Animals

All procedures were carried out in accordance with the National Institutes of Health guidelines for animal care and use and were approved by the Administrative Panel on Laboratory Animal Care of Stanford University (protocol number 30183). For the scFLARE experiments, adult wild-type male C57BL/6 mice 12–20 weeks old (Jackson Laboratory, strain number 000664) were used. For the two-photon microscopy and closed-loop optogenetic experiments, PCP4-Cre male and female mice 12–24 weeks old (RIKEN, strain number RBRC05662) were used.

All surgeries were conducted under aseptic conditions using a small-animal stereotaxic instrument (Leica Biosystems). Mice were anesthetized with isoflurane (5% for induction, 1.5–2.0% after) in the stereotaxic frame for the entire surgery, and their body temperature was maintained using a heating pad.

Kainate injection in mice

Intrahippocampal kainic acid injections were performed as described previously21. Briefly, mice were placed under isoflurane anesthesia and given local anesthetic, 0.5% bupivacaine, at the site of incision. Kainic acid (60 nl, 20 mmol l−1 in saline; Tocris Bioscience) was injected into the dorsal hippocampus (from the bregma: −2.0 mm anterior-posterior (AP), +1.25 mm medial-lateral (ML), −1.6 mm dorsal–ventral (DV)). The above protocol was modified for intra-amygdala kainic acid injections in which 100 nl, 20 mmol l−1 kainic acid in saline was injected into the right basolateral amygdala (from the bregma: −1.2 mm AP, +3.3 mm ML, −4.2 from dura DV). For both intrahippocampal kainic acid and intra-amygdala kainic acid, kainic-acid-induced status epilepticus after injection was allowed to self-terminate. For experiments conducted in the setting of acute seizures, animals were allowed to recover for 2 h (scFLARE experiments in Fig. 1c,d and Extended Data Fig. 3b) or returned to the vivarium for at least 2 weeks to allow for the emergence of chronic spontaneous seizures (scFLARE experiments in Extended Data Fig. 2b, and all calcium imaging and closed-loop optogenetic experiments).

Virus infusion in mice

For scFLARE experiments, the hippocampus was targeted using the following coordinates from the bregma: −2.3 mm AP, +1.5 mm ML and −1.35 mm DV. Adeno-associated viruses 1/2 (AAV1/2s) carrying scFLARE2 (Addgene 158700) and tetracycline response element driven expression of either mCherry or eGFP fluorescent reporters (TRE-mCherry (Addgene 92202) or TRE-eGFP (Addgene 89875)) were gifts from M. Sanchez and A. Ting, and were injected using a 10 μl microsyringe with a beveled 33-gauge microinjection needle (Nanofil; World Precision Instruments (WPI)). A total volume of 1.5 µl of virus was injected (750 nl at −1.35 DV and 750 nl at −1.55 DV) at a rate of 100 nl min−1 using a microsyringe pump (UMP3; WPI) and its controller (Micro4; WPI). After each injection, the needle was raised 100 μm for an additional 10 min to allow for viral diffusion at the injection site and then slowly withdrawn.

For two-photon and closed-loop optogenetic experiments, the FC was targeted using the following coordinates from the bregma: −1.9 mm AP, +0.15 mm ML and −1.85 mm DV. For two-photon experiments, 60 nl of AAV5-syn-FLEX-jGCaMP8f-WPRE (a gift from the GENIE Project, Addgene 162379) was injected. For closed-loop optogenetic experiments, 60 nl of either AAV1-hSyn1-SIO-stGtACR2-FusionRed (a gift from O. Yizhar, Addgene 105677) or AAV5-EF1a-DIO-mCherry (University of North Carolina Vector Core) was injected. After each injection, the needle was slowly withdrawn.

Mice were selected for viral injections and experimentation with no particular order to avoid systematic biases. Expression was verified after each experiment, and only mice with clear expression were used for further analyses.

scFLARE labeling in mice

Light was delivered 6–7 days following viral injection. For light delivery, the optical fiber implant was connected to a 473 nm diode-pumped solid-state laser (Shanghai Laser & Optics Century). Mice were allowed 2 h to recover from implant surgery before light delivery. For kainate experiments, closed-loop seizure detection and light delivery were carried out as previously described22. Briefly, LFP recording electrodes (PlasticsOne) for kainate-injected animals were connected to an electrical commutator (PlasticsOne) routed to an amplifier (BrownLee 410, Automate Scientific), and in turn connected to a digitizer (USB-6221, National Instruments) and a computer running custom MATLAB recorder and seizure detection software. When a seizure was detected, the software enabled light delivery. Animals in all groups receiving light had one single session of 10 mW 473 nm light delivered in 2 s pulses every 6 s (33% duty cycle), for a total of 10 min. Animals were euthanized and perfused 18–24 h after the end of light administration.

In vivo two-photon calcium imaging

Within 1 week of virus injection, mice were anesthetized with isoflurane and secured into a stereotaxic frame. We then inserted a 4-mm-long 0.5-mm-diameter gradient-index (GRIN) relay lens (Inscopix), which was lowered with a stereotaxic arm at a 5° angle (to avoid the superior sagittal sinus) to a target of −1.9 mm AP, +0.15 mm ML and −1.7 mm DV, which is at the bottom of the corpus callosum and just above the FC. After at least a week of recovery, a small craniotomy was performed over the ipsilateral hemisphere (−2 mm posterior, −2 mm lateral to the bregma—marked with a permanent marker during the previous surgery) under isoflurane anesthesia and mice were transferred to a floating ball where they woke up. A bipolar twisted wire (A-M Systems, catalog number 795500) tungsten electrode with gold amphenol pin connectors was slowly lowered into the CA1 hippocampal subfield while LFP was monitored for the occurrence of ictal and interictal spiking activity. Once the electrode was at an ideal depth (maximum spiking amplitude), it was secured in place with dental cement and the mouse was returned to its home cage.

Mice were habituated to the imaging setup (a treadmill consisting of a 2-m-long belt) and head fixation for at least two 20 min sessions before experimentation. Ipsilateral CA1 LFP during imaging was amplified 1,000× (A-M systems 1700) and digitized at 10 kHz. Imaging was performed on a two-photon microscope (Neurolabware) equipped with a pulsed infrared laser (Mai Tai, Spectra-Physics) tuned to 920 nm, GaAsP PMT detectors (Hamamatsu) and a ×16 water immersion objective (0.8 NA, 3.0 mm WD; WI, Nikon) and recorded at 15.49 frames per second (Scanbox.org).

Calcium imaging data were processed and analyzed using Python scripts. Motion correction was performed using the HiddenMarkov2D function of SIMA23. Binary regions of interest were drawn manually around jGCaMP8f-expressing cells visible on an average intensity projection image of motion-corrected movies. Next, the fluorescence intensity traces were extracted for each region of interest by averaging the included pixel intensities within each frame. Changes in fluorescence intensity (DF/F) traces were obtained as described previously24. LFP traces were automatically processed to detect ictal and interictal spiking as described24.

Closed-loop seizure detection and light delivery

Within 1–2 weeks of virus injection, mice were anesthetized with isoflurane and secured into a stereotaxic frame. We then performed a small craniotomy and inserted optical fibers (0.37 NA, low OH, 200 μm diameter; ThorLabs) terminated in 1.25 mm ceramic ferrules (Kientec Systems) to a target of −1.9 mm AP, +0.15 mm ML and −1.7 mm DV, which is at the inferior aspect of the corpus callosum and just superior to the FC. During the same surgery, another small craniotomy was performed over the ipsilateral hemisphere (−2 mm AP, +2 mm ML) and a bipolar depth electrode (PlasticsOne) was implanted to a depth of −2 mm DV to detect seizures from the CA1.

Following the implant procedure, animals were connected through an electrical commutator (PlasticsOne) to a Brownlee 410 amplifier; signals were digitized by an NI USB-6221-BNC digitizer (National Instruments) sampled at 500 Hz and analyzed in real time using a PC running a custom MATLAB seizure detection algorithm as previously described. Animals were also connected to a fiber-coupled diode laser (Shanghai Laser & Optics Century) with 473 nm wavelength to activate the GTaCR2 opsin. Optical patch cords (Thorlabs, Doric Lenses) directed the laser light to the mouse through an optical commutator (Doric Lenses), and were terminated in a 1.25 mm ferrule, which was connected to the implanted optical fiber with a ceramic split sleeve (Precision Fibre Products). Laser power at the source was 0.5 mW.

Continuous LFP monitoring established the presence of spontaneous recurrent seizures in individual animals, at which time an experimenter used custom MATLAB software to identify features of the early ictal electrographic signal to be used in triggering the real-time closed-loop seizure detection software. For semiautomatic analysis of spike clusters (that is, seizures) as previously described21,22, the custom MATLAB program used different detection criteria provided by the experimenter for LFP spikes (including filtering, amplitude threshold, width and template matching), LFP spike clusters (including interspike interval and intercluster interval) and artifact rejection (including different filters and signal features), which were combined using Boolean logic. The experimenter verified and, if necessary, corrected all processed files on their detection accuracy of seizure starts and ends. Spike clusters with an interspike interval of less than 1 s were included as seizures. Seizure duration values used for analysis were taken from time of trigger for the closed-loop detector to the end of the seizure. Seizures were not considered ended until the spiking rate fell below one spike every 2 s. The experimenter was blinded to whether a mouse was in the experimental or control group when selecting the inclusion and exclusion criteria and adjusting thresholds to optimally detect seizures.

Mouse perfusion, histology and imaging

After all data were collected for each mouse, the animals were euthanized by being deeply anesthetized with a mixture of ketamine and xylazine (80–100 mg kg−1 ketamine, 5–10 mg kg−1 xylazine; intraperitoneal) and transcardially perfused with 10 ml of 0.9% sodium chloride solution followed by 10 ml of cold 4% paraformaldehyde dissolved in phosphate buffer solution. The excised brains were held in a 4% paraformaldehyde solution for at least 24 h before being sectioned into 60 μm slices using a vibratome (Leica VT1200S, Leica Biosystems). For immunostaining, the slices were incubated in blocking buffer containing 1% bovine serum albumin and 0.5% Triton-X in tris-buffered saline (TBS) for 1 h at room temperature, then incubated with rabbit anti-PCP4 antibody (1:200; Sigma HPA005792) in TBS containing 1% bovine serum albumin and 0.5% Triton-X overnight at 4 °C. The slices were subsequently washed in TBS (4 × 10 min) before being incubated in anti-rabbit secondary antibodies conjugated to Alexa Fluor 647 (1:1,000; Thermo Fisher Scientific A-21245) for 2 h at room temperature. Afterward, the slices were washed in TBS (4 × 10 min) before being mounted on glass slides and covered with a coverslip using Vectashield Antifade Mounting Medium (Vector Laboratories). Imaging was performed on a Zeiss LSM 800 confocal microscope using a ×10 or ×20 objective, and a z-stack of 5–7 images was taken.

Patient selection

Patients 1–4 were considered clinically to have TLE of uncertain laterality and precise anatomical origin, and patient 5 was believed to have an occipital focal cortical dysplasia with subsequent involvement of her mesial temporal lobe. Per routine clinical protocols in our institution, these patients underwent bilateral sEEG recordings after giving informed patient consent. Access to the resulting data followed research procedures approved by the Stanford institutional review board (IRB 70482). Of note, many of the patients also belonged to a cohort that underwent sampling of thalamic targets, as detailed in our previous study25. Patient 6 underwent placement of sEEG electrodes as part of routine clinical care owing to seizure recurrence after LITT amygdalohippocampectomy; thus, an electrode was placed in his residual posterior hippocampus as part of normal seizure evaluation. Before the sEEG recordings, all patients completed a comprehensive set of evaluations, including detailed clinical history, neurological examination, neuropsychological assessment, structural magnetic resonance imaging (MRI) and scalp EEG monitoring. Patients completed additional imaging and neurophysiological studies as needed for presurgical planning, including functional MRI for language mapping, fluorodeoxyglucose positron emission tomography (PET) study and high-density electrical source imaging. The sex and gender of human research participants were based on self-reporting. Sex and gender were not considered in the study design. Due to the low number of participants, gender-based analysis was not performed.

Electrode trajectory planning

The approximate locations and number of electrodes, along with their trajectories, were planned in a multidisciplinary surgical epilepsy conference with a detailed review of presurgical data leading to the clinical hypotheses of most likely seizure onset zones. High-resolution T1, and T1 post-contrast imaging, were used for planning. To sample the posterior hippocampal tail containing the FC, the usual posterior hippocampal target was adjusted further posterior–medially, with the electrode entry point in the inferior temporal gyrus and optimized for a safe, subventricular trajectory. We used only reduced-diameter (0.86 mm) electrodes (Ad-Tech Medical) to help ensure minimal disruption to tissue.

Intraoperative workflow

Patients 1–6 were brought to the operating room where general endotracheal anesthesia was induced. Five bone fiducials were placed. A volumetric intraoperative O-arm (Medtronic) computed tomography (CT) scan was obtained with the fiducials. The image data set was then merged with the preoperative CT and T1 pre- and post-contrast MRI scans. The patient was placed in a Leksell head holder and positioned supine. The ROSA robot (Zimmer Biomet) was then attached to the Leksell adapter and registered to the patient’s head using the bone fiducials. Registration was accepted once <0.5 mm accuracy was achieved. The head was then prepped in the usual fashion. For each percutaneous trajectory, the ROSA robot was positioned coaxially. A small vertical stab incision was made with a number 15 blade. A 2.4 mm drill bit was then introduced through the ROSA drill guide, and the drill guide lowered coaxially all the way down to the scalp. Once through the inner table of the skull, a bolt (bone anchor) was placed. A reduced-diameter (0.8 mm) obturating stylet was passed slowly to create the trajectory. Once the stylet was passed to depth and then removed, a reduced-diameter (0.86 mm) electrode was passed to target depth, the inner stylet was removed and the electrode was tightened into the bolt cap.

Co-localization of electrodes

A thin-cut CT scan of the head was obtained after electrode implantation to confirm the absence of intracranial hemorrhage. In addition, the CT images were co-registered to MRI data for verification of the trajectory. The electrode coordinates in the native anatomical space were carefully inspected for every single electrode contact and manually labeled by a neurologist and anatomist based on the individual brain’s morphology and landmarks.

Intracranial recording and identification of ictal patterns

Signals were collected from multiple-contact depth electrodes with a center-to-center contact spacing of 3 mm. A continuous EEG signal was acquired with a digital Nihon Kohden EEG machine at a sampling rate of 1,000 Hz, in combination with continuous video recording. High-frequency filter, time constant and voltage sensitivity settings were adjusted to optimize visual detection of high-frequency oscillations (typically at 300 Hz high-frequency filter, 0.001 s time constant, 10 µV sensitivity). A sEEG bipolar montage including all channels was used for signal detection. Channels with excessive artifacts obscuring EEG signals were excluded from the analysis. All seizures captured were reviewed for onset zones, which were determined by visual analysis by the primary inpatient epilepsy team. Ictal onset signals identified by the epileptologists were inclusive of various morphologies, such as pathologic high-frequency oscillations, evolving fast activity, rhythmic spikes or rhythmic spike–waves. To examine the relationship between seizure activity in the FC and that in the anterior hippocampus, an epileptologist read all files (blinded to whether each LFP trace was from the FC or another region), and when clear ictal patterns were separated by 50 ms or more, a differential onset was described.

Fiber-optic insertion for LITT

Patient 6 underwent endotracheal general anesthesia in the MRI suite and was placed in a supine position with a bump and head rotation, with his left ear superior. His head was secured to the table of the interventional MR scanner by four skull pins, and the exposed temporal region was clipped, prepped and draped, with the top of his pinna bent inferiorly to expose a low temporal entry site, where a sterile self-adhesive fiducial grid (Clearpoint Neuro) was placed.

Targeting trajectories were planned using a Clearpoint Neuro workstation, and the grid was punctured with a trocar to mark the entry site. The grid was removed, a scalpel was used to make a stab incision and the Clearpoint frame base was affixed to the skull via self-tapping screws. A series of planning scans were obtained to align the frame along the correct trajectory. An MRI-compatible hand drill was then used to make a burr hole through the stab incision, and a ceramic rod was partially inserted to confirm the correct trajectory. Next, a Visualase cooling cannula (with stiffening stylet) was inserted through a reducing cannula, its placement confirmed with imaging. The stylet was removed and the optical fiber was inserted. The laser fiber and cooling lines were connected to the Visualase workstation, and temperature safety limits were set relative to the thermometric monitoring image, in the inferior lateral thalamus (particularly the lateral geniculate nucleus), basal ganglia and lateral mesencephalon, to automatically terminate laser delivery if these structures exceeded 45 °C. The initial lesion was made in the hippocampal tail remnant during real-time MRI thermometry. The laser fiber was then retracted in approximately 1 cm increments, and several overlapping focal ablations formed a tubular ablation zone encompassing the hippocampal tail remnant and FC.

Data analysis, software and code

Software used for data acquisition included Zen Blue (Zeiss LSM 800 confocal microscope image acquisition), Matlab R2019b (LFP recording, 2p imaging) and Nihon Kohden Neuroworkbench Version 08–11 (patient EEG). Data analysis was performed using the following software: Matlab R2019b, Pycharm Community Edition 2018.2.5, ImageJ 1.53, Graphpad Prism 9, OriginPro 2021b, Python 3.9.7, Pandas 1.3.4, Scipy 1.7.1, Statsmodels 0.13.2, Pingouin 0.5.2 and Seaborn 0.12.1.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.