α2δ-1 Signaling Drives Cell Death, Synaptogenesis, Circuit Reorganization, and Gabapentin-Mediated Neuroprotection in a Model of Insult-Induced Cortical Malformation

Visual Abstract


Introduction
Developmental cortical malformations (DCMs) are characterized by an anatomically disorganized cortex and are associated with difficult to treat epilepsy (Crino, 2005;Takano et al., 2006). There are multiple DCM subsyndromes, including tuberous sclerosis, focal cortical dysplasia (FCD), and polymicrogyria. Both genetic and environmental causes of DCMs exist, with individual subsyndromes having their own etiology. These include mutations in genes responsible for neuronal proliferation, migration, and apoptosis (Barkovich et al., 2012), as well as neonatal insults including pre-and perinatal hypoxia, viral infection, and traumatic injury (Leventer et al., 2008;Blümcke et al., 2011). DCMs associated with neonatal insult (polymicrogyria and FCD Type IIId) may be more amenable to therapeutic intervention, as their precipitating insult is often rapidly clinically documented.
To better understand the molecular and cellular changes associated with early life insult-induced cortical network reorganization, we used the neonatal freeze-lesion (FL) model. FL recapitulates key features of polymicrogyria and FCD Type IIId (Luhmann, 2016) including insultinduced etiology, cell death, astrocytosis, anatomic reorganization, and spontaneous seizures (Sun et al., 2016;Williams et al., 2016). Additionally, the FL model allows study of the changes that occur between insult and network dysfunction in a stereotyped model. Briefly, on the day of birth (postnatal day 0, P0), a freezing probe is applied to the exposed skull, generating a hypoxic insult. This results in cell death and the formation of a microgyrus and a surrounding paramicrogyral zone (PMZ). The PMZ is hyperexcitable and generates epileptiform network activity beginning approximately two weeks after injury (Jacobs et al., 1996;Jacobs et al., 1999;Andresen et al., 2014). Identifying the molecular, cellular, and network level changes that lead to brain dysfunction after neonatal cortical insult are paramount to developing novel therapeutic approaches.
Recently, we reported that thrombospondin (TSP)/␣2␦-1 signaling is increased in the cortex following FL (Andresen et al., 2014). TSP promotes excitatory synaptogenesis during development (Christopherson et al., 2005) by binding to ␣2␦-1 (Eroglu et al., 2009). ␣2␦-1 acts as an auxiliary subunit of voltage gated calcium channels, but its synaptogenic function appears to be independent of calcium channel association (Eroglu et al., 2009). Recent genetic studies have implicated ␣2␦-1 mutations in human polymicrogyria (Vergult et al., 2015). Of note, gabapentin (GBP) blocks the interaction of TSP and ␣2␦-1 and reduces ␣2␦-1-mediated excitatory synaptogenesis (Eroglu et al., 2009). There is reason to suspect that brain injury reactivates TSP/␣2␦-1, leading to hyperexcitation. In vivo GBP treatment attenuates injury-induced increases in synaptogenesis (Li et al., 2012;Li et al., 2014;Takahashi et al., 2017) following adult brain injury, and reduces excitatory synaptic transmission and cortical hyperexcitability following FL (Andresen et al., 2014). Interestingly, GBP is also neuroprotective in models of spinal cord injury Kale et al., 2011), cortical trauma (Li et al., 2012) and stroke (Traa et al., 2008;. It is unknown, however, whether ␣2␦-1 activity contributes to cell death or if GBP's neuroprotective effects are mediated by its actions on ␣2␦-1. In this study, we examined the role of ␣2␦-1 in FLassociated pathologies (cell death, cortical reorganization, astrocytosis, excessive synaptic excitation, and network hyperexcitability) and GBP-mediated neuroprotection. We used both pharmacological blockade and genetic deletion of ␣2␦-1 and found that both reduced apoptotic cell death following FL. Attenuating ␣2␦-1 signaling pharmacologically or genetically helped restore normal cortical structure following FL, reduced astrocyte reactivity, and attenuated FL-induced increases in excitatory synapse number and glutamatergic synaptic transmission. The vast majority of GBP's effects were absent in mice lacking ␣2␦-1, confirming that GBP acts largely through ␣2␦-1. Our studies of synapse number, however, suggest GBP may also have ␣2␦-1-independent effects. Additionally, cortical hyperexcitability is only partially rescued in FL mice deficient in ␣2␦-1, consistent with other pathways contributing to dysfunction following neonatal cortical insult. To enhance the rigor of this study we also employed a linear mixed model statistical approach and found that GBP and ␣2␦-1 have robust effects on FL pathology. Overall, our study suggests that ␣2␦-1 may be involved in driving multiple pathologies associated with insult-induced DCMs, as well as in mediating the neuroprotective effects of GBP.

FL
Experimental microgyri in primary somatosensory cortex (right hemisphere) were induced in P0 WT or ␣2␦-1 KO mouse pups by freeze lesioning as described previously (Andresen et al., 2014). Briefly, animals were anesthetized by hypothermia, an incision into the scalp was made, and a copper probe cooled to Ϫ50°C to Ϫ60°C was placed onto the exposed skull for 5 secs. Sham operated littermates were generated by leaving the probe at room temperature. After freeze-lesioning, the incision was closed using surgical glue, and pups were warmed and returned to the dam. All FL surgeries were performed by the same experimenter to reduce variability in the procedure, who was blinded to the genotype/treatment group.

Drug treatments
For drug treatment experiments, mice were treated with once daily intraperitoneal injections of either GBP (200 mg/kg) or vehicle (sterile injection saline) from P1 to P7.

Behavioral tests
Ten-week-old male mice were used for behavioral testing. Behavioral assays were performed in the Tufts Center for Neuroscience Research Animal Behavior Facility. Animals were tested on a battery of tests, starting with the least stressful test (open field) and moving to more stress-ful tests (spontaneous alteration and rotarod) on subsequent days.

Open field test
Mice were individually placed into the center of a 40 ϫ 40 cm open field apparatus with 16 ϫ 16 equally spaced photocells (Hamilton-Kinder). The total distance traveled, and the amount of time spent in the center of the open field, as well as the total number of beam breaks, were measured during the 20-min test using MotorMonitor software (Hamilton-Kinder).

Spontaneous alteration
Mice were placed in a three arm T-maze. After being placed in the starting arm, the barrier was removed to allow the rodent free choice to enter either the left or right arm. After entering an arm, a barrier was placed to keep the animals in that arm for 30 s. The mouse was returned to the home cage for 1 min between trails and a total of 10 trials were performed. Generally, rodents will engage in exploratory behavior in the maze and alternate between the arms. The percentage of trials in which the animal spontaneously alternated arms was reported.

Rotarod
Motor coordination and balance was measured in the mice using an acceleration paradigm on a five position rotarod (ENV 577M, Med Associates). Mice were trained at a constant, slow speed of 14 rpm until they could successfully stay on the beam for 5 min. Testing was conducted using a fast acceleration paradigm (4Ϫ40 rpm over 5 min). Mice were tested over the course of three trials (30 min apart) and latency to fall and the RPM at which mice fall was reported.

Electroencephalogram (EEG) recording
Naïve, 10-week-old WT and ␣2␦-1 KO littermates were implanted for EEG recording based on a Tufts Institutional Animal Care and Use Committee approved protocol. Briefly, animals were anesthetized (100 mg/kg ketamine and 10 mg/kg xylazine). An incision was made on the scalp, the surface of the skull was dried, and four burr holes were drilled with care to avoid puncturing the dura. Four 0.1Љ stainless steel screws (Pinnacle Technologies) were gently screwed into the drilled holes. A common reference electrode was placed anterior to the left side of the intraparietal bone, and two EEG electrodes were placed in the left front bone (for front/motor cortex) and in the right parietal bone (for somatosensory cortex). A separate ground electrode was also placed anterior to the right side of the intraparietal bone. The screws were then attached to a headmount (Pinnacle Technology), which was fixed using dental cement. Following surgery, the animals were given 7 d to recover. Chronic 24-h EEG recordings were acquired using a 100ϫ gain preamplifier high-pass filtered at 1.0 Hz (Pinnacle Technology) with video monitoring for two weeks. LabChart Pro software (ADInstruments, RRID:SCR_001620) was used for data acquisition and analysis.

Preparation of acute brain slices
Cortical brain slices containing sensorimotor cortex (400 M) were prepared from mice of either sex. Briefly, mice were anesthetized with isoflurane, decapitated, and the brains were rapidly removed and placed in chilled (4°C) low-Ca, low-Na slicing solution consisting of 234 mM sucrose, 11 mM glucose, 24 mM NaHCO 2 , 2.5 mM KCl, 1.25 mM NaH 2 PO 4 , 10 mM MgSO 4 , and 0.5 mM CaCl 2 , equilibrated with a mixture of 95% O 2 /5% CO 2 . The brain was glued to the slicing stage of a Vibratome 3000 sectioning system and slices were cut in a coronal orientation. The slices were then incubated in 32°C oxygenated aCSF (126 mM NaCl, 2.5 mM KCl, 1.25 mM NaH 2 PO 4 , 1 mM MgSO 4 , 2 mM CaCl 2 , 10 mM glucose, and 26 mM NaHCO 2 ) for 1 h, and then allowed to cool to room temperature and subsequently used for recording.

Field recordings
Slices were placed in an interface chamber maintained at 34°C, superfused with oxygenated aCSF at 2 ml/min and cortical projections were stimulated with a tungsten concentric bipolar electrode at the Layer VI, white matter boundary. Electrical stimulation consisted of 10 -50 A, 100-s pulses at 30-s intervals delivered by a stimulus isolator (World Precision Instruments). Glass micropipettes (resistance Х 1 M⍀) were filled with aCSF and placed in Layer V ϳ200 m from the site of injury in the PMZ of FL animals (or comparable area of cortex in sham-injured animals) directly above the stimulation electrode. Electrophysiological data were recorded with an Axon Multiclamp 700A amplifier and Digidata 1322A digitizer (sampling rate ϭ 20 kHz) with Lab Chart software (AD Instruments, RRID:SCR_001620). Threshold stimulation intensity was identified as the minimum amount of current required to elicit a detectable cortical field potential response (Ն0.05 mV).

Whole-cell patch-clamp recordings
Slices were placed in the recording chamber of an Olympus Bx51 microscope with continual superfusion of oxygenated aCSF maintained at 32°C (2 ml/min). Layer V pyramidal neurons were visually identified with infrared differential interference contrast microscopy and wholecell patch-clamp recordings were made with a borosilicate glass electrode (3-5 M⍀) filled with 140 mM CsMs, 10 mM HEPES, 5 mM NaCl, 0.2 mM EGTA, 5 mM Qx314, 1.8 mM MgATP, and 0.3 mM NaGTP; pH 7.25. The recording electrode was placed ϳ200 m from the site of injury in the PMZ of FL animals, or comparable cortical area in sham-injured animals. Data were collected using an axon Multiclamp 700B amplifier, Digidata 1440A digitizer and pClamp software. Miniature EPSCs (mEPSCs) were recorded at a holding potential of Ϫ70mV in the presence of 1M TTX. Only recordings with an access resistance that varied Ͻ20% were accepted for analysis.

Electrophysiological data analysis
Field recordings were analyzed using Lab Chart (AD Instruments, RRID:SCR_001620), pClamp (Molecular Devices, RRID:SCR_011323) and MATLAB software (RRID: SCR_001622). Traces were recorded at threshold stimu-lation, the minimum stimulation required to elicit a detectable response. Each sweep was analyzed for epileptiform activity to calculate the percentage of epileptiform activity per slice. The area under the curve was used to determine the integrated network activity and was calculated by integrating the extracellular field potential during the first 1000 ms following initial stimulation. mEPSC recordings were analyzed using Clampfit (Molecular Devices, RRID: SCR_011323) and Mini Analysis (Synaptosoft, RRID: SCR_002184). Recordings of 60 -120 s were analyzed for average amplitude and interevent frequency.

Synapse counting
Tissue preparation, staining and analysis was adapted from the protocol from Ippolito and Eroglu (Ippolito and Eroglu, 2010). Brains were fixed as above and sectioned at 14 m. Brain sections were blocked using 20% normal goat serum in PBS for 1 h at room temperature. Vglut1 (1:2500, Millipore, RRID:AB_2301751), PSD95 (1:500, Invitrogen, RRID:AB_2533914), and GFAP (1:1000, Abcam, RRID:AB_296804) were diluted in PBS with 0.3% Triton X-100 and 10% normal goat serum. Cortical sections were incubated with diluted primary antibodies for 42 h at 4°C. Appropriate secondary antibodies (The Jackson Laboratory) were diluted in PBS with 0.3% Triton X-100 and 10% normal goat serum and added to cortical sections for 2 h at room temperature. Slices were mounted using fluoromount-G (Southern Biotech) and imaged with a Nikon A1R confocal microscope. Images were collected with a 100ϫ oil immersion objective and maximum intensity projections (MIPs), of three serial optical sections at 0.175-m steps, were generated. Five MIPs were generated and analyzed per section. These were averaged to give one value per brain section. Synapses were quantified in ImageJ (RRID:SCR_003070) using the Puncta Analyzer plug in.

Terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assay
Brains were prepared by transcardial perfusion with PBS followed by overnight fixation with 4% paraformaldehyde. Fixed brains were sectioned at 40 m using a Thermo Fisher Microm HM 525 cryostat. TUNEL was detected using an ApopTag peroxidase in situ apoptosis detection kit (Millipore). Briefly, endogenous peroxidase activity was blocked by incubation of the sections with 3% H 2 O 2 for 5 min. TdT enzyme provided by the kit was used to label dUTP nick ends with DAB as the substrate. Nuclei were counterstained with methyl green and slices were mounted using Permount (Fisher Scientific). Tissue sections were evaluated by conventional bright-field microscopy and analyzed in ImageJ. The color deconvolution plug-in was used to isolate the DAB component of the staining. TUNEL-positive cells were quantified using the puncta analyzer in ImageJ. A uniform threshold and puncta size was applied across all images. The density of TUNEL-positive cells was quantified based on the number of TUNEL-positive cells in the region of interest which spanned from pial surface to the cortex white matter border and ϳ500 m on either side of the site of lesion.

Experimental design and statistical analysis
For all studies, control experiments were performed including sham injury and age-matched vehicle treatment. Additionally, WT littermates were used as controls in experiments testing the effect of genetic deletion of ␣2␦-1. Multiple litters were used for each experiment to control for litter to litter differences. GBP versus vehicle and sham versus FL was randomized by litter, ensuring a mix of treatments and genotypes throughout all studies. The distribution of genotypes followed a Mendelian distribution. The sex of the mice was ϳ50% male and 50% female. Animals were randomly assigned to one of four treatment groups. The experimenter was blind to genotype at the time of FL and drug treatment. Data analysis was not blinded, but many assays use automated, quantitative analysis which helps to reduce potential bias in many experiments. The majority of experiments in this study had three between-subject factors being tested: sham versus FL, WT versus ␣2␦-1 KO, and vehicle versus GBP treatment. Each experimental group for electrophysiology and immunohistochemistry experiments consisted of three to four animals, with eight different experimental groups (WT sham ϩvehicle, KO sham ϩvehicle, WT sham ϩGBP, KO sham ϩGBP, WT FL ϩvehicle, KO FL ϩvehicle, WT FL ϩGBP, and KO WT ϩGBP). Exact animal number and slice/cell number is noted in the results. Sample size has been calculated to ensure power Ͼ0.8 for ␣ ϭ 0.05. Power was calculated using G‫ء‬Power Software. All experiments used animals of the same age between groups.
For comparison between two experimental groups, a Student's t test was used. To control for multiple comparisons (with three or more experimental groups), a oneway ANOVA was used. For comparison of cumulative probabilities, a Kolmogorov-Smirnov test (KS test) was used. For all experiments, we compared WT FL to WT Sham, WT FL to WT FL ϩ GBP, WT FL to KO FL, and KO FL to KO FL ϩ GBP. To control for multiple comparisons, a Holm-Bonferroni correction was used. All data are presented as box-whisker plots Ϯ SEM, with average values indicated by X on the plots. Exact P values are indicated in the results, unless they were Ͻ0.001; values of p Ͻ 0.05 for ␣ ϭ 0.05 or less were considered statistically significant. OriginLab and MATLAB software was used to perform the statistical analyses.
In addition, to perform more rigorous statistical analysis linear mixed-effects models were fitted with random intercepts to account for the correlation between repeated measurements on the same mouse. To standardize the modeling procedure and facilitate comparison across the experiments, fixed effects always included main effects for FL, drug treatment, and genetic KO of ␣2␦-1, as well as interaction terms between FL and drug treatment, FL and genetic KO of ␣2␦-1, and drug treatment and genetic KO of ␣2␦-1. In other words, sham, vehicle, and WT served as reference categories. t values Ͼ 1.96 and Ͻ Ϫ1.96 we considered to be statistically significant.

Drugs and reagents
All salts and glucose for buffers were obtained from Sigma-Aldrich. GBP was obtained from Abcam.

Pharmacological attenuation of ␣2␦-1 signaling decreases cell death following FL
GBP treatment has been shown to reduce cell death in models of posttraumatic epilepsy (Li et al., 2012) and stroke (Traa et al., 2008), suggesting that ␣2␦-1 signaling may contribute to insult-induced cortical cell death. Here, we tested whether attenuating ␣2␦-1 signaling via in vivo GBP treatment was neuroprotective in the FL model. As described in previous studies, FL was performed in C57BL/6J mice on the day of birth (P0). To identify dead and dying cells in the neonatal cortex, we used TUNEL labeling and activated caspase-3 immunolabeling (Watanabe et al., 2002). We found that FL induced significant cell death in the cortex at P7 ( Fig. 1; caspase-3 immunoreactivity: sham, 4.1 Ϯ 1.0 cells per 0.1 mm 2 , n ϭ 11 sections from four animals; FL, 61.5 Ϯ 4.4 cells per 0.1 mm 2 , n ϭ 10 sections from three animals; Mann-Whitney, U (19) ϭ 0, p Ͻ 0.001; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.001; TUNEL: sham, 0.5 Ϯ 0.2 cells per 0.1 mm 2 , n ϭ 7 sections from three animals; FL, 59.1 Ϯ 16.6 cells per 0.1 mm 2 , n ϭ 8 sections from four animals; Mann-Whitney, U (13) ϭ 0, p Ͻ 0.001; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.001). The majority of cell death was seen at the site of FL and in the surrounding cortical tissue. Other developmental time points were examined and the peak expression of cell death markers was seen at approximately P5-P7 (data not shown). Therefore, the effects of ␣2␦-1 signaling on cell death were quantified at P7.

Genetic deletion of ␣2␦-1 attenuates FL-induced cell death
Next, we examined whether genetic deletion of ␣2␦-1 recapitulated GBP's neuroprotective effects after FL. Mice lacking the ␣2␦-1 gene CACNA2D1 (␣2␦-1 KOs) were generated and the absence of the ␣2␦-1 protein was confirmed by Western blot analysis of total cortical homogenate (data not shown). These mice showed no obvious phenotype and developed normally with regard to size and weight. ␣2␦-1 KO mice were recently reported to have a higher risk of developing diabetes due to changes in pancreatic ␤-cell function (Mastrolia et al., 2017). ␣2␦-1 KOs showed no change in open field activity or time spent in the center of the open field and performed normally on the rotarod and alternating T-maze test (data not shown). No behavioral or electrographic seizures were detected in ␣2␦-1 KOs subjected to two weeks of 24/7 video/EEG monitoring (data not shown).
Attenuating ␣2␦-1 signaling reduces FL-induced cortical malformation Because genetic and pharmacological inhibition of ␣2␦-1 signaling reduced cell death, we hypothesized that similar manipulations would decrease FL-induced cortical reorganization. To test this hypothesis, we examined anatomic changes induced by FL in GBP-treated and ␣2␦-1 KO animals. Because FL creates a 3-dimensional cortical malformation, we prepared serial cortical brain sections from P28 FL animals that encompassed the entire FL (Fig.  2D). We identified areas of malformation, characterized by large-scale cortical disorganization, by examining NeuN (marker of neurons) immunolabeled sections. The area of malformed, or missing, cortex was quantified throughout the reconstructed cortex. When FL resulted in loss of cortical tissue, this loss was quantified by drawing a line continuous with the adjacent pial surface and calculating the area of predicted tissue loss. The area of missing cortex was added to the area of malformed cortex to calculate lesion area per slice. The area was calculated in serial sections through the volume of the lesion and used to estimate the total lesion volume. Consistent with our cell death data, the volume of the lesion was decreased in FL animals treated with GBP from P1 to P7 ( Fig. 2A,B; FL: 1.65 Ϯ 0.13 mm 3 , n ϭ three animals vs FL ϩ GBP: 1.25 Ϯ 0.03, n ϭ three animals; one-way ANOVA, F (3) ϭ 8.01, p ϭ 0.003; post hoc Bonferroni test, ␣ ϭ 0.05). ␣2␦-1 KO mice also showed reduced lesion size compared to WT FL ( Fig.  2A,B; 1.08 Ϯ 0.13 mm 3 , n ϭ 4 and 1.16 Ϯ 0.06 mm 3 , n ϭ 4 for vehicle and GBP-treated animals, respectively; post hoc Bonferroni test, ␣ ϭ 0.05). These results demonstrate that reducing ␣2␦-1 signaling with either GBP treatment or ␣2␦-1 KO reduces the areas of cortical malformation following FL. Similar to cell death data, treating ␣2␦-1 KO animals with GBP did not result in additional reductions in cortical malformation ( Fig. 2A,B; 1.16 Ϯ 0.06 mm 3 , n ϭ 4). This again suggests that GBP's neuroprotective effects are mediated by ␣2␦-1.

FL induces ␣2␦-1-dependent laminar reorganization of the cortex
In order to better understand FL-induced anatomic changes, we examined the expression of layer specific markers in the cortex at P28. CUX1 and CTIP2 are transcription factors that preferentially label superficial (LII/III) and deep (LIV/VI) cortical layers, respectively (Insolera et al., 2014). Consistent with published data, CTIP2ϩ cells are abundant in deep cortical layers ( Fig. 3A; 588 Ϯ 52 per mm 2 , n ϭ 11 sections from four animals), and found sporadically in outer layers ( Fig. 3A; 26 Ϯ 5 per mm 2 ) in the sham-injured cortex. Conversely, CUX1ϩ cells are more abundant in the outer layers and are sparser in deep layers ( Fig. 3A; 743 Ϯ 31 vs 119 Ϯ 26 per mm 2 ). In contrast to the normal six-layered cortex, FL creates a four-layered MZ. Layer i is continuous with cortical Layer I (lower case ϭ FL nomenclature, upper case ϭ standard cortical layer nomenclature), and is the superficial most layer. It is thin and contains few cell bodies. Layer ii is neuron-rich and makes up the bulk of the MZ (Dvorák and Feit, 1977) and contains many cell bodies. Layer iii is a neuron-free area, and makes a clear border between superficial and deep cortical layers. Layer iv is identified as the deepest layer of cell bodies above the white matter. We defined layer i and ii as the superficial layers of the MZ, and layer iv as the deep layer. Studies postulate that neurons in MZ layer ii are similar to Layer II/III neurons and that Layer IV-VI cortical neurons are lost, but the expression of layer-specific markers following FL has not been examined. We found that layer ii of the MZ is predominately composed of CUX1ϩ cells and layer iv of CTIP2ϩ cells, in line with previous predictions. Interestingly, there is a decreased density of CUX1ϩ cells in MZ layer ii (Fig.  3A,B; 285 Ϯ 34 per mm 2 , n ϭ 7 sections from three animals; t test, t (16) ϭ 9.40, p Ͻ 0.001, compared to sham-injured Layer II/III; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.001) and CTIP2ϩ cells in MZ layer iv (Fig. 3A,C; 240 Ϯ 31 per mm 2 ; t test, t (15) ϭ 6.47, p Ͻ 0.001, compared to Layer V-VI; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.001). Additionally, the ratio of superficial layer (layer ii or II/III) to deep layer (iv or VVI) is significantly higher in FL MZ cortex ( Fig. 3D; 7.26 Ϯ 1.87, n ϭ 7 sections from three animals) compared to sham-injured cortex (1.12 Ϯ 0.06, n ϭ 11 sections from four animals; t test, t (16) ϭ Ϫ4.17, p Ͻ 0.001; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.001) reflecting the loss of deep layer cells and concurrent expansion of the outer layer in FL cortex.
Based on these findings, we next examined how attenuating ␣2␦-1 signaling affects layer-specific changes in the MZ. Sham and FL animals were treated with GBP, as described above, and CTIP2ϩ and CUX1ϩ cells were identified at P28. GBP treatment restored the density of CUX1ϩ cells in the outer layers to sham-injured levels (Fig. 3A,B; 668 Ϯ 32 per mm 2 , n ϭ 21 sections from four animals; Mann-Whitney, U (23) ϭ 0, p Ͻ 0.001, compared to vehicle-treated FL; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.001) and significantly increased the density of deep layer CTIP2ϩ cells (Fig. 3A,C; 111 Ϯ 14 per mm 2 ; t test, t (26) ϭ Ϫ3.43, p Ͻ 0.001, compared to vehicle-treated FL; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.001). Consistent with a significant reduction of lesion volume, FL animals treated with GBP have a significantly smaller superficial/deep layer ratio of 2.93 Ϯ 0.37 ( Fig. 3D; n ϭ 21 sections from four animals; t test, t (26) ϭ 3.53, p Ͻ 0.001; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.001) and the MZ more closely resembles the normal six-layered cortex.

Total neuron number is normal in the FL cortex
Because the expression of CUX1 and CTIP2 was altered by FL, we wondered whether neuronal density was also affected. To test this, we performed NeuN staining and counted the total number of neurons in the cortex. Confocal images of the outer and deep cortical layer were used to determine neuronal density, in sham-injured cortex and in the MZ and PMZ of the FL cortex. In comparison to the dramatic changes found in CUX1 and CTIP2 staining following FL, no significant reduction in total neuron number was observed (data not shown). These results suggest that overall neuronal density is normal in the FL cortex at P28, despite changes in CUX1ϩ and CTIP2ϩ cell density.

Attenuating ␣2␦-1 signaling inhibits FL-driven increases in synaptic excitation
Given the changes in excitatory synapse number, we next assessed the role of ␣2␦-1 signaling on synaptic excitation by examining mEPSCs in the FL cortex. mEP-SCs were recorded from Layer V pyramidal neurons ϳ200 m from the site of injury in the PMZ of FL animals, or comparable cortical area in sham-injured animals. Neurons were held at -70 mV and perfused with normal aCSF containing 1 m TTX. We found no significant differences in resting membrane potential, rise or decay kinetics, or the amplitude of the mEPSC events between groups (Fig.  5B). Similar to our previous study (Andresen et al., 2014), we found that FL induced a rise in mEPSC frequency (Fig.  5A,C; 0.79 Ϯ 0.19 Hz, n ϭ 14 cells from three animals; Mann-Whitney, U (20) ϭ 22.5, p ϭ 0.002; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.05) compared to sham-injured animals (0.23 Ϯ 0.05 Hz, n ϭ 8 cells from three animals). GBP treatment significantly reduced the mEPSC frequency (0.29 Ϯ 0.09 Hz, n ϭ 13 cells from three animals; Mann-Whitney, U (25) ϭ 142.5, p ϭ 0.013; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.05). ␣2␦-1 deletion led to a non-significant reduction in mean mEPSC frequency (0.30 Ϯ 0.04 Hz, n ϭ 10 cells from three animals; Mann-Whitney, U (22) ϭ 103, p ϭ 0.055) following FL. GBP treatment had no effect on mEPSC frequency in ␣2␦-1 KO mice. The cumulative probability distribution of the interevent interval (IEI) was also calculated to more rigorously assess changes in mEPSC frequency. The leftward shift in the IEI cumulative probability in WT FL cortex, compared to WT sham, shows an increased frequency of excitatory synaptic events following FL (Fig. 5D; KS test, D (21) ϭ 0.71, p ϭ 0.0002; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.001. The IEI cumulative probability is shifted rightward in both GBP treated WT FL and ␣2␦-1 KO FL animals (KS test,D (26) ϭ 0.82, p ϭ 0.0001 and KS test, D (22) ϭ 0.64, p ϭ 0.0009 compared to WT FL; Holm-Bonferroni correction for multiple comparisons, ␣ ϭ 0.001), demonstrating that both GBP treatment and ␣2␦-1 deletion reduces synaptic excitation.

Linear mixed effects modeling of data to increase statistical rigor
Based on the call to increase rigor and reproducibility in preclinical studies, we performed a more advanced statistical analysis of all our results. The goal of this secondary level of analysis was to identify results that may require further studies to ensure their robustness and reproducibility. We employed a linear mixed effects model (Laird and Ware, 1982) that (1) includes mouse-specific random intercepts to account for correlation due to multiple measures in the same mouse, (2) allows us to look at the interactions between manipulations (FL, drug treatment, and genetic manipulation), and (3) encompasses the entire study design within a single statistical approach. For all studies described above, regression coefficients from the linear mixed-effects models were calculated (Table 1) and indicate the degree to which different experimental conditions alter the mean outcomes for each assay. The GBP main effect, for example, measures the change in mean outcome after GBP treatment specifically in sham-injured WT mice, though the effect also applies to all experimental groups that received GBP. The FL/GBP interaction term, on the other hand, measures the additional difference in the effect of GBP in FL versus sham-injured mice. Results varied by experiment, but generally agree with statistical analysis performed using Student's t test/Mann-Whitney tests with a Bonferroni correction for repeated measures. The most robust effects based on linear mixed effects modeling were the (1) FL main effect, and the interactions among (2) FL/GBP, (3) FL/KO, and (4) GBP/KO. This indicates that (1) FL has robust effect across assays, (2/3) GBP and ␣2␦-1 KO have important effects in FL animals, and that (4) the effects of GBP are largely lost in ␣2␦-1 KO mice. Importantly, there are some results that have less statistical significance based on the linear mixed effects model. Less robust results include the effects of ␣2␦-1 KO on Figure 6. ␣2␦-1 KO partially rescues network hyperexcitability following FL and is insensitive to GBP treatment. A, Representative evoked cortical field EPSP traces recorded from the PMZ of acute cortical slices prepared from P21 to P28 WT sham, ␣2␦-1 Ϫ/Ϫ sham, WT FL, ␣2␦-1 Ϫ/Ϫ FL, WT FL ϩ GBP, and ␣2␦-1 Ϫ/Ϫ FL ϩ GBP animals. B, Box-whisker plot of percentage epileptiform sweeps per slice, ‫␣ءءء‬ ϭ 0.001 (Holm-Bonferroni multiple-comparison correction). C, Box-whisker plot of integrated area under the curve per slice, ‫␣ء‬ ϭ 0.05, ‫␣ءء‬ ϭ 0.01, and ‫␣ء‬ ϭ 0.05 (Holm-Bonferroni multiple-comparison correction). circuit hyperexcitability after FL, the effects of GBP treatment on the recovery of CTIP2, and the effects of ␣2␦-1 KO on the recovery of CUX1. These specific findings may warrant further study to ensure their robustness before moving forward with any clinical applications. In general, this additional statistical analysis shows that GBP and ␣2␦-1 have robust effects across multiple FL-associated pathologies.

Discussion
Treatment with GBP, a clinically used anticonvulsant and antagonist of ␣2␦-1, attenuates pathologic changes following experimental neonatal and adult brain insult (Traa et al., 2008;Li et al., 2012;Andresen et al., 2014). ␣2␦-1 (a calcium channel auxiliary subunit) and its synaptogenic ligand, TSP, are increased following brain trauma, suggesting injury-induced activation of ␣2␦-1 signaling. Defining the precise role of ␣2␦-1 in the pathology of brain injury has been difficult, however, as GBP is known to directly bind to both ␣2␦-1 and ␣2␦-2, and has indirect effects on other systems (NMDA receptors; Kim et al., 2009), sodium channels (Yang et al., 2009), and protein kinase C (Maneuf and McKnight, 2001;Yeh et al., 2011). Here, we used mice lacking ␣2␦-1 to establish its role in pathology associated with FL, and to determine the site of action of GBP's neuroprotective effects. We show, for the first time, that ␣2␦-1 plays a direct role in cell death, anatomic reorganization, synaptogenesis, astrocytosis, expression of layer-specific cortical markers, and network hyperexcitability in the FL model of cortical malformation. By genetically deleting ␣2␦-1, we established that it mediates GBP's neuroprotective effects. Our findings show ␣2␦-1 has an important role in injury-induced cell death and synaptogenesis, and suggest that ␣2␦-1 activity may be involved with pathogenesis associated with acquired DCMs.
GBP treatment is neuroprotective in models of cortical deafferenation (Li et al., 2012), neonatal ischemia (Traa et al., 2008), and ischemic brain injury (Williams et al., 2006), likely due to attenuation of TSP/␣2␦-1 signaling. Human studies show elevated TSP-1 in plasma after head Table 1. Linear mixed effects modeling table of coefficient, SE, and t value calculated from the linear mixed effect model for data from: TUNEL assay (Fig. 1), caspase-3 staining (Fig. 1), CUX1 and CTIP2 staining (Fig. 3), superficial/deep layer area ratio (Fig. 3), synapse counting (Fig. 4), fEPSP %epileptiform (Fig. 6), and fEPSP area (Fig. 6 trauma (Wang et al., 2016), human endothelial cells upregulate TSP-1 after oxidative stress (Ning et al., 2011), and TSP1 and TSP2 are increased in rodent models of stroke (Lin et al., 2003). Likewise, ␣2␦-1 expression is upregulated in injury models in both the central and peripheral nervous system (Boroujerdi et al., 2008;Li et al., 2012;Andresen et al., 2014), leading to an overabundance of both ligand and receptor. Beyond changes in expression, ␣2␦-1's subcellular localization is altered in areas of neuronal death following chemoconvulsive seizures , suggesting subtle changes in ␣2␦-1 signaling may be relevant to cell death and epileptogenesis. How ␣2␦-1 signaling leads to cell death is not clear, but calcium-dependent caspase-mediated apoptosis, appears to be involved. Loss of ␣2␦-1 decreases the duration of calcium channel currents in cardiac myocytes (Fuller-Bicer et al., 2009) and calcium channel density (Patel et al., 2013) in sensory neurons. Conversely, over expression of ␣2␦-1 increases neurotransmitter release following action potential firing in the hippocampus (Hoppa et al., 2012), which could leader to greater excitation. We see two ways in which ␣2␦-1 might contribute to injury-induced cell death. First, by increasing excitatory synapse formation and enhancing excitation, ␣2␦-1 activity likely potentiates glutamate-induced excitotoxicity. Second, by enhancing calcium currents and intracellular calcium levels, ␣2␦-1 may promote caspase activation. Both aspects of ␣2␦-1 may contribute to lesion-induced cell death, and undoubtedly occur in parallel with other forms of cell death that do not involve ␣2␦-1. Future studies will be required to identify which aspects of ␣2␦-1 function lead to injury-induced cell death and if certain cell types are more vulnerable to ␣2␦-1-mediated cell death.
In addition to driving cell death, FL also induces robust astrocytosis (Bordey et al., 2001;Campbell and Hablitz, 2008;Dulla et al., 2013;Armbruster et al., 2014). Here, we show that ␣2␦-1 signaling plays an important role in driving reactive astrocytosis. GFAP immunoreactivity was greatly reduced by attenuating ␣2␦-1 signaling, and no additional effect of GBP was seen in ␣2␦-1 KO animals. Although astrocytes express ␣2␦-1 at a low-level, it is much more abundantly expressed by neurons (Zhang et al., 2014). This suggests that decreased astrocyte reactivity may be a secondary effect of reduced cell death and subsequent neuroinflammation, rather than a direct effect on ␣2␦-1 expressed by astrocytes. How ␣2␦-1 contributes more broadly to neuroinflammation, macrophage lineage cells, neuro-immune interactions, and cell death signaling warrants future investigation.
␣2␦-1 activity following FL also promotes lesion formation and disrupts cortical lamination (shown by CUX1 and CTIP2 staining), in both the MZ and PMZ. Previous reports describe the neuronal architecture of the PMZ as normally laminated. Our results confirm that although gross level changes in cortical structure and total neuron number do not occur in the PMZ, more subtle differences may exist, as indicated by the reduction in the density of in CUX1ϩ and CTIP2ϩ neurons. How this change is linked to FL-induced pathologies is unknown. CTIP2 or CUX1 are both transcription factors (Molyneaux et al., 2007;Sansregret and Nepveu, 2008) that control the expression of proteins related to neuronal morphologic development (Arlotta et al., 2005;Li et al., 2010). Whether changes in the density of CUX1ϩ and CTIP2ϩ neurons are associated with known changes in synaptic connectivity or neuronal morphology remains to be seen.
In addition to the established role of TSP/␣2␦-1 signaling in developmental synaptogenesis, our work also supports a role for ␣2␦-1 in pathologic synaptogenesis. ␣2␦-1 KO and in vivo treatment with GBP attenuated injuryinduced increases in both anatomic and functional synapses, in agreement with previous studies (Li et al., 2012;Andresen et al., 2014). Although functional synapses were not affected by GBP treatment in ␣2␦-1 KO mice, there was a minor reduction in the abundance of structural synapses. Other ␣2␦ subunits may mediate these effects. ␣2␦-1, ␣2␦-2, and ␣2␦-3 are all widely expressed in the CNS, with each isoform having a distinct distribution (Cole et al., 2005). Genetic deletion of ␣2␦-1 does not grossly alter the expression of other ␣2␦ isoforms (Fuller-Bicer et al., 2009), however, regional compensation or injuryspecific alterations may still exist and be relevant to FLinduced pathologies. GBP can bind to ␣2␦-2 (Gee et al., 1996) and ␣2␦-3 has been shown to play a synaptogenic role in drosophila (Kurshan et al., 2009). Taken together, these results suggest that other ␣2␦ isoforms may account for the GBP effect on synapse number in the ␣2␦-1-KO FL cortex. Additionally, it is unknown whether changes in other calcium channel auxiliary subunits occur following genetic deletion of ␣2␦-1 or neonatal injury. Also important to note, only vGlut1ϩ synapses, the predominant vGlut in the cortex, were quantified here. TSP/␣2␦-1 signaling drives the formation of vGlut2 containing synapses during normal development (Eroglu et al., 2009). Conversely, GBP treatment reduces injury-induced increases in vGlut1-containing excitatory synapses (Li et al., 2012). Our studies further support a role of ␣2␦-1 in driving vGlut1-containing synapse formation after injury, in addition to its role in driving vGlut2-containing synapses during development. This suggests injury increases intracortical excitation, mediated by vGlut1-containing synapses (Fremeau et al., 2004). Changes in vGlut2containing synapses, calcium channel trafficking (Hendrich et al., 2008), and homeostatic plasticity may underlie the discrepancy in the effects of GBP in ␣2␦-1 KOs on functional versus structural synapses.
Finally, inhibiting ␣2␦-1 signaling reduced cortical network hyperexcitability after FL. While genetic deletion of ␣2␦-1 attenuated FL-driven synaptogenesis, similar to GBP treatment, the effects on network hyperexcitability were intermediate to those seen with GBP treatment. Furthermore, GBP treatment had no effect on the residual epileptiform activity in the ␣2␦-1 KO animals, despite decreasing anatomic synapses. This suggests that constraining excitatory synaptogenesis is not sufficient to completely prevent hyperexcitability in the ␣2␦-1 KO cortex, and that deletion of ␣2␦-1 may cause compensatory changes that predispose the cortex to injury-induced network pathology. Genetic deletion of ␣2␦-1 did not alter the amplitude or kinetics of excitatory currents, which argues against changes in postsynaptic receptors. Therefore, presynaptic changes, via altered voltage-gated calcium channel function, or changes in neuroinflammatory response may promote hyperexcitability in the ␣2␦-1 KO cortex. Additionally, ␣2␦-1 KO may cause differences brain size, vascularization, neuronal morphology, and more, which may be relevant to FL-induced pathology.
To examine the robustness of our findings, we employed a linear mixed effect model, a statistical approach used in clinical studies. This approach revealed the effects of FL were extremely robust and the effects of GBP treatment and ␣2␦-1 KO in FL animals varied by assay, but were largely robust as well. Minor discrepancies between a standard statistical approach and the linear mixed effect model include the effects of GBP treatment and ␣2␦-1 KO on the expression of layer specific markers and of ␣2␦-1 KO on network hyperexcitability following FL. This approach illustrated areas that require more study before translational application; overall, however, the model confirmed that ␣2␦-1 signaling plays a critical role in much of the pathogenesis of the FL model.
Overall, our results show that significant ␣2␦-1-mediated pathologies occur in the FL model and a growing number of studies suggest that ␣2␦-1 signaling may contribute to the development of epilepsy. Overexpression of ␣2␦-1 in control animals leads to epileptiform EEG discharges and behavioral arrests, which are rescued by acute treatment with an anticonvulsant (Faria et al., 2017). The ␣2␦-1 ligand TSP-1 shows increased genetic variation in patients with idiopathic/genetic generalized epilepsies. Interestingly, TSP-1 mRNA levels are reduced in the ventrobasal thalamus of WAG/RiJ rats, a genetic model absence epilepsy (Santolini et al., 2017). Additionally, human mutations in ␣2␦-1 have been linked to polymicrogyria (Vergult et al., 2015). Whether cell death plays a role in these cases is unknown. Furthermore, the functional consequences of these mutations are not known, but gain of function mutations could increase calcium currents, increase cell death, and drive pathologic excitatory synaptogenesis. Also, although attenuating ␣2␦-1 signaling may reduce cell death and network hyperexcitability, it may compromise functional recovery. Reestablishing synaptic connectivity after injury is a critical aspect of rehabilitation and attenuating TSP1/2 signaling after stroke reduces behavioral recovery (Liauw et al., 2008). Care must be taken, therefore, in manipulating TSP/␣2␦-1 signaling, as it plays a role in both the adaptive and pathologic response to insult. Given these caveats, parsing apart the relationship between cell death, synaptogenesis, and epileptiform activity will enable us to optimize treatment strategies that target ␣2␦-1 signaling in neonatal injury and DCMs.