Pannexin 1 Regulates Network Ensembles and Dendritic Spine Development in Cortical Neurons

Visual Abstract

Several lines of evidence suggest that Panx1 could regulate the formation of neuronal networks or network ensembles, which are groups of spontaneously coactive neurons. Ensembles are emerging as the functional building blocks of cortical activity that underlie sensorimotor integration and learning and memory (for review see Harris et al., 2003;Miller et al., 2014;Carrillo-Reid et al., 2015;Arce-McShane et al., 2016). The formation of synapses plays a major role in the development of network ensembles, providing the structural basis for higher network connectivity (Jung and Herms, 2014; for review, see Hoshiba et al., 2017;Frank et al., 2018). In the rodent cortex, Panx1 transcript levels peak around the time of birth, and then markedly decline during the first four postnatal weeks (Ray et al., 2005;Vogt et al., 2005). This decrease in Panx1 coincides with the critical period for the formation of microscopic protrusions emanating from glutamatergic pyramidal neurons called dendritic spines (Schlaggar et al., 1993;for review, see O'Leary et al., 1994;Grutzendler et al., 2002;Trachtenberg et al., 2002;Hensch, 2004;Holtmaat et al., 2005), which receive the majority of excitatory inputs in the brain (for review, see Nimchinsky et al., 2002;Alvarez and Sabatini, 2007;Yuste, 2011). Panx1 regulates neurite growth (Wicki-Stordeur and Swayne, 2013) and interacts with collapsinresponse mediator protein 2 (Crmp2; Wicki-Stordeur, 2015;Xu et al., 2018), a stable synaptic protein (Heo et al., 2018) that regulates spine development (Zhang et al., 2016).
To understand how Panx1 regulates cortical neuron development, we used a multilevel approach involving analyses of network ensembles, synaptic protein expression and dendritic spines in mice with global and glutamatergic-neuron specific Panx1 KO. Panx1 KO cortical cultures showed increased network ensemble formation. Moreover, Panx1 KO cortical synaptosomes exhibited significantly increased expression of excitatory synapse markers (PSD-95, GluA1, and GluN2A) and significantly increased cortical neuron dendritic spine densities. Together our results suggest that Panx1 regulates network ensemble formation by acting as a brake for dendritic spine formation. tems (154534PK, ThermoFisher Scientific) coated with PDL (P6407, Millipore-Sigma) for MTT assays. The medium was replaced with Neurocult medium (STEMCELL Technologies, 05700) supplemented with SM1 and L-glutamine, P/S, and gentamicin (0.1 mg/ml; G1397, Millipore-Sigma). From 4 days in vitro onward, partial (half) the medium was replenished with new BrainPhys maturation medium (Bardy et al., 2015) supplemented with SM1 and Cytosine ␤-D-arabinofuranoside (ara-C, C1768, Millipore-Sigma) every third day.

Immunostaining and spiny protrusion analysis in cultured neurons
Primary cortical neurons were fixed in 4% EM-grade paraformaldehyde solution pre-warmed to 37°C for 10 min, washed in PBS and permeabilization with 0.25% Triton X-100 in PBS (PBST) for 10 min at RT, washed again with PBS and then blocked with 10% donkey serum (DS; 017-000-121, Jackson ImmunoResearch), 1% BSA, and 22.52 mg/ml glycine in PBST for 30 min at RT. Following blocking, cultures were incubated with primary antibodies in 1% BSA, and 5% DS in PBST overnight at 4°C, washed in PBS three times (10 min each), and incubated with secondary antibodies and Alexa Fluor 555 phalloidin (A34055, Invitrogen) in PBST supplemented with 1% BSA, and 5% DS for 1 h at RT. After three washes (10 min), coverslips were mounted on microscope slides using VECTASHIELD Antifade Mounting Medium (H-1000, Vector Laboratories). Hoechst 33342 (H3570, Invitrogen) was used as nuclear stain. For the analysis of spiny protrusions and PSD-95-positive dendritic spines, high-resolution (2048 ϫ 2048, pixel size 0.090 m) images of neurons were captured using a Leica SP8 confocal microscope (63ϫ/1.20 NA). The same acquisition parameters were maintained for all cells across all separate cultures within an experiment. Dendritic spines were defined as actin-enriched protrusions ranging from 0.4 to 10 m in length that emanated directly from the dendritic shaft. Using ImageJ, the longest dendrite of each cell was selected and defined as the primary neurite. Within the primary neurite, a 20 m segment from the distal tip of the primary neurite was traced and dendritic spines within the segment were traced with individual regions-of-interest (ROIs); spine density was defined as the number of spines per 10 m and was calculated by multiplying the total number of spines traced by 0.5. For cell-type characterization of neuronal cultures, coverslips were stained with the protocol described above and primary antibodies (MAP2, Gad67, and GFAP) were incubated overnight at 4°C, followed by three 10 min washes in PBS, secondary antibody and Hoechst incubation at room temperature, and three more 10 min washes before mounting the coverslips with VECTASHIELD. Images (1024 ϫ 1024, pixel size 0.568 m, 0.34 mm 2 ) were captured with a Leica SP8 confocal microscope (20ϫ/0.7 NA). The proportion of astrocytes and inhibitory cells were calculated based on GFAP and Gad67 immunoreactivity relative to the total amount of cells (MAP2-positive cells ϩ GFAP-positive cells). The proportion of excitatory cells was determined from MAP2-positive/Gad67-negative relative to the total amount of cells. Representative images were uniformly adjusted with Gaussian blur (2 pixels), and mild uniform adjustments to levels and contrast were made using Photoshop CS6 Extended suite (Adobe Systems).

Neuronal network analysis in primary cortical neuron cultures
For Ca 2ϩ imaging experiments, neuronal cultures 12-14 days in vitro were washed with HBSS and incubated in in BrainPhys maturation medium supplemented 4 M Fluo-4 AM (F14201, ThermoFisher Scientific) for 40 min at 37°C, 5% CO 2 , and 95% humidity. Coverslips were washed, transferred to a 35 mm dish containing BrainPhys without phenol red (05791, STEMCELL Technologies), and incubated in the dark for 30 min at 37°C, 5% CO 2 , and 95% humidity to allow for complete de-esterification of the probe. The dish was then mounted onto a heated chamber held at 37°C, 5% CO 2 and images were acquired every 5 s for 10 min (pixel dwell time 36 ns, streamed at 7.41 Hz, exposure/frame capture time 135 ms, 120 frames) with a laser-scanning microscope (Leica SP8) using 471 nm laser illumination (constant 5% laser power) and a 20ϫ objective (NA 0.70). Three fields-of-view (FOVs, 1024 ϫ 1024, pixel size 0.455 m) were analyzed per coverslip. ROIs were drawn around each soma within each FOV. The raw fluorescence intensity values over time within each ROI were extracted using the Leica Application Suite Software v3.1.3.16308 (Leica Microsystems); the background signal was determined in areas lacking neurons and was subtracted from all the intensity records; fluorescence intensity values were exported as .csv files. Two to three coverslips across three independent cultures were used for this analysis (WT ϭ 8 coverslips across 3 independent cultures; KO ϭ 7 coverslips across 3 independent cultures). Note that cells exhibiting a range in fluorescence intensity values limited to within 10% of maximal fluorescence intensity across the entire recording (i.e. fluorescence intensity of 90% of maximum or greater) were removed from all subsequent analyses, resulting in a total of 27/1044 cells removed across WT coverslips (2.9%) and 66/1155 cells removed across Panx1 KO coverslips (5.7%). This exclusion criteria was used to remove cells with abnormally high fluorescence values that could confound our analysis; however, note that it might potentially also eliminate cells with very small calcium transients, skewing our results toward more active cells. Then, the extracted .csv files were processed using the Fluorescence Single Neuron and Network Analysis Package, FluoroSNNAP (https://www.seas.upenn. edu/~molneuro/software.html; University of Pennsylvania), an open-source, interactive plugin for MATLAB (MATLAB R2014a, the MathWorks) for ⌬F/F 0 conversion from raw F data, spike probability inference, and network ensemble analysis. Once the raw fluorescence .csv file was imported, the analysis package generated a mock image or stack by randomly placing all the traced ROIs contained in the .csv file, which served to interact with the imported data by selecting individual ROIs and visualizing their time-varying traces. By selecting the option "Convert raw fluorescence data to deltaF/F" the difference in fluo-rescence (⌬F/F 0 ) was computed by taking the average of all the pixels within each ROI (raw fluorescence trace) and subtracting each value with the mean of the Ͻ50% values in the previous 10 frames (adjustable parameter), and then dividing that product by the mean of the lower 50% values in the previous 10 frames (Patel et al., 2015). Selection of the module "Infer underlying spike probability" calculated the spike probability of each individual ROI using a fast, non-negative deconvolution method developed by Vogelstein et al. (2010). This inferred spike probability algorithm represents neuronal activity better than ⌬F/F (Vogelstein et al., 2010;Miller et al., 2014). Network ensembles, defined as the group of co-activate neurons in a high-activity frame, were calculated by thresholding spike probability data to 3 SD above zero, determined from spike probabilities of the entire population in each FOV; this serves to identify active cells not confounded by noise. Values above the threshold were set to 1, and those below the threshold were set to 0. Then, these binary activity data were shuffled 1000 times to identify the statistically significant number of groups of coactive neurons, using a significant level of p Ͻ 0.05 (Miller et al., 2014). For distribution analyses the median raw Fluo-4 AM fluorescence intensity (in this case defined as baseline fluorescence) and the difference between maximum and minimum fluorescence intensity values (in this case defined as ⌬F, fluorescence intensity range) across all frames were obtained for each ROI (cell) collected in the present study and plotted as relative frequency distributions. Violin plots were generated in RStudio, and distributions were compared using the nonparametric Mann-Whitney U test.

MTT cell viability assay
Cell viability was evaluated in WT and Panx1 KO neuronal cultures Nunc TM LabTek TM Chamber Slide TM using the Vybrant ® MTT ([3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide]) cell proliferation Assay Kit (V13154, ThermoFisher Scientific) following the manufacturer's instructions. Briefly, 12 mM of MTT stock solution (Component A) were prepared by adding 1 ml of PBS to a 5 mg vial of MTT; 20 L of this 12 mM MTT solution was added to each well containing neurons bathed in 180 L of fresh BrainPhys without Phenol Red (05791, STEMCELL Technologies) and incubated for 4 h at 37°C and 5% CO 2 ; wells without neurons were used as negative controls. After this step, ¾ of the medium were removed and 100 l of DMSO were added in, mixing thoroughly the contents of the well and incubating for 10 min at 37°C. Next, the resulting solution was mixed once again, and the absorbance was read at 540 nm using a microplate reader (Infinite PRO microplate reader, Tecan Life Sciences). All absorbance values represent the average of nine scans per well and were normalized to blank wells (wells without neurons). Six wells per culture per group were used for this assay (n ϭ 3 per group).

Synaptosome preparation and Western blotting
Synaptic proteins were extracted using Syn-PER Synaptic Protein Extraction Reagent (87793, ThermoFisher Scientific) according to the manufacturer's instructions. Briefly, WT and Panx1 KO P14 and P29 cortices were dissected and weighed and then submerged in ice-cold Syn-PER reagent (1 mL/100 mg) supplemented with protease inhibitor cocktail (P8340, Millipore-Sigma). After homogenization on ice, 10 -20% of the homogenate was stored at Ϫ80°C; the remaining of the homogenate was centrifuged at 1200 ϫ g for 10 min at 4°C. The pellet was discarded, and the supernatant transferred to a new tube, for a new round of centrifugation at 15,000 ϫ g for 20 min at 4°C, obtaining synaptosomes. This pellet was resuspended in Syn-PER reagent using 150 L per 100 mg of brain tissue. This synaptosome suspension was stored in 5% (v/v) DMSO at Ϫ80°C until analysis. On the day of analysis, 50 L of the synaptosome suspension was placed in a new tube and centrifuged to collect the pellet. Protein was extracted by adding 200 L of PBSbased RIPA lysis buffer (1% IGEPAL, 0.5% sodium deoxycholate, 0.1% SDS, supplemented with PI cocktail, PMSF and Na orthovanadate) followed by incubation on ice for 30 min. Samples were heated to 95-100°C for 10 min in Laemmli sample buffer, DTT and ␤ϪME before loading 10 g of protein per lane onto 10% PAGE gels (TGX Stain-Free FastCast Acrylamide Kit 161-0183, Bio-Rad) and protein separation was achieved by application of 200 V. Following electrophoresis, gels were exposed to 30 s UV light (G-box imager) to obtain the TGX Stain-Free signal (total protein) and then transferred to polyvinylidene fluoride (PVDF) for 1 h at 100 V. Following this, the TGX Stain-Free signal was captured by UV light (5 s), rinsed with deionized water for 30 s, blocked in 5% skim milk in PBS supplemented with 0.1% Tween 20, incubated with primary antibodies at 4°C overnight, and secondary antibodies for 1 h at RT after three washes in PBST. The immunoreactive bands were visualized by enhanced chemiluminescence and quantified using ImageJ (http:// imagej.nih.gov/ij/).

Experimental design and statistical analysis
For ex vivo analysis (diolistic labeling of dendritic spines) WT and Panx1 KO groups consisted of equal numbers of male and female mice. Note that separate analyses of male and female groups revealed no sexspecific differences in the overall effects and as such the sexes were combined. For in vitro experiments, appropriate controls are clearly identified in detail in the figures and figure legends. Treatment timelines and all other relevant details are described in Results and in the figure legends and where appropriate, illustrated on the figures themselves. Researchers were blinded to the identity of the treatment/experimental groups at all stages of the analysis, except for Western blot analysis. Data are presented as mean Ϯ SEM. Significance comparisons were calculated using unpaired Student's t test, one-way ANOVA and two-way ANOVA for grouped analyses. Bonferroni's correction was used for multiple comparisons when appropriate. When interactions were statistically significant while using two-way ANOVA for grouped analysis, simple effect ANOVAs with multiple comparison were performed using Bonferroni's correction. For non-normally distributed data, we used nonparametric tests. Details of normality tests can be found in Table 1. Statistical significance was determined by p Ͻ 0.05 in all tests used in the present study. Data were analyzed using GraphPad Prism version 6.0d (GraphPad Software), and RStudio v1.1.463 (RStudio). Significance is denoted as ‫ء‬p Ͻ 0.05, ‫‪p‬ءء‬ Ͻ 0.01, ‫‪p‬ءءء‬ Ͻ 0.001, ‫‪p‬ءءءء‬ Ͻ 0.0001. Results of statistical tests are described in detail in the Table 1; superscript letters throughout the results section and figure legends indicate the corresponding statistic in the table.

Increased network ensembles and altered Ca 2؉ dynamics in Panx1 KO cortical neurons
To determine the impact of Panx1 on network connectivity, we performed Fluo-4 AM Ca 2ϩ imaging in primary cortical neuron cultures from WT and Panx1 KO mice ( Fig.  1). Spontaneous developing networks in cultured cortical neurons exhibit self-sustaining bursts lasting a few hundred milliseconds occurring at 0.05-0.1 Hz between days in vitro 8 (DIV8) and DIV21 (Habets et al., 1987;Murphy et al., 1992;Maeda et al., 1995;Tibau et al., 2013). Considering these characteristics and other experimental factors (minimization of photo toxicity, imaging multiple FOVs), we imaged DIV12-14 cultures at 0.2 Hz (pixel dwell time ϭ 36 ns; total frame capture time ϭ 135 ms) for 10 min (120 frames). To tease out the effects of Panx1 KO on network properties we performed computational modeling of our Ca 2ϩ imaging data using the MATLAB based open-source package, FluoroSNNAP (Fluorescence Single Neuron and Network Analysis Package). Fluo-roSNNAP allowed us to determine the number and properties of network ensembles, which are defined as a group of neurons that undergo a statistically significant degree of coactivation neurons. These ensembles were identified by their contribution to a so-called "high-activity frame" characterized by a statistically significant proportion of activated neurons (Miller et al., 2014;Patel et al., 2015). Within this algorithm, statistically significant (p Ͻ 0.05) high activity frames were identified by comparing the mean activity level of a given frame with a computationally-derived activity threshold calculated using the inferred spike activity data of each cell permutated 1000 times across the entirety of the recording period). Figure 1A depicts Fluo-4 and FluoroSNNAP analyses from exemplary FOV and exemplary cells from WT (left side) and Panx1 KO (right side) cortical neuron cultures. Figure 1Ai depicts two sequential Fluo-4 fluorescence frames captured from an exemplary WT FOV (left side) and an exemplary Panx1 KO FOV (right side) cultured cortical neurons. An increase in Fluo-4 fluorescence intensity in exemplary WT Cell 75 highlighted in high-activity Frame 27 (on the right) was evident by comparison with the preceding frame (26). Similarly, an increase in Fluo-4 fluorescence intensity in exemplary Panx1 KO Cell 85 highlighted in high-activity Frame 30 (on the right) was evident by comparison with preceding frame (29). The FluoroSNNAP-computed ⌬F/F (Fig. 1Ai, middle) and inferred activity (Fig. 1Ai, bottom) of these two exemplary cells is also shown across all frames. Figure 1Aii depicts the percentage of coactive neurons (top) and cell-specific spike activity (raster plot; bottom) across all 120 frames from the WT (left) and Panx1 KO (right) exemplary FOV. In this exemplary FOV, there are more red crosses (spikes from cells participating in a network ensemble) in the Panx1 KO raster plot. Note that total of 27/1044 cells (2.9%) were removed from WT (2.9%) and 66/1155 cells (5.7%) were from Panx1 KO coverslips according to our exclusion criteria (cells exhibiting sustained fluorescence intensity at 90% of maximum or greater). Consistent with this exemplary data, overall, Panx1 KO cultures exhibited a significant increase in network ensembles ( Fig. 1Bi; p ϭ 0.0014 a1 ) and number of cells per ensemble ( Fig. 1Bii; p Ͻ 0.0001 a2 ), as well as a significant increase in core ensembles (coactive groups of neurons active in more than one network ensemble; Fig. 1Ci; p ϭ 0.0071 b1 ). These observed effects on network properties were conserved with or without the excluded cells. We then looked at raw fluorescence intensity values from Fluo-4 AM labeled primary cortical neurons and plotted the median (defined as baseline for this analysis) and the difference between the maximum and minimum fluorescence intensity values (⌬F, fluorescence intensity range) for each neuron recorded during our imaging sessions. Panx1 KO neurons exhibited a significant increase in the baseline intensity of Ca 2ϩ transients ( Fig. 1D; p Ͻ 0.0001 c ) and range of fluorescence compared with WT neurons (Fig. 1E; p Ͻ 0.0001 d ). Additionally, we examined cell-type composition and cell viability. WT and Panx1 KO DIV12-13 cortical neuron cultures were composed of highly similar percentages of excitatory neurons (ϳ80%), inhibitory neurons (ϳ16%), and astrocytes (2-4%; Fig. 1F; p ϭ 0.9702 e4,8 , p ϭ 0.7500 e5,9 , p ϭ 0.1026 e6,10 , respectively). The low percentage of interneurons is consistent with previous data from this developmental time point (Habets et al., 1987;Benson et al., 1994;Frega et al., 2014;Johnson et al., 2015). Similarly, cell viability assessed by the conversion of MTT to formazan (MTT assay) was not significantly different between the two groups ( Fig.  1G; p ϭ 0.9089 f ). Together, these data suggest that Panx1 KO enhances functional connectivity of developing networks cortical neurons.

Panx1 Is enriched in synaptic compartments
To confirm expression of Panx1 in synaptic compartments, P14 cortical synaptosome fractions were prepared and validated by enrichment for PSD-95, and exclusion of the astrocyte protein GFAP by Western blotting (Fig. 2A). The synaptosome fractions demonstrated specific enrichment of Panx1 ( Fig. 2Aiii; p ϭ 0.0093 g6,8 ). Western blot analysis of whole cortical lysates from WT (C57BL/6J) mice revealed a dramatic drop in Panx1 expression from P7 to P14, and further, from P14 to P29 ( Fig. 2Bii; p Ͻ 0.0001 g2 , p ϭ 0.0006 g3 ), consistent with previous reports demonstrating peak Panx1 transcript expression at embryonic day (E)18 followed by a precipitous postnatal decreased (Ray et al., 2005;Vogt et al., 2005).

Increased PSD-95 and altered postsynaptic receptor expression in Panx1 KO cortical synaptosomes
Similar to the changes observed in Panx1 expression in whole cortex lysates, Panx1 expression in cortical synaptosomes dropped markedly (ϳ80%) between P14 and  3A,B). Consistent with the rapid development of dendritic spines in the first month of postnatal life (Miller, 1986;Zuo et al., 2005;Romand et al., 2011), expression of PSD-95 increased significantly between P14 and P29 ( Fig. 3B; p Ͻ 0.0001 h3 ). Somewhat unexpectedly, PSD-95 was further increased in Panx1 KO synaptosomes relative to synaptosomes from age matched WT controls (P14, p Ͻ 0.0001 h4 ; P29, p ϭ 0.0220 h5 ). These changes were accompanied by significant increases in GluA1 and GluN2A in Panx1 KO synaptosomes ( Fig. 3B; p ϭ 0.0082 i2 and p ϭ 0.0126 l2 , respectively). Interestingly, the developmental increase in GluN1 was more pronounced in WT synaptosomes (p ϭ 0.0009 m7 ), whereas GluN2B levels in Panx1 KO synaptosomes were higher at P14 and showed a more marked developmental decline at P29 (p ϭ 0.0488 o3 , p ϭ 0.0240 o5 ). Because the elevated PSD-95 levels and altered expression of glutamate postsynaptic receptor subunits in Panx1 KO synaptosomes could result from changes in the number of dendritic spines, we next investigated the impact of Panx1 KO on the density of dendritic spines in cortical neurons ex vivo.

Increased dendritic spine densities in cortical neurons from Panx1 KO mice
Based on our finding that synaptosomal PSD-95 expression was selectively increased in age-matched Panx1 KO cortical synaptosomes, we tested the hypothesis that Panx1 regulates dendritic spine development. We used the fluorescent lipophilic dye, DiI, which allows for relatively sparse labeling of somatosensory layer 5 pyramidal neurons (Fig. 4A). As predicted, spine densities from the apical dendritic tuft of layer 5 pyramidal neurons were significantly higher in Panx1 global KO mice than age matched WT controls at both P14 and P29 (Fig. 4Bi,Bii; P14, p ϭ 0.0014 p1 ; P29, p Ͻ 0.0001 q1 ). These changes were consistent with the increased synaptosome PSD-95 expression levels observed at P14 and P29. Next, because Panx1 has also been detected in astrocytes (Huang et al., 2007;Iglesias et al., 2009), microglia (Burma et al., 2017), and several vascular system cell types (Begandt et al., 2017) in various contexts, we also generated a conditional glutamatergic neuron-specific (Emx-1 IRES-Cre ; Panx1 f/f ) Panx1 KO (Panx1cKO E ). Consistent with our results from the global Panx1 KO, spine densities in Panx1cKO E were significantly higher than in Panx1 f/f con-  Fig. 4Biii; p ϭ 0.0104 r1 ). The Cre-based recombination in the Emx1-expressing lineage begins as early as E10.5 (Gorski et al., 2002). Western blot analysis (Fig.  4Biv) of Panx1cKO E and control (Panx1 f/f ) cortical (Cx) and cerebellar (Cb) lysates demonstrated marked reduction of Panx1 immunoreactivity in Panx1cKO E cortical lysates, confirming Panx1 KO in cortical excitatory neurons (comprising the majority of cortical tissue). Notably, mean spine lengths were not significantly different in either Panx1 KO line, suggesting the additional spines do not represent abnormally long spines or filopodia (Miller and Peters, 1981;Ziv and Smith, 1996;Zuo et al., 2005; for review, see Sala and Segal, 2014). Panx1 KO primary cortical neurons grown in culture for DIV12-14 exhibited significantly higher densities of dendritic protrusions resembling dendritic spines (Fig. 4C). A similar proportion of dendritic protrusions co-localized with PSD-95 in both WT and Panx1 KO cortical neurons and spine lengths were not significantly different between groups. Together these results suggest that deletion of Panx1 in glutamatergic cortical neurons increases spine density in a cellautonomous way.

Discussion
To our knowledge, this is the first study connecting Panx1 to the structural development of dendritic spines. We observed similar spine lengths and proportions of spines expressing PSD-95 in WT and Panx1 KO cortical neurons, suggesting Panx1 KO does not simply induce a selective proliferation of immature spines, but rather increases the number of spines with very similar properties to those found in WT cortical neurons. Our results suggest that this increase in dendritic spine density underlies the larger number of network ensembles observed in Panx1 KO cortical cultures. These findings are consistent with recent evidence demonstrating that incorporation of a cell into a network ensemble requires the development of spines and synapses (Jung and Herms, 2014; for review, see Hoshiba et al., 2017;Frank et al., 2018). While cortical cultures, in which we performed our network analysis, are known to contain abundant autaptic connections, these are also highly abundant within the developing rodent neocortex. Lübke et al. (1996) reported that autaptic contacts are found in most layer 5 cortical neurons in situ in the developing rat neocortex (92% of all coupled neurons; 80% of all cells analyzed). Among others, a recent report from Yin et al. (2018) confirmed that autapses occur in layer 5 pyramidal neurons in developing mouse prefrontal cortex and human frontal lobe (acute brain slices) that persist into adulthood, promoting neuronal responsiveness, burst firing and coincidence detection. Thus, not only might Panx1 KO impact autapses in culture but also potentially in vivo. It is also important to note that our continued determine network ensemble properties. Ai, Confocal micrographs of exemplary FOVs of WT and Panx1 KO (labeled KO) demonstrating Fluo-4-derived Ca 2ϩ activity from low and high activity frames (as indicated), along with the FluoroSNNAP output ⌬F/F (middle) and inferred spikes (bottom) from the identified WT (75) and KO (85) cells. Aii, Percentage of active neurons in each frame from the example FOVs (top). The red line indicates the threshold for a statistically significant number of coactive cells in a frame used by FluoroSNNAP (3 SD). Raster plots of WT and KO example FOVs (bottom) generated from thresholded spike probability data. Spikes from cells participating in a network ensemble are shown in red. The exemplary high activity frames and cells from A are also highlighted in red. B, Network ensemble data from WT and Panx1 KO DIV12-14 primary neuron cultures. Bi, The mean number of network ensembles was increased in Panx1 KO cultures (WT: 4.0 Ϯ 0.6, KO: 7.6 Ϯ 0.7 network ensembles; t (13) ϭ 4.1, p ϭ 0.0014 a1 ; n ϭ 7-8 coverslips from 3 independent cultures; ‫‪p‬ءء‬ Ͻ 0.01). Bii, The number of cells involved in network ensembles was also increased in Panx1 KO neurons (WT: 5.0 Ϯ 0.6, KO: 8.5 Ϯ 0.6 cell per ensemble; t (13) ϭ 4.4, p Ͻ 0.0001 a2 ; n ϭ 20 -21 network ensembles from 3 independent cultures; ‫‪p‬ءءءء‬ Ͻ 0.0001). C, Core network ensemble data from WT and Panx1 KO DIV12-14 primary neuron cultures. Ci, The mean number of core ensembles (co-activated neurons participating in more than one ensemble) was increased in Panx1 KO cultures (WT: 1.2 Ϯ 0.3, KO: 2.7 Ϯ 0.5 core ensembles; t (39) ϭ 2.8, p ϭ 0.0071 b1 ; n ϭ 20 -21 network ensembles from 3 independent cultures; ‫‪p‬ءء‬ Ͻ 0.01). Cii, The number of cells forming a core ensemble was not significant different between the analyzed groups (WT: 3.1 Ϯ 0.3, KO: 3.7 Ϯ 0.3 cells per core ensemble; t (30) ϭ 1.3, p ϭ 0.1968 b2 ; n ϭ 12-20 core ensembles from 3 independent cultures; n.s., not significant). D, Distributions and violin plots of resting and total change (maximum minus minimum) of Fluo-4 fluorescence intensities in DIV12-14 primary cortical neuronal cultures. Di, Frequency distributions of Fluo-4 Ca 2ϩ indicator dye fluorescence intensities of WT (red) and Panx1 KO (blue) revealed a right shift toward higher median Ca 2ϩ levels at baseline (defined the as raw median fluorescence intensity value for each neuron; WT median ϭ 37, n ϭ 1017 cells; KO median ϭ 58.50, n ϭ 1089 cells; p Ͻ 0.0001 c ; Mann-Whitney U ϭ 316,969; data compiled from 7 to 8 coverslips from 3 independent cultures per condition; ‫‪p‬ءءءء‬ Ͻ 0.0001). Dotted lines represent the mean of each distribution; a.u., arbitrary units. E, Similarly, the difference between the maximum and minimum fluorescence intensity values (⌬F, fluorescence intensity range) was right-shifted and significant larger in Panx1 KO neurons (WT median ϭ 16, n ϭ 1017 cells; KO median ϭ 25, n ϭ 1089 cells; p Ͻ 0.0001 d ; Mann-Whitney U ϭ 294,294; data compiled from a total of 7-8 coverslips across 3 independent cultures per condition; ‫‪p‬ءءءء‬ Ͻ 0.0001). Dotted lines represent the mean of each distribution; a.u., arbitrary units. F, WT and Panx1 KO cortical neuronal cultures have a similar cell-type composition. Fi, Representative images of WT and Panx1 KO cortical neurons labeled with the pan-neuronal marker MAP2, interneuron marker Gad67, and the astrocytic marker GFAP. Scale bar, 100 m. Fii, The proportion of excitatory neurons, inhibitory neurons, and astrocytes was similar between groups (WT excitatory neurons ϭ 81.4% Ϯ 1.3%, KO excitatory neurons ϭ 79.8% Ϯ 1.6%, p ϭ 0.9702 e8 ; WT inhibitory neurons ϭ 17.1% Ϯ 1.3%, KO inhibitory neurons ϭ 15.2% Ϯ 1.0%, p ϭ 0.7500 e9 ; WT astrocytes ϭ 1.5% Ϯ 0.4%, KO astrocytes ϭ 4.9% Ϯ 1.0%, p ϭ 0.1026 e10 ; simple-effect ANOVA with Bonferroni's multiplecomparison test, n ϭ 16 FOV from 2 independent cultures; n.s., not significant). G, WT and Panx1 cortical neurons exhibited similar cell viability. Conversion of MTT to formazan (absorbance measured at 540 nm) was not significant between groups (WT ϭ 100% Ϯ 2.5%, KO ϭ 98.62% Ϯ 8.5%; p ϭ 0.9089 f ; t (4) ϭ 0.128; n ϭ 3 independent cultures; n.s., not significant). Data are presented as mean Ϯ SEM.
understanding of the contribution of spine development and synaptic strengthening to spontaneous network development is still relatively limited and our experiments did not address whether the ϳ20 -30% increases in spine density we observed ex vivo and in vitro equated directly to 20 -30% increases in the number of synapses. Finally, given that not all synapses/cells are recruited to shape the development of neuronal network ensembles, and because our current understanding of the recruitment process is limited (Hoshiba et al., 2017), in the absence of more sophisticated methodology, we are unable to predict which Panx1 KO cells might be engaged in enhanced coupling. Moreover, rescue experiments in which Panx1 is re-expressed in control and Panx1 KO neurons are now needed to determine whether the role of Panx1 in regulating spine formation and network ensembles is direct.
The current study expands on previous findings relating to synaptic plasticity in Panx1 KO mice by targeting a different region of the brain, the cortex, and by focusing on potential developmental contributions. Previous studies looked primmarily at the CA1 region of the hippocampus at 1 month of age or older (Prochnow et al., 2012;Ardiles et al., 2014). These studies showed increased CA1 long-term potentiation (Prochnow et al., 2009;Ardiles et al., 2014;Gajardo et al., 2018) as well as a reduction in LTD (Ardiles et al., 2014) associated with Panx1 KO; albeit, these effects were observed uniquely in adult animals [3 months for the Prochnow et al. (2009) study, 9 -12 Figure 2. Panx1 is enriched in synaptic compartments. A, Synaptic protein extraction and isolation revealed Panx1 enrichment in cortical synaptic compartments. Ai, Protocol for synaptosome preparation from dissected cortical tissue using SynPer. Aii, Western blot of subcellular fractionations obtained from a P14 WT brain and probed with PSD-95 (top), Panx1 (second panel), and GFAP (third panel), with Stain-Free (total protein) at the bottom, demonstrating enrichment of PSD-95 in the P3 fraction (synaptosomes) and exclusion of GFAP (negative control). Aiii, Quantification of Panx1 enrichment in synaptic compartments as determined by higher immunoreactivity in P3 (synaptosomes) relative to homogenate. As expected, PSD-95 was also enriched in P3 (Panx1, p ϭ 0.0093 g6,8 ; PSD-95, p Ͻ 0.0001 g5,7 ; simple-effect ANOVA with Bonferroni's multiple-comparison test; n ϭ 3 animals; ‫‪p‬ءء‬ Ͻ 0.01, ‫‪p‬ءءءء‬ Ͻ 0.0001). B, Panx1 cortical expression is developmentally down regulated. Bi, Western blot of WT dissected whole cortical tissues from P7-P63 animals, probed with Panx1 (top), and Stain-Free (total protein) at the bottom. Bii, Panx1 expression decreased with age (age: F (3,8) ϭ 365.9, p Ͻ 0.0001 h1 ; n ϭ 3 animals per group; ‫‪p‬ءءءء‬ Ͻ 0.0001) with levels markedly dropping from P7 to P14 (p Ͻ 0.0001 h2 ; P14 -P29, p ϭ 0.0006 h3 ; P29 -P63, p ϭ 0.9604 h4 ; one-way ANOVA with Bonferroni's comparison test; n ϭ 3 animals per age group; ‫‪p‬ءءء‬ Ͻ 0.001; ‫‪p‬ءءءء‬ Ͻ 0.0001; n.s., not significant). Data are presented as mean Ϯ SEM. months for the Ardiles et al. (2014) study]. Ardiles et al. (2014) found that hippocampal Panx1 expression levels were greatly reduced between young (1 month) and older (9 -12-month-old) animals, which is consistent with our findings over our earlier age range, and suggest that the decline in Panx1 levels begins in the early postnatal period and continues on with increasing age. More recent work showed that Panx1 channels are strongly active under ictal conditions in human brain cortical tissue from epilepsy patients and in the CA1 region of the hippocampus following kainic acid seizure induction (Dossi et al., 2018), suggesting that in pathologic conditions Panx1 is positively correlated with excitability; although the mechanistic underpinnings, such as the possible interneuronal or glial contributions to this effect, have not yet been fully resolved. Relatedly, because Panx1 has been detected in Expression levels for each protein were normalized to total protein and expressed as a percentage of WT P14 values; n ϭ 5 animals per group analyzed in five independent experiments. Panx1 significantly decreased from P14 to P29 in WT cortical synaptosomes (P14 ϭ 100 Ϯ 9.4%; P29 ϭ 13.4 Ϯ 1.2%, p Ͻ 0.0001 i3,4,7 ; simple effect ANOVA with Bonferroni's multiple-comparison test; ‫‪p‬ءءءء‬ Ͻ 0.0001). No Panx1 signal was detected in Panx1 KO cortical synaptosomes. PSD-95 significantly increased with age in both WT and Panx1 KO, and was also significantly higher in Panx1 KO relative to WT within age-matched controls (age: F (1,16) ϭ 37.4, p Ͻ 0.0001 j3 ; genotype: F (1,6 ϭ 175.8, p Ͻ 0.0001 j2 ; interaction: F (1,16) ϭ 4.2, p ϭ 0.0570 j1 ; two-way ANOVA with Bonferroni's multiple-comparison test; WT P14 ϭ 100 Ϯ 8.5%, KO P14 ϭ 179.2 Ϯ 9.1%, p Ͻ 0.0001 j4 ; WT P29 ϭ 248.5 Ϯ 9.0%, KO P29 ϭ 287.9 Ϯ 11.8%, p ϭ 0.0220 j5 ; ‫ء‬p Ͻ 0.05, ‫‪p‬ءءءء‬ Ͻ 0.0001). GluA1 and GluN2a also exhibited age-matched increases in expression in Panx1 KO cortical synaptosomes (GluA1: genotype, F (1,16)   Average spine density was significantly higher in Panx1 KO (WT, 13.7 Ϯ 0.7 spines per 10 m; KO, 17.2 Ϯ 0.5 spines per 10 m, p ϭ 0.0014 p1 ; t (14) ϭ 3.9, unpaired t test, n ϭ 8 animals per genotype; ‫‪p‬ءء‬ Ͻ 0.01). Average spine length was not significantly different multiple cell types and has been associated with cell death processes (for review, see Sandilos and Bayliss, 2012;Thompson, 2015;Swayne and Boyce, 2017), we analyzed the cellular composition and viability of our cortical cultures. WT and Panx1 KO cultures were comprised of similar percentages of excitatory neurons, inhibitory neurons and astrocytes. The majority of cells in both WT and Panx1 KO cultures were excitatory neurons (ϳ80%). Inhibitory neurons, although less abundant (ϳ16% in our cultures for both WT and Panx1 KO), play an important role in shaping cortical networks (Lu et al., 2017). Panx1 has been detected in both excitatory and inhibitory neurons (Ray et al., 2005;Vogt et al., 2005;Zoidl et al., 2007), and this is consistent with our Western blot data from Panx1cKO E (glutamatergic-specific Panx1 KO) and control (Panx1 f/f ) lysates. Cortical control (Panx1 f/f ) lysates exhibited a minor residual Panx1 immunoreactivity, which likely reflects expression in inhibitory neurons. While together our results suggest that the impact of Panx1 KO on dendritic spine formation is cell-autonomous (glutamatergic neurons), a potential contribution of inhibitory neuron Panx1 to network ensemble development remains to be determined.
The mechanisms governing spine formation and plasticity are poorly understood (for review, see Yoshihara et al., 2009;Murakoshi and Yasuda, 2012), and are developmental ageand brain region-specific, making it difficult to directly compare these previous studies with our own. For example the major chloride extruder in neurons, KCC2, differentially regulates brain-derived neurotrophic factor (BDNF)-dependent dendritic spine development in CA1 and somatosensory neurons during the first postnatal week (Awad et al., 2018). This study suggests molecular mechanisms of dendritic spine formation in different brain regions might not be completely generalizable. Moreover, mounting evidence implicates Panx1 as a possible chloride permeable channel (Ma et al., 2012;Nomura et al., 2017) and thus this differential regulation and of a chloride extruder could directly impact on Panx1 in spine formation; although it remains to be confirmed whether Panx1 channel function itself is implicated in the regulation of spine development. Further, BDNF, which regulates the brain region-specific effect of KCC2 (Awad et al., 2018), is expressed at higher levels in the hippocampus than in the cortex (Awad et al., 2018), suggesting fundamental differences in baseline levels of key molecular effectors of synaptic plasticity.
Altogether, our novel findings presented here have implications for understanding neurodevelopment and diseases involving changes in spines. Alterations in dendritic spine densities have been described in a variety of neuropsychiatric disorders (Swann et al., 2000;Kulkarni and Firestein, 2012;Forrest et al., 2018). Of note, a recent study identified a human PANX1 variant with multi-organ developmental abnormalities associated with marked intellectual disability (Shao et al., 2016). Moreover, single nucleotide polymorphisms affecting Panx1 expression levels have been implicated in autism spectrum disorder (Davis et al., 2012). Therefore, understanding the developmental role of Panx1 could provide important insights into variations in normal brain development as well as risk of neuropsychiatric disease.