Localized network hyperconnectivity leads to hyperexcitability after injury

Brain injury increases the risk of the development of epilepsy. Axonal sprouting and synaptogenesis are homeostatically mediated responses by neurons to injury-induced de-afferentation and de-efferentation. Rewiring which occurs due to axonal sprouting and synaptogenesis may alter network excitability and lead to epilepsy. Excitatory and inhibitory connectivity are both subject to homeostatic rewiring. Thus, post-sprouting hyperexcitability cannot be simply explained as result of altered inhibitory/excitatory balance. In this work, we show computationally and experimentally that hyperconnected local networks are created by homeostatic rewiring near the injury site. These local networks are characterized by altered system dynamics despite preservation of excitatory/inhibitory balance. Hyperconnected local networks have a lower threshold for burst initiation, and generate spontaneous bursts, which in turn ignite seizure-like activity in the larger network. Our findings demonstrate a novel network mechanism of hyperexcitability and seizure generation due to maladaptive recovery after injury.


Introduction 21
Brain injury increases the risk of the development of epilepsy 1, 2 . Severe brain injury, such as 22 penetrating head trauma, is associated with particularly high risk of epilepsy 3 . One of the mechanisms 23 that may link injury and epilepsy is neural network reorganization, which at cellular level is represented 24 by axon sprouting and synaptogenesis 4 . Axon sprouting has been observed in the hippocampus of 25 patients with temporal lobe epilepsy, and in animal models of acquired chronic epilepsy [5][6][7][8][9] . 26 Axon sprouting can be induced by injury that involves axotomy and deafferentation 10-13 . The 27 brain reorganizes in response to a wide array of pathologies in order to recover normal function [14][15][16] . 28 Sprouting has also been observed in the peripheral nervous system. Maladaptive reorganization of 29 peripheral axons after injury may lead to involuntary muscle contractions 17 or neuropathic pain 18 . Axon 30 sprouting and synaptogenesis in cortex and hippocampus can also be maladaptive. Reorganization of the 31 mossy fiber pathway increased recurrent connectivity in the hippocampus of epileptic patients 5, 6 . 32 Transection of axons of pyramidal neurons in CA3 subregion of the hippocampus resulted in axon 33 sprouting and hyperexcitability 19,20 . Partial isolation of a region of neocortex led to reorganization of 34 excitatory connectivity of layer V neurons, and emergence of spontaneous epileptiform bursts 21 . 35 Sprouting of excitatory axons may cause hyperexcitability by creating or increasing excitatory feedback. 36 However, sprouting mossy fibers innervate not only excitatory granule cells, but also inhibitory 37 parvalbumin-expressing interneurons 22,23 . Excitatory synaptic drive to hilar inhibitory neurons increased 38 after injury 24 . Surviving somatostatin inhibitory interneurons also sprout axons and form new synapses 39 between hilus and granule cells 25 , and between CA1 and granule cells 26 . Therefore, inhibitory feedback 40 rises in parallel with increases in excitatory feedback after injury, suggesting that hyperexcitability is not 41 simply a consequence of increased excitatory drive to surviving neurons. 42 To conceptually understand the development of hyperexcitability, we can consider the case of a 43 pyramidal neuron located close to the site of injury (such as stroke or penetrative head trauma) that results 44 correlated with hyperexcitability, with a lower threshold to ignition of a population burst compared to an 77 un-confined culture. This occurred despite the scaling up of the strength of inhibitory synapses. In 78 cultures where density of neurons (as opposed to size of the network, as in our data) was varied, scaling 79 of excitatory and inhibitory synapses also occurred, with excitatory/inhibitory balance preserved for all 80 densities 30 . 81 82 Hyperexcitability in rate-based computational model 83 To understand heightened excitability of confined networks, we carried out computer simulations 84 of ring-shaped networks of rate-based neurons. We built networks containing different numbers of 85 neurons, but with the same number of pre-synaptic excitatory and inhibitory connections per neuron ( Fig.  86 2(A)). This resulted in the scaling of synaptic weights with network size -smaller networks had higher 87 average synaptic weights than larger networks (Fig. 2(B)), and higher probability of a synapse existing 88 between a pair of neurons ( Fig. 2(C)) (n = 3 networks of each size). These results were in line with the 89 findings from dissociated cortical cultures. We then stimulated different numbers of excitatory neurons 90 and allowed network activity to stabilize. We found that average firing rate either went back to zero, or 91 containing 5%, 25%, 50%, 75%, or 95% of total neurons. After removing neurons, their connections and 151 connections passing through the removed portion of the sphere, different rewiring algorithms were used. 152 Neurons with reduced connectivity compensated by making new connections, modeling structural 153 homeostatic drive. Three different sigmoid distributions of connection probability vs distance were used 154 to model recovery: i) short critical distance, ii) intermediate critical distance, iii) critical distance same as 155 during development (long) (Fig. 3 (A, B)). Re-wiring was terminated after total pre-synaptic weight 156 reached pre-lesion set point. Re-wiring resulted in increased average pre-synaptic weights in the 157 perilesional region (Fig. 3(B)). Inhibitory/excitatory synaptic ratio was not affected by homeostatically 158 driven re-wiring ( Fig. 3 (C)). Eight neurons were selected at the north pole of an intact sphere or the edge 159 of a lesioned sphere and the south pole. Tight network clustering 34 in the microregion at the lesioned 160 sphere edge was observed, contrasting with the loose network clustering in equivalent microregion in the 161 intact sphere, or at the south pole ( Fig. 3(D)). 162 Connectivity within the microregion near the lesion was significantly higher than connectivity 163 between the microregion and the rest of the network, for all lesion sizes except 95% (Fig. 3(E), n = 3 164 networks generated with different random seeds). No difference of in/out connectivity was observed for 165 equivalent microregion in the intact network (n = 3). Lack of difference in connectivity after 95% lesion 166 is due to the small size of the remaining network, with largest distance between any two neurons 167 comparable to critical re-wiring distance. 168 We then examined the functional outcome of different lesion sizes and re-wiring constraints. wide bursts, indicating increase in hyperexcitability ( Fig. 3 (F-H)). Re-wiring with the shortest critical 172 distance resulted in more bursts (Fig. 3(F)) than re-wiring with relaxed critical distance ( Fig. 3(G, H)). P-173 values are reported in Fig. 3 -figure supplements 1-3. 174

Population bursts initiate in the perilesional region after re-wiring 175
We explored the effects of rewiring in the perilesional portion of the network on the initiation of 176 population bursts. Out of 14 population events (bursts) that spontaneously occurred in a network with 177 25% loss, 13 events initiated in the perilesional region. In contrast, event initiation was evenly distributed 178 in the intact network (11 events) (Figure 4 (A-D)). We have also observed focused bursting in the 179 perilesional region (Fig. 4 (E, F)). These sub-population events may represent failed ignitions, similar to 180 those observed in the coupled rate-based network (Fig. 2(K)). 181 Healing the hyperexcitability at the lesion edge by replacing lesioned neurons 182 We then investigated whether it was possible to rescue (heal) the effect of re-wiring on the 183 network in the perilesional region. We removed all the connections to the lesioned area, but left the 184 neurons in place ( Fig. 4 (G, H)). This provided neurons in the perilesional region with additional targets 185 for re-wiring (under short sprouting constraint). The average synaptic weight post sprouting remained 186 comparable to the rest of the network and no strong edge effect was developed ( Fig. 4 (G, H)). 187 Excitability of the 'healed' network after re-wiring was significantly lower than lesioned network and was 188 not significantly different from the intact network ( Fig. 4 (I)), (n = 3 networks, each condition). 189

Smaller network size correlates with hyperexcitability of organotypic cultures 190
We created organotypic cultures of different size to experimentally validate findings in 191 computational models (Fig. 5 (A-D)). The number of neurons in these cultures spanned two orders of 192 magnitude, from ~1000 neurons in the smallest to ~130,000 neurons in largest cultures ( Fig. 5 (E, F)). 193 The number of ictal-like events was inversely related to culture size, with smallest cultures having 194 significantly more ictal-like events than the largest cultures ( Fig. 5(G)). 195

Axon sprouting and functional synaptogenesis in organotypic cultures 196
We tracked the growth of axons sprouted by neurons in whole hippocampal cultures. We found 197 that axons continued growing rapidly from DIV 2 to DIV 5 ( Fig. 5 -figure supplement 1 (A, B)). We did 198 not quantify axon growth at later DIVs, but axons continued growing until at least DIV 14. In order to 199 assess whether sprouted axons can form functional connections, two organotypic slices were cultured 200 together. Slices were placed at distances ranging from 0.5 mm to 1.7 mm from each other. Cultures  Cultures at distances larger than 1.5 mm failed to synchronize. These findings show extensive axon 204 sprouting by neurons in organotypic cultures that leads to functional re-wiring. 205

Clustering of the network in the perilesional region 206
We next examined whether axon sprouting results in local hyperconnectivity in organotypic 207 hippocampal cultures, as predicted by the computational model (Fig. 3). Injury was experimentally 208 simulated by cutting off different portions of a cultured slice (Fig. 6 (A)), mimicking computational 209 model ( Fig. 3 (A, B)). After allowing the slices to re-wire for at least 8 days, we examined the network in 210 the microregion immediately next to the cut (perilesional region) and the larger sub-regional network 211 ( Fig. 6 (A)). We found higher connectivity in the microregion near the cut compared to an equivalent 212 microregion in a control culture (Fig. 6 (B)), for cuts delivered to CA3c, CA1, Subiculum (Sub), or cortex 213 (Ctx). Clustering of the network 34 in the microregion significantly increased in cut versus control cultures 214 After delivery of a cut to CA3c, the cut edges of the slice were placed into close proximity with 220 each other. Activity in CA3 and DG in the 'healed' slice was significantly more correlated than in the cut 221 slice ( Fig. 7 -figure supplement 1), p = 0.002 (n = 7 cut cultures and n = 6 healed cultures). This 222 suggested that 'healed' cultures re-established synaptic connectivity that was severed by the cut. Network 223 clustering in the 'healed' cultures was significantly lower than clustering in 'cut' cultures (p = 0.0046 for 224 CA3c-DG, and p = 0.0049 for CA3c-CA3b), but not significantly different than clustering in controls (p = 225 0.7711 for CA3c-DG, and p = 0.1457 for CA3c-CA3b) (Fig. 7). N = 8, including 2 technical replicates 226 for 4 cultures were used for each group. Relaxation of clustering suggests that 'healing' reduced 227 hyperconnectivity in the perilesional region, similar to computational results in Fig. 4(H). 228

Epileptiform activity initiates in perilesional region 229
We then examined the effect of cut-induced network reorganization on excitability. We found 230 that perilesional region (near cut delivered to Ctx or subiculum) was predominantly the initiation site for 231 slice-wide ictal-like events (Fig. 8(A-C)). In cultures with cuts at both subiculum and CA3, ictal-like 232 activity initiated in both perilesional regions. 'Healing' the CA3 cut abolished ictal initiation near CA3, 233 with ictal events predominantly initiating in subiculum ( Fig. 8 (A-C)). Z-scores and corresponding p 234 values and sample numbers are shown in Fig. 8 -figure supplement 1. We also examined population 235 events that did not propagate to the entire culture (interictal-like activity, Fig. 8(D)). Energy of interictal 236 activity was significantly higher in the perilesional regions (

Perilesional region has increased excitatory and inhibitory synaptic drive 240
We delivered optical stimulation to the perilesional region and an equivalent region in control 241 cultures ( Fig. 8(I)). Evoked activity in the perilesional region was significantly higher compared to 242 control. Significant difference persisted after application of bicuculline, but was abolished after 243 application of bicuculline together with NBQX and D-APV ( Fig. 8(J)). Biological replicates were as 244 follows: vehicle control n = 5(5) and lesioned n = 9(5), bicuculline and bicuculline + D-AP5 + NBQX 245 control: n = 11 (6) and lesioned n = 11(6), where n = recordings(cultures). Increase in evoked response 246 due to bicuculline was significantly higher in perilesional region compared to control, while decrease in 247 evoked response due NBQX and D-APV was also significantly higher in perilesional region compared to 248 control ( Fig. 8(K), p = 5.776*10 -4 , and 0.009, t-test performed on n = 50 randomly drawn differences (bic 249 -veh, and veh -(NBQX+D-APV)) from data in Fig. 8(J)). This suggests that excitatory and inhibitory 250 synaptic drive was higher in the perilesional region. Axon sprouting and network reorganization in slice cultures are not homogeneous.
A 266 hippocampal slice is disconnected from the surrounding tissue at its horizontal top and bottom planes, as 267 well as at the transverse cut through subiculum.
Surviving neurons near the cut are the most 268 disconnected. Ability of these neurons to establish long-distance connectivity is limited by absence of 269 normal synaptic partners and developmental axonal guidance cues. Homeostatic connectivity drive and 270 inability to form long-distance connections can be expected to result in a hyperconnected local network in 271 the region near the transverse cut (perilesional region). We observed this experimentally, and found that 272 formation of hyperconnected local network occurred near the cut delivered to subiculum, entorhinal 273 cortex, CA1, or CA3. Hyperconnectivity was abolished by restoring the number of available synaptic 274 partners. This 'healing' experiment also demonstrated that hyperexcitability of the local network was not 275 due to non-network effects of the cut such as inflammation or changes in intrinsic neuronal excitability. 276 Increases in excitatory and inhibitory responses to stimulation are consistent with formation of 277 hyperconnected network in the perilesional region. The local networks generated by the cut in entorhinal 278 cortex, subiculum, CA3, and CA1 all exhibited a higher occurrence of local population events. Ictal-like 279 events were frequently, but unreliably initiated by local activity in the hyperconnected microregion. 280 These experimental results are consistent with the computational result where limitation of the distance of 281 axon sprouting after network lesion led to formation of hyperconnected, seizure initiating local network. 282 Hyperexcitability of the local hyperconnected network may be due to a lower threshold for local network 283 ignition, as described above for small isolated networks. Coupling of small, local hyperconnected 284 network, and a larger network, results in the local network functioning as an unreliable igniter of the 285 larger network. The unreliability may be due to variability in the small network ignition trajectory, 286 which in turn gives a chance for the inhibitory neurons in the larger network to activate faster than 287 excitatory neurons and prevent full network ignition ( Fig. 2(K)). This is consistent with the concept of 288 In dissociated cortical cultures, the strength of newly formed synapses inversely scales with culture 294 density 30 . This is indicative of a homeostatic process that leads to formation of more synapses with the 295 same partner if fewer partners are available (low density culture), or fewer synapses with the same partner 296 if more partners are available (high density culture). We found that changing the size of dissociated 297 culture, and thus the number of available synaptic partners, also leads to the same effect. The ratio of 298 inhibitory to excitatory synapses was preserved in both low and high density cultures 30 , consistent with 299 our finding that both excitatory and inhibitory synapses scale up in confined networks. Hyperexcitability 300 in confined networks is therefore more likely to be a function of altered network dynamics, rather than 301 altered inhibitory/excitatory balance. One of the mechanisms of homeostasis is preservation of activity 302 level in neurons. Pharmacological suppression of activity leads to an increase in excitatory synaptic 303 conductivity. This functional homeostasis is paralleled by an increase in number of synaptic contacts -304 structural homeostasis 42 . Injury that results in disruption of neural connectivity may lower neural 305 activity, and thus provide homeostatic impetus for synaptogenesis and strengthening of existing synapses. 306 However, activity can also result in synaptic pruning, and the role of activity in rewiring of inhibitory 307 connections is less clear. We have therefore elected to use excitatory and inhibitory synaptic drive as a 308 homeostatic set point in our computational models. This set point may be the result of neuronal activity 309 levels, or a consequence of homeostasis of neuronal metabolism, synaptic protein synthesis and 310 degradation. Cell signaling kinases that play a role in regulation of metabolism and protein synthesis, 311 such as receptor tyrosine kinases, PI3K, and mTOR, have been considered as antiepileptic targets 43,44 . 312 Pharmacological inhibition of mTOR resulted in reduction of axon sprouting, suggesting that this kinase 313 may be involved in maintenance of structural homeostasis, potentially through its role in regulating 314 protein translation. However, reduction of axon sprouting did not always prevent epileptogenesis 45 . This 315 may be due to differential role of mTOR inhibition on inhibitory and excitatory sprouting 46 . Our results 316 suggest an alternative interpretation: reduction in the length of sprouted axons may lead to increased local 317 synaptogenesis, which leads to formation of local, hyperconnected network with a lower ignition 318 threshold. This may reduce anti-epileptogenic effects of mTOR inhibition. Interventions that reduce 319 density of local sprouted axonal collaterals may be more beneficial from this point of view. 320 The goal of epilepsy resection surgery is the removal of a portion of the brain that is seizure-321 genic. Resection surgery is highly effective, with majority of patients seizure-free one year after the 322 input to neuron j is < threshold, or equal to slope * (input -threshold) or maximum firing rate max_rate. 372 Slope for neuron_type = inhibitory neurons (inh_slope) is higher than the slope of the neuron_type = 373 excitatory neurons (exc_slope) to reflect their higher firing rate and lower spike frequency adaptation. 374 Neurons are arranged at equal distances along a perimeter of a circle with radius that scales with In the case of the coupled network, two sub-networks are placed at distance equal to half of the 385 radius of the small network, and 1-2% of total excitatory synaptic weight of the smallest sub-network is 386 randomly removed from both sub-networks. Synapses are then replaced with trans-network connections, 387 using the same iterative process as above. 388 Stimulation was delivered as a single time step increase in rate, while noise was delivered as 389 randomly occurring pulses with width of a single time step, average frequency of 0.5 Hz, and specified 390 amplitude.Simulation parameters are shown in  The time delay from the presynaptic neuron firing an action potential and inducing a postsynaptic 485 potential is calculated based on the arc distance between neuron i and j and is as follows: 486 = distance( , ) × + ,where axon delay = 0.05ms/µm, and AP delay = 5ms 487 We introduced noise (spontaneous activity) by making each neuron fire action potentials as a 488 Poisson train with specified average frequency. All simulation runs were 10 seconds long. 6-well tissue culture plates were transferred to an interface chamber (Bioscience Tools) connected 504 to a temperature controller maintaining temperature at 37 C and a blood gas providing 5% CO 2 , 21% 505 Oxygen, balanced Nitrogen (Airgas). Tungsten microwires with 50 μm diameter non-insulated tip were 506 placed under CA3 or CA1 neuronal layers and extracellular field potentials were recorded for 45 minutes 507 via high-impedance multiple-channel pre-amplifier stage (PZ2-64, Tucker Davis Technologies) connected 508 to a RZ2 amplifier (Tucker Davis Technologies). Signals were filtered with a band-pass filter (1 Hz -3 509 KHz, gain x1000) and sampled at 6 KHz rate. Recorded signals were analyzed using OpenX software 510 RRID:Addgene_100854). Culture dishes were placed into the mini-incubator while keeping the 522 temperature at 37 C and supplementing with blood gas, on a fluorescent inverted microscope stage 523 (Olympus). Recordings were 15 to 30 minutes long. CCD camera and 4X objective was used to record 524 fluorescent changes at frame rates ranging from 1.59 seconds/frame to 0.16 seconds/frame (~0.6-6 Hz). 525 Videos were then analyzed using ImageJ and MATLAB. 526 527

Image Processing 528
Regions of interest were drawn for different cultures and mean gray values were calculated for 529 each frame and recorded in .txt files using ImageJ drawing tools and Plugins. Values were then imported 530 to MATLAB and baseline was calculated for the optical signal using the asymmetric least square 531 smoothing method 56 . Signal to baseline ratio was calculated by the following equation: 532 where F0 is the baseline. In order to identify paroxysmal activities, a simple thresholding method was 534 used. The threshold was greater than three times the standard deviation of baseline activity. For calcium 535 intensity analysis, all the data points above the threshold were summed for each region of interest (ROI) 536 and normalized to the activity of the whole culture. For seizure-like event detection we used the same 537 criteria we used for electrical recordings; the optical signal above the threshold must last for at least 10 538 seconds to be counted as a seizure. 539 540

Axon growth tracking 541
Slices were cultured in 6-well plates and phase contrast images were taken with a 20X objective 542 from 2 DIV to 5 DIV. Two sets of images were collected on each day 2 to 4 hours apart. Trackable axon 543 terminals were marked by arrows, and the length by which axons grew was measured. 544

Correlation-based Network Analysis 546
Optical recordings (250 msec/frame) of jRGECO1a fluorescence in control, cut, and healed 547 cultures on DIV 8 -12 were used. Activity was analyzed in a region of interest encompassing 548 hippocampal or cortical sub-region located near the cut, or an equivalent region of interest in control 549 cultures. Region of interest was split into bins of 2x2 or 3x3 pixels, and ΔF/F timeseries of approximately were then calculated for r between bins within the microregion (r in-in ), and between bins within 554 microregion and bins outside of microregion(r in-out ). Adjacency matrices for each sub-region were 555 populated with r values between appropriate bins. Network based on this adjacency matrix was then 556 plotted using 'force-directed layout' option in Matlab graph plot command. Effect of edge weights (r 557 coefficients) was selected to be inverse; in other words, largest edge weights produced strongest attractive 558 force and thus closest spacing between nodes (bins). Bins belonging to the microregion, and edges 559 between them, were highlighted in red on the resulting graph plot to show the relationship between sub-560 regional and microregional networks. 561

Detection of ictal initiation zone 562
Each culture was recorded at least three times between 8 and 25 DIV. Multiple ROIs within each 563 culture type were selected. ROIs considered were: subiculum, CA1, CA2, CA3b, CA3c-proximal, CA3c-564 distal, DG-infra, DG-crest, DG-supra for hippocampal, lesioned, and healed cultures. Two extra ROIs 565 were considered for Ctx+hipp cultures as well: Ctx-sub and perirhinal-Ctx. We then traced the evolution 566 of calcium activity for each ROI within the culture. A zero-phase low pass filter was applied to the optical 567 signal to remove noise. An event is considered ictal if the optical trace for all the ROIs (generalized 568 events within hippocampus) was above the threshold for more than 10 second without dropping below 569 threshold for more than 0.5 seconds. However, for the Ctx+hipp cultures activity in the perirhinal-Ctx and 570 Ctx+sub ROIs were exempt from this requirement as they are not present in other culture types. In 571 addition, in lesioned cultures ROIs located at CA3c-distal and DG regions were also exempted from the 572 requirement as in some cases they are distally located and act as separate network. Initiation zone 573 probability was calculated as the number of ictal events originating in a given ROI over the total number 574 of ictal events that were evaluated. 575

Energy of Interictal Activity 576
Energy was defined as the area under the curve (a.u.c) of the ΔF/F(t) during interictal period. Glucose 10 mM, CaCl 2 2 mM, MgCl 2 1 mM, and Na 3 PO 4 1.2 mM. Simultaneous optical stimulation of 584 ChR2 and optical recording of jRGECO1a fluorescence were carried out by using a dual-deck inverted 585 fluorescence microscope (Olympus IX73) equipped with X-cite LED illuminator on the lower deck and a 586 digital micromirror device-based patterned illuminator (Mightex Polygon400) on the upper deck. For 587 these experiments, 20X objective was used and light stimuli were delivered via patterned illuminator to a 588 narrow rectangle at the desired location on a cultured slice. Stimulation was delivered as 10 blue light 589 pulses at 0.02 Hz with 20 ms pulse width, with power of 18 mW/mm 2 . Peak evoked ΔF/F values were 590 measured within a 1 sec time window after stimulation, and average response was calculated. Bicuculline 591 which is a GABAA receptor antagonist was also tested both individually and also in a combination with 592 two glutamatergic blockers. D-AP5 which is an NMDA receptor antagonist, and NBQX which is an 593 AMPA receptor antagonist were used as blockers of glutamatergic synaptic activity. Similarly, 10 pulses 594 at 0.02Hz with 20ms pulse width were delivered and peak values were measured within a 1sec time

Immunofluorescent Staining 602
Cultures were fixed in 4% Paraformaldehyde for 2 hours, removed from substrate, washed and 603 permeabilized with 0.3% Triton X-100 (Sigma-Aldrich) in phosphate-buffered saline (PBS) for 2 hours 604 on a shaking platform at RT. Cultures were then blocked in 10% goat serum in PBS for 1 hr. All the 605 primary antibodies were applied to the culture for 48 hours on a shaking platform at 4 C. Anti-NeuN 606 conjugated to Alexa Fluor 555 (MAB377A5, Millipore) was applied at 1:500 dilution. Anti-Beta-III 607 Tubulin (ab78078, Abcam) was at dilution of 1:1000. Alexa Fluor 488 was used as secondary antibody 608 for Anti-Beta-III Tubulin at dilution of 1:1000. Optical stacks were imaged from the entire thickness of 609 the cultures using Zeiss confocal microscope with 40X or 25X objectives (Zeiss LSM 880, Germany). 610 Optical slices were taken with 1 μm intervals. Images were then processed in ImageJ. For neuronal 611 counting, we slightly modified the existing 3D watershed technique from 57 to detect the nuclei of 612 pyramidal cells in CA1 and CA3 layers and granule cells in DG neuronal layer. 613

Statistics 614
Statistical tests were used as described in the Results section. Nonparametric tests were used for 615 data that were not normally distributed. A statistically significant difference was defined as p < 0.05. At                              Overall, 9 ctx+hipp cultures from 3 animals, 8 hipp cultures from 3 animals, 7 lesioned cultures from 3 animals, and 4 healed cultures from 3 animals were used for this experiment.