao and B. -C. Jiang did miR-29c staining. H. Jeong, L. -J.Wu, J. Kim, and S. B. Oh wrote the paper. All authors discussedthe results and conclusions, and reviewed the manuscript.

Appendix A. Supplemental Digital Content

Supplemental Digital Content associated with this article can befound online at http://links.lww.com/PAIN/A202, http://links.lww.com/PAIN/A203, http://links.lww.com/PAIN/A204, http://links.lww.com/PAIN/A205, http://links.lww.com/PAIN/A206, http://links.lww.com/PAIN/A207, and http://links.lww.com/PAIN/A208.

Article history:Received 18 August 2015Received in revised form 6 November 2015Accepted 14 December 2015Available online 5 January 2016

References

[1] Beggs S, Trang T, Salter MW. P2X4R1microglia drive neuropathic pain.Nat Neurosci 2012;15:1068–73.

[2] Berta T, Park CK, Xu ZZ, Xie RG, Liu T, Lu N, Liu YC, Ji RR. Extracellularcaspase-6 drives murine inflammatory pain via microglial TNF-alphasecretion. J Clin Invest 2014;124:1173–86.

[3] Bettoni I, Comelli F, Rossini C, Granucci F, Giagnoni G, Peri F, Costa B.Glial TLR4 receptor as new target to treat neuropathic pain: efficacy ofa new receptor antagonist in a model of peripheral nerve injury in mice.Glia 2008;56:1312–19.

[4] Bhalala OG, Srikanth M, Kessler JA. The emerging roles of microRNAs inCNS injuries. Nat Rev Neurol 2013;9:328–39.

[5] Butovsky O, Jedrychowski MP, Moore CS, Cialic R, Lanser AJ, GabrielyG, Koeglsperger T, Dake B, Wu PM, Doykan CE, Fanek Z, Liu L, Chen Z,Rothstein JD, Ransohoff RM, Gygi SP, Antel JP, Weiner HL. Identificationof a unique TGF-beta-dependent molecular and functional signature inmicroglia. Nat Neurosci 2014;17:131–43.

[6] Calvo M, Bennett DL. The mechanisms of microgliosis and pain followingperipheral nerve injury. Exp Neurol 2012;234:271–82.

[7] Campbell JN, Meyer RA. Mechanisms of neuropathic pain. Neuron 2006;52:77–92.

[8] Chaplan SR, Bach FW, Pogrel JW, Chung JM, Yaksh TL. Quantitativeassessment of tactile allodynia in the rat paw. J Neurosci Methods 1994;53:55–63.

[9] Chiu IM, Morimoto ET, Goodarzi H, Liao JT, O’Keeffe S, Phatnani HP,Muratet M, Carroll MC, Levy S, Tavazoie S, Myers RM, Maniatis T. Aneurodegeneration-specific gene-expression signature of acutelyisolated microglia from an amyotrophic lateral sclerosis mouse model.Cell Reports 2013;4:385–401.

[10] Clark AK, Yip PK, Grist J, Gentry C, Staniland AA,Marchand F, Dehvari M,Wotherspoon G, Winter J, Ullah J, Bevan S, Malcangio M. Inhibition ofspinal microglial cathepsin S for the reversal of neuropathic pain. ProcNatl Acad Sci U S A 2007;104:10655–60.

[11] Costigan M, Belfer I, Griffin RS, Dai F, Barrett LB, Coppola G, Wu T,Kiselycznyk C, Poddar M, Lu Y, Diatchenko L, Smith S, Cobos EJ,Zaykin D, Allchorne A, Gershon E, Livneh J, Shen PH, Nikolajsen L,Karppinen J, Mannikko M, Kelempisioti A, Goldman D, Maixner W,Geschwind DH, Max MB, Seltzer Z, Woolf CJ. Multiple chronic painstates are associated with a common amino acid-changing allele inKCNS1. Brain 2010;133:2519–27.

[12] Costigan M, Scholz J, Woolf CJ. Neuropathic pain: a maladaptiveresponse of the nervous system to damage. Annu Rev Neurosci 2009;32:1–32.

[13] DeLeo JA, Rutkowski MD, Stalder AK, Campbell IL. Transgenicexpression of TNF by astrocytes increases mechanical allodynia ina mouse neuropathy model. Neuroreport 2000;11:599–602.

[14] Dixon WJ. Efficient analysis of experimental observations. Annu RevPharmacol Toxicol 1980;20:441–62.

[15] Dworkin RH, Backonja M, Rowbotham MC, Allen RR, Argoff CR, BennettGJ, Bushnell MC, Farrar JT, Galer BS, Haythornthwaite JA, Hewitt DJ,Loeser JD,MaxMB, SaltarelliM,Schmader KE, Stein C, ThompsonD, TurkDC, Wallace MS, Watkins LR, Weinstein SM. Advances in neuropathicpain: diagnosis, mechanisms, and treatment recommendations. ArchNeurol 2003;60:1524–34.

[16] Eberwine J, Sul JY, Bartfai T, Kim J. The promise of single-cellsequencing. Nat Methods 2014;11:25–7.

[17] Esumi S, Wu SX, Yanagawa Y, Obata K, Sugimoto Y, Tamamaki N.Method for single-cell microarray analysis and application to gene-expression profiling of GABAergic neuron progenitors. Neurosci Res2008;60:439–51.

[18] Gao YJ, Ji RR. Chemokines, neuronal-glial interactions, and centralprocessing of neuropathic pain. Pharmacol Ther 2010;126:56–68.

[19] Giordan M. A two-stage procedure for the removal of batch effects inmicroarray studies. Stat Biosci 2013;6:73–84.

[20] Grace PM, Hutchinson MR, Maier SF, Watkins LR. Pathological pain andthe neuroimmune interface. Nat Rev Immunol 2014;14:217–31.

[21] Griffin RS, CostiganM, Brenner GJ, MaCHE, Scholz J, Moss A, AllchorneAJ, Stahl GL, Woolf CJ. Complement induction in spinal cord microgliaresults in anaphylatoxin C5a-mediated pain hypersensitivity. J Neurosci2007;27:8699–708.

[22] Hagino Y, Kariura Y, Manago Y, Amano T, Wang B, Sekiguchi M,Nishikawa K, Aoki S, Wada K, NodaM. Heterogeneity and potentiation ofAMPA type of glutamate receptors in rat cultured microglia. Glia 2004;47:68–77.

[23] Hanisch UK, Kettenmann H. Microglia: active sensor and versatileeffector cells in the normal and pathological brain. Nat Neurosci 2007;10:1387–94.

[24] Hathway GJ, Vega-Avelaira D, Moss A, Ingram R, Fitzgerald M. Brief, lowfrequency stimulation of rat peripheral C-fibres evokes prolongedmicroglial-induced central sensitization in adults but not in neonates.PAIN 2009;144:110–18.

[25] Haynes SE, Hollopeter G, Yang G, Kurpius D, Dailey ME, Gan WB, JuliusD. The P2Y12 receptor regulates microglial activation by extracellularnucleotides. Nat Neurosci 2006;9:1512–19.

April 2016·Volume 157·Number 4 www.painjournalonline.com 975

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.

[26] Hickman SE, Kingery ND, Ohsumi TK, Borowsky ML, Wang LC, MeansTK, El Khoury J. The microglial sensome revealed by direct RNAsequencing. Nat Neurosci 2013;16:1896–905.

[27] Hylden JL,Wilcox GL. Intrathecal morphine inmice: a new technique. EurJ Pharmacol 1980;67:313–16.

[28] Irizarry RA, Bolstad BM, Collin F, Cope LM, Hobbs B, Speed TP.Summaries of Affymetrix GeneChip probe level data. Nucleic Acids Res2003;31:e15.

[29] Ji RR, Suter MR. p38 MAPK, microglial signaling, and neuropathic pain.Mol Pain 2007;3:33.

[30] Jung S, Aliberti J, Graemmel P, Sunshine MJ, Kreutzberg GW, Sher A,Littman DR. Analysis of fractalkine receptor CX(3)CR1 function bytargeted deletion and green fluorescent protein reporter gene insertion.Mol Cell Biol 2000;20:4106–14.

[31] Kettenmann H, Hanisch UK, Noda M, Verkhratsky A. Physiology ofmicroglia. Physiol Rev 2011;91:461–553.

[32] Kim SH, Chung JM. An experimental model for peripheral neuropathyproduced by segmental spinal nerve ligation in the rat. PAIN 1992;50:355–63.

[33] Kim YH, Back SK, Davies AJ, Jeong H, Jo HJ, Chung G, Na HS, Bae YC,Kim SJ, Kim JS, Jung SJ, Oh SB. TRPV1 in GABAergic interneuronsmediates neuropathic mechanical allodynia and disinhibition of thenociceptive circuitry in the spinal cord. Neuron 2012;74:640–7.

[34] Kobayashi K, Yamanaka H, Fukuoka T, Dai Y, Obata K, Noguchi K.P2Y12 receptor upregulation in activated microglia is a gateway of p38signaling and neuropathic pain. J Neurosci 2008;28:2892–902.

[35] Kolodziejczyk AA, Kim JK, Svensson V, Marioni JC, Teichmann SA. Thetechnology and biology of single-cell RNA sequencing. Mol Cell 2015;58:610–20.

[36] Lacroix-Fralish ML, Tawfik VL, Tanga FY, Spratt KF, DeLeo JA.Differential spinal cord gene expression in rodent models of radicularand neuropathic pain. Anesthesiology 2006;104:1283–92.

[37] Lawson LJ, Perry VH, Dri P, Gordon S. Heterogeneity in the distributionand morphology of microglia in the normal adult mouse brain.Neuroscience 1990;39:151–70.

[38] Lehmann SM, Kruger C, Park B, Derkow K, Rosenberger K, Baumgart J,Trimbuch T, EomG, HinzM, Kaul D, Habbel P, Kalin R, Franzoni E, RybakA, Nguyen D, Veh R, Ninnemann O, Peters O, Nitsch R, Heppner FL,Golenbock D, Schott E, Ploegh HL, Wulczyn FG, Lehnardt S. Anunconventional role for miRNA: let-7 activates Toll-like receptor 7 andcauses neurodegeneration. Nat Neurosci 2012;15:827–35.

[39] Livak KJ, Schmittgen TD. Analysis of relative gene expression data usingreal-time quantitative PCR and the 2(-Delta Delta C(T)) Method. Methods2001;25:402–8.

[40] Lutz BM, Bekker A, Tao YX. Noncoding RNAs: new players in chronicpain. Anesthesiology 2014;121:409–17.

[41] McDonald MK, Tian Y, Qureshi RA, Gormley M, Ertel A, Gao R, AradillasLopez E, Alexander GM, Sacan A, Fortina P, Ajit SK. Functionalsignificance of macrophage-derived exosomes in inflammation andpain. PAIN 2014;155:1527–39.

[42] Mika J, Zychowska M, Popiolek-Barczyk K, Rojewska E, Przewlocka B.Importance of glial activation in neuropathic pain. Eur J Pharmacol 2013;716:106–19.

[43] Nikitin A, Egorov S, Daraselia N, Mazo I. Pathway studio—the analysisand navigation of molecular networks. Bioinformatics 2003;19:2155–7.

[44] Pan Z, Zhu LJ, Li YQ,Hao LY, Yin C, Yang JX, GuoY, ZhangS, Hua L, XueZY, Zhang H, Cao JL. Epigenetic modification of spinal miR-219expression regulates chronic inflammation pain by targetingCaMKIIgamma. J Neurosci 2014;34:9476–83.

[45] Park CK, Xu ZZ, Berta T, Han Q, Chen G, Liu XJ, Ji RR. ExtracellularmicroRNAs activate nociceptor neurons to elicit pain via TLR7 andTRPA1. Neuron 2014;82:47–54.

[46] Pocock JM, Kettenmann H. Neurotransmitter receptors on microglia.Trends Neurosci 2007;30:527–35.

[47] Ren K, Dubner R. Interactions between the immune and nervous systemsin pain. Nat Med 2010;16:1267–76.

[48] Salter MW, Beggs S. Sublime Microglia: expanding roles for theguardians of the CNS. Cell 2014;158:15–24.

[49] Sandberg R. Entering the era of single-cell transcriptomics in biology andmedicine. Nat Methods 2014;11:22–4.

[50] Sayed D, Abdellatif M. MicroRNAs in development and disease. PhysiolRev 2011;91:827–87.

[51] Strong JA, XieW, Coyle DE, Zhang JM. Microarray analysis of rat sensoryganglia after local inflammation implicates novel cytokines in pain. PLoSOne 2012;7:e40779.

[52] Tanga FY, Nutile-McMenemy N, DeLeo JA. The CNS role of toll-likereceptor 4 in innate neuroimmunity and painful neuropathy. Proc NatlAcad Sci U S A 2005;102:5856–61.

[53] Tozaki-SaitohH, TsudaM,Miyata H, Ueda K, Kohsaka S, Inoue K. P2Y12receptors in spinal microglia are required for neuropathic pain afterperipheral nerve injury. J Neurosci 2008;28:4949–56.

[54] Tsuda M, Shigemoto-Mogami Y, Koizumi S, Mizokoshi A, Kohsaka S,Salter MW, Inoue K. P2X4 receptors induced in spinal microglia gatetactile allodynia after nerve injury. Nature 2003;424:778–83.

[55] Ulmann L, Hatcher JP, Hughes JP, Chaumont S, Green PJ, ConquetF, Buell GN, Reeve AJ, Chessell IP, Rassendren F. Up-regulation ofP2X4 receptors in spinal microglia after peripheral nerve injurymediates BDNF release and neuropathic pain. J Neurosci 2008;28:11263–8.

[56] von Schack D, Agostino MJ, Murray BS, Li Y, Reddy PS, Chen J, ChoeSE, Strassle BW, Li C, Bates B, Zhang L, Hu H, Kotnis S, Bingham B, LiuW, Whiteside GT, Samad TA, Kennedy JD, Ajit SK. Dynamic changes inthemicroRNA expression profile revealmultiple regulatorymechanisms inthe spinal nerve ligation model of neuropathic pain. PLoS One 2011;6:e17670.

[57] Woolf CJ. Central sensitization: implications for the diagnosis andtreatment of pain. PAIN 2011;152(3 suppl):S2–15.

[58] Yang L, Zhang FX, Huang F, Lu YJ, Li GD, Bao L, Xiao HS, Zhang X.Peripheral nerve injury induces trans-synaptic modification of channels,receptors and signal pathways in rat dorsal spinal cord. Eur J Neurosci2004;19:871–83.

[59] Zhuang ZY, Kawasaki Y, Tan PH, Wen YR, Huang J, Ji RR. Role of theCX3CR1/p38 MAPK pathway in spinal microglia for the development ofneuropathic pain following nerve injury-induced cleavage of fractalkine.Brain Behav Immun 2007;21:642–51.

976 H. Jeong et al.·157 (2016) 964–976 PAIN®

Copyright � 2016 by the International Association for the Study of Pain. Unauthorized reproduction of this article is prohibited.