Subjects were anesthetized with pentobarbital and a dorsal midline incision was made. Connective tissue was dissected and a C5 laminectomy was performed. Subjects were given unilateral cervical contusion injuries using a New York University/Multicenter Animal Spinal Cord Injury Study device with a 2.0 mm head dropped from an impact height of 6.25 mm. These procedures have been shown to produce a focal lesion with minimal contralateral spread, allowing use of the contralateral side as an internal control (Gensel et al., 2006). Given the regional effect of injury, microscope field was used as the unit of analysis, yielding sample sizes of n = 17 for spinal cord injury and n = 21 for control. To provide an accurate estimate of total synaptic AMPAR, each field was sampled in three-dimensions (3D) with a laser scanning microscope (yielding 608 total images) and quantification was performed on the mean three-dimensional colocalization for each field. The experiment was replicated at 90 min and 3 h to provide a time course analysis (n = 4 subjects with SCI tested as a within-subject variable) (Fig. 1).

Subjects were given stereotactic nanoinjections (35 nl) into the T9-10 ventral horn as previously described (Hermann et al., 2001) using compressed air micropressure applied to pulled glass pipettes (tip diameter, 30 μm; 30° bevel; Radnoti). Albumin (1 μM) was used as a control because it has molecular weight similar to rat recombinant TNFα (R&D systems). For biochemical experiments, subjects (n = 5) were given nanoinjections of either TNFα (1 μM) or albumin into four evenly spaced sites within a 750 μm length of spinal cord (Fig. 2A; supplemental Fig. 1, available at www.jneurosci.org as supplemental material). For imaging experiments, subjects (n = 4) received a single TNFα dose (0.01, 0.1, or 1 μM), and each subject received a contralateral control injection of albumin (supplemental Fig. 2A, available at www.jneurosci.org as supplemental material). The selected TNFα doses have been shown to be insufficient to induce cell death alone but enhance glutamate-mediated excitotoxicity in the spinal cord (Hermann et al., 2001). To localize injections (supplemental Fig. 2B, available at www.jneurosci.org as supplemental material), FluoroRuby (Invitrogen) was included in the solutions.

Subjects were given unilateral cervical spinal cord injuries as described above and then placed in a spinal stereotaxic for intraparenchymal treatment with human recombinant soluble TNFα receptor (R&D systems). The dura was opened under a dissecting scope and then two nanoinjection pipettes (50 μm tip diameter, 30° bevel, Radnoti) were positioned under stereotaxic control to target the rostral and caudal edges of the lesion. Pipettes were mounted on a stereotaxic arm at a fixed space of 7.2 mm apart, with the visible contusion hematoma placed equidistant between the tips. Drug or vehicle was delivered into the spinal parenchyma at a depth of 1.4 mm at the midline, at a dose of 10 μg/ml dissolved in PBS containing 0.1% BSA. Solutions were delivered in 30 nl bursts over 5 min per pipette for 60 min (subjects were killed 90 min after injury), resulting in a total dose of 7.2 ng soluble TNFα receptor 1 (sTNFαR1) in 720 nl of solution per subject. To achieve a balanced design, nanoinjection of sTNFαR1 was compared against two control groups: vehicle nanoinjection, and a noninjected contusion lesion (n = 4).

Sixty minutes after TNFα or albumin nanoinjections, 7.5 mm of spinal cord was extracted under deep anesthesia, snap-frozen on dry ice, and placed in a −80°C freezer for later processing. The fractionation procedures were adapted from previous work with mouse spinal cord (Galan et al., 2004). Frozen spinal cord was homogenized with 30 passes of a “Type B” pestle in a dounce homogenizer (Kontes) followed by 5 passes through a 22 gauge needle in ice-cold buffer, pH 7.5, containing 10 mM Tris, 300 mM sucrose, and a complete mini protease inhibitor mixture (Roche). Crude homogenate was centrifuged at 5000 RCF for 5 min at 4°C. Supernatant (S1) was further fractionated at 13,000 RCF for 30 min. Supernatant (S2) was transferred to a new tube and pellet (P2) was resuspended in PBS (50 μl) containing protease inhibitor. All samples were sonicated and stored at −80°C for later processing. N-cadherin was used to confirm plasma membrane enrichment (supplemental Fig. 1, available at www.jneurosci.org as supplemental material).

Total protein was determined with BCA (Pierce) and quantified with a plate reader (Tecan; GeNios). Samples from each subject were separately mixed with cold Laemmli sample buffer (62.5 mM Tris HCl, pH 6.8, 25% glycerol, 2% SDS, 5% β-mercaptoethanol) and immediately loaded into precast 12.5% Tris-HCl polyacrylamide gels (Bio-Rad). Samples from albumin- and TNFα-treated subjects were loaded pairwise on adjacent lanes (40 μg/lane) in a semirandomized manner to counterbalance for regional variability within the gel. To minimize potential for bias, the experimenter was unaware of condition during gel loading. Spinal cord fractions and a positive control of brain homogenate (supplemental Fig. 1, available at www.jneurosci.org as supplemental material) were electrophoresed in SDS running buffer (Bio-Rad; 25 mM Tris, 192 mM glycine, 0.1% SDS, pH 8.3). Molecular weights were confirmed with kaleidoscopic precision protein standard (Bio-Rad). Proteins were transferred to nitrocellulose membrane in cold tris-glycine buffer (25 mM Tris, 192 mM glycine, 20% methanol, pH 8.3). Membranes were blocked for 1 h at room temperature with 5% (w/v) nonfat dry milk in Tween-tris-buffered saline (TTBS) [0.1% (v/v) Tween 20, 50 mM Tris, 150 mM NaCl, pH 7.6] and then incubated overnight at 4°C with primary antibodies in 1% milk TTBS. Membranes were washed 3 × 10 min with TTBS and then incubated with goat anti-rabbit or goat anti-mouse horseradish peroxidase-conjugated antibodies (Pierce) for 1 h at room temperature in 1% milk TTBS. Membranes were washed 3 × 10 min with TTBS (0.05% Tween), incubated with ECL substrate (West Femto; Pierce), and exposed to x-ray film (Phenix). After blotting GluR1, membranes were stripped with 0.2 N NaOH for 5 min, rinsed with nanopure water and then reprobed for GluR2. A second stripping step was performed before probing with N-cadherin. The success of stripping was confirmed by ECL. Actin (BD transduction laboratories) was probed as a loading control after N-cadherin visualization (supplemental Fig. 1, available at www.jneurosci.org as supplemental material).

Films were digitized using a Perfection 1200 photo scanner (Epson) and quantified with Kodak 1d software. Bands were detected in an automated manner by the software with supervision by experimenter who was unaware of experimental condition. Mean intensity of each band was measured relative to the mean background immediately above and below the band. Exposure time was optimized for each protein to maximize the difference between band intensity and background. To control for potential loading-induced variance in total AMPAR levels while remaining sensitive to differential changes in subunit composition, the GluR1:GluR2 intensity ratio was used as the primary outcome. Statistical analysis of the actin density was used to confirm equal total protein loading.

For imaging experiments, subjects were killed under deep anesthesia by transcardiac perfusion with 0.9% saline followed by 4% paraformaldehyde. Lengths of spinal cord (30 mm) centered on the injection sites or lesion were removed and postfixed overnight (<18 h) in 4% paraformaldehyde. Tissue was cryoprotected in 30% sucrose for 2 d, cut into 10 mm blocks, flash frozen on dry ice, and placed in a −80°C freezer for later processing. Tissue was embedded in OCT and sectioned 20 μm horizontally, dividing adjacent sections across four sets of slides.

Antibody labeling was performed on fixed tissue sections from a full set of experimental conditions simultaneously using a high-throughput staining station (Sequenza; Thermo Scientific). Before antibody application, tissue was blocked and permeabilized for 1 h with 5% normal goat serum and 0.3% Triton X-100. Sections were incubated overnight at room temperature with a solution containing mouse monoclonal antibody for presynaptic synaptophysin (1:200) and rabbit polyclonal anti-AMPAR antibodies (1:200) directed at the C-terminal intracellular domains of GluR1 or GluR2 (Millipore). cFos was labeled with a rabbit primary (Oncogene) at 1:5000 coincubated with synaptophysin to identify neurons. GluR1, GluR2, and cFos were applied to serial slide sets allowing regional specificity of the effects to be compared within a given subject. After a wash with 2 ml of PBS, all slides were incubated for 1 h at RT with the same freshly diluted fluorescent secondary antibody solution containing 1:100 Alexa 488 goat anti-rabbit (Invitrogen) and Alexa 633 goat-anti-mouse. These procedures labeled AMPARs or cFos as Alexa 488 and synaptophysin as Alexa 633. Slides were briefly rinsed with 2 ml PBS and coverslipped with Vectashield containing DAPI (4′ ,6-diamidino-2-phenylindole) (Vector Laboratories). Each immunolabel had three negative controls: no primary, and each individual primary with the incorrect secondary. Confocal microscopy on these controls revealed virtually no detectable label above threshold.

For contusion tissue, a region of interest was selected using c-Fos within the penumbra of the expanding lesion. Although motoneurons do not label strongly for c-Fos, labeling in smaller neurons (Fig. 1B) was used as a bioassay for activation in the region of imaged motoneurons. Confocal stacks were sampled within the c-fos positive region both rostral and caudal to the injury site. Using microscope stage coordinates, control samples were selected to be directly contralateral to injury samples. For nanoinjection experiments, one large motoneuron was sampled every 100 μm along the rostrocaudal axis of each ventral horn in horizontal sections for a 600 μm length of spinal cord rostral and caudal to the injection center (supplemental Fig. 2, available at www.jneurosci.org as supplemental material). For all experiments, large ventral motoneurons (defined by diameter >40 μm) were selected under wide-field fluorescence by researchers who were blinded to experimental condition. Cells were selected by the characteristic pattern of presynaptic synaptophysin outlining the plasma membrane, thereby ensuring that researchers were blinded to the postsynaptic AMPA receptor levels at the time of cell selection. If several motoneurons existed at a given sampling site, a single cell was selected at random.

Confocal stacks were generated for large motoneurons using a Zeiss 510 META laser scanning confocal microscope with a 63× objective (NA = 1.4) and a 2× zoom. Bandpass filter and laser settings were optimized on control tissue and then held constant for the duration of the experiment, providing consistent, near-complete spectral separation of synaptophysin (Alexa 633), FluoroRuby (Texas Red), and AMPARs (Alexa 488). Confocal slices (1 μm) over-sampled at 0.5 μm z-intervals were deblurred using 3D-blind iterative deconvolution (AutoQuant). Iteration number for the deconvolution algorithm was determined using a random subset of images and then held constant for each label for the duration of the study (iteration, 2 for GluR1; 4 for GluR2). Sequential use of confocal and deconvolution allowed higher resolution than achievable by either technique alone, providing more precise localization of receptor puncta (supplemental Fig. 2, available at www.jneurosci.org as supplemental material).

MetaMorph software (Molecular Devices) was used to quantify the number of fluorescently labeled receptor puncta on the plasma membrane using custom-designed macro programs. All measures represent the number of pixels above a predetermined threshold that was established on control tissue and then held constant throughout the study. The first macro measured, for the three-dimensional z-series of each field, the total number of AMPAR positive pixels (reflecting both membrane and intracellular receptors) and AMPAR/synaptophysin colocalized pixels (reflecting synaptic AMPARs in the neuropil). Similar measurement procedures have been shown to be sensitive to subcellular localization of receptors in vitro (Beattie et al., 2002). A second macro screened confocal stacks to identify the optical section with maximal synaptophysin/AMPAR colocalization. A blinded researcher supervised the algorithm, dropping planes that were inappropriately selected based on staining artifacts. The motoneuron plasma membrane was traced on the selected optical plane using the synaptophysin channel and a subroutine generated a 2 μm-wide cutout representing plasma membrane area (see Fig. 4C,D). MetaMorph then measured the proportion of this plasma membrane area labeled with AMPAR puncta alone (representing extrasynaptic receptors) and colocalized with synaptophysin (representing synaptic receptors).

Nissl staining was performed using 0.5% cresyl violet (Sigma) that was dissolved in dH₂O and filtered. Tissue sections from each experimental condition were mounted on the same slide, and staining was performed on the entire set at the same time, thereby minimizing systematic variance in staining. Tissue was briefly rinsed in dH₂O and then stained with cresyl violet for 10 min. After a 5 min wash in 95% EtOH, tissue was differentiated in 95% EtOH containing a small amount of glacial acetic acid (10 drops/180 ml). Differentiation was checked under a microscope by a researcher who was blinded to experimental condition. Tissue was serially washed in 100% EtOH (2 × 3 min), serially placed in Xylene (4 × 3 min), and coverslipped with DPX mounting medium (Sigma).

To quantify the extent of sparing after drug treatment we used a slide scanning system mounted on a Zeiss Axioplan 2 microscope. A motorized stage under precise digital feedback control allowed us to build, in an automated manner, full montage reconstructions of horizontal tissue sections through the ventral horn from high-power images (see Figs. 5, ,6).6). We then designed an image analysis algorithm that counted all Nissl-positive objects above a preset threshold within the tissue section using a moving, 100 μm sampling window. This produced >367,000 data points that were exported into SPSS statistical software where the data were filtered by size to generate counts for Nissl-positive cells in the ventral horn with a mean diameter >40 μm. Verification by expert histologists (MSB and JCB) indicated that the selected cells met morphological and size criteria of large ventral motoneurons.

Previous in vitro work has revealed that AMPAR trafficking can be detected optically by simultaneously imaging the presynaptic vesicular protein synaptophysin and AMPARs, which are expressed postsynaptically (Lissin et al., 1998; Stellwagen et al., 2005). To test whether altered AMPAR trafficking occurs in spinal cord injury we applied a similar analysis method to perfused, fixed, and permeabilized tissue sections from a preclinical model of spinal contusion injury. Accurate subcellular localization of AMPAR puncta within fixed tissue sections was achieved using a serial imaging methodology of high-resolution confocal microscopy followed by 3 d blind-iterative deconvolution (supplemental Fig. 2, available at www.jneurosci.org as supplemental material). Given that TNFα levels in spinal tissue become rapidly elevated after CNS trauma (Wang et al., 1996), we hypothesized that altered AMPAR trafficking would occur at early time points after spinal cord injury. Testing this hypothesis is challenging, in part, because cells that have the most robust increases in plasma membrane AMPAR may be selectively killed, thereby obscuring detection of AMPAR trafficking. However, if changes in AMPAR trafficking occur in a temporally and spatially graded manner, then interfering with AMPAR trafficking represents a valuable therapeutic target for stopping early secondary injury and limiting expansion of the lesion. To test this important hypothesis, we used a recently established unilateral model of cervical spinal cord injury (Gensel et al., 2006). This model is unique in that it provides the uninjured contralateral side as a control for surface AMPAR levels, allowing quantitative comparisons within a single subject (Fig. 1A). At early postinjury time points, spinal contusion is marked by an area of frank lesion at the site of impact with an expanding penumbra of secondary damage that moves centrifugally over time (Crowe et al., 1997; Beattie, 2004). Based on the hypothesis that AMPAR trafficking contributes to the early wave of secondary cell death, we examined tissue just beyond the frank lesion border for evidence of AMPAR trafficking at 90 min and 3 h after spinal cord injury. Because TNFα strongly activates cFos in the region of highest glutamate vulnerability (Hermann et al., 2001), we used cFos labeling to define a region of interest for high-resolution confocal imaging of AMPAR puncta. cFos labeling on the injury side was observed in a narrow ring just beyond the region of frank damage, in tissue that otherwise appeared healthy (Fig. 1B). Confocal analysis of large motoneurons in this region indicated that spinal cord injury produced a marked increase in total synaptic AMPAR within the neuropil (Fig. 1C,D) as measured by a high-throughput image analysis algorithm that was sensitive to synaptic and total AMPAR levels within the tissue. Although the increase in synaptic AMPAR was significant at both postinjury time points (Fig. 1E)(p < 0.001), the effect was more pronounced at 90 min than 3 h, mirroring the time course of early TNFα release in the spinal cord after contusion injury (Wang et al., 1996).

Recent in vitro work has shown that rapid translation of AMPAR protein in the dendrites can lead to increased insertion of new GluR1 at synaptic sites in hippocampal cultures (Ju et al., 2004). To test whether spinal cord injury increases synaptic AMPARs by increasing GluR1 protein levels, we performed ANCOVA to statistically test whether variance in the total number of GluR1 puncta could account for the injury-induced increase in synaptic GluR1. ANCOVA indicated that the increase in synaptic AMPAR was not driven by a change in total cellular AMPAR (p < 0.0001 after correcting for total GluR1), implicating a receptor trafficking mechanism rather than protein synthesis/degradation.

To test whether TNFα could contribute to SCI-induced AMPAR trafficking we delivered TNFα or the vehicle (albumin) to uninjured subjects and evaluated AMPAR trafficking in vivo using a biochemical assay (Fig. 2A). Modest plasma membrane enrichment via low speed centrifugation has been used to detect AMPAR trafficking in the spinal cord in vivo (Galan et al., 2004; Whitlock et al., 2006). The presence of plasma membrane was indicated by N-cadherin (Fig. 2B). Because AMPARs that contain the GluR1 subunit but lack GluR2 have been linked to increased calcium permeability (Hollmann et al., 1991) and excitotoxicity (Vandenberghe et al., 2000), we tested the relative levels of these subunits in the plasma membrane using a stripping-reprobing protocol. Sequential immunolabeling of the same Western blots (Fig. 2C) revealed a significant increase in the GluR1:GluR2 ratio in the plasma membrane (p < 0.05, Mann-Whitney U), indicating an increase in the proportion of GluR2-lacking receptors in TNFα-treated subjects.

Although the biochemical fractionation assay revealed in vivo trafficking of GluR2-lacking AMPARs by TNFα, it leaves two important questions unanswered: (1) Does TNFα-induced AMPAR trafficking occur in clinically relevant populations of neurons? (2) Does TNFα induce divergent effects on GluR1 and GluR2 subunits within the same neurons, thereby predisposing these cells to excitotoxic cell death? To address these issues we performed immunohistochemistry and high resolution image analysis of large ventral horn motoneurons. These cells are exquisitely sensitive to excitotoxicity (Vandenberghe et al., 2000), and their loss contributes to functional deficits in spinal cord injury and amyotrophic lateral sclerosis (Grossman et al., 2001b; Sun et al., 2006). Subunit levels were evaluated using a modification of the imaging procedures that we used to evaluate AMPAR trafficking after SCI (Fig. 1). TNFα and vehicle (albumin) were delivered on contralateral sides within the same subjects (Fig. 3). FluoroRuby was included in the injection solutions, providing a region of interest for high-resolution confocal microscopy (supplemental Fig. 2, available at www.jneurosci.org as supplemental material). Differential regulation of GluR1 levels at somatic and dendritic synapses has been reported in hippocampal cell culture (Grooms et al., 2006). If TNFα alters synaptic AMPAR levels in motoneuron dendrites, this would indicate a greater number of surface receptors available for glutamate binding, and therefore, greater potential for excitotoxicity. Superimposition of differential interference contrast (DIC) microscopy (Fig. 3A) and confocal images (Fig. 3B) revealed that many AMPAR-positive synapses existed on structures within the neuropil that have the morphology of cross-cut dendrites (Fig. 3C). To test whether TNFα-altered AMPAR levels at these synapses in vivo, we used the same image analysis algorithm that we developed to quantify colocalization of AMPARs and synaptophysin within the three-dimensional confocal field after SCI. Analysis revealed that TNFα increased the total levels of synaptic GluR1 in a dose-dependent manner (p < 0.001) (Fig. 3D-J), whereas GluR2 did not change (Fig. 3K), indicating an increase in GluR2-lacking receptors. At higher doses, TNFα increased GluR2-lacking receptors across a broad tissue region, even extending to the contralateral side where the vehicle control was injected. This effect manifested statistically as a significant dose by side interaction by mixed-factorial ANOVA (p < 0.05) (supplemental Fig. 3, available at www.jneurosci.org as supplemental material).

ANCOVA was used to test whether increased total GluR1 protein levels were responsible for the TNFα-induced enhancement of synaptic GluR1. The dose-dependent effect of TNFα on synaptic GluR1 was robustly maintained after correcting for total protein levels (p < 0.0001), suggesting that the effect of TNFα on synaptic GluR1 was specifically driven by redistribution of AMPARs to synaptic sites rather than an increase in total GluR1. These data indicate that the targets for therapeutic intervention may be the specific mechanisms underlying AMPAR trafficking rather than mechanisms in control of protein synthesis and degradation, just as we saw in the spinal cord injury experiments above.

To test whether TNFα contributes to AMPAR trafficking after spinal cord injury, we nanoinjected a soluble form of human recombinant TNFα receptor (sTNFαR1) to reduce functional TNFα levels acutely after a unilateral cervical spinal cord injury. Synaptic AMPAR levels were measured by confocal analysis of the penumbra as identified by registering GluR1/synaptophysin-labeled sections to adjacent dorsal and ventral sections which were respectively labeled for Nissl and cFos. As observed in the first experiment (Fig. 1B), cFos labeling on the injury side occurred in a ring just beyond the region of frank damage, in tissue that otherwise appeared healthy (Fig. 5A). Blinded digital image analysis revealed that the diameter of the cFos-positive ring was not significantly altered by sTNFαR1 treatment (p < 0.05), confirming that cFos positivity could serve as a consistent landmark for identifying the lesion penumbra. High-resolution confocal analysis of GluR1 and synaptophysin suggested that sTNFαR1 nanoinjection reduced synaptic AMPAR levels in the neuropil after SCI (Fig. 5B). Automated quantification through the three-dimensional stacks of large motoneurons revealed a significant reduction of synaptic AMPAR numbers in sTNFαR1 treatment relative to control conditions (albumin vehicle and noninjected SCI controls). This effect was significant after correcting for the total GluR1 numbers with ANCOVA, suggesting that sTNFαR1 altered AMPAR trafficking to synapses without directly impacting total GluR1 protein levels (p < 0.001). These data strongly suggest that TNFα contributes to SCI-induced trafficking of AMPARs to the synaptic plasma membrane, providing a potential therapeutic target for reducing excitotoxic cell death after CNS trauma.