Closed-Loop Distributed Detection and Stimulation for the Prevention of Spontaneous Seizure in a Chronic Rodent Model of Temporal Lobe Epilepsy

Background and Hypothesis

Drug-resistant temporal lobe epilepsy (TLE) affects roughly one-third of the 50 million people worldwide living with epilepsy (Asadi-Pooya et al., 2017). Beyond the occurrence of seizures, drug-resistant TLE carries a sudden unexpected death in epilepsy (SUDEP) rate of 6.3-9.3 per 1000 person-years, alongside restrictions on driving, employment, and repeated hospitalizations (Salanova et al., 2021). The current clinical standard and only FDA-approved closed-loop neuromodulatory therapy is the NeuroPace responsive neurostimulation (RNS) system, which uses two leads with 8 contacts each for recording and stimulating. RNS achieves a 75% median seizure reduction at 9 years; however, only 18% of patients reach complete seizure freedom (Nair et al., 2020; Razavi et al., 2020). This is best explained as a mismatch between the disorder and intervention as TLE is a network-level disorder, with seizures originating in the hippocampus and propagating through the amygdala, thalamus, and temporal neocortex along defined white-matter pathways (Bartolomei et al., 2025; Ahmadi et al., 2021). Functional network analysis of intracranial EEG shows that degree and betweenness centrality in the seizure-onset zone change seconds before clinical onset (Brohl et al., 2024), and that spatiotemporal network reconfigurations precede seizures across large-scale circuits (Fruengel et al., 2020). Therefore, effective intervention should monitor the entire network, not just 1-2 focal points.

Recent hardware advances have made distributed recording feasible. The Neurograins platform demonstrated coordinated 48-node recording with an architecture that can scale up to 770 nodes (Lee et al., 2021), and a companion distributed stimulation platform was demonstrated in vivo (Lee et al., 2025). Ultrasound-powered implants extend wireless power delivery beyond 5 cm of neural tissue (Piech et al., 2020), which could enable power delivery to hippocampal and thalamic depths. Through class projects in other classes (EBME 407 and MATH 419), I have proposed an R21 and validated initial claims using computational simulations to develop a computational decision layer that connects the recording and stimulation platforms. This includes a sequential probability ratio test based on node detectors operating within a < 10 µW power envelope, validated on the SWEC-ETHZ intracranial EEG dataset. The following proposal assumes that the experiments proposed in the R21 are validated and successful. What remains untested is whether the closed-loop distributed detection and stimulation actually prevents or aborts seizures in a living, seizing brain, and whether it does so better than the current clinical standard (RNS).

We propose to test this in the intrahippocampal kainic acid (KA) rat model of chronic TLE, which produces spontaneous, recurrent, hippocampal-onset seizures with secondary network involvement and well-characterized propagation patterns. The primary hypothesis is that distributed multi-site detection with network-level fusion driving closed-loop stimulation will reduce spontaneous seizure frequency by ≥50% relative to sham, and by a significantly greater margin than focal single-site detection and stimulation at similar stimulation parameters, over a four-week treatment period. Secondary hypotheses are that network-level fusion will detect seizures with shorter latency from electrographic onset than any single node, and that chronic stimulation under the proposed protocol will not produce histological tissue damage beyond the electrode-insertion baseline.

Methods

48 male and female Sprague-Dawley rats (250-350 g) will be used, justified by the power analysis below. All procedures will be approved by the Case Western Reserve University Institutional Animal Care and Use Committee (IACUC) before initiation.

Procedure 1: Chronic TLE Induction, Multisite Implantation, and Closed-loop Treatment

Rats will undergo stereotaxic hippocampal injection of KA (0.4 nmol in 50 nL of sterile saline; coordinates relative to bregma: AP -5.6 mm, ML +4.5 mm, DV -5.0 mm, targeting dorsal CA1) under isoflurane anesthesia (4% induction, 1.5-2.5% maintenance) with subcutaneous buprenorphine (0.05 mg/kg) for analgesia (Paxinos & Watson 7th ed.). Injection will be delivered at 25 nL/min via Hamilton syringe with the needle left in place for 5 mins post-injection to limit reflux. Animals will be monitored for 6 hours post-induction. Following a 4-week latent period, continuous video-EEG (single skull-screw electrode, 256 Hz) will confirm the emergence of spontaneous recurrent seizures, and only animals with >2 confirmed seizures over a 7-day baseline will advance. A second survival surgery will implant 6 recording and stimulation arrays to target the ipsilateral CA1, the ipsilateral CA3, the contralateral CA1, the ipsilateral basolateral amygdala, the ipsilateral anterior nucleus of the thalamus, and the ipsilateral entorhinal cortex. Electrodes will be commercial silicon probes (impedance 100-500 kΩ at 1kHz) bonded to a tethered headstage compatible with continuous LFP recording at 2 kHz per channel and biphasic charge-balanced stimulation. Coordinates for implantation will be obtained from Paximos & Watson and verified for each animal using post-mortem histology. Headcaps will be secured with skull screws and dental acrylic and protected with a 3D-printed housing. After one week post-surgical (allowing for the rodent to recover), animals will be randomized using a stratified scheme by baseline seizure frequency into three arms (n=16 per arm). The sham group will receive a recording but no stimulation, the distributed group will receive 6-site SPRT detection with hub-level Bayesian fusion, and the focal group will receive single-site detection and stimulation in the ipsilateral CA1 region, mimicking RNS. Stimulation parameters in both active arms are identical, biphasic, charge-balanced, 100–200 µs/phase, ≤120 µA, ≤10 Hz burst, constrained to Shannon k ≤ 1.75 with charge density <30 µC/cm²/phase (Shannon, 1992). The primary endpoint will be seizure frequency (#/day) over the final 14 days of treatment, scored offline by blinded reviewers. Secondary endpoints include seizure duration, generalization rate, detection latency from onset, and intervention success (termination of the seizure within 10 seconds).

Procedure 2: Behavioral Assessment

Behavioral testing will assess hippocampal-dependent memory and seizure-related behavior at three timepoints: pre-induction, post-latent period, and end of treatment. The water maze will be used to assess spatial memory using a 1.5 m circular pool with a hidden 10 cm platform in a fixed quadrant. Four trials will be performed per day for 5 days, with a single 60s probe trial on day 6 to measure time in the target quadrant. Open-field testing will assess anxiety and locomotor function using the center-time fraction and total distance traveled. Spontaneous seizure severity will be continuously scored over the course of the experiment using the modified Racine scale (1-5) by blinded observers reviewing EEG data. All behavioral videos will be analyzed using DeepLabCut, with the experimenter blinded to treatment.

Procedure 3: Histology and Tissue Safety Assessment

At the endpoint of the trial, animals will be anesthetized, and transcardially perfused with

0.9% saline, followed by 4% paraformaldehyde in a 0.1 M phosphate buffer. Brains will be extracted, post-fixed for 24 hours, cryoprotected in 30% sucrose, and sectioned coronally at 40 µm on a microtome. Adjacent series will be processed for Nissl staining using cresyl violet to assess neuronal density; immunohistochemistry for NeuN to quantify neuron counts via unbiased stereology; GFAP to quantify astrogliosis; and Iba1 to quantify microglial activation.

Quantification, including cell counting, will be performed on >5 sections per animal spanning the implant tract by a blinded observer. The primary histological comparison is between implant-adjacent tissue and contralateral homologous tissue. This will test the safety hypothesis that chronic stimulation under our protocol does not produce damage beyond the insertion baseline.

Statistical Analysis

A sample size of 16 was chosen per arm to detect a 50% mean reduction in seizure frequency between the distributed and sham arms, assuming a within-arm coefficient of variation of 70%, at α = 0.05 with 80% power, accommodating ~25% attrition. The primary analysis will use a negative binomial generalized linear mixed-effects model with rat as a random effect and treatment arm as a fixed effect applied to the daily seizure counts over the final 14 treatment days. The primary contrast is distributed vs sham, and the secondary contrast is distributed vs focal. Behavioral and histological endpoints will be analyzed by a mixed-effects ANOVA with Tukey HSD post-hoc correction, and Benjamini-Hochberg applied across the secondary endpoint.

Safety and Ethical Concerns

Because this study includes animals, it requires approval from multiple regulatory bodies. An IACUC protocol will be filed with Case Western Reserve University, taking into account the 3 Rs (replacement, reduction, and refinement). Replacement is not an option under the central hypothesis, as epilepsy tests require an animal that is alive and behaving. Reduction is addressed through formal power analysis and the use of a within-animal repeated-measures design. Refinement is addressed through the endpoints, which include humane euthanasia. Kainic acid is a controlled biological toxin that requires institutional biosafety registration and operation from trained personnel. All KA handling will occur in a BSL-2 area with necessary PPE. Buprenorphine and diazepam handling will be in adherence with the DEA schedule III/IV requirements. All surgical personnel will be trained and certified by the institutional veterinary staff to perform stereotaxic surgery prior to the experiments starting. Stimulation safety will be ensured by meeting the Shannon criterion (k < 1.75) and charge density ceiling (Shannon 1992). The closed-loop firmware will include safety measures to monitor the cumulative charge per electrode for 24-hour periods, with automatic shutoff when thresholds are exceeded. All data will be registered on the Open Science Framework before unblinding, including all endpoints. PHS Policy and OLAW compliance will be maintained throughout.

Limitations and Troubleshooting

The intrahippocampal KA model produces seizures that are typically unilateral, while in human TLE often spreads bilaterally as the disease progresses. The contralateral electrode placement in CA1 partly addresses this by detecting inter-hemispheric propagation and identifying animals that demonstrate this seizure behavior. Additionally, the rodent hippocampal network anatomy and seizure dynamics differ from the human anterior temporal neocortex; as such, this study must be framed as a pre-emptive study that may lead to non-human primate studies as the true clinical surrogate.

The detector thresholds determined via the aforementioned R21 simulations may not translate well to the rat LFPs, as they are trained on human intracortical EEG datasets. These signals may have different spectral content, signal-to-noise ratio, and seizure dynamics. To mitigate this, we will use a pilot cohort of 3-5 animals to recalibrate the algorithmic thresholds before the main study begins. Additionally, chronic recording in the freely behaving rodent has a relatively large attrition rate, with 10-20% of specimens being excluded due to broken wires, headcap loss, and implant infection. This was factored into the sample-size estimate and will be tracked through a ‘non-completion’ category.

References

[1] Ahmadi, N., Bagheri, M., Akhbari, M., Tafakhori, A., Joghataei, M. T., & Mehrpour, M. (2021). Patterns of seizure spread in temporal lobe epilepsy are associated with distinct white matter tracts. Frontiers in Neurology, 12, 758638.

[2] Asadi-Pooya, A. A., Stewart, G. R., Abrams, D. J., & Sharan, A. (2017). Prevalence and incidence of drug-resistant mesial temporal lobe epilepsy in the United States. World Neurosurgery, 99, 662–666.

[3] Bartolomei, F., et al. (2025). The different subtypes of temporal lobe seizures networks. Revue Neurologique.

[4] Bröhl, T., Rings, T., Pukropski, J., von Wrede, R., & Lehnertz, K. (2024). The time-evolving epileptic brain network: Concepts, definitions, accomplishments, perspectives. Frontiers in Network Physiology, 3, 1338864.

[5] Fruengel, R., Bröhl, T., Rings, T., & Lehnertz, K. (2020). Reconfiguration of human evolving large-scale epileptic brain networks prior to seizures. Scientific Reports, 10, 21921.

[6] Lee, J., Leung, V., Lee, A. H., Huang, J., Asbeck, P., Mercier, P. P., Shellhammer, S., Larson, L., Laiwalla, F., & Nurmikko, A. (2021). Neural recording and stimulation using wireless networks of microimplants. Nature Electronics, 4, 604–614.

[7] Lee, J., et al. (2025). Wireless network of distributed implantable microelectronic neural stimulators. Nature Communications.

[8] Nair, D. R., Laxer, K. D., Weber, P. B., Murro, A. M., Park, Y. D., Barkley, G. L., Smith, B. J., Gwinn, R. P., Doherty, M. J., Noe, K. H., Zimmerman, R. S., Bergey, G. K., Anderson, W. S., Heck, C., Liu, C. Y., Lee, R. W., Sadler, T., Duckrow, R. B., Hirsch, L. J., … Morrell, M. J. (2020). Nine-year prospective efficacy and safety of brain-responsive neurostimulation for focal epilepsy. Neurology, 95(9), e1244–e1256.

[9] Paxinos, G., & Watson, C. (2014). The rat brain in stereotaxic coordinates (7th ed.). Academic Press.

[10] Piech, D. K., Johnson, B. C., Shen, K., Ghanbari, M. M., Li, K. Y., Neely, R. M., Kay, J. E., Carmena, J. M., Maharbiz, M. M., & Muller, R. (2020). A wireless millimetre-scale implantable neural stimulator with ultrasonically powered bidirectional communication. Nature Biomedical Engineering, 4, 207–222.

[11] Razavi, B., Rao, V. R., Lin, C., Bujarski, K. A., Patra, S. E., Burdette, D. E., Loring, D. W., Meador, K. J., Pati, S., Hope, O. A., Smart, O., Gerard, E. E., Rolston, J. D., Rutecki, P. A., Templer, J. W., Sun, F. T., Skarpaas, T. L., & Morrell, M. J. (2020). Real-world experience with direct brain-responsive neurostimulation for focal onset seizures. Epilepsia, 61(8), 1749–1757.

[12] Salanova, V., Sperling, M. R., Gross, R. E., Irwin, C. P., Vollhaber, J. A., Giftakis, J. E., & Fisher, R. S. (2021). The SANTÉ study at 10 years of follow-up. Epilepsia, 62(6), 1306–1317.

[13] Shannon, R. V. (1992). A model of safe levels for electrical stimulation. IEEE Transactions on Biomedical Engineering, 39(4), 424–426.