Multifunctional hydrogel electronics for closed-loop antiepileptic treatment

Abstract

Closed-loop strategies offer advanced therapeutic potential through intelligent disease management. Here, we develop a hydrogel-based, single-component, organic electronic device for closed-loop neurotherapy. Fabricated out of conductive hydrogels, the device consists of a flexible array of microneedle electrodes, each of which can be individually addressed to perform electrical recording and control chemical release with sophisticated spatiotemporal control, thus pioneering a smart antiseizure therapeutic system by combining electrical and pharmacological interventions. The recorded neural signal acts as the trigger for a voltage-driven drug release in detected pathological conditions predicted by real-time electrophysiology analysis. When implanted into epileptic animals, the device enables autonomous antiseizure management, where the dosing of antiepileptic drug is controlled in a time-sensitive, region-selective, and dose-adaptive manner, allowing the inhibition of seizure outbursts through the delivery of just-necessary drug dosages. The side effects are minimized with dosages three orders of magnitude lower than the usage in approaches simulating existing clinical treatments.

Hydrogel electronics enables smart antiseizure management by a closed-loop therapeutic system.

INTRODUCTION

In an intelligent therapeutic system, medication should be delicately controlled for optimal treatment over prolonged periods. The concept features a dose-adaptive and time-sensitive intervention on an as-needed basis, achieving better effectiveness and more assessable outcome with external intervention (1, 2). Closed-loop strategies composing of continuous signal monitoring and feedback-based modulation offers on-demand and precise management of chronic and acute pathological conditions, such as seizure (3), diabetes mellitus (4, 5), Parkinson’s disease (6), arrhythmia (7), etc. The detected signal could trigger an immediate response to sudden disease conditions and subsequently engage therapeutic measures based on subtle pathological indications. However, very limited closed-loop systems are clinically available for treating brain diseases. For antiepilepsy therapy, the first and only clinically available closed-loop treatment uses an electrical recording and stimulation system for treating adults with drug-resistant focal epilepsy (8). The lack of success with closed-loop controlled pharmacological interventions shows the urgent need for a rational approach that combines an electrical modulation and pharmacological intervention to achieve superior efficacy (9).

Toward these goals, many efforts thus far have been focused on the modular integration of different components—such as electrical circuits, drug releasing channels, biophysical sensors, and data analytical units—to develop miniaturized devices through advanced fabrication techniques (10). These integration of these components enables coordinated operation to achieve therapeutic efficacy. For example, an optoelectronic neuromodulation system was fabricated to monitor and control bladder dysfunction in vivo (11); a neural probe based on integrated electrophoretic ion pump and recording electrodes was designed for treating neurological disorders (12); an implantable drug capsule and wearable device were combined for triggered subcutaneous drug release (13). While useful, these devices are mostly based on integrative miniaturization of electronic and mechanical systems and are sometimes limited by the bulky size and inadequate temporal-spatial resolution for physiological sensing and modulation. The fabrication complexity and the lack of smart interventional function also hinder their wide use for long-term applications (14, 15). To bridge these gaps, the combined utilization of organic electronics and stimulus-responsive materials is emerging as a promising solution for the creation of smart multifunctional devices with a relatively straightforward design to accommodate closed-loop intelligent therapeutic systems.

Stimulus-responsive hydrogel, having excellent biocompatibility and adjustable mechanical properties, can be made to release prefilled inclusions in response to specific triggers, including electrical signals (16, 17), magnetic field changes (18), and chemical reactions (19). When placed in the human body, this responsiveness can be used to create novel multifunctional devices, in which physiological sensation, pharmaceutical treatment, and regulatory feedback are all performed by a single component. Particularly, electroactive hydrogels with excellent electrical sensitivity and mechanical tunability hold great potential for neuroengineering applications in the complex in vivo environment. As organic materials, most conductive hydrogels are polymers containing charge carrier dopants, such as poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS), polypyrrole, and polyaniline, which have been extensively studied and used for biomedical sensing (20–22). Conductive hydrogels can also be synthesized with adjustable modulus, mechanical strength, and excellent machinability, giving them notable advantages in device fabrication by allowing for direct molding and three-dimensional (3D) printing (23, 24). These features also provide the possibility to accommodate unique designs of a device for optimal therapeutic performance, especially for invasive applications. Through the utilization of both the electrical properties and stimulus-responsive nature, electroactive hydrogels offer an opportunity to produce next generation closed-loop therapeutic devices with elegant design for the intelligent management of high-impact diseases.

This study developed a hydrogel bioelectronics (hydroElex) platform for intelligent disease monitoring and closed-loop antiseizure treatment in epileptic rodents. The hydroElex device, made of conductive hydrogels, was fabricated as a flexible electronic array of individually addressable microneedle electrodes, which can be closely interfaced with curved brain tissues to continuously perform electrophysiological recordings (Fig. 1). The recorded neural signal then acts as the feedback to trigger a voltage-driven drug release in detected pathological conditions, where seizures are predicted by real-time electrophysiological analysis. As a proof of concept, when implanted into 4-aminopyridine (4-ap)–induced epileptic animals, the hydroElex device was used to implement antiseizure management through a closed-loop feedback strategy. With the help of real-time monitoring of anomalous neural activity, the outburst of seizure is well-controlled by the hydroElex device in a time-sensitive, location-selective, and dose-adaptive manner. We believe that the single-component, multifunctional hydroElex device opens paths toward in vivo closed-loop neurotherapies by using organic electronics and artificial intelligence (AI). Such a strategy can be further extended to develop precise and autonomous pharmaceutical administration in a broad range of chronic diseases and medical conditions to achieve smart health management.

Fig. 1. Schematic diagram of a closed-loop bioelectronic system for adaptive antiepileptic treatment.

The hydroElex device with an array of drug-loaded and hydrogel-based microneedle electrodes was applied for multiplexed electrophysiology recording and controlled drug release. The recorded neural spiking signal acts as closed-loop feedback to trigger a voltage-driven drug release in detected pathological conditions, where seizure occurrence is predicted by real-time electrophysiology analysis. The closed-loop bioelectronic system includes a hydroElex device that works as a biosignal sensor and responsive interface for drug intervention, signal acquisition system, data storage, and data processing package (blue). Red molecules represent antiepilepsy drugs for the treatment. Green and yellow cells represent neurons in cortical tissues.

RESULTS

Synthesis of the electroactive stimuli-responsive hydrogels

The hydroElex device was fabricated as an array of in situ polymerized microneedle electrodes on a flexible thin film, where each of the electrodes can be independently addressed and function as a signal recording and drug-releasing unit (Fig. 1A). To prepare the electroactive hydrogel, PEDOT:PSS dispersions [~0.7% (w/v)] were mixed with N-(3-sulfopropyl)-N-methacroyloxyethyl-N,N-dimethylammonium betaine (DMAPS) to form a primary crosslinking network. The mixing was followed by the addition of crosslinker diacrylate-functionalized poly(ethylene glycol)-b-poly(propylene glycol)-b-poly(ethylene glycol) (PF127-DA), bis-acrylamide (MBAA), and photoinitiator Irgacure 2959 (Fig. 2A). Upon ultraviolet (UV) exposure, the precursor solution underwent in situ polymerization to form the ready-to-use conductive hydrogels (fig. S1), which contained a tuneable microscale porous network and showed no cytotoxicity (Fig. 2B, also see fig. S2). After polymerization, the resulted hydrogel was examined by Fourier transform infrared (FTIR) spectroscopy. Compared to the specific precursor components (DMAPS and PF127-DA), the disappearance of the peaks at 1639 and 1721 cm⁻¹ showed the break of C═C and C═O bonds in DMAPS and PF127-DA, respectively; the newly formed C─H bond and addition of S-phenyl group was indicated by signal at 2836 and at 1081 cm⁻¹ in the resulted DMA/PEDOTS–interpenetrating network (IPN) hydrogel (Fig. 2C) (25, 26). These characteristic peaks confirm successful hydrogel crosslinking and a physical inclusion of the PEDOT:PSS inserts. The polymerized hydrogel exhibits excellent structural stability (Fig. 2D). After 48 hours in phosphate-buffered saline (PBS) at 37°C, the volume swelling from lyophilized format was about 60% lower (210.92 ± 17.27% versus 490.56 ± 15.30%) than the commonly used polyacrylamide (PAAm) hydrogel (fig. S1). The degradation of the DMA/PEDOTS-IPN hydrogel is also slower, and only minor weight loss (26.33 ± 3.2%) was observed after a 24-day incubation in a biomimetic environment (Fig. 2E). The structural and compositional stability of the electroactive hydrogel makes it a suitable material for long-term in vivo usage. The characterization of hydrogel conductivity showed a positive correlation between the conductivity and the DMAPS concentration [10 to 40% (w/v)], ranging from 1.22 ± 0.01 to 15.14 ± 0.58 S/m (Fig. 2F). For a series of hydrogels with various DMAPS formulations, the sample was stretchable up to a strain of 430%, corresponding to elastic modulus ranging from 9.96 ± 0.14 kPa to 35.14 ± 0.14 kPa (Fig. 2G, more details in table S1), which is comparable to the modulus measured from brain tissues (3.15 to 10 kPa) (27). In contrast to other commonly used materials (e.g., platinum, tungsten, and PEDOT:PSS; Fig. 2H), a matching modulus ensures better integration with the host neural tissues with minimal contact resistivity (28).

Fig. 2. Synthesis and characterization of the DMA/PEDOTS-IPN conductive hydrogel.

(A) Workflow of synthesis procedures of the DMA/PEDOTS-IPN hydrogel. (B) Photogram and scanning electron microscopy (SEM) images of the hydrogels. The SEM images (ii, iii, and iv) show the microscale structure of the boxed region at different scales. Scale bar, 1 cm (i); 100 μm (ii); 15 μm (iii); 5 μm (iv). (C) FTIR spectra showing the chemical components of the hydrogels. (D) Measurement of the swelling volume change of the hydrogels (n = 3). (E) Characterization of long-term degradation of the hydrogels by measuring the weight loss over a 25 day period (n = 3). (F) Characterization of the conductivity of the hydrogels with different formulations of DMAPS content (n = 5). (G) Tensile test of the hydrogels. (H) Comparison of Young’s modulus among neural tissues, the DMA/PEDOTS-IPN hydrogel-based electrodes, and some commonly used electrodes for recording neural signals. The conductive materials and the nerve tissues are shaded in blue or yellow, respectively. The error bars indicate the SD. *P < 0.05. N.S., not significant, by Student’s t test.

Fabrication of the hydroElex devices

To implement a closed-loop strategy, the hydroElex device used bimodal, bidirectional signal flow at each microneedle electrode, enabling both physiological monitoring and pharmaceutical intervention. Each microneedle electrode, serving as a basic functional unit, can be individually addressed through separate wiring. The overall design of the hydroElex chip was inspired by the well-established and Food and Drug Administration (FDA)–approved Utah array (29). The device fabrication involves a convenient four-step protocol (Fig. 3A, more details in Materials and Methods). In our design, the pyramidal shape of the microneedle electrode further improves the placement and fixation of the device in the tissues after being implanted in an animal brain (Fig. 3B, also see fig. S3). The height of microneedles was measured at 673.02 ± 13.08 μm, and the tip sharpness was measured at 20.85° ± 2.18°, providing an easier access to deeper cortical tissues with minimized damage, which is important for the surveillance of abnormal neural activity and rapid release of drug molecules to a wider range of brain cells. The transparency of the hydrogel microneedles also facilitated their alignment and assembly on the polyethylene terephthalate (PET) substrate to guarantee proper wiring (Fig. 3, C and D). The hydroElex device was fabricated in such a way so as to promote a highly stable electrode substrate interface through chemical anchoring. This is to ensure that the device is capable of withstanding the mechanical strain of implantation without losing functionality. This capability was demonstrated by the fact that the device was able to withstand bending tests of at least 100 cycles without any detectable damage. The electrical properties of the hydrogel circuit part were also preserved as demonstrated by the fact that the multicycle deformation did not cause any change in the electrical resistance (fig. S4). A layer of SU-8 photoresist was used as the insulation layer to seal the circuits, while only the tips of the electrodes were exposed for biological interfacing (Fig. 3, E and F, and fig. S5). The mechanical strength of the hydroElex device was characterized using a force gauge, showing a failure level beyond 0.07 N per needle for a unloaded electrode and 0.05 N per needle for a drug-loaded [1 mg/ml; γ-aminobutyric acid (GABA) in PBS] one, respectively (Fig. 3G). In both cases, the microneedle electrodes could penetrate the brain tissues without breaking.

Fig. 3. Fabrication of the implantable hydroElex device.

(A) Schematic of a four-step fabrication process to fabricate the hydroElex device. (B) SEM images showing the side view of the casted hydrogel electrodes and the quantification of the height and the tip sharpness of the microneedle morphology (n = 6). Scale bar, 200 μm. The error bars indicate the standard deviation. (C) An overview photograph of a flexible hydroElex device. Scale bar, 0.5 cm. For the enlarged view, scale bar, 2 mm. (D) A close-up photograph of a hydroElex device with individually addressable hydrogel electrodes. Scale bar, 1 mm. (E) Top-view SEM image of the hydrogel electrodes coated with an SU-8 insulation layer. Scale bar, 500 μm. (F) SEM image showing an enlarged view of a single hydrogel electrode with (left) or without (right) insulation coating. Scale bar, 150 μm. (G) Test of mechanical strength the hydrogel electrodes with or without drug loading (GABA). (H) Exploded view of the hydroElex device integrated with a microfluidic refill component, consisting of a SU-8 insulation layer, a conductive hydrogel-based electrode array, an Cr/Au circuit layer, a soft PET substrate, and a PDMS-based soft microfluidic layer as drug supplementary channels to refill each electrode in a programmable format. (I) The photograph of a hydroElex with microfluidic channels. The hematoxylin dye (red) flows into the microfluidic channels to visualize the refilling of the selected electrodes (blue). Scale bar, 1 mm. (J) Deformation of a hydroElex device under large mechanical stress. Scale bar, 2 mm. (K) Photograph of a mouse implanted with a hydroElex device. Scale bar, 1 cm.

For long-term usage in managing chronic diseases, a drug storage or refill mechanism was designed and integrated. The hydroElex device was equipped with a microfluidic layer with multiple microscale channels for drug supplementation (fig. S6). Specifically, the microfluidic channels (80 μm by 100 μm in cross section) were fabricated in a soft polydimethylsiloxane (PDMS) layer (350 μm in thickness), which was bonded to the hydrogel microelectrode array layer (fig. S7). A special adapter cap was also included to ensure a stable and secure microfluidic supply of the drug without the need for repeated surgical procedures or replacement of the implantable device (Fig. 3, H to J). The device highlights the addition of programmable hydrogel electrodes in which therapeutics can be repeatedly supplemented through microfluidic channels (fig. S6, B and C), providing a solution for prolonged usage in chronic applications. Still, the multifunctional device was sufficiently compact for use in a freely moving mouse (Fig. 3K and fig. S8).

Electrical characterization

To characterize the electrochemical stability of the hydroElex device, cyclic voltammetry (CV) was performed. The voltammogram remained to be well overlapped even after 100 sweeping cycles in PBS (Fig. 4A). The device stability was further assessed by measuring the charge storage capacity (CSC), which was only slightly increased after 20 cycles, and was plateaued at a level of less than 2% increment after 100 cycles (35.11 ± 0.3 mC/cm²; Fig. 4B). When placed in PBS, the electrochemical impedance of the device showed a steady decrease against increasing scanning frequency from 1 to 1000 Hz (Fig. 4C) and beyond (fig. S9). The impedance stabilized at a level around 100 kilohm after being tested for more than 7 days (Fig. 4C), which is comparable to the traditional tungsten electrode (0.1 to 5 megohm). At different frequencies, a similar response in the square waveform recording was observed, when comparing the hydroElex device to tungsten electrodes (Fig. 4D). The low impedance of the hydroElex device (0.01 to 0.1 megohm at 1000 Hz; Fig. 4C) is particularly suitable for recording low-frequency local field potential (LFP) signal with a large sensing coverage as intended in this study.

Fig. 4. Functional characterization of the hydroElex device.

(A) Cyclic voltammogram of a hydroElex device in PBS. (B) Quantification of the CSC of increasing CV cycles (n = 3). (C) Impedance of a hydroElex device at different frequencies (n = 3). The inset shows the impedance measured at 1000 Hz after a long-term incubation in PBS (n = 3). (D) Representative recordings of square waveform using the hydroElex device (middle) and its comparison to a tungsten electrode (right). (E) Diagram of the ex vivo experiment for characterizing the voltage-driven drug release from a hydroElex device. The molecules were released into a block of agarose, which was collected and analyzed using spectroscopy. (F) Cyclic GABA release from a hydrogel electrode at a fixed voltage stimulation (200 mV for 3 min). The examined parameters include release amount per cycle (orange), cumulative release percentage (red), and remaining percentage (blue) (n = 3). (G and H) Triggered release of GABA and VPA from a single electrode as a function of various triggered voltages (G) and durations (H) (n = 3). The chemical was loaded at 1 mg/ml. The error bars indicate the SD.

Voltage-driven release of biomolecules

To evaluate the voltage-driven release of biomolecules ex vivo, a block of agarose gel [1.5% (w/w) in PBS] was used as a biomimetic substrate to mimic the device-tissue interface. In this assay, the released biomolecules were collected and then analyzed by spectroscopy (Fig. 4E). A neural inhibitor, GABA, was preloaded within the porous hydrogel microneedles. The encapsulated biomolecules can be triggered to release upon the application of a supplied voltage (movie S1), which is synergically affected by mechanisms including electroosmosis, electrophoresis, and the hydrogel stress gradient, etc. (30). Diffusion-based passive release (without voltage stimulation) of the drug molecules was intended to be minimized by our material development of the DMA/PEDOTS-IPN hydrogel with proper scaffold structure and charge distribution. While a slight leakage release was observed over the prolonged characterization (10 days; fig. S10), especially in the first 24 hours of the assay, the leakage of GABA molecules was observed to be associated with the encapsulation dose and was insufficient to change the physiological condition. For example, given an encapsulation of GABA at 1 mg/ml, the cumulative escape of GABA was only 3.23 ± 0.27 ng after 24 hours, which is insufficient to alter any spontaneous neural activity (31, 32). A single triggered release from the hydroElex device would be at least three orders of magnitude more than the release from passive diffusion within the same temporal window (fig. S10), suggesting a drug administration process dominated by an active voltage-driven process. In a cyclic releasing experiment up to 50 cycles (200 mV, 3 min), the device showed excellent electrochemical responsiveness, although the absolute releasing dose was observed to drop due to a reduction of the remnant GABA molecules in the microneedle hydrogel electrodes. After 50 cycles, the cumulative drug release from a hydroElex device was around 77.55% of the preloaded amount (Fig. 4F). Without any triggers, less than 5% of the preloads can be released even after an assay of 10 days long (fig. S10), suggesting an excellent drug encapsulation and a well-controlled drug administration using the hydroElex device.

Also, a dynamic control of the releasing profile can be programmed by adjusting the amplitude and duration of the applied voltage triggers. The DMA/PEDOTS-IPN hydrogel-based electrodes can sustain a voltage up to 1000 mV to trigger a pulsed release of GABA (3.34 ± 0.27 ng) or sodium valproate (VPA; 21.64 ± 6.35 ng) (Fig. 4G). At a fixed trigger voltage of 200 mV varying from 1 to 5 min, the release showed a positive correlation with the trigger duration, ranging from 2.78 to 3.31 ng for GABA or from 3.99 to 14.61 ng for VPA (Fig. 4H). In addition to GABA and VPA, similar controllable releasing dynamics was also observed for other molecules (fluorescein; fig. S11). Accordingly, by using a hydroElex device, a consistent or flexible drug administration dose can be compensated by programmed adjustment of the electrical trigger parameters (e.g., voltage and duration).

Electrophysiology recording and in vivo pharmaceutical intervention

To realize closed-loop seizure control, the hydroElex’s function of neurophysiology recording and in vivo pharmaceutical interventions was tested, respectively. The neurophysiology recording function of hydroElex was firstly tested in a rodent seizure model, which was established by the injection of 4-ap (25 mM, 500 nl) to one side of the somatosensory cortex of a mouse brain to produce occasional seizure outburst and propagation (33). After an initial evaluation of 4-ap–induced abnormal neural spiking, of the nine active electrodes, three nearest microneedle channels (Ch1, Ch2, and Ch3) with an offset of 2, 3, and 4 mm posterior to the injection site were selected for surveillance purpose in this experiment (Fig. 5A and fig. S12). Upon the injection of 4-ap, a surge of neural spiking was observed by electrical recording of LFP and was processed by the associated Fourier power spectrum (Fig. 5B). The multichannel recording provides a spatial and temporal differentiation to show a spatial propagation of induced high-frequency spiking activity from the 4-ap injection site, which was first detected by Ch1 and subsequently by Ch2 and Ch3 (Fig. 5C). Because of the immediate proximity to the injection site, Ch1 recording was selected as the major source for the feedback signal to capture the onset of a seizure outburst, which would otherwise spread to wider cortical regions without any pharmaceutical intervention (Fig. 5D). Similar recording was acquired using multiple tungsten electrodes, suggesting a comparable recording performance as the widely used tungsten-based electrophysiology system (fig. S13). For predictive feedback to control the drug administration process and to combat a monolithic propagation of a seizure event from the origin point, a quantitative analysis of the LFP recording was derived to improve the antiseizure management. For example, the power analysis confirmed an up-regulation of neural spiking activity at various oscillation frequencies (δ for 1 to 4 Hz, θ for 4 to 8 Hz, α for 8 to 13 Hz, β for 13 to 30 Hz, and γ for 30 to 70 Hz; fig. S14), which is consistent with the documentation by previous studies (34). The in vivo electrical recording by a hydroElex device proves the feasibility of effective and accurate temporospatial surveillance of neural physiology.

Fig. 5. Combined function of electrophysiological recording and triggered pharmaceutical intervention in a hydroElex implant.

(A) Illustration of the implantation of hydroElex device and 4-ap injection site on a mouse brain. (B) Representative LFP recordings (gray) and associated power spectrum before (top) and after (bottom) seizure induction by 4-ap injection. (C) Representative LFP recordings from three adjacent electrodes with increasing distance from the seizure induction site (Ch1, 2 mm; Ch2, 3 mm; Ch3, 4 mm). Red arrowheads indicate the appearance of induced abnormal high-frequency spiking activity. (D) Statistical analysis of seizure-like events (SLEs) in three different channels before (Ctrl) and after 4-ap injection (SLE, n = 6). (E) Timeline of in vivo drug intervention for antiseizure management by hydroElex when fully developed epileptic activity were detected by the same device (hydroElex clearance). (F) LFP recording and the associated power spectrum showing the antiseizure efficacy. (G) Quantification of the SLEs in the epileptic animals at different stages (“SLE,” “Healthy state,” and “hydroElex clearance”; n = 6). (H) Analysis of the amplitude and duration of abnormal β oscillation (13 to 30 Hz) at different stages. The error bars denote the SD. n = 6. *P < 0.05 by Student’s t test.

Thereafter, in vivo pharmaceutical intervention by hydroElex was evaluated in epileptic mice. A seizure-like event (SLE) was characterized by at least a 6% increase of the power of β oscillation, which was used as the threshold feature to trigger a voltage-driven GABA release (voltage of 200 mV for 3-min trigger, ~3 ng of GABA). Drug release was executed after an SLE was fully developed and was detected from multiple channels. Upon the drug release, the number of evoked SLEs in this “clearance” condition was significantly reduced (20.6 ± 1.6%) in comparison to the pretreatment stage (Fig. 5, F and G). The duration of high-power β oscillation (6% above baseline) was also reduced as the result of curbed seizure progression (Fig. 5H). Sole application of a 200-mV voltage stimulation through a blank (no GABA loading) hydroElex electrode cannot provide any therapeutic benefit to counter-act the seizure onset (fig. S15). In a more active “prevention” mode, the trigger threshold was determined as 3.5% increase of the power of β oscillation, which was shown to further improve the antiseizure efficacy by preventing the deteriorating development of a seizure outbreak (fig. S16). These results emphasize the timely on-demand pharmaceutical intervention in controlling a seizure outbreak and lay down the foundation of multimodal interventional function for an ultimate closed-loop strategy.

Closed-loop antiseizure management

As the proof of concept for closed-loop seizure management that combines electrical recording and feedback-based adaptable drug administration, we demonstrated the feasibility and effectiveness of the hydroElex device in an epileptic mouse model (Fig. 6A). With real-time signal monitoring and processing, hydroElex could give an on-demand and adaptive dosage of drug release to manage upcoming SLEs (Fig. 6, B and C, also see fig. S17). In the self-adaptive adjustment of the drug administration experiments, epileptic symptoms were suppressed, emphasizing the importance of the temporal, spatial, and dosing control of drug administrations in a therapeutic practice, which can be delicately achieved by the predictive processing of the feedback signal in this closed-loop neurotherapy. In addition to a real-time extraction of the power analysis of the β oscillation as the trigger for drug releases (fig. S18), the drug dose was also adjusted by varying the duration of the voltage trigger to compensate for the surged power variation (β oscillation) closely related to upcoming SLEs (Fig. 6C, also see GABA case in fig. S17). Specifically, different drug dosages (for VPA, 8.6 to 15.9 ng) were autonomously applied to control seizure outbreaks based on the increment of pathological β oscillation power. Starting from a minimal dose in response to a threshold of 3.5% increment above the baseline, the drug dosage increases by 3 ng with every 1% power increment, up to 6% power change, which determines the maximal drug dosage (Fig. 6C). While an electrode was engaged for drug releasing, the function (recording or drug releasing) of the other electrodes were not affected in a multichannel hydroElex device (fig. S19), providing the opportunity for complex pharmaceutical intervention at multiple seizure occurrence sites.

Fig. 6. The closed-loop strategy for seizure control.

(A) Schematic diagram for hydroElex-assisted closed-loop seizure management based on a real-time multisite LFP recording and adaptive drug delivery on epileptic rodents. (B) Representative raw data showing the whole process of a closed-loop adaptive anti-seizure management. Continuous real-time surveillance of neurophysiology was performed using a hydroElex implant. Abnormal surge (>3.5%, indicated by red arrowheads) of β oscillation in the power spectrum was extracted as the feedback feature to trigger adaptive dosing of VPA. The animal’s recovery was also monitored by neurophysiological recording after a series pharmaceutical intervention (five VPA administrations), which were autonomously engaged and stopped. (C) Adaptive dosing control by varying voltage stimulation (200 mV, 3 to 6 min) based on the change of β oscillation amplitude to trigger drug release from an individual electrode of the hydroElex device. (D and E) Comparison of different administration strategies for antiseizure management using antiepileptic drug, VPA. The efficacy is evaluated by the frequency of SLE [(D) n = 6] and abnormal β oscillation [(E) n = 5]. The error bars denote the SD. *P < 0.05 by Student’s t test.

For benchmarking, the hydroElex closed-loop system was further compared to several clinical practices for treating epilepsy with the use of a common antiepileptic drug, VPA. The drug is typically administrated by oral intake or intraperitoneal injection to prevent or treat seizure outburst and can also be directly infused into the central nervous system (CNS) by implantable microfluidic devices (35). Accordingly, intraperitoneal injection before (intraperitoneal prevention, ~500 μg per mouse) or after seizure outburst (intraperitoneal clearance, ~2500 μg per mouse), and direct CNS infusion (100 μg per mouse) were designed and executed to simulate different clinical VPA administration strategies, as a benchmark for the hydroElex device (fig. S20). Consistent with our results with GABA, the closed-loop seizure monitoring and intelligent VPA administration by using a hydroElex device successfully inhibited seizure outburst after 4-ap induction by multiple preventive releases of small doses of VPA at nanogram scale (Fig. 6, B and C), achieving the best therapeutic effects with substantially lower total dose around 50 ng, which is at least three orders of magnitude lower than the usage in other conditions with different VPA administrations. Both the SLE activity and the high-power β oscillation were significantly lower at the tested doses (Fig. 6, D and E). The hydroElex-based closed-loop strategy adaptively uses multiple small doses to achieve significantly better therapy. The animals in hydroElex group was completely rescued from any seizure outbreak (>6% increment of pathological β oscillation; Fig. 6, B to E).

While VPA is generally well tolerated by the recipients, lower dose can further reduce the drug’s side effect by reducing potential hepatoxicity, gastrointestinal disturbances, platelet disorders, tremors, etc. (fig. S21) (36, 37). The efficacy improvement would be even more evident from a translational perspective to large models (e.g., human brain). The hydroElex platform precisely controls drug release dosage (up to 16 ng of VPA for a 6-min delivery at 200 mV; Fig. 6C) for effective seizure control in mouse brains, which is equivalent to an approximate dose of 3 μg for a 60-kg human subject (38). This is still three orders of magnitude lower than the regular oral intake of VPA in the range of 2.5 to 10 mg every day.

In summary, the hydroElex device and its associated closed-loop treatment system have been demonstrated to maintain epileptic animals as free from seizures in a low-dose manner by maximizing the efficacy of the drug that is delivered through the high-resolution spatial and temporal control of the drug release and dynamically adjusting the treatment based on the response of the physiological environment. The ability to treat the intended symptoms in a highly specific way is the core concept of intelligent therapeutics, which is demonstrated by the hydroElex device.

Long-term biocompatibility and chronic drug supplement

Considering the long-term use and potential clinical translation, we further tested the biocompatibility of the hydroElex implant over a 3-week experimental period. Structurally, the DMA/PEDOTS hydrogel-based device was generally stable, only minor degradation was observed over a long-term stay in a biological environment (fig. S25). It is likely that the degraded hydrogel polymers breakdown into large fragments, instead of small chemical molecules; so the fragment diffusion is largely restricted by the tight device-tissue interfacing, and we did not identify any hydrogel-related compounds in the fluid environment surrounding the electrode at different experimental stages (figs. S23 and S24). Even at a high temperature of 42°C, the crosslinking of the hydrogel was well maintained throughout the experimental window.

The change of mechanical properties over long-term usage was then evaluated. We implanted the hydroElex device into a piece agarose gel (1.5% (w/w) in artificial cerebrospinal fluid (aCSF) to simulate the in vivo implantation condition and incubate the setup at 37°C for 21 days. The swelling of the hydrogel electrode mostly happened in the early stage after implantation and gradually reached a plateau at 124.3 ± 23.0% after 21 days (fig. S25), implying an averaged water uptake about ~0.04 ± 0.005 μl per electrode, which translates to an extremely low uptake rate (0.002 μl/day) and would not disturb the electrolyte balance at the implantation site, considering cerebrospinal fluid (CSF) perfusion rate at 0.35 μl/min (39). With a soft modulus and similar bending stiffness to the brain tissue, the low water uptake volume and slight volume swelling would not cause any mechanical stress to surrounding brain tissues for long-term in vivo usage (40). Electrically, both the ex vivo (13.34 ± 2.80 S/m at day 0 versus 16.38 ± 4.21 S/m at day 21) and in vivo (13.34 ± 2.80 S/m at day 0 versus 14.02 ± 2.37 S/m at day 21) characterization showed that there was no significant conductivity change over the experimental period (fig. S26). The electrical recording performance was also maintained throughout the long-term implantation in a brain (fig. S26).

Functionally, the hydrogel electrodes on the hydroElex device exhibited well-preserved structural and morphological integrity when taken out from the mouse brain after recording (>3 hours; Fig. 7A). At the implantation site, the trace of tissue penetration by the hydrogel electrodes could be observed, demonstrating the ability of the electrodes to penetrate into the tissue of interest (fig. S22). The preservation of healthy nerve cells at the device-tissue interface demonstrates good biocompatibility and that the device implantation was minimally invasiveness, which was further evidenced by a chronic assay up to 4 weeks long (Fig. 7, B to G). In comparison to the sham control (no implant), there were no significant changes in microglia, astrocyte, or neuron recruitment to the regions near the penetration site. We also characterized the level of the immune response by measuring the expression of an inflammatory biomarker, tumor necrosis factor CD68. The hydroElex device caused almost negligible inflammatory response over the 1-month period, although a slightly more inflammation was observed at the early stage (2 weeks) after implantation (Fig. 7, D and H).

Fig. 7. Long-term in vivo biocompatibility of the hydroElex.

(A) Photogram of a hydroElex device after being used in a mouse brain. Scale bar, 500 μm. (B) Fluorescence microscopic images of a coronal cortical slice with a hydroElex electrode insertion track [Iba1, red; glial fibrillary acidic protein (GFAP), green; NeuN, blue]. Scale bar, 100 μm. (C) Fluorescent microscopic images of the cortical regions surrounding a hydrogel electrode 4 weeks after a hydroElex device implantation. Scale bar, 50 μm. (D) Staining for inflammatory biomarker, CD68, in coronal tissues isolated from a brain implanted with a hydroElex for 4 weeks. Scale bar, 50 μm. (E to H) Statistical analysis of different cell component changes related to the device in vivo biocompatibility, including microglia (E), astrocytes (F), neuron (G), and CD68⁺ cells (H). For (E) to (H), n = 5. Error bars denote the SD. N.S. by Student’s t test.

DISCUSSION

In this study, we demonstrate a closed-loop strategy for intelligent neurotherapy using multifunctional organic bioelectronics. The core component is based on a single-component and multifunctional organic device, hydroElex, which can be used to perform electrophysiology recording and to intelligently control in vivo drug release. The two different functions are connected by a feedback mechanism in the closed-loop treatment, where the recorded electrophysiological signal is analyzed and extracted as a trigger to control a voltage-driven drug release process. The pharmaceutical intervention conversely rescues the brain tissue from abnormal conditions (e.g., seizure outbreak), as reflected by continuous surveillance of neural activities. In epileptic animals, the closed-loop strategy was successfully used for intelligent antiseizure management, in which the timing and dose of pharmaceutical intervention were adaptively determined by a predictive analysis of cortical spiking conditions.

The overall design of the hydroElex system was inspired by the well-established and FDA-approved Utah array, which is a microelectrode system consist of multiple microneedle-shaped metal electrodes for electrical recording of neural signal in both research and clinical applications (29). The choice of a microneedle structure over a planar configuration is motivated by the multiple functions to be performed by a hydroElex device, including electrical modulation, drug loading and release, as well as brain tissue interfacing. Similar to the Utah array, the microneedle electrodes are easily fabricated as a high-density array, which allows for good spatial resolution and multiplexity for electrical recording and stimulation. Our design of a pyramid-shaped electrodes can effectively penetrate brain tissue and also provides a fixation mechanism to secure the device placement after implantation. For pharmaceutical modulation, the microneedle electrode also benefits the drug loading and release capacity when compared to a planar morphology. While the tissue penetration step may cause damages to the host brain, it also provides the opportunity to access neuronal cells at different cortical layers electrically or chemically (Fig. 7), which would be critically important for fast engagement of the pharmaceutical intervention. Aiming for drug delivery in the middle depth of the mouse cortex, the electrodes were fabricated to a height (~670 μm) approximately half of that of the typical mouse cortex (~1200 μm). The tight interface between the microneedle electrode and host tissue aids to further localize the therapeutic treatment and maintain a high local therapeutic concentration by restricting the spread via diffusion of the delivered drugs. The hydroElex device can be adapted to other configurations for different applications. For example, planar electrodes could be fabricated on flexible substrates to make nerve cuff devices, so that the single-component hydrogel electronics can be used for signal recording and therapeutic intervention on peripheral nerve or the spinal cord (41, 42).

From a material perspective, the hydroElex device was fabricated from a photo-polymerizable conductive IPN hydrogel using the biocompatible DMAPS and PEDOT:PSS as the major components. Many other conductive hydrogels are formed by multiple steps of polymerization of structural monomers and conductive molecules to balance the conductivity and mechanical stability (43–46). In contrast, the use of DMAPS simplifies the synthesis process and forms a hydrogel framework through a one-step in situ polymerization. The material design fully uses the functional groups on the end of DMAPS chains to improve the electrical and mechanical properties of the resulted hydrogels. The sulfonic acid group was used to electrostatically interact with PEDOT⁺; the acrylate group was used to form a stable polymer network with the PF127-DA and MBAA. Therefore, simply adjusting the DMAPS content is sufficient to vary the hydrogel crosslinking structure to render varying mechanical strength for tissue penetration, stable structure to resist deformation, and small micropores for drug encapsulation (Fig. 2). At the microscale, the electroactive hydrogel electrodes are strong enough to penetrate cortical brain tissue without compromising the structural integrity. At the macroscale, the hydroElex device has a flexible construction with a modulus matching that of brain tissues (28), therefore ensuring proper integration with complex geometries of brain surfaces for better signal transmission. Furthermore, minimizing the mechanical mismatch between the implant device and neural tissue improves the performance of the device, as an electrical signal recorder and a drug delivery mechanism, by reducing gliosis, inflammation, and promoting neuronal preservation (47, 48). In this study, the DMA/PEDOTS hydrogel-based device was fabricated with a modulus of 13.73 ± 0.01 kPa, which is closely comparable to brain tissues of 3.15 to 10 kPa (27) and greatly contributes to the prevention of glial scar formation at the electrode-tissue interfaces. The SU-8 used in the insulation layer has a higher modulus than that of the neural tissue; however, through a reduction of the thickness of the SU-8 to the microscale, the mechanical matching between the implanted device and the host tissues was maintained. It has been reported that SU-8 neural probes of 800 nm in thickness have a bending stiffness similar to that of neural tissue and do not induce any glial scarring or acute immune response when implanted in a rat’s cortex (49). Similarly, in our study, the thickness of SU-8 insulation layer was a few micrometers (μm). We were therefore still able to see that the hydroElex exhibited very good long-term biocompatibility. Minimal glial scar or inflammatory responses were identified in close proximity to the electrodes 2 and 4 weeks after the implantation (Fig. 7). In the future, viscoelastic encapsulation layer can be further explored for even softer electrical insulation. For example, alginate-based, physically entangled, viscoelastic material or hydrophilic polyurethane-based hydrogel can be considered for fabricating a next generation of hydroElex (50, 51).

For better electrical performance, a denser crosslinked network of the conductive hydrogel can be created. This is because it increases the ionic strength that can weaken the electrostatic attraction between PEDOT⁺ and PSS⁻, which increases the conductivity through the π-π stacking of PEDOT⁺ (52). Because of its good carrier mobility and electrochemical stability, PEDOT:PSS was used as a dopant in the hydrogel to improve the electrical performance. The increased conductivity is especially important for resolving high-frequency electrophysiological signal. Comparing the recording performance of the hydroElex to the gold-standard tungsten electrodes (fig. S13), the signal power and signal-to-noise ratio showed similarly good performance. This was observed at all neural oscillation frequencies for two types of electrodes. For controlling the therapeutic release using the hydroElex device, a working voltage less than 200 mV was sufficient to drive necessary drug release. This triggering voltage is substantially lower than the required level (usually >800 mV) in some previous studies, which were based on electrophoretic techniques to achieve a similar releasing rate, and warrants a safe and long-term operation in brain tissues (53, 54). The electroactive release is likely achieved through a synergy contributed by multiple mechanisms. When an electrical field is established across the electrodes, an electrophoresis effect would drive the movement of negatively charged molecules (e.g., GABA and VPA) toward the anode to facilitate their release. Meanwhile, also effected by the electric field, the electroosmotic flow will be induced in the anode-to-cathode direction (55). In addition, the DMA/PEDOTS-IPN hydrogel itself could undergo slight deswelling or swelling subject to the stress gradient established upon the application of the electrical trigger (30), which could favor drug release, or vice versa (fig. S27). While the three processes are likely to happen simultaneously to affect the movement of ions, water, or charged molecules in an electric field (56, 57), subsequently resulting in the triggered release, yet the exact contribution from electrophoresis, electroosmosis, or hydrogel stress gradient remains to be resolved by further biophysical investigations.

Recently, many efforts have been made to create miniaturized, multifunctional, neural interfaces for in vivo brain modulation using chemical, electrical, optical, or mechanical stimuli (58, 59). For example, a series of devices incorporating microfluidic ion pump (μFIP) and multiple electrodes (neural probe or electrocorticography device) were reported for on-demand drug delivery and local signal recording (12, 60, 61). Most of these devices are based on modular integration of different parts to perform different functions; the addition of signal recording components to μFIP could increase the surgical invasiveness and complexity for implantation and also complicate the fabrication process. Alternatively, our hydroElex device consists of a single component and relies on a combinatory utilization of the electrical, mechanical, and structural properties of a conductive hydrogel to perform different functions. When fabricated as an array of microneedle electrodes on a flexible substrate, every single electrode can be individually addressed to conduct electrical recording and to control drug releasing, which enables closed-loop therapeutic strategies. The single-component design helps improve the device biocompatibility by avoiding the use of different materials as it is common in modular devices; it also leads to a more compact size for in vivo applications and markedly simplifies the device fabrication by using a one-step polymer casting process, which would otherwise involve complex layer-by-layer protocols (62).

As a preferable solution for neural training, rehabilitation, and disease management, the key to a closed-loop strategy is a feedback mechanism that can act bimodally between physiology surveillance and therapeutic intervention to enable intelligent tuning of the timing and dose of associated drug administration. In this framework, a timely on-demand pharmaceutical intervention could take effect before the development of wide-spread pathological symptoms to avoid irreversible organ-level tissue damages. In our proof-of-concept demonstration for antiseizure management, the occasional outburst was continuously monitored by a hydroElex device, from which predictive indicators for seizure onset was derived to trigger a drug release to disrupt the seizure progression. In the efforts to mitigate a series of seizure outbursts over several hours long, the amplitude of observed SLE varies; therefore, a fixed drug dose may suffer from under- or overdosing risks, where underdosing may be ineffective to prevent seizure progress and overdosing may cause aversive short- or long-term side effects, such as drug resistance and desensitization, etc. (1, 63). The feedback-based dose adaption further makes a closed-loop disease management advantageous over traditional methods, especially for treating chronic diseases. As demonstrated in this study, for each detected SLE onset, a necessary drug dose was determined and triggered based on a real-time analysis of the neural activity of multiple aspects. This includes spiking frequency, SLE amplitude, and previous drug delivery outcomes, ultimately achieving the optimal therapeutic effects with minimal drug doses. In addition to a temporal selectivity, the LFP recording by an array of hydrogel electrodes can track a multisite changes of the neural activity. Accordingly, a compatible scope of pharmaceutical release can be engaged by specific electrodes where a local pathology is detected. The electrode structure was designed to allow a direct access to four cortical layers spanning around ~600 μm below the dura surface from layer I to V, which is reported to be closely related to seizure propagation and treatment (33). A localized engagement of pharmaceutical intervention provides another dimension with spatial selectivity to precisely control and minimize necessary drug dose.

Looking forward to intelligent neurotherapies, we introduced some AI components into the demonstrated antiepileptic system, which primarily resides in the autonomous detection of pathological brain signals and adaptive engagement of pharmacological intervention. Specifically in our study (Fig. 6), different dosages of VPA were autonomously applied to control seizure outbreaks based on the increment of pathological β oscillation power. In this study, the engagement of individual electrode closest to the seizure point was sufficiently effective to prevent a potential seizure outbreak (Fig. 6). For larger brains, clustered and scaled engagement of multiple electrodes for pharmaceutical intervention can be readily implemented as part of the AI-facilitated closed-loop strategy. The topography of electrophysiology sensing could be treated as an input for training an AI model to derive the corresponding drug release pattern. Furthermore, advanced AI algorithms could also benefit the closed-loop operations in two aspects: (i) for predicting pathological brain signals and (ii) for analyzing the post-intervention efficacy. Machine learning approaches—such as latent variable, support vector machine, and neural networks—could leverage their strength in pattern analysis to systematically study the pathological features associated with multidimensional electrophysiological signals before any seizure onset. Even for more complex brain diseases, such as depression, the associated subtle electrophysiological pathology might also be resolved by machine learning and then used to engage drug intervention. Equally important in a closed-loop system, the trained AI model can then be applied to evaluate the post-intervention efficacy for accurate and timely feedback generation. Recent studies in the prediction of human seizure also reveal possible feedback mechanisms by detecting preictal pathological signals (64), which can reasonably serve as the trigger to control therapeutic intervention in an adaptive and closed-loop manner. The combination of these findings with AI could promote closed-loop therapy for different diseases.

To unleash the full potential of the hydroElex technology, it is important to recognize its current limitations for further improvement. For example, we observed that VPA release exhibited a much larger responsive dynamic range to the electrical trigger than GABA, likely due to combined effects from the molecular size and charge of the drug. Therefore, tailored stimulation programming according to specific drug molecule can be made for improved releasing control. Further optimization of the charge distribution and doping strategy in the conductive hydrogel will also be useful. The microneedle electrodes were fabricated with fixed dimensions, which limits the spatial resolution in depth and regional specificity. As distinct cortical layers may be differentially associated with seizure generation, signals from cross-nucleus interaction among multiple brain regions could be an important feature for evaluating disease pathology or therapeutic efficacy. The hydroElex device design could be improved by taking considerations of the geometrical and geographic factors of a brain (65). The inclusion of electrically switchable chemical sealing by advanced control of ionic exchange at the tissue-deceive interface could prevent unwanted chemical leakage from the porous hydrogel electrodes. In addition, although the soft hydrogel-based systems can seamlessly interact with brain tissues with minimal inflammatory responses, the biocompatibility could be further improved by surface functionalization with bioactive coatings to enhance the long-term stability significantly (66). A wireless transmission component could also enable more powerful remote analysis of the electrophysiology signal for on-chip closed-loop surveillance and feedback (67).

From a translational perspective, this study provides a proof of concept for an antiepileptic neurotherapy by local infusing therapeutics directly to the neural tissues, which was implemented by using the hydroElex implant. This is particularly important for some drugs that are not permeable to the blood-brain barrier and cannot be administrated via oral or circulatory routes. For use with human subjects, the hydroElex technology needs to be scaled up to accommodate a significantly larger brain. One immediate extension is to build larger devices with a higher throughput of electrode addressability to make it sufficient to conduct effective and accurate temporospatial surveillance of neural physiology and to provide on-time pharmaceutical intervention during seizure occurrences over a large brain surface area. Accordingly, the device would require longer and thinner electrodes for access tissues at different depth and spatial reorganization for region-specific coverage. Proper optimization of the drug loading and release dosage would be needed to accommodate the requirement of a much higher capacity.

Although the pathological relevance of the 4-ap–induced seizure model is limited, it is a model that can mimic the sporadic seizure outbreak and is widely used in early-stage exploration of antiepileptic management (68). We believe that the results acquired in this research demonstrate a creative solution on how to use a closed-loop strategy for controlling seizure outbreaks using a hydrogel-based implantable device. This shows the promise for coping with more complex epileptic pathologies with further development in the future. Beyond the use in cortical brain, the hydroElex technology can also be extended to implement antiseizure management engaged via remote neural intervention, thus a chemical approach would be preferred due to the existence of the multiple circulatory options (e.g., blood and CSF). CSF perfusion rate can reach 0.35 μl/min (39), which would be quite efficient to bring chemical compounds to different parts of deep brain regions if the drug release is triggered somewhere in spinal cord (54, 69). In this sense, the closed-loop strategy combining electrical and chemical modalities shows great promise for more advanced therapeutic systems. Besides the demonstrated antiepileptic application, we believe that the closed-loop strategy has a great potential to be extended for smart health management in many other diseases and conditions, such as heart dysregulation, Parkinson’s diseases, bionic prosthesis, etc.

MATERIALS AND METHODS

The materials used for fabricating the hydroElex device are mostly acquired from Sigma-Aldrich, and the reagents used for cell and animal related experiments are mostly from Thermo Fisher Scientific. The materials from other vendors are specified accordingly otherwise.

Synthesis of crosslinker PF127-DA

Pluronic F-127 (PF127) was acylated following a previously reported protocol (25). Briefly, 5 g of PF127 and 0.07 g of triethylamine (J&K Chemical) were dissolved in 25 ml of anhydrous dichloromethane in ice bath with N₂ protection. A total of 600 μl of acryloyl chloride was slowly added to the solution above in a dropwise fashion over 20 min. The mixture was then stirred for 24 hours, and the reaction temperature was gradually increased to room temperature (RT). The resulted product was concentrated by rotational evaporation to remove the solvent and then dialyzed (cutoff molecular weight, 3500 Da) against deionized (DI) water for 3 days. The purified product was obtained by lyophilization, and the chemical structure of PF127-DA was confirmed by FTIR spectroscopy.

Synthesis of DMA/PEDOTS-IPN hydrogels

To prepare the DMA/PEDOTS-IPN hydrogel precursor solution, DMAPS was slowly added to the filtered PEDOT:PSS dispersion under constant stirring to induce gelation of PEDOT:PSS. The crosslinker PF127-DA, MBAA, and photoinitiator Irgacure 2959 (3% of total solid mass) were dissolved in cold DI water and mixed with the PEDOT:PSS solution. DMA/PEDOTS-IPN hydrogels [14% PF127-DA (w/v)] of four different DMAPS concentrations [10, 20, 30, and 40% (w/v)] were prepared in this study (table S1). The polymerization of hydrogel precursor solution was induced by UV exposure (30 mW/cm², 365 nm, Blak-Ray) for 25 minutes. Specifically, the hydrogels were synthesized using DMAPS as a backbone, to which conductive PEDOT:PSS was electrostatically linked, rendering an IPN structure for improved conductivity and mechanical strength. The DMA/PEDOTS-IPN30 hydrogel was used for the subsequent experiments and hydroElex device fabrication.

Material characterization

Details regarding the structural, chemical, mechanical, and electrical characterization of the DMA/PEDOTS-IPN hydrogel are included in the Supplementary Materials.

Fabrication of the hydroElex device

The fabrication of hydroElex devices involves four steps, including substrate circuit patterning, polymer linker coating, microneedle casting, and insulation packaging. The PET substrate (100 μm in thickness) was cleaned in ethanol (100%) by ultrasonication. To layout the control circuits, a standard photolithography-based lift-off process was used to pattern an Cr/Au (10 and 100 nm, respectively) circuit on a PET substrate (20), which was subsequently cleaned with acetone, ethanol, and DI water and dried in air. The cleaned substrates were treated with oxygen plasma for 10 min and then placed in 5 ml of a silane solution [100 ml of DI water, 10 μl of acetic acid with pH 3.5, and 2 weight % of 3-(trimethoxysilyl) propyl methacrylate (TMSPMA)] to incubate overnight at RT (70). The TMSPMA coating on the substrate can form covalent bonding with the DMA/PEDOTS-IPN hydrogel for a stable connection. The substrates were rinsed with ethanol and dried before further processing. To make DMA/PEDOTS-IPN hydrogel microneedle by replicate molding, the hydrogel solution was poured into a PDMS mold and degassed. The mold is a 6 × 6 microneedle array with a pitch distance of 1 mm. Each electrode was designed to be a pyramidal microneedle with a 250 μm in base width. The final height of an electrode was around ~700 μm. A linker-conjugated substrate was aligned against a hydrogel-loaded mold. After removing extra hydrogel solution, the setup was exposed to UV irradiation (30 mW/cm², 365 nm, 25 min) to induce polymerization. The resulted microneedles were revealed after a careful separation of the mold from the substrate. The devices were briefly rinsed in PBS to remove the unreacted monomers and dried in air overnight. For proper insulation, a layer of SU-8 2002 photoresist (MicroChem) was coated to cover all circuit regions but the microneedles. The coated devices were baked at 105°C for 2 min and exposed to UV (30 mW/cm²) for 2 min. The final devices were tested for conductivity among different conductive routes to rule out circuit defects.

For long-term use, the hydroElex was integrated with a microfluidic PDMS layer for drug supplement. To be specific, the substrate was drilled with laser micromachining to create the micro holes (75 μm in diameter) for solution flow. Subsequently, similar procedures were conducted to drill the Cr/Au patterning on the PET substrate. Then, the back of the PET substrate was subsequently treated with oxygen plasma and (3-aminopropyl) triethoxysilane solution [5% (v/v) in DI water] to form a linker coating layer. The patterned PDMS layer was bonded with PET substrate covalently through plasma activation. The following microneedle casting process was similar to the above version without the microfluidic layer.

Voltage-driven drug release assay

For ex vivo characterization of the voltage-drive drug release from a hydroElex device, a customized manipulator platform was built to control the placement of the device on a block of agarose gel [1.5% (w/w), in PBS or aCSF] to mimic the device-tissue interaction. A SourceMeter (Keithley 2612B System) was used to provide a fixed voltage bias (50 to 1000 mV) to trigger molecular release from the hydrogel electrodes (connected to cathode) to the agarose (connected to anode). Followed by a triggered delivery, all the agarose gel was collected and filtered by running through an RNA spin column, which was centrifuged to collect the supernatant containing the released molecules. For the release of fluorescein, the fluorescent intensity of the supernatant was measured using a spectrometer (Molecular Devices). For triggered release of GABA, the collected supernatant was first amplified using an enzyme-linked immunosorbent assay–GABA detection kit (Abnova) and subsequently analyzed by the spectrometer.

Voltage-driven in vitro cyclic releasing assay

For the in vitro cyclic releasing characterization, the setup is the same with the voltage-drive drug release. A hydroElex device was interfaced with a block of agarose gel [1.5% (w/w), in PBS or aCSF]. The hydrogel electrodes were connected to the cathode and the agarose was connected to the anode. A constant voltage (200 mV) was supplied by a SourceMeter (Keithley 2612B) for the designated trigger duration. After each release cycle finished, there was a 10-min interval to reset and to simulate realistic recovery scenarios. Then, 1, 3, 5, 10, 20, 30, and 50 cycles were conducted, respectively. Subsequently, the agarose gels were filtered, and the supernatant was collected by centrifugation for further molecular analysis.

HPLC analysis

The amount of triggered release of VPA was determined by high-performance liquid chromatography (HPLC). A total of 200-μl sample containing released VPA was collected, and then, 100 μl of 1,1-cyclohexanedicarboxylic acid and 100 μl of H₂SO₄ (3 mol/liter) solution were added and homogenized. A total of 2 ml of n-hexane was added and mixed by vortexing. The mixture was centrifuged at 14,000 rpm for 10 min to collect the supernatant, to which 200 μl of derivatization reagent (2-bromo-p-acetophenone) and 20 μl of catalyst (triethylamine) were added. VPA derivatization was conducted at 60°C for 10 min in N₂, and the dried sample was dissolved in acetonitrile. The purified derivative was stored for further analysis. Quantification of VPA was performed with gradient elution at a flow rate of 1.0 ml/min, and 100 μl of the sample was injected. The column temperature was kept at 30°C, and the retention time was 12 min. Acetonitrile/water [75:25 (v/v)] was used as the mobile phase. The VPA derivatives were detected using a fluorescence detector at an emission wavelength of 255 nm.

Animals

All animal experiments were conducted following protocols approved by the Animal Ethical Committee of City University of Hong Kong (CityU) and Department of Health of Government of HKSAR [(20-105) in DH/HT&A/8/2/5/Pt.2]. Male and female C57BL/6 mice weighted between 20 and 25 g were acquired from Laboratory Animal Research Unit of CityU and were maintained following the standard guidelines.

Surgical operations for LFP recordings and long-term implantation

The mice were anesthetized by intraperitoneal injection of pentobarbital (40 mg/kg, Ceva Santé Animale). Another intraperitoneal injection of atropine (3.25 mg/kg) was performed to suppress bronchial secretion that may cause suffocation. After the mouse was mounted on a stereotaxic apparatus (Narishige), an incision was made on the scalp to expose the skull. A square craniotomy window (3 mm by 3 mm) was created at cortical region [window center at −2.2 mm anterior/posterior (A/P) and 1.5 mm medial/lateral (M/L)] to allow the placement of a hydroElex device. A small access hole (0.5 mm in diameter, 0.8-mm A/P, and 0.2-mm M/L) was drilled on the skull to insert a glass micropipette for 4-ap injection. Two metal screws were secured on the skull near bregma and cerebellum region to serve as the ground electrodes for both electrophysiological recording and voltage-driven drug release. Throughout an experiment, a heating pad was placed beneath the animal to maintain a stable temperature at 37°C.

In the experiments to assess the long-term biocompatibility of a hydroElex device, the mice underwent the same surgical procedures. After the implantation, dental cement was carefully applied to seal the surgical region. After a recovery period of 2 or 4 weeks, the mice were euthanized for histology analysis.

Electrophysiology recording

To perform electrophysiology recording, the hydroElex device was connected to a flexible flat cables socket soldered on a customized circuit board for convenient connection to a preamplifier [Tucker-Davis Technologies (TDT)]. Specifically, the hydroElex implant was directly connected to a mini-DB26 headstage (pre-amplifier). The headstage was connected to the PZ5 neurodigitizer for signal amplification and digitization. The LFP signal was recorded using Synapse software (TDT). Data from all channels are transmitted to a RZ base station (TDT) for further processing via a single fiber optic connection. During recording, the sampling frequency was set to 1017 Hz, while a high-pass filter and a low-pass filter were set with a cutoff frequency of 1 and 200 Hz, respectively. A customized MATLAB program was developed to process the recorded LFP signal. Power spectrum density (PSD) was calculated by fast Fourier transformation with a multiband analysis method. In the power spectrum analysis using the built-in “spectrogram” function in MATLAB, the signal of frequency around 50 Hz (48 to 52 Hz) was omitted from the analysis. PSDs in six distinct frequency bands were then summed, i.e., 1 to 4 Hz (δ-band), 4 to 7 Hz (θ-band), 7 to 13 Hz (α-band), 13 to 30 Hz (β-band), and 30 to 70 Hz (γ-band).

In vivo drug intervention in rodent seizure animals

Seizures were characterized as high-frequency, high-voltage, and synchronized neural spiking. The early onset of an SLE was indicated by a 3.5% or greater power increase of the β oscillation of LFP (13 to 30 Hz), and an SLE was defined as a more than 6% power increase of β oscillation. For controlled drug release, an electrical stimulator (A.M.P.I) was used to provide constant voltage stimuli ranging from 50 to 400 mV at selected electrode to drive the release of biomolecules (GABA and VPA). In GABA-releasing procedures in vivo, the drugs are released from electrodes to intervene disorders when the power of β oscillation increased by more than 6% above control baseline (healthy state), defined as “clearance” treatment.

For VPA clinical administrations, in the “intraperitoneal prevention” group, VPA [25 mg/kg (500 μg per mouse)] was intraperitoneally injected 1 hour before seizure induction by 4-ap; in the “intraperitoneal clearance” group, VPA [125 mg/kg (2500 μg per mouse)] was intraperitoneally injected after seizure induction; in the “CNS injection” group, 100 μg of VPA (5 mg/kg) was microinjected to the 4-ap lesion area 1 hour before seizure induction. All these three kinds of VPA administration methods are adapted as control groups to evaluate the efficacy of hydroElex-assisted closed-loop treatment method.

Closed-loop feedback system settings and adaptive treatment

To visualize and analyse the electrophysiology signal, an interactive MATLAB program was developed using the software development kit provided by TDT. The recorded LFP signal was processed in real time with statistics of various parameters, including the moving average of LFP amplitude, mean power, frequency spectrum, etc. A baseline neural activity was first derived from 30-min LFP recording before the seizure induction. Right after 4-ap injection, the recording and real-time analysis were carried out simultaneously to monitor occasional outburst of seizure events. Upon the detection of a seizure onset, a therapeutic drug release was triggered with an adaptive dose, M, determined by Eq. 1

$$M=f_{\text{delivery}}(T,V)\times ∆{\text{Power}}_{\text{LFP}}\times \mathrm{\epsilon }$$ (1)

where f(T, V) is the calibrated releasing curve in response to a voltage stimulus, T is the time of delivery, V is the voltage to drive a molecular release, Power_LFP is moving average of signal power in β oscillation, and ε is a coefficient derived from post-delivery assessment.

In this part, when the real-time time power of β oscillation increased by 3.5% than control (baseline neural activity), the hydroElex closed-loop system would initiate the procedure of drug intervention. The subsequent drug release amounts are set upon various power increase, as shown in fig S17 and Fig. 6. Specifically, different dosages of VPA ranging from 8.6 to 15.9 ng were autonomously applied to control seizure outbreaks based on the increment of pathological β oscillation power. Starting from a minimal dose in response to a threshold of 3.5% increment above the baseline, the drug dosage increases by 3 ng with every 1% power increment, up to 6% power change, which determines the maximal drug dosage.

Histochemistry evaluation

For biocompatibility assay, brain tissue damage and immune responses at the implantation site were assessed in brain tissues isolated from animals with or without the implantation of a hydroElex device. All the animals were transcardially perfused with ice-cold PBS and subsequently 4% paraformaldehyde (PFA; in PBS). The brain with an implant was harvested and further fixed in 4% PFA overnight, followed by a dehydration in 30% sucrose (in PBS) until settlement. The brain samples were later sectioned to coronal slices (50 μm in thickness) using a cryosectioning instrument (CryoStar NX70, Thermo Fisher Scientific) and then stored at −80°C for later use.

For immunostaining, all procedures follow a standard protocol. The brain slices were blocked in tris-buffered saline (TBS) supplemented with 0.5% triton and 3% bovine serum albumin for 2 hours. The samples were then incubated with primary antibodies (1:250) diluted in block solution at 4°C overnight. The used primary antibodies include anti-Iba1 (ab48004, Abcam), anti-NeuN (ab104224, Abcam), anti–glial fibrillary acidic protein (ab4674, Abcam), and anti-CD68 (ab201340, Abcam). Afterward, the brain slices were gently rinsed with PBS and incubated with secondary antibodies (1:500) diluted in TBS at RT for 2 hours. The secondary antibodies used include anti-mouse immunoglobulin G (IgG) H&L (Alexa Fluor 647, Abcam), anti-goat IgG H&L (Alexa Fluor 555, Abcam), and anti-chicken IgY H&L (Alexa Fluor 488, Abcam). Followed by a rinse in PBS, the samples were stained with 4′,6-diamidino-2-phenylindole (1:500 dilution in TBS, 20 min at RT) to label cell nucleus. After further rinse in PBS, the stained brain slices were mounted for confocal microscopy imaging (Leica SP8). The fluorescence intensity was normalized to the mean value of the sham control.

Statistical analysis

All the experimental data in this study were subjected to statistical analysis and were expressed as means ± SD. Statistical significance (*P < 0.05) was determined using Student’s t test, unless otherwise specified in the text.

Supplementary Materials

The PDF file includes:

Supplementary Note

Figs. S1 to S27

Legends for movies S1 and S2

Table S1

Other Supplementary Material for this manuscript includes the following:

Movies S1 and S2