w w w . e l s e v i e r . c o m / l o c a t e / p a i n

PAIN�

154 (2013) 2782–2793

Multimodal assessment of nervous and immune system responsesfollowing sciatic nerve injury

0304-3959/$36.00 � 2013 International Association for the Study of Pain. Published by Elsevier B.V. All rights reserved.http://dx.doi.org/10.1016/j.pain.2013.08.016

⇑ Corresponding author. Tel.: +1 617 869 8193; fax: +1 847 938 5286.E-mail address: jaymin.upadhyay@abbvie.com (J. Upadhyay).

Loren Lasko a, Xin Huang b, Martin J. Voorbach a, La Geisha R. Lewis c, Jason Stavropoulos d,Julie Carriker d, Terese R. Seifert a, Scott J. Baker a, Jaymin Upadhyay a,⇑a Integrated Sciences and Technology, AbbVie Inc., North Chicago, IL, USAb Exploratory Statistics, AbbVie Inc., North Chicago, IL, USAc Neuroscience Discovery, AbbVie Inc., North Chicago, IL, USAd Comparative Medicine, AbbVie Inc., North Chicago, IL, USA

Sponsorships or competing interests that may be relevant to content are disclosed at the end of this article.

a r t i c l e i n f o a b s t r a c t

Article history:Received 2 June 2013Received in revised form 13 August 2013Accepted 15 August 2013

Keywords:Sciatic nerveMRIFDG-PETIL-1bFractional anisotropy

Subsequent to peripheral nerve compression and irritation, pathophysiological processes take placewithin nervous and immune systems. Here, we utilized a multimodal approach to comprehend periph-eral and central soft tissue changes as well as alterations occurring in systemic analytes following uni-lateral chronic constriction injury (CCI) of the sciatic nerve in rodents. Using magnetic resonanceimaging and [18F]-2-fluoro-2-deoxy-D-glucose (FDG) positron emission tomography, we demonstratedrobust structural abnormalities and enhanced FDG uptake within the injured nerve and surroundingmuscle, respectively. To assess whether central morphological changes were induced by nerve injury,diffusion tenor imaging was performed. A decrease in fractional anisotropy in primary motor cortexcontralateral to the injury site was observed. Evaluation of a panel of circulating cytokines, chemokines,and growth factors showed decreased levels of interleukin-1b and Fractalkine in CCI animals. Areaunder the receiver operating curve (ROC) calculations of analyte levels, imaging, and behavioral endpoints ranged from 0.786 to 1, where behavioral and peripheral imaging end points (eg, FDG uptakein muscle) were observed to have the highest discriminatory capabilities (maximum area underROC = 1) between nerve injury and sham conditions. Lastly, performance of correlation analysis involv-ing all analyte, behavioral, and imaging data provided an understanding of the overall associationamongst these end points, and importantly, a distinction in correlation patterns was observed betweenCCI and sham conditions. These findings demonstrate the multidimensional pathophysiology of sciaticnerve injury and how a combined analyte, behavioral, and imaging assessment can be implemented toprobe this complexity.

� 2013 International Association for the Study of Pain. Published by Elsevier B.V. All rights reserved.

1. Introduction

In the presence of sciatic nerve compression or irritation, symp-toms of pain, numbness, weakness, and loss of normal motor con-trol can be experienced. Additionally, aberrant inflammatory andimmune responses at systemic or local levels can coexist withnerve trauma, and also contribute to the experienced symptoms.The existence of pain and proper sensorimotor function along withthe above-mentioned pathologies can be indicative of sciatica, anervous system condition influenced by occupational, routinephysical activity and genetic factors [29]. In some patients, sciaticaand the associated symptoms are resolved over time following a

conservative treatment regimen [35]. However, in cases wherethe disease state and symptoms, particularly pain, become chronic,resolving the condition over a long-term period can at times be dif-ficult, despite implementation of invasive procedures such as epi-dural steroid injections or surgery [19].

Today, magnetic resonance imaging (MRI) of the periphery iscommonly used to assist in diagnosing chronic sciatica [37], as wellas in preclinical settings where models of sciatica are evaluated[3,4,15,38]. From clinical MRI investigation specifically, previouswork suggests discordance between MRI observations and symp-toms experienced (or lack thereof) by patients. Jensen et al. andel Barzouhi et al. have reported that anatomical abnormalities(eg, scar tissue, protrusions, and extrusions) observed in repeatlumbar spine MRI were not predictive of chronic symptomatic vsasymptomatic sciatica in patients, nor was it indicative of

L. Lasko et al. / PAIN�

154 (2013) 2782–2793 2783

treatment outcome [14,21]. Thus, a discrepancy exists betweenMRI-based detection of peripheral pathology, such as qualitativeproof of neuropathy and symptoms of sciatica (eg, chronic pain).

The objective of the current study was to implement a multi-modal approach in order to better comprehend the disease pro-cesses and symptoms of sciatic nerve injury. In the rodentchronic constriction nerve injury (CCI) model, pain was assessedbehaviorally, after which, blood samples were collected in orderto quantify a panel of 26 analytes associated with pain and inflam-mation. To measure structural and regional metabolic demandalong the distribution of the sciatic nerve, T2-weighted MRI and[18F]-2-fluoro-2-deoxy-D-glucose positron emission tomography(FDG-PET)/computed tomography (CT), respectively, were used.Lastly, diffusion tensor imaging (DTI) of the brain was performedto determine if structural plasticity was induced by nerve injury.

Nervous and immune system functionality deviate from theirrespective baseline states in response to peripheral nerve injury[34]. Thus, we tested the hypothesis that structural and metabolicproperties quantified along the distribution of an injured sciaticnerve as well as within the central nervous system parallel theactivity of immune system regulators relevant to pain and inflam-mation. Structural properties of the injured nerve, structural andmetabolic properties within leg muscle, systemic levels of interleu-kin (IL)-1b, mechanical allodynia, and fractional anisotropy (FA)within the primary motor cortex were observed to best differenti-ate CCI rats from sham controls. Correlation analysis performedamongst analyte, behavioral, and imaging data demonstrated asso-ciations and clustering amongst these distinct end points. By calcu-lating the area under the receiver operating curve (ROC) andperforming least absolute shrinkage and selection operator regres-sion analysis, between-group classifier assessment and predictivemodeling of end points were respectively enabled.

2. Method

2.1. Animals

All studies were conducted in accordance with InstitutionalAnimal Care and Use Committee guidelines and the National Insti-tutes of Health Guide for Care and Use of Laboratory Animals. Abb-Vie facilities are accredited by the Association for the Assessmentand Accreditation of Laboratory Animal Care. Adult male Spra-gue-Dawley rats (Charles River, Portage, MI, USA) were used(n = 16; 250-325 g at start of imaging). The current study con-tained one sham cohort (n = 8) and one CCI model cohort (n = 8).Data from one sham animal were not collected in this study.

2.2. CCI Surgery

Initially, rats were anesthetized with isoflurane (4% to induceand 2%-3% to maintain). The incision site was sterilized usingethanol and 10% povidone-iodine solution prior to and after sur-geries. CCI of the sciatic nerve in rats was produced by followingthe method of Bennett and Xie [6]. The right common sciaticnerve was isolated at mid-thigh level, and loosely ligated by 4chromic gut (4-0) ties separated by an interval of �1 mm; forsham rats, the right common sciatic nerve was isolated but notligated. All animals were allowed to rest and then placed in acage with soft bedding for 2 weeks prior to further experimentalprocedures.

2.3. Behavioral testing

Sensitivity to mechanical stimulation was blindly tested 2 and3 weeks post surgery with tactile allodynia using calibrated von

Frey filaments (Stoelting, Wood Dale, IL, USA), as previously de-scribed [10]. The maximum force applied was 15 g, a force thatnormally does not evoke a response in a naive rat. Rats with scores65 g were considered allodynic, confirming successful modelcreation.

2.4. Blood collection

Blood was collected (0.5 mL) 2 weeks following surgery andprior to FDG injection. Following room temperature exposure for�45 minutes, each sample was spun at 4�C and 1000 g for 10 min-utes. The serum layer was siphoned off and frozen at �80�C. Serumsamples were run through a panel of 26 analytes (AssayGate,Ijamsville, MD, USA).

2.5. Peripheral MRI data acquisition

Animals were fasted for a minimum of 6 hours prior to imagingdata acquisition. Each subject was anesthetized with 2%-3% isoflu-rane. A 4.7T Bruker Pharmascan (Karlsruhe, Germany) was used fordata acquisition. Each animal was secured headfirst to the MRI bedwith hind paws slightly extended and away from the body. A heat-ing pad was placed beneath the animal to maintain 37�C body tem-perature and a pressure respiration pad (SA Instruments, StonyBrook, NY, USA) was used to monitor animal health for the entireduration of the scan. A 16-cm (inner diameter) volume coil (Bru-ker) was used to image the hip and lower limbs, covering the dis-tribution of the sciatic nerve. The incision site was used as an initiallocalization point for positioning within the coil. Final positioningwas adjusted using a 3-dimensional reference scan. A fat-sup-pressed, T2-weighted rapid acquisition with relaxation enhance-ment (RARE) scan was used for structural imaging of theperiphery. T2-weighted RARE scan parameters: RARE factor = 3, #of dummy scans = 2, # of averages = 8, flip angle = 180�, # ofslices = 20, effective echo time (TE) = 28 ms, temporal resolution(TR) = 4000 ms, spatial resolution = 0.205 mm � 0.205 mm �1.5 mm.

2.6. Peripheral PET/CT data acquisition

Approximately 500 micro-Curie (lCi) of FDG (purchased fromIBA Molecular, Dulles, VA, USA) in 100-200 lL were injected viatail vein 50-80 minutes before PET/CT imaging and just prior toMRI data acquisition. Once MRI data acquisition was completed,animals were kept secured on the MRI bed and transferred to aPET/CT station. Animals were imaged on a docked PET/CT station(Siemens Inveon microPET/CT, Knoxville, TN, USA). Positioning ofeach animal was carried out such that the field of view encom-passed the hips and lower limbs. CT images were acquired at80 kV and 500 uA, with an exposure time of 210 ms, and 200steps. CT images were reconstructed using filtered back projec-tion with a Shepp-Logan filter. PET was a 7-minute scan andreconstructed using an ordered subsets expectations maximiza-tion 2D algorithm.

2.7. Peripheral MRI and PET/CT data analyses

All peripheral MRI and FDG-PET data analysis was performedusing VivoQuant 1.20 (inviCRO, LLC, Boston, MA, USA). Initially,each subject-specific PET/CT dataset was co-registered to therespective MRI dataset using a semi-automated, rigid body regis-tration process. Subsequent to quality assurance/quality controlof co-registration output, regions of interests (ROI) were maskedand positions using the MRI datasets given spatial resolution aswell as soft and hard tissue specificity (Fig. 1A).

Fig. 1. Definition of regions of interest (ROI) utilized in peripheral magnetic resonance imaging (MRI) and [18F]-2-fluoro-2-deoxy-D-glucose positron emission tomography(FDG-PET) data analysis. (A) The 3 peripheral ROIs (surgery site, sciatic nerve and muscle) evaluated in MRI and FDG-PET data analyses are defined on a T2-weighted rapidacquisition with relaxation enhancement (RARE) coronal slice from a representative chronic constriction injury (CCI) subject. All ROIs were defined and manually masked onthe MRI datasets and transformed to the PET-CT (computed tomography) space as required. (B) The sciatic nerve is shown on the contralateral side (relative to surgery site). Abroad area along the sciatic nerve was manually defined along its length and on multiple slices (�3-4 slices). On the ipsilateral side, the sciatic nerve ROI was restricted tobelow the ligation site. (C) In CCI animals, the muscle, hyperintensity region (blue ROI) identified on each animal and on the ipsilateral side was masked. For all other muscleROIs (CCI contralateral, sham ipsilateral, and sham contralateral), a fixed-volume, cylindrical ROI (light-blue ROI) was positioned in a muscle region comparable to wheremuscle hyperintensity was observed and masked in CCI animals. The reference region ROI utilized in all analysis for normalization purposes is depicted in green. (D) In CCIanimals, the surgery site (green ROI) was identified in �3-4 coronal slices. Fixed-volume ROIs were used in all other cases to define the ‘‘surgery site’’ (CCI contralateral, shamipsilateral, and sham contralateral). (E) The signal intensity profile across the sciatic nerve is shown. Such a signal intensity profile along the sciatic nerve and across multipleslices was obtained to calculate the signal intensity (maximum and mean), area under the curve (AUC) and full width half maximum (FWHM) metrics. (F) Using CT data, thespine ROI was identified. Here it can be observed that the MRI field of view was much smaller than that of the CT scan.

2784 L. Lasko et al. / PAIN�

154 (2013) 2782–2793

2.7.1. Sciatic nerve ROIIn 2-dimensional coronal MRI slices, line segments of 3.9-

4.1 mm in length were positioned perpendicular to the main orien-tation of the nerve, centered on the nerve at regular intervals andalong the entire length of the nerve. This procedure was carried outin 3-4 slices where the sciatic nerve was present. For CCI animals,the portion of the sciatic nerve containing sutures was not in-cluded in the sciatic nerve ROI. A script was then used to join eachline segment and later interpolated to create a 3-dimensionalnerve ROI, which was subsequently used for all MRI and PET dataanalyses (Fig. 1B).

2.7.2. Muscle ROIIn 2-dimensional coronal MRI slices for each CCI animal, the re-

gion of hyperintensity within muscle was manually segmented(blue ROI; Fig. 1C). For the contralateral (opposite anatomical loca-tion where CCI and sham surgeries were performed) muscle area inCCI animals as well as the ipsilateral and contralateral muscle re-gion in sham animals, fixed cylindrical volume muscle ROIs werepositioned across slices. In all CCI and sham animals, reference tis-sue ROIs (depicted in green) used for normalization procedures inMRI and PET data quantification were placed in the contralateralmuscle region.

2.7.3. Surgery site ROIAn ROI (depicted in red) was also placed around the sutures in

CCI animals in 2-dimensional MRI slices where the sutures werevisible (Fig. 1D). This ROI was then dilated radially for 10 voxels(green ROI) to capture tissue in the surrounding surgery site in or-der to determine if a washout effect would be present. A referencesurgery site ROI of similar size as the dilated region was also gen-erated for the contralateral sciatic nerve. For comparison, mocksurgery site ROIs for sham animals were also produced.

In Fig. 1E, a plot of signal intensity across a line segment posi-tioned over the sciatic nerve is shown. MRI-based metrics usedto define the sciatic nerve signal are also defined (full width halfmaximum [FWHM]), minimum signal intensity, and maximum sig-nal intensity). The sciatic nerve metrics reported below were basedon the 3-dimensional ROI.

2.7.4. Spine ROISpine ROIs were segmented from the CT data (Fig. 1F). First,

bone was segmented using the connected thresholding tool andalgorithm within VivoQuant 1.2. Lumbar vertebrae were identifiedand designated as the ‘‘upper spine’’ ROI, relative to the sacral ver-tebrae (blue ROI), which were designated as the ‘‘lower spine’’ ROI(blue-green ROI). Both ROIs, upper and lower spine, were dilated

L. Lasko et al. / PAIN�

154 (2013) 2782–2793 2785

individually and iteratively in single-voxel-diameter increments toaccount for possible partial volume effects in the PET data. Giventhe MRI field of view, signal intensity within the spine was con-fined to the lower-spine ROI, while PET data analysis was per-formed in both the lower and upper spine ROIs. Moreover, the L4and L5 region of the spine was specifically evaluated.

2.8. DTI of brain

2.8.1. PerfusionDTI data were collected at week 3 post CCI and sham surgeries,

and was performed ex vivo. Prior to DTI data acquisition, whole-animal perfusion was performed with a perfusion pump. Theanimal was anesthetized with an intraperitoneal injection ofketamine/medetomidine mixture. The animals’ thorax was openedand a 20-gauge needle was inserted directly into the protrusion ofthe left ventricle and extended 5 mm. A slow, steady flow of20 mL/min of 0.9% saline solution was started. Then the atriumwas cut and solution flowed freely. When blood had been clearedfrom the body, the solution was changed to 500 mL of 4%paraformaldehyde.

2.8.2. DTI data acquisitionAfter fixation, all DTI data were collected on a 4.7T Bruker Phar-

mascan MRI magnet, with a 72-mm-diameter transmit volume coiland a quadrature surface receiver coil. Fat-suppressed DTI datawere collected using a single-shot spin echo sequence. DTI scanparameters: # of dummy scans = 4, # of segments = 4, # ofaverages = 2, # of slices = 15, TE = 23 ms, TR = 3800 ms, spatialresolution = 0.312 mm � 0.312 mm � 1.0 mm, # of diffusion direc-tions = 30, # of nondiffusion weight = 5, b-Value = 670 s/mm2,d = 5.0 ms, D = 9.5 ms.

2.8.3. DTI analysisAll DTI data analysis was performed using the FMRIB Software

Library software packages (http://www.fmrib.ox.ac.uk/fsl/). Man-ual brain extraction was performed for each DTI dataset. For theDTI dataset, the B0 image was used for brain extraction and co-reg-istration. Before calculation of the diffusion tensors and DTI met-rics (ie, FA), an eddy current correction was performed for eachDTI dataset. Subsequently, each FA dataset was co-registered toin-house MRI rat atlas using a 12-degrees-of-freedom affine trans-formation [20]. Once atlas-based brain ROIs were mapped to eachFA map, mean FA values were extracted separately from left andright brain regions known to play a role in pain and sensory motorprocessing (ie, thalamus, caudate-putamen, hypothalamus, cingu-late, primary motor cortex, primary somatosensory cortex, second-ary motor cortex, and secondary somatosensory cortex) [32].

2.9. Analyte data analysis

2.9.1. Analyte assayAnalyte levels in serum were measured using a multiplex Lum-

inex bead-based immunoassay assay with a Bio-Plex 200 BeadReader System (AssayGate). Microparticles were dyed with differ-ing concentrations of 2 fluorophores to generate distinct bead sets.Each bead set was coated to capture an antibody specific to eachanalyte. Captured analytes were detected using a biotinylateddetection antibody and streptavidin-phycoerythrin. The bead ana-lyzer was a dual-laser, flow-based sorting and detection platform.One laser was bead-specific and determines which analyte wasbeing detected. The other laser determines the magnitude of phy-coerythrin-derived signal, which was in direct proportion to theamount of analyte bound. The panel of 26 analytes assessed inthe study included: KC-GRO, monocyte chemoattractant protient(MCP)-1, Leptin, vascular endothelial growth factor (VEGF),

endothelial growth factor (EGF), macrophage inflammation protein(MIP)-1a, regulated on activation, normal t cell expressed andsecreted (RANTES), tumor necrosis factor (TNF)-a, interferon(IFN)-c, granulocyte-colony stimulating factor (G-CSF), Eotaxin,macrophage inflammation protein (MIP)-2, LIX, Fractalkine,interferon-c-inducible protein (IP)-10, Interleukin (IL)-1a, IL-1b,IL-2, IL.4, IL-5, IL-6, IL-10, IL-12 (p70), IL-13, IL-17 and IL-18.

2.9.2. Analyte concentration determinationProtein concentrations of the samples were determined by a 5-

parameter logistic regression algorithm with analysis of the med-ian fluorescence intensity readings of an 8-point protein standardcurve. Once a regression equation was derived, the fluorescenceintensity values of the standards were treated as unknowns andthe concentration of each standard was calculated to measure as-say recovery. Samples were tested in duplicate along with positiveand negative controls on each bead plate, which allowed for assayquality assurance.

2.10. Statistical analysis

2.10.1. Statistical analysis for imaging dataStatistical analyses were performed using t-tests or Wilcoxon

tests as appropriate, given the data. For paired analysis (ie, contra-lateral and ipsilateral comparisons within a single cohort), the dif-ference in the samples of interest was evaluated for normalityusing the Jarque-Bera test. If normality was passed, a paired t-testwas used. Otherwise, a Wilcoxon signed-rank test was used. Forindependent samples analysis (ie, CCI and sham comparisons fora single region), each vector of samples was tested for normalityusing the Jarque-Bera test. If both vectors of samples passed nor-mality, data were tested for equal variance using a v2 variance test.An independent-samples t-test was performed using the appropri-ate variance assumption. If normality failed for either cohort, theWilcoxon rank-sum test on the medians was employed. For mostparameters of interest, including all those shown explicitly in thisreport, normality was preserved and t-test values are reported. Thelatter also holds true for DTI as well as analyte (IL-1b) cross-groupcomparisons.

2.10.2. Classifier analysisTo assess discrimination capabilities (classifying CCI and sham

experimental cohorts) for each of the end points, the area underthe ROC was calculated, while confidence intervals for the area un-der the ROCs were estimated from DeLong’s method [16]. Further,we built penalized linear regression models using analyte endpoints as inputs for predicting end points with perfect ROCs (areaunder ROC = 1). Furthermore, for the current dataset we derivedthe relative importance of each analyte end point for predictingthe designated outcome. Specifically, a total of 1000 iterationsand cross-validations were performed, and the least absoluteshrinkage and selection operator penalization was used for vari-able selection in each iteration [31].

2.10.3. Multimodal correlation analysisPairwise correlations of analyte, behavioral, and imaging mea-

surements for all animals were calculated by the Spearman’s rankcorrelation coefficient. The correlation matrix was presented as acell plot in order to show correlations among variables. Variableswere ordered by increasing magnitudes of element-wise arctan-gent of the first 2 eigenvectors of the correlation matrix such thatvariables with comparable correlations were clustered together.The same correlation matrices were then calculated separatelyfor CCI or sham groups. The element-wise comparisons of correla-tions between the 2 groups were performed based on the Fisher’s Z

2786 L. Lasko et al. / PAIN�

154 (2013) 2782–2793

transformation of the correlation, while also controlled for multi-plicity [5].

3. Results

3.1. Mechanical allodynia

At weeks 2 and 3 and relative to the sham condition, CCIanimals showed significant (week 2: P < 3E-05 and week 3:P < 1E-06) mechanical allodynia based on paw withdrawal re-sponses to von Frey stimulation (Fig. 2). All behavioral painassessments were performed on the hind paw ipsilateral to thesurgery site.

3.2. Peripheral MRI of the sciatic nerve, muscle, and spine

Structural MRI in sham animals qualitatively revealed healthysciatic nerves on both ipsilateral and contralateral sides (Fig. 3A,top row). At week 2, pathology of the sciatic nerve was revealedin all CCI animals (Fig. 3A, bottom row). CCI animals showed signif-icantly increased normalized mean signal intensities in the ipsilat-eral sciatic nerve (Fig. 3B, vs contralateral CCI sciatic nerve:P < 0.0008, vs ipsilateral sham sciatic nerve: P < 0.0001). Similarstatistically significant trends were observed when quantifying sci-atic nerve thickness at FWHM (Fig. 3C) and area under the curve(Fig. 3D) measurements.

In CCI animals, a robust hyperintensity within the muscle re-gion (primarily the ipsilateral quadricep) could be observed(Fig. 4A–B). Quantification of the muscle region hyperintensity inCCI animals supported the consistent qualitative observationsmade in the MRI data (vs contralateral CCI muscle: P < 8E-07, vsipsilateral sham muscle: P < 5E-08). Significant increases or de-creases in MRI signal intensity were not observed in upper spine,lower spine, L4, or L5 ROIs.

3.3. Peripheral FDG-PET of the sciatic nerve, muscle, and spine

An increased FDG-PET signal overlapping with the MRI hyperin-tensity muscle region was observed in CCI animals (Fig. 4C–D, vscontralateral CCI muscle: P < 0.001, vs ipsilateral sham muscle:P < 0.0003). An increase in the FDG uptake was also observed inthe ipsilateral CCI sciatic nerve; however, spillover originating

Fig. 2. Mechanical allodynia in chronic constriction injury (CCI) animals. Blindedassessment of mechanical allodynia was performed in all animals prior to imagingsessions at weeks 2 and 3. The CCI cohort showed significant hypersensitivity tomechanical stimulation at both time points compared to the sham cohort. Errorbars represent standard error. ⁄⁄P < 1E-06, ##P < 3E-05.

from the surrounding pathological muscle was visually observed.The spillover effect was considered to result from the inherent spa-tial resolution of the FDG-PET data. Thus, this sciatic nerve endpoint was not interpreted further. A significant increase in FDG up-take was observed within the surgery site (red region denoted inFig. 1D) of the CCI cohort (Fig. 4E–F, vs contralateral CCI muscle:P < 0.04, vs ipsilateral sham muscle: P < 0.04). When the surgerysite ROI was dilated to encompass a greater area of soft tissue, sig-nificant increases or decreases in the FDG-PET signal were not ob-served, indicating an increase in metabolic activity local to themain surgery site. Moreover, significant change in metabolic activ-ity was not observed in upper spine, lower spine, L4, or L5 ROIs.

3.4. Brain DTI

DTI data collected at week 3 post CCI and sham surgeries re-vealed significantly reduced FA changes in CCI animals (Fig. 5).Specifically, while a decreasing trend (vs sham: P < 0.08) in FAwas revealed in the contralateral primary somatosensory cortex(S1), a significant (vs sham: P < 0.02) decrease in FA was observedin the contralateral primary motor cortex (M1). Significant changesin FA were not observed for other brain regions evaluated (eg, thal-amus, caudate-putamen, hypothalamus, cingulate, secondarysomatosensory cortex, and secondary motor cortex).

3.5. Systemic analyte levels

Compared to the sham condition, Fractalkine plasma levels(Fig. 6A) showed an insignificant decreasing trend (P < 0.12) inthe CCI cohort, while plasma IL-1b levels (Fig. 6B) were signifi-cantly decreased (P < 0.03). No other analyte levels in the panelof 26 showed trends of similar magnitude or significance. By calcu-lating the area under the ROC for all 26 analytes, IL-1b alone dem-onstrated predictive capabilities (Fig. 6C).

3.6. Classification analysis and predictive modeling

In Table 1, a summary index of discrimination assessment asdefined by the area under the ROC is given, which shows eachend point’s ability to classify animals between the 2 experimentalcohorts. In this classifier table, only those end points possessingsignificant group-level differences are reported. Behavior at week2, behavior at week 3, Max MR Signal Intensity in Nerve, AUC ofNerve, Mean MR Signal Intensity in Muscle, and Mean FDG-PETSignal in Muscle were observed to well separate between the 2treatment groups (area under ROC = 1). Thus, these 6 end pointswere identified as ‘‘perfect classifiers’’ for this specific dataset. Sub-sequently, using the 6 ‘‘perfect classifiers’’ shown in Table 1, pre-dictive models using all 26 analytes were used to predict eachperfect classifier. Of all the non-analyte-analyte interactions, IL-1b showed the highest capability of predicting mean MR SignalIntensity in Muscle (Table 2).

3.7. Multimodal correlation analysis

Correlation patterns amongst 26 analytes, 2 behavioral, and 8imaging end points were observed (Fig. 7). The correlation resultsshown in Fig. 7 were obtained by combining data from all shamand CCI animals. In Fig. 7, approximately 4 clusters of variableswere detected that were highly associated (Cluster #1: MIP-2,Eotaxin, G-CSF, IL-1b, iL12p70, IL-18, Fractalkine; Cluster #2:IL-13, IFN-c, IP-10, IL-6, IL-4, IL-1a; Cluster #3: IL-17, IL-5, TNF-a,IL-10, EGF, RANTES, IL-2, MIP-1a, VEGF, Leptin, MCP-1, KC-GRO;Cluster #4: all the behavioral and imaging variables). Variableswithin Clusters #1, #2, and #3 are in positive correlation, whileClusters #1 and #2 appear to be in negative correlation. Behavioral

Fig. 3. Pathology of the sciatic nerve at week 2 following chronic constriction injury (CCI) surgery. (A) Representative sham (top row) and CCI (bottom row) magneticresonance imaging (MRI) datasets are shown, where MRI images were thresholded to identify contra- and ipsilateral sciatic nerves and superimposed on the subject-specificcomputed tomography (CT). Qualitatively, a thickening of the ipsilateral (respective of surgery site) sciatic nerve as well as the ligation sites in the CCI condition can beobserved. (B) In CCI animals, the normalized mean signal intensity of the ipsilateral sciatic nerve (quantification performed below ligation site) was significantly highercompared to the contralateral nerve (##P < 0.0008) and ipsilateral sciatic nerve in sham animals (⁄⁄P < 0.0001). Similar trends were observed when quantifying sciatic nervethickness (C) at full width half maximum (FWHM) (##P < 0.0004 [vs CCI contralateral]; ⁄P < 0.016 [vs sham ipsilateral]) and area under the curve (D) (##P < 7E-05 [vs CCIcontralateral]; ⁄⁄P < 7E-05 [vs sham ipsilateral]).

L. Lasko et al. / PAIN�

154 (2013) 2782–2793 2787

readouts at weeks 2 and 3 in conjunction with brain-FA (contralat-eral M1) demonstrated high positive correlation, while other imag-ing variables were positively correlated. Behavioral readouts atweek 2 and week 3 showed strong negative correlations with otherimaging variables.

Correlation matrices specific to the CCI and sham groups havebeen given in Supplemental Fig. 1. Moreover, element-wise com-parisons showed which correlations significantly differed (FalseDiscovery Rate (FDR), q < 0.05) between the CCI and sham matri-ces (Fig. 8). Particularly, significant changes in correlation pat-terns involving several interleukins (yellow voxels) wereobserved.

4. Discussion

Unilateral constriction and irritation of the sciatic nerve elicitedcomplex responses in peripheral and central structures in conjunc-tion with changes in pain behavior and circulating analytes. Theselocalized and systemic measures proved to accurately discriminatebetween nerve injury and healthy states as determined by classifi-cation analysis. In addition to elucidating how analyte, behavioral,and imaging-based end points are associated amongst each other,

examination of correlation patterns between the distinct variablesprovided signatures of the peripheral nerve injury and shamconditions.

4.1. Mechanical allodynia and peripheral pathology

Behavioral pain and peripheral imaging measures best illus-trated the effects of sciatic nerve injury. CCI animals possessed sig-nificant mechanical allodynia at weeks 2 and 3, and calculation ofarea under the ROC identified behavioral pain end points as ‘‘per-fect classifiers’’ between CCI and sham conditions. Systemic expo-sures of TNF-a, Fractalkine, and IL-1b showed predictivecapabilities of the extent of mechanical allodynia present. With re-spect to imaging end points, group-level comparisons and classifieranalysis showed that MRI signal intensity of the sciatic nerve andmuscle, sciatic nerve thickness, and FDG uptake in muscle definedwell the peripheral pathology following sciatic nerve injury. Inter-estingly, these peripheral imaging end points showed a substantialanticorrelation with behavioral pain measures; however, thedynamics between pain behavior and MRI- or FDG-PET-basedend points did not change significantly between CCI and shamconditions.

Fig. 4. Pathology within muscle at week 2 following chronic constriction injury (CCI) surgery. T2-weighted rapid acquisition with relaxation enhancement (RARE) magneticresonance imaging (MRI) (A, B) and [18F]-2-fluoro-2-deoxy-D-glucose positron emission tomography (FDG-PET) (C, D) revealed pathology in the muscle regions, whileincrease in the FDG-PET signal was also observed at the surgery site (E, F) at week 2 following CCI surgery. (A) In the CCI computed tomography (CT)+MRI dataset, ahyperintense region within the muscle (denoted in green) can be observed. (B) Quantification of the muscle region MRI signal intensity across cohorts as well as ipsilateraland contralateral muscle regions corroborate qualitative findings (##P < 8E-07 [vs CCI contralateral]; ⁄P < 5E-08 [vs sham ipsilateral]). (C) An increase in the FDG-PET signaloverlapped with the MRI hyperintensity region in CCI animals. (D) Compared to MRI finding in the muscle region, similar comparisons stemming from the FDG-PET end pointwere still significant, but less robust (##P < 0.001 [vs CCI contralateral]; ⁄P < 0.0003 [vs sham ipsilateral]). Sham contralateral showed an increase (#P < 0.02) compared to theCCI contralateral FDG-PET end point in muscle. (E, F) Similarly, a significant increase in the FDG-PET signal was observed in the surgery site region (#P < 0.04 [vs CCIcontralateral]; ⁄P < 0.04 [vs sham ipsilateral]).

Fig. 5. Brain fractional anisotropy (FA) changes in chronic constriction injury (CCI) animals at week 3. (A) Coronal, diffusion direction encoded FA maps from a single subjectare shown. The locations of primary somatosensory (S1) and motor (M1) regions were mapped based on an internal atlas defined by Paxinos and Watson. Diffusionorientation: Green: superior-inferior, Blue: anterior-posterior, Red: left-right. (B) Relative to sham, an insignificant decreasing trend (P < 0.08) in contralateral (to surgery site)primary somatosensory cortex was observed, while, as shown in (C), significant (#P < 0.02) decrease in FA was observed in the contralateral primary motor cortex.

2788 L. Lasko et al. / PAIN�

154 (2013) 2782–2793

Fig. 6. Decrease in plasma analyte levels in chronic constriction injury (CCI) animals. (A) Relative to sham, Fractalkine plasma levels showed an insignificant decreasing trend(P < 0.12) in the CCI cohort, while a significant (#P < 0.03) decrease in plasma interleukin (IL)-1b levels was observed in CCI animals (B). No other analyte levels in the panel of26 showed trends of similar magnitude. (C) The area under the receiver operating curve (ROC) was calculated for IL-1b to determine its ability to classify animals into the 2experimental cohorts. Of the 26 analytes, only IL-1b demonstrated predictive capabilities.

Table 1Endpoint classifier ability.

End point p-Value T-statistic

ROC 95%Lower

95%Higher

Plasma IL-1b 0.03 2.44 0.839 0.628 1Behavior at week 3⁄⁄ 1.00E-06 14.53 1 1 1Behavior at week 3⁄⁄ 3.00E-05 9.14 1 1 1Brain FA 0.02 2.63 0.786 0.517 1Max MR signal intensity

in nerve⁄⁄1.43E-05 6.79 1 1 1

Mean MR signal intensityin nerve

0.0001 4.37 0.964 0.881 1

AUC of nerve⁄⁄ 7.00E-05 6.26 1 1 1FWHM (mm) of nerve 0.016 2.77 0.821 0.574 1Mean MRI signal intensity

in muscle⁄⁄5.00E-08 11.24 1 1 1

Mean FDG-PET signal inmuscle⁄⁄

0.0003 4.61 1 1 1

Mean FDG-PET signal insurgery site

0.04 2.3 0.839 0.612 1

As a summary index of discrimination assessment, the area under the ROC wascalculated for each endpoint to assess its ability to classify animals between the twoexperimental cohorts. For peripheral imaging based measures, discriminationassessment was performed between CCI ipsilateral and sham ipsilateral measures.Behavior at week 2, behavior at week 3, Max MR Signal Intensity in Nerve, AUC ofNerve, Mean MR Signal Intensity in Muscle and Mean FDG-PET Signal in Musclewere observe to well-separate between the 2 cohorts (area under ROC = 1). Thus,these 6 endpoints were identified as ‘perfect classifiers⁄⁄’ for this specific dataset.

Table 2Predictive modeling between analytes and perfect classifiers.

Analyte Behaviorat week 2

Behaviorat week 3

Max MRsignalintensityin nerve

Areaundercurveofnerve

Mean MRsignalintensityin muscle

MeanFDG-PETsignal inmuscle

KC.GRO 0.017 0 0 0 0 0MCP-1 0.112 0.004 0 0 0 0Leptin 0 0.004 0 0.05 0 0VEGF 0.013 0 0 0 0 0MIP-1a 0.005 0 0.016 0.063 0 0IL-2 0 0 0 0.064 0 0RANTES 0 0.003 0 0 0 0EGF 0 0 0 0 0 0IL-10 0 0 0.042 0 0 0TNF-a 0.62 0.004 0.04 0.11 0.043 0IL-5 0 0.004 0 0 0 0IL-17 0.05 0 0.018 0 0 0IL-1a 0.115 0.004 0.04 0.059 0 0IL-4 0.112 0 0.04 0.059 0 0IL-6 0.027 0.001 0 0.023 0 0IP-10 0 0 0 0 0 0IFN-c 0 0 0 0 0 0IL-13 0.05 0.001 0 0.085 0 0Fractalkine 0.728 0.093 0.038 0.059 0.035 0IL-18 0.017 0 0 0.002 0 0IL-12

(p70)0.112 0.004 0 0.059 0 0

IL-1b 0.674 0 0.041 0.286 0.922 0G.CSF 0.013 0 0.018 0 0 0Eotaxin 0.418 0.004 0.038 0.074 0.022 0MIP-2 0.058 0.001 0 0 0 0LIX 0 0.001 0 0 0 0

Predictive models using all 26 analytes were used to predict each perfect classifiershown in Table 1. Of all of the non-analyte–analyte interactions, IL-1b showed highprediction capability of the mean MR Signal Intensity in Muscle, while IL-1b,Fractalkine and TNF-a showed more moderate predictive capabilities for painbehavior (mechanical allodynia) at week 2.

L. Lasko et al. / PAIN�

154 (2013) 2782–2793 2789

4.1.1. Structural and metabolic changes in sciatic nerveStructural changes observed in the sciatic nerve (below surgery

site) of CCI animals included a unilateral increase in MRI signalintensity as well as increased thickness. Multiple intra- and ex-tra-axonal factors (eg, axonal density, axonal diameter, myelinthickness, and myelin density) could contribute to these relatedMRI findings in the sciatic nerve bundle. However, the presenceof edema within the endo- and epineurium of injured nerve is con-sidered to drive the increases in MRI signal intensity and nervethickness [4,12]. Certainly, early observations of decreased waterdiffusion occurring parallel to the main axonal axis support theidea of edema prolonging transverse relaxation times within in-jured nerve [2,30].

A significant yet subtle increase in FDG uptake was observed atthe nerve injury site in CCI animals. Similarly, an increase in FDGuptake was observed within the sciatic nerve below the surgerysite; however, this could be the result of spillover from the adja-cent muscle. Nonetheless, Behera et al. recently reported

significant uptake of FDG in sciatic nerve and surrounding musclein the spared nerve injury model of neuropathic pain [3].

4.1.2. Neurogenic muscular pathologyOur MRI findings in muscle ipsilateral to sciatic nerve injury are

in accord with previous work, which demonstrated prolongedtransverse relaxation times or increased MRI signal intensity with-in muscle [3,27,33]. In addition to the longer transverse relaxationtimes in muscle, Wessig et al. [38] also demonstrated a concurrent

Fig. 7. Correlation patterns amongst and between analyte, behavioral, and imaging end points. Pairwise correlations of 26 analytes, 2 behavior, and 8 imaging end points forall 15 animals were calculated by the Spearman’s rank correlation coefficient. The correlation matrix is presented as a cell plot that shows the correlations among variables ona scale from red (+1) to blue (�1). From the correlation matrix, it can be observed that there are �4 clusters of variables that are highly correlated (Cluster #1: MIP-2, Eotaxin,G-CSF, IL-1b, iL12p70, IL-18, Fractalkine; Cluster #2: IL-13, IFN-g, IP-10, IL-6, IL-4, IL-1a; Cluster #3: IL-17, IL-5, TNF-a, IL-10, EGF, RANTES, IL-2, MIP-1a, VEGF, Leptin, MCP-1,KC-GRO; Cluster #4: all the behavioral and imaging variables).

2790 L. Lasko et al. / PAIN�

154 (2013) 2782–2793

spontaneous electrical activity in leg muscles (up to 9 weeks postnerve injury) as measured by electromyography as well as sub-stantially increased capillary enlargement or intravascular spacein muscle – the latter of which increases local blood volume andhence, MRI signal intensity in muscle. Taking together vascularand electrophysiological changes in muscle, the increase in FDG-uptake observed in regions of muscle MRI hyperintensity can beexplained.

4.2. Altered analyte properties following peripheral nerve injury

4.2.1. IL-1b and Fractalkine expressionOf the panel of 26 analytes whose systemic exposure was mea-

sured at the week 2 time point, IL-1b levels were significantly de-creased in the CCI cohort as compared to the sham condition, anddemonstrated discriminatory capabilities between CCI and shamconditions. IL-1b is a proinflammatory cytokine that plays animportant role in generating and sustaining inflammatory andneuropathic pain states [17]. IL-1b is released by immune cellsand can be detected directly by nociceptors [7]. In turn, thisparticular cytokine facilitates a hypersensitive pain state as well

as mediates inflammatory responses, both of which were wellassociated with the pathology of sciatic nerve injury. Comparedto sham, a decreasing trend in Fractalkine (CX3CL1) was alsoobserved. Fractalkine is a chemokine active during the inflamma-tory responses that co-exists with vascular pathology, as well asis detected and activated at the endothelium [1,39]. Furthermore,the change in systemic levels of Fractalkine can be considered inthe context of muscle vasculature changes following denervation[38] in conjunction with the increase in FDG uptake and MRI signalintensity measured in muscle. Interestingly, both IL-1b and Fractal-kine exposure levels demonstrated predictive capabilities of painbehavior at week 2, while IL-1b was highly predictive of MRI signalintensity in muscle ipsilateral to the surgery site – an observationthat may relate to IL-1b modulation of vasculature and endothelialcells during pathological and inflammatory responses [28].

The exact cause of systemic IL-1b and Fractalkine downregula-tion in the CCI cohort could not be conclusively determined in thepresent study. It is possible that this systemic downregulation mayresult from a substantial concentration of either analyte being del-egated at the site of injury [11]. Given past work by Echeverryet al., the breakdown of the blood–brain and blood–spinal cord

Fig. 8. Comparison of correlation patterns between sham and chronic constriction injury (CCI) animals. Pairs of analyte, behavioral, and imaging end points whose correlationwas significantly altered by the sciatic nerve injury are depicted in yellow (false discovery rate; q < 0.05). Black voxels indicate correlation patterns not significantly alteredbetween CCI and sham conditions.

L. Lasko et al. / PAIN�

154 (2013) 2782–2793 2791

barriers following peripheral nerve injury could facilitate perme-ation of IL-1b specifically from vasculature to brain and spinal cordsites [13]. Feedback mechanisms may have also been active in or-der to balance proinflammatory and antiinflammatory processes[25].

4.2.2. Correlation patterns of analytesCorrelation analysis of the 26 analytes revealed distinct groups

of cytokines and chemokines. Of particular interest was the factthat IL-1b and Fractalkine were clustered within the same group,and the correlation amongst several interleukins was especiallystrong. Upon a qualitative view of the analyte correlation patternsbetween CCI and sham conditions, it was clear that peripheralnerve injury undoubtedly disrupted the basal relationshipsamongst the cytokines, chemokines, and growth factors investi-gated herein. IL-17, a major proinflammatory cytokine, was onecytokine whose correlation was significantly altered with severalother analytes (FDR, q < 0.05: IL-1a, IL-6, IP-10, IL-13, Eotaxin;not surpassing the multiplicity correction: IL-4, IL-12, IFN-c andMIP-2). IL-17 drives the expression of many other cytokines and

chemokines, as well as other proinflammatory mediators such asprostaglandins and matrix metalloproteases [18,23]. Thus, the syn-ergistic relationships between IL-17 and other analytes may ex-plain the significant change in correlation following sciatic nerveinjury. It is believed that assessing the relationships amongst dis-tinct metrics (ie, analyte levels) in a longitudinal manner may offerfurther insights as to how distinct metrics relate to each other orare altered between peripheral nerve injury and healthy states.

4.3. Decreased anisotropy in motor cortex

It is well known that trauma to peripheral nerves can initiatefunctional and structural alterations within the central nervoussystem through a series of complex feed-forward and feedbackneural mechanisms [22,24], where the characteristics of centralplasticity can depend upon the location of peripheral injury [36].The sensorimotor system, in particular, undergoes functional andstructural changes following peripheral nerve injury [8,26,28],and our DTI-based observations at week 3 are consistent with ear-lier studies with respect to modulated neuroanatomical structures.

2792 L. Lasko et al. / PAIN�

154 (2013) 2782–2793

In the current study, the CCI cohort possessed a significant de-crease in FA within contralateral (relative to nerve injury site)M1, while a decreasing trend in anisotropy was present in contra-lateral S1. While previous neuroimaging studies involving nerveinjury or neuropathic pain patients have utilized DTI to probewhite matter microstructure in humans, here we implementedthe method to morphologically characterize gray matter regionsimplicated in sensorimotor and pain processing. The observed de-creases in FA suggest changes in cytoarchitecture, perhaps in layersII and III, where axons are predominantly oriented in the superior-inferior direction (corresponding to green voxels in Fig. 5A, see alsowork by Budde and Frank [9]). To comprehend the underlying nat-ure of decreased FA values in contralateral M1 specifically, histo-logical studies or longitudinal measurement of intraneuronalmetabolite diffusion in conjunction with cortical thickness assess-ment may be of value [40]. Furthermore, the significant anticorre-lative (r = �0.75, P = 0.001) behavior observed between measuressuch as the FDG uptake in muscle and contralateral M1 FA stronglysuggests a link between distinct peripheral and central physiolog-ical mechanisms.

DTI-based measurements were significantly different in contra-lateral M1, yet this effect was not as robust compared to peripheralmeasures such as sciatic nerve thickness. The more subtle FA dif-ferences between CCI and sham conditions may have resulted inpart from DTI data being acquired at week 3 post CCI and shamsurgeries. It could be possible that at 3 weeks, central processesspecific to this model may correspond with initial stages of mor-phological or cytoarchitectural changes. Performing longitudinalstudies beyond the 3-week time point may enable a better compre-hension of the trajectory of central morphological properties fol-lowing peripheral nerve injury.

4.4. Conclusion

This study used a combination of analyte, behavioral, and imag-ing assessments to understand the pathophysiological signature ofsciatic nerve injury. As a result of this multimodal approach, thephysiological response to nerve injury was reflected in peripheraland central soft tissues, as well as in the expression of circulatingcytokines, chemokines, and growth factors. Such an approach ina clinical setting, involving sciatica patients over time, could fur-ther determine the utility of this multimodal approach with re-gards to its diagnostic value.

Conflict of interest

All of the authors are employees of AbbVie Inc.

Acknowledgements

The study design, conduct, and funding were provided by Abb-Vie Inc. AbbVie Inc. participated in the interpretation of data, re-view, and approval of the manuscript.

Appendix A. Supplementary data

Supplementary data associated with this article can be found, inthe online version, at http://dx.doi.org/10.1016/j.pain.2013.08.016.

References

[1] Bazan JF, Bacon KB, Hardiman G, Wang W, Soo K, Rossi D, Greaves DR, ZlotnikA, Schall TJ. A new class of membrane-bound chemokine with a CX3C motif.Nature 1997;385:640–4.

[2] Beaulieu C, Does MD, Snyder RE, Allen PS. Changes in water diffusion due toWallerian degeneration in peripheral nerve. Magn Reson Med1996;36:627–31.

[3] Behera D, Jacobs KE, Behera S, Rosenberg J, Biswal S. (18)F-FDG PET/MRI can beused to identify injured peripheral nerves in a model of neuropathic pain. JNucl Med 2011;52:1308–12.

[4] Bendszus M, Wessig C, Solymosi L, Reiners K, Koltzenburg M. MRI of peripheralnerve degeneration and regeneration: correlation with electrophysiology andhistology. Exp Neurol 2004;188:171–7.

[5] Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical andpowerful approach to multiple testing. J R Stat Soc Series B Stat Methodol1995;57:289–300.

[6] Bennett GJ, Xie YK. A peripheral mononeuropathy in rat that producesdisorders of pain sensation like those seen in man. PAIN� 1988;33:87–107.

[7] Binshtok AM, Wang H, Zimmermann K, Amaya F, Vardeh D, Shi L, Brenner GJ, JiRR, Bean BP, Woolf CJ, Samad TA. Nociceptors are interleukin-1beta sensors. JNeurosci 2008;28:14062–73.

[8] Borsook D, Becerra L, Fishman S, Edwards A, Jennings CL, Stojanovic M,Papinicolas L, Ramachandran VS, Gonzalez RG, Breiter H. Acute plasticity in thehuman somatosensory cortex following amputation. Neuroreport1998;9:1013–7.

[9] Budde MD, Frank JA. Examining brain microstructure using structure tensoranalysis of histological sections. Neuroimage 2012;63:1–10.

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

[11] del Rey A, Yau HJ, Randolf A, Centeno MV, Wildmann J, Martina M, BesedovskyHO, Apkarian AV. Chronic neuropathic pain-like behavior correlates with IL-1beta expression and disrupts cytokine interactions in the hippocampus.PAIN� 2011;152:2827–35.

[12] Does MD, Snyder RE. Multiexponential T2 relaxation in degeneratingperipheral nerve. Magn Reson Med 1996;35:207–13.

[13] Echeverry S, Shi XQ, Rivest S, Zhang J. Peripheral nerve injury alters blood-spinal cord barrier functional and molecular integrity through a selectiveinflammatory pathway. J Neurosci 2011;31:10819–28.

[14] el Barzouhi A, Vleggeert-Lankamp CL, Lycklama a Nijeholt GJ, Van der KallenBF, van den Hout WB, Jacobs WC, Koes BW, Peul WC. Magnetic resonanceimaging in follow-up assessment of sciatica. N Engl J Med2013;368:999–1007.

[15] Ghanouni P, Behera D, Xie J, Chen X, Moseley M, Biswal S. In vivo USPIOmagnetic resonance imaging shows that minocycline mitigates macrophagerecruitment to a peripheral nerve injury. Mol Pain 2012;8:49.

[16] Hanley JA, Hajian-Tilaki KO. Sampling variability of nonparametric estimatesof the areas under receiver operating characteristic curves: an update. AcadRadiol 1997;4:49–58.

[17] Honore P, Wade CL, Zhong C, Harris RR, Wu C, Ghayur T, Iwakura Y, DeckerMW, Faltynek C, Sullivan J, Jarvis MF. Interleukin-1alphabeta gene-deficientmice show reduced nociceptive sensitivity in models of inflammatory andneuropathic pain but not post-operative pain. Behav Brain Res2006;167:355–64.

[18] Iwakura Y, Ishigame H. The IL-23/IL-17 axis in inflammation. J Clin Invest2006;116:1218–22.

[19] Jacobs WC, van Tulder M, Arts M, Rubinstein SM, van Middelkoop M, Ostelo R,Verhagen A, Koes B, Peul WC. Surgery versus conservative management ofsciatica due to a lumbar herniated disc: a systematic review. Eur Spine J2011;20:513–22.

[20] Jenkinson M, Bannister P, Brady M, Smith S. Improved optimization for therobust and accurate linear registration and motion correction of brain images.Neuroimage 2002;17:825–41.

[21] Jensen MC, Brant-Zawadzki MN, Obuchowski N, Modic MT, Malkasian D, RossJS. Magnetic resonance imaging of the lumbar spine in people without backpain. N Engl J Med 1994;331:69–73.

[22] Kaas JH, Collins CE. Anatomic and functional reorganization of somatosensorycortex in mature primates after peripheral nerve and spinal cord injury. AdvNeurol 2003;93:87–95.

[23] Kolls JK, Linden A. Interleukin-17 family members and inflammation.Immunity 2004;21:467–76.

[24] Navarro X, Vivo M, Valero-Cabre A. Neural plasticity after peripheral nerveinjury and regeneration. Prog Neurobiol 2007;82:163–201.

[25] Olas K, Butterweck H, Teschner W, Schwarz HP, Reipert B. Immunomodulatoryproperties of human serum immunoglobulin A: anti-inflammatory and pro-inflammatory activities in human monocytes and peripheral bloodmononuclear cells. Clin Exp Immunol 2005;140:478–90.

[26] Pawela CP, Biswal BB, Hudetz AG, Li R, Jones SR, Cho YR, Matloub HS, Hyde JS.Interhemispheric neuroplasticity following limb deafferentation detected byresting-state functional connectivity magnetic resonance imaging (fcMRI) andfunctional magnetic resonance imaging (fMRI). Neuroimage2010;49:2467–78.

[27] Polak JF, Jolesz FA, Adams DF. Magnetic resonance imaging of skeletal muscle.Prolongation of T1 and T2 subsequent to denervation. Invest Radiol1988;23:365–9.

[28] Puhlmann M, Weinreich DM, Farma JM, Carroll NM, Turner EM,Alexander Jr HR. Interleukin-1beta induced vascular permeability isdependent on induction of endothelial tissue factor (TF) activity. JTransl Med 2005;3:37.

[29] Stafford MA, Peng P, Hill DA. Sciatica: a review of history, epidemiology,pathogenesis, and the role of epidural steroid injection in management. Br JAnaesth 2007;99:461–73.

[30] Takagi T, Nakamura M, Yamada M, Hikishima K, Momoshima S, Fujiyoshi K,Shibata S, Okano HJ, Toyama Y, Okano H. Visualization of peripheral nerve

L. Lasko et al. / PAIN�

154 (2013) 2782–2793 2793

degeneration and regeneration: monitoring with diffusion tensortractography. Neuroimage 2009;44:884–92.

[31] Tibshiranit R. Regression shrinkage and selection via the lasso. J R Stat SocSeries B Stat Methodol 1996;58:267–88.

[32] Tracey I, Dickenson A. SnapShot: pain perception. Cell 2012;148:e2.[33] Viddeleer AR, Sijens PE, van Ooyen PM, Kuypers PD, Hovius SE, Oudkerk M.

Sequential MR imaging of denervated and reinnervated skeletal muscle ascorrelated to functional outcome. Radiology 2012;264:522–30.

[34] von Hehn CA, Baron R, Woolf CJ. Deconstructing the neuropathic painphenotype to reveal neural mechanisms. Neuron 2012;73:638–52.

[35] Vroomen PC, de Krom MC, Slofstra PD, Knottnerus JA. Conservative treatmentof sciatica: a systematic review. J Spinal Disord 2000;13:463–9.

[36] Wall JT, Xu J, Wang X. Human brain plasticity: an emerging view of themultiple substrates and mechanisms that cause cortical changes and related

sensory dysfunctions after injuries of sensory inputs from the body. Brain ResBrain Res Rev 2002;39:181–215.

[37] Wassenaar M, van Rijn RM, van Tulder MW, Verhagen AP, van der Windt DA,Koes BW, de Boer MR, Ginai AZ, Ostelo RW. Magnetic resonance imaging fordiagnosing lumbar spinal pathology in adult patients with low back pain orsciatica: a diagnostic systematic review. Eur Spine J 2012;21:220–7.

[38] Wessig C, Koltzenburg M, Reiners K, Solymosi L, Bendszus M. Muscle magneticresonance imaging of denervation and reinnervation: correlation withelectrophysiology and histology. Exp Neurol 2004;185:254–61.

[39] White GE, Greaves DR. Fractalkine: a survivor’s guide: chemokines asantiapoptotic mediators. Arterioscler Thromb Vasc Biol 2012;32:589–94.

[40] Wood ET, Ronen I, Techawiboonwong A, Jones CK, Barker PB, Calabresi P,Harrison D, Reich DS. Investigating axonal damage in multiple sclerosis bydiffusion tensor spectroscopy. J Neurosci 2012;32:6665–9.