Device fabrication
The devices were fabricated on two substrates: 4″ silicon wafer (p-doped, <100>, single-side polished) and indium-tin-oxide (ITO) coated glass substrate. Silicon wafers were used as bought from University Wafers and a sacrificial layer of titanium and aluminum (100 nm and 200 nm) was deposited using electron-beam evaporation. Thereafter, the wafers were plasma treated (oxygen, 150 W, 60 s) to improve the wettability for spin-coating process. The ITO-coated glass substrates were bought from Sigma, and cleaned by sonicating in deionized (DI) water, acetone and iso-propyl alcohol, respectively, for 10 min each and plasma treated (oxygen, 150 W, 60 s) before spin-coating. PEDOT:PSS (Ossila) was used as bought. 5% dimethyl sulfoxide (DMSO, Sigma, to increase the conductivity), 0.2% (3-glycidyloxypropyl)trimethoxysilane (Sigma, for higher water stability) and 0.2% Dynol 604 (to improve the wettability of the solution while spin-coating) were added to the PEDOT:PSS solution and stirred for 30 min before use. After mixing, the solution was filtered through a 0.2-µm polyethersulfone syringe filter. Thereafter, this solution was spin-coated (1,000 rpm, 60 s) on the substrates (silicon or ITO-coated glass substrate) and annealed at 120 °C for 10 min. The preparation of the organic semiconducting blends P3HT:PCBM or PCPDTBT:PCBM was performed in a glovebox under nitrogen atmosphere. The binary blends were prepared at least 6 h before spin-coating them on the Si wafers or ITO-coated glass substrate. P3HT:PCBM was prepared in a ratio of 1:0.83 with the overall concentration of 22 mg ml−1. Similarly, PCPDTBT:PCBM was prepared in a ratio of 1:2 with the overall concentration of 22 mg ml−1. While P3HT:PCBM was spin-coated at 800 rpm (100 ± 12 nm), PCPDTBT:PCBM was spin-coated at 1,200 rpm (80 ± 15 nm) for 60 s (only 1 polymer blend at a time) on top of PEDOT:PSS and annealed at 130 °C for 30 min. Once the spin-coating was done, the samples were transferred to the electron-beam evaporation machine for deposition of the top metal electrode (titanium (Ti): 50 nm, 1 Å s−1). For samples requiring titanium nitride (TiN) (for improved electrode–electrolyte interface), it was deposited after Ti deposition using reactive sputtering (30 nm, 0.27 Å s−1, argon:nitrogen at a 3:1 ratio). Once Ti/TiN was deposited, wafers were kept under vacuum until the lithography process. For photolithography, the samples were dehydrated at 100 °C for 30 min to increase the adhesion of the photoresist (AZ3312). Photoresist was spin-coated (3,000 rpm, 60 s) on the substrates, prebaked at 100 °C for 60 s and exposed using MLA150 (Heidelberg, dose 130 mJ cm−2 at 375 nm). The samples were postbaked at 100 °C for 60 s and then developed using AZ300-MIF developer for 60 s. After developing the photoresist, samples were hard baked on a hotplate for 30 min at 100 °C. For patterning the devices, the substrates were then transferred for dry etching of Ti/TiN and the organic polymers in an Oxford-100 reactive ion etcher. Ti/TiN was etched using SF6 plasma (200 W, 3.5 min). After the Ti/TiN etch, the samples were sonicated in acetone for 5 min to remove the photoresist and its residuals. After this, oxygen plasma was used to etch away the organic polymers as well as the left-over photoresist (Ti/TiN served as a hard mask for this step). The wafer was cleaved into small chips (20 mm × 20 mm) and stored under a nitrogen environment for further use.
Device characterization setup
A custom-built probe station was used for characterizing the devices (200 µm to 5 µm in size). External laser sources (50 mW, 520 nm, Thorlabs; 100 mW, 785 nm, Coherent) were coupled into the probe station for providing a bottom illumination onto the devices from the PEDOT:PSS side. For the ex vivo measurements (792 nm laser, HJ Optronics was used) the light passed through the brain before being incident on the devices. The micromanipulators were connected to a potentiostat (CompactStat, Ivium Technologies) for measuring the current–voltage curves. An upright optical microscope was used to locate the devices on the substrates and aligned with the bottom light source each time before recording the measurements.
Device releasing and collection
The devices fabricated on the silicon wafers had to be released to create free-floating devices for Circulatronics technology. For this, the wafer was cleaved into smaller chips and put inside a 5-ml glass vial. 1.5 ml of diluted TMAH solution (2.7% v/v) was used for a 20 mm × 20 mm silicon chip. The glass vial was then put in a sonication bath for 10 min to etch away the sacrificial aluminum layer and collect the devices. Thereafter, the devices were rinsed multiple times (using a custom vacuum-based filtration setup) in deionized water to get rid of trace amounts of TMAH. The devices were collected and stored for further experiments (Extended Data Fig. 3).
SPICE simulations
SPICE simulations were performed to investigate the SWED’s operation in free-floating form in the extracellular environment and to understand the effect of seal resistance (RSEAL) on the induced transmembrane potentials. The circuit model used for the simulations included several key components. Experimentally measured current–voltage characteristics of a SWED were fitted using third-order polynomials, a linear resistance and a shunt resistance. The electrode–electrolyte interface impedances for the two device terminals were extracted using Electrochemical Impedance Spectroscopy measurements performed with a potentiostat (IviumStat) and incorporated into the circuit model to accurately represent the electrode–electrolyte interface: RTiN/CTiN − 2.14 GΩ/214.3 pF, RPEDOT:PSS/CPEDOT:PSS − 1.6 GΩ/583 pF. The cell membrane was modeled as a parallel RC circuit, consisting of the membrane’s resistance and the capacitance. The resistance and capacitance values (RCELL, CCELL) for the cell were taken from the literature23 to ensure a realistic representation of the neural membrane. RSEAL represents the leakage resistive path between the SWED and the cell, which can vary depending on the interface conditions. In the simulations, RSEAL was varied from 1 kΩ to 100 MΩ to simulate different interface conditions and to study its effect on the induced transmembrane potentials. The simulations were carried out using the LTspice XVII software (Linear Technology). The operating point of the SWEDs and the induced transmembrane potentials were determined for each RSEAL value by analyzing the simulation results. From the simulations, we estimated that the SWEDs’ operation point stabilizes at approximately 400 pA and 147 mV (Supplementary Fig. 2b) in the extracellular environment. Moreover, the simulations shed light on the relationship between RSEAL and the induced transmembrane potentials, indicating that high RSEAL values—such as 100 MΩ (typical of neural interfaces with multi-electrode arrays)—enable SWEDs to induce transmembrane potentials exceeding the threshold voltage (13.5 ± 2.8 mV, determined experimentally using whole-cell patch clamp, Supplementary Fig. 10 and Methods section ‘Threshold voltage estimation for neurons’) required for neuronal activation (Supplementary Fig. 2c).
Threshold voltage estimation for neurons
To determine the threshold voltage necessary for action potential initiation in primary embryonic rat hippocampal neurons, we used whole-cell patch-clamp recordings. Neurons cultured in vitro were subjected to constant current injections, and the transmembrane potential (VM) was measured in current-clamp mode. The threshold for action potential generation was identified by analyzing the dVM/dt for a nonlinear increase indicative of ion channel activation. Specifically, during the initial period of charge injection, we calculated the mean and standard deviation of dVM/dt when it was expected to remain constant. The onset of nonlinearity, marking the neural threshold, was determined when dVM/dt exceeded the mean by 2.5 times the standard deviation, and continuing to rise thereafter. This method yielded an estimated action potential threshold of 13.5 mV ± 2.8 mV across the sampled neurons (n = 9), providing a quantitative basis for understanding neuronal excitability under controlled stimulation conditions.
NIR light transmittance measurements
A continuous-wave diode laser, operating at 792 nm, was used for transmittance evaluations. Brain slices of varied thicknesses were obtained from a perfused mice, while a tissue phantom was synthesized following the protocol used in ref. 40. For measurements, the sample (brain slice or brain phantom) was placed between the laser source and a photodetector. Transmittance was quantified by comparing light intensities with and without the sample.
Creation of cell–electronics hybrids
SWEDs functionalization
SWEDs (5–10 million at a concentration of ~10 million per ml) were immersed in a 10% v/v (3-aminopropyl)triethoxysilane (APTES) solution (Sigma) in 100% ethanol overnight at room temperature. Then these SWEDs were annealed for 2 h at 75 °C to promote the cross-linking of the APTES molecules on PEDOT:PSS. Then, they were rinsed multiple times in deionized water to remove loosely physiosorbed APTES molecules. After this, the APTES-functionalized SWEDs were incubated with equimolar solution of N-succinimidyl-4-((5-aza-3,4:7,8-dibenzocyclooct-1-yne)-5-yl)-4-oxobutyrate (Broadpharma) (1 mM solution in PBS-1×) and NHS-Cy3 (1 mM, lumiprobe) for 2 h at room temperature. The SWEDs were constantly stirred during the incubation periods. After the reaction was completed, the SWEDs were rinsed multiple times in PBS-1× and stored at 4 °C until further use.
Cell functionalization
Wehi-265.1 cells (American Type Culture Collection) (10–20 million at a concentration of ~2 million per ml) were incubated with succinimidyl 2-azidoacetate (Thermo Scientific) (100 µM in complete cell medium: Dulbecco’s Modified Eagle Medium (Sigma) + 10% fetal bovine serum (American Type Culture Collection) + 1% penicillin streptomycin (PS, Sigma)) solution for 2 h at 37 °C. Thereafter, the cells were rinsed and plated in complete cell medium.
Cell–electronics hybrid creation
After the cell functionalization process, the functionalized SWEDs (5–10 million) were incubated with the cells (10–20 million) in complete cell medium (5 ml) to allow the attachment of the SWEDs to the cell membrane for 2 h at 37 °C. Periodic agitation was provided to ensure that the SWEDs did not settle down in the petri dish and to increase the SWED attachment efficiency. For experiments requiring the cells to be fluorescent, NHS-Cy5 (100 µM, Lumiprobe) and Qtracker 705 (QT705, Invitrogen) were added after incubating SWEDs with the cells for 2 h and were allowed to stain the cells according to the manufacturer’s protocol.
Cell–electronics hybrid sorting
Cell capture and isolation were performed using FACS. After incubation, a single-cell suspension was created by filtering the solution from a 35-µm nylon mesh and stored on ice. The single cells were loaded onto the sorter (Sony MA900-1) and passed as a stream in droplets, in front of a laser. Cells with SWEDs attached to them showed higher scattering and fluorescence signal (SWEDs were functionalized with NHS-Cy3), and a double gating strategy was used to isolate the cell–electronics hybrids with a purity of 92.4% ± 5.2% (Fig. 3c and Extended Data Fig. 4). When fixing and embedding for imaging, the cells attached to the SWEDs were fixed and embedded on a 35-mm glass-bottom petri dish (no. 1.5). Note that the cells were fixed for imaging purposes only. Live cells were used for i.v. injection. For fixing the cells, 4% paraformaldehyde (PFA, EMS) was added to cells suspended in PBS-1× (cell concentration 2 million per milliliter) in equal volume (to make the final concentration of 1 million per ml) and incubated for 10 min at room temperature. Thereafter, the fixed suspended cells were rinsed with PBS-1× (at least 3 times) to remove any fixing reagent from the solution. The rinsed cells were plated on a poly-l-lysine (Sigma) coated 35 mm glass-bottom petri dish to make them adherent to the glass surface for at least 24 h. Once the cells had adhered to the surface, the PBS-1× solution was aspirated, and the cells were embedded in agarose (5% w/v in deionized water) solution before imaging.
Imaging
The cell–electronics hybrids were imaged using a Zeiss Crossbeam 540 SEM and a focused ion beam with a serial sectioning done at 20-nm resolution (Fig. 3a). A FV1200 Olympus confocal microscope was used for fluorescent imaging. Cells were stained with FAM-DBCO dye and SWEDs showed auto-fluorescence owing to P3HT molecules. Z-stack images of the hybrids (cells in yellow, SWEDs in green) were taken simultaneously to confirm the SWED attachment to the cell (Fig. 3b).
Transmigration assay
To determine the stability of cell–electronics hybrids during transmigration across the vascular endothelium, we conducted transmigration assays using a modified Boyden chamber kit (ECM558, Millipore Sigma). The Boyden chamber system is a two-chamber setup with a porous membrane providing an interface between the upper and lower chambers. The upper side of the cell culture insert was coated with fibronectin to support optimal attachment and growth of murine endothelial cells (BALB-5023, Cell Biologics), which were cultured on top of the porous membrane to simulate the endothelial barrier. The integrity of the endothelial monolayer was confirmed by measuring the trans-endothelial electrical resistance using EVOM2 equipment (WPI). The experiment began only when the measured impedance was found to be above 120 Ω cm2, ensuring a confluent monolayer of endothelial cells had formed.
To activate the endothelial cells and mimic an inflammatory environment, the cells were treated with TNF at a concentration of 50 ng ml−1 overnight. Following the activation of endothelial cells with TNF, each of the suspensions—SWEDs or monocytes or cell–electronics hybrids—were added separately to the upper chamber and were allowed to transmigrate toward a chemoattractant (monocyte chemoattractant protein 1 (MCP-1), 200 ng ml−1, Invitrogen) in the lower chamber through the endothelial cell-coated porous membrane. After an incubation period of 48 h, the cells in the lower chamber were quantified using a hemocytometer and confocal microscopy to determine the population and the transmigration rates of monocytes, SWEDs and hybrids and evaluate the stability of the hybrids.
Subjects used for animal experiments
Male and female Balb/C mice (Taconic) aged 7–12 weeks were maintained with a 12-h light/dark cycle and provided food and water ad libitum, also during the duration of the experiment. Animal husbandry and all experimental procedures were approved by the Massachusetts Institute of Technology Committee on Animal Care.
Stereotactic injection for the inflammation model
Note that here the stereotactic injection is done only to create the inflammation model. In actual applications of this technology, inflammation will already be present in the target diseased brain regions and our technology does not require any surgery. All mice surgeries were performed under aseptic conditions on a stereotaxic frame. Mice were anesthetized using isoflurane (1–4%). Analgesics were provided after anesthesia: subcutaneous injection of buprenorphine sustained release (1.0 mg kg−1) before surgery and lidocaine (0.5%, 2–4 mg kg−1) was injected subcutaneously at the incision site before making the incision. Coordinates used for the intracranial injection into the ventrolateral thalamic nucleus region relative to bregma were established according to the Allen Brain Atlas41 as follows: anterior–posterior −1.6 mm, medial–lateral +1.0 mm and dorsal–ventral −3.5 mm. A dental drill was used to create an opening in the skull and 1.0 μl of fluorescein isothiocyanate (FITC)-conjugated LPS (5 µg µl−1) was injected into the target region. A 10-μl nanofil syringe with a 33-gauge beveled needle was used for the injection. Injection speed of 100 nl min−1 was maintained using a micro-syringe pump and its controller. The syringe was left positioned for 10 min inside the brain before injecting LPS as well as before withdrawal from the brain. Skin tissue was closed with an adhesive and sutures, and the mouse was allowed to recover on a heat pad.
i.v. injection
After the LPS injection, the mice were allowed to recover on a heat pad and transferred to a cage before the next set of experiments. The cell–electronics hybrids solution was prepared using FACS in the meantime for delivering them via i.v. (retro-orbital or tail-vein) injection. Cells were thereafter stained with NHS-Cy5/QT705 (Invitrogen, Thermo Fisher) according to the manufacturer’s protocol. The sorted cell–electronics hybrids were then rinsed and resuspended in PBS at a concentration of 20 million per ml. The final volume of the injection was kept at 100 µl. The entire process was done under sterile conditions. The prepared solution was i.v. injected 6 h after the LPS injection. The mice were allowed to recover after that and kept under observation until they were euthanized.
Perfusion and imaging
Anesthetized mice were perfused transcardially with 4% PFA in PBS-1× 72 h after the i.v. injection. Brains were harvested and stored in 4% PFA overnight. Then 50-μm thick sections of the brain were sliced coronally using a vibratome (Leica VT1000 S) and imaged using a confocal microscope (Nikon 1AR Ultra-fast confocal microscope) for locating the LPS injection site as well as the distribution of the cell–electronics hybrids. LPS was conjugated with FITC-dye and the cells were stained with NHS-Cy5/QT705 to allow for their simultaneous imaging in separate fluorescent channels.
Quantifying cell–electronics hybrids in tissue
The quantification of cell–electronics hybrids was carried out by quantifying the content of Ti (as the SWEDs contain Ti layers) using inductively coupled plasma-mass spectrometry (Agilent 7900). Specifically, the brain was digested in nitric acid (70%, trace-metal grade) and hydrogen peroxide (30%, trace-metal grade) in a ratio of 5:1 using UltraWave microwave digestor (Milestone) (Fig. 3g and Supplementary Fig. 11). The entire solution was then diluted with milli-Q water to a final concentration of 3% v/v of nitric acid. One blank and six standards (0, 1 ppb to 100 ppm in steps of 10×) were used for calibration of Ti concentration and 103Rh was used as the internal standard during these measurements.
To quantify the number of implanted SWEDs (PEDOT:PSS|PCPDTBT:PCBM|Ti), the Ti content corresponding to a single SWED was calculated. This was based on the amount of Ti present in the SWED (50 nm Ti deposited on a 10-µm diameter device using an electron-beam deposition tool). The intrinsic baseline Ti content in the brain for control animals (without undergoing any administration of SWEDs) was subtracted from the Ti content in the brain for the experimental animals, and the result was divided by the Ti content of a single SWED to estimate the number of implanted SWEDs.
Assessment of cell–electronics hybrid localization
We used a two-channel imaging system to investigate the localization of self-implanted cell–electronics hybrids in relation to the target region, indicated by LPS image intensity using a logistic regression model (P(1|X) = 1/exp(−β1X + β0)). Regions of interest (ROIs) were identified where the cell–electronics hybrids were located. For the actual data from our experiment, 50% of pixels from these ROIs were randomly selected and mapped to corresponding LPS intensities. An equivalent number of pixels from the remaining region, without identified cell electronics, were randomly selected to collectively create the training dataset.
Similarly, the rest of the untrained dataset (50% of pixels from the identified ROIs and an equivalent number of pixels from the remaining region, without identified cell–electronics and not selected in training dataset) were used to create the prediction dataset. For the prediction, a threshold of 0.5 was used for the classification. The final values of the logistic regression analysis (β1, β0 and accuracy) were obtained by averaging the results from 1,000 repetitions of the fitting for each image.
In the control, the mapping between the LPS intensity values and the regions with or without identified cell–electronics hybrids was shuffled. A logistic regression model was then used to evaluate the influence of target region on the localization of cell–electronics hybrids, with the beta coefficient (β1) and accuracy values to assess the strength of this relationship and the fit of the model, respectively.
To visually compare the experimental and predicted localizations of cell–electronics hybrids, predicted images were generated (Extended Data Fig. 6). The experimental data included an image of FITC-conjugated LPS and an image showing the distribution of self-implanted cell–electronics hybrids. The predicted image was generated using the trained logistic regression model, showing the probability of cell–electronics hybrids localization based on the LPS intensity.
In vitro patch-clamp experiments
For our in vitro electrophysiology experiments, we used a carefully designed protocol to investigate the interaction between SWEDs and neurons. Hippocampal neuron cultures (grown on glass coverslips) were transferred to glass-bottomed petri dishes containing fresh media and placed on an inverted microscope (Nikon Ti2-E) equipped for patch-clamp recordings and imaging after 14 days in vitro. Released SWEDs were drop-casted in the culture and neurons with SWED on top of them were identified for patching using an inverted microscope.
Patch pipettes (4–5 MΩ resistance) were prepared using a micropipette puller (P-1000, Sutter). The intracellular solution for whole-cell recordings consisted of K-gluconate, NaCl, CaCl2, MgCl2, EGTA, HEPES, Mg-ATP and Na-ATP, with pH and osmolarity adjusted to physiological levels. Voltage-clamp and current-clamp recordings were conducted using a Multiclamp amplifier and digitized for analysis (Molecular Devices). Optical illumination was achieved using a diode laser (Doric) positioned on top of the targeted neuron interfacing with the SWED. Light control and synchronization with electrophysiology recordings were managed through custom software and hardware triggers. The generated action potentials show precise temporal correlation to optical pulse offset with a consistent latency of 174 ms ± 52 ms (Extended Data Fig. 7b,c). In addition, the observed neural response pattern (as shown in Extended Data Fig. 7) of initial hyperpolarization followed by depolarization leading to action potentials aligns with established literature on capacitive neuron-device coupling42,43 and postinhibitory rebound mechanism44, while the TiN interface properties likely contribute to the observed stimulation efficacy45.
In vivo c-Fos modulation and analysis
In this study, 18 mice were allocated into two distinct groups to investigate the effects of self-implanted hybrids, monocytes and NIR light on neural activation. The first group, comprising nine mice, received i.v. injections of cell–electronics hybrids while the second group (nine mice) received i.v. injections of cells (without the SWEDs). After a recovery period of 72 h, 4 mice from the first group and 5 mice from the second group were randomly selected for optical stimulation. The selected mice were anesthetized following previously established protocols, with their heads securely fixed. Optical stimulation was administered using a 792 nm wavelength laser (HJ Optronics), delivering a pulse sequence of 15 mW mm−2 intensity, 10-ms pulse width, at a frequency of 20 Hz for a duration of 20 min. To allow for adequate c-Fos protein induction, these mice remained anesthetized for an additional 90 min poststimulation before being euthanized. The remaining mice from the two groups did not receive optical stimulation but were subjected to the same conditions of housing, habituation and a 90-min anesthetization period before euthanasia.
After perfusion, brain was extracted and postfixed in 4% PFA overnight for all the animals. Coronal brain sections (50 µm) were prepared using a vibratome (Leica VT1000 S) and stored in PBS-1×. For IHC, tissues were incubated overnight with polyclonal rabbit antibodies raised against c-Fos protein (1:500, ABE457, sigma). c-Fos-positive cells were visualized using immunofluorescence with donkey anti-rabbit Alexa Fluor 647 (1:500, Thermo Fisher; 2 h of incubation at room temperature). All the cell nuclei were stained with Hoechst (Invitrogen). The stained slices were imaged using a slide scanner (TissueFAXS SL) and a confocal microscope (Nikon Ti, CSU-X1 confocal module) with a ×40 objective. For c-Fos-positive cell counting and quantification, an initial mask was drawn by a researcher who was blinded to the experiments, to delineate the target region (characterized by nuclei pattern). The mask was then extended by 50 µm in all directions to account for activation of neighboring neurons. Cell Profiler software was used to identify all c-Fos positive cells within the masked region, and results were expressed as cells per square millimeter (Fig. 4f). An adaptive thresholding approach based on the Otsu algorithm was implemented to segment the cells, with a c-Fos-positive cell being defined as having a mean intensity of the ROI 2.25 times greater than the average background of the entire image and confirmed by colocalization with the cell nucleus.
For the radial distribution of c-Fos positive cells, an initial mask was used to delineate the target region boundary. Then, polygonal ring masks (based on this boundary shape) with 20-µm radial width and varying radial distances were generated using MATLAB. Cell Profiler software with an adaptive Otsu algorithm, was used for cell identification as described above and any ROIs from c-Fos positive cells within these rings were counted and reported per unit area. The distribution of these cells was plotted as a function of radial distance with zero being at the boundary of the target region and positive and negative values corresponding to regions outside and inside the target, respectively (Fig. 4g).
Single-unit recording
Mice with or without self-implanted cell-electronic hybrids (or monocytes) in the ventrolateral thalamic nucleus region were initially sedated with 3% isoflurane in a mixture of air, and subsequently maintained anesthetized by 1–2% isoflurane after being immobilized in a stereotactic frame. Body temperature was maintained at 37 °C with a heat pad. The epicranium was incised to expose the skull. A craniotomy, 1 mm in size was performed at 1.60 mm posterior to the bregma and 1.00 mm right of the midline (at the LPS injection site). Commercially available 4 × 4 matrix electrodes (50 µm in diameter, impedance ranging between 0.5 MΩ and 2 MΩ) with 200-µm spacing were used to confine the recording site to 800 µm × 800 µm. The electrode array was stereotactically implanted vertically at the inflamed region with the tips of the recording electrodes extending 3.3 mm to 3.7 mm into the brain tissue from the surface. Recorded signals were monitored with Doric Neuroscience Studio V6. Action potentials from electrophysiological signals were amplified 2,000 times with high-pass filtering at 0.25 kHz. The resultant waveform was sampled at 30 kHz and low-amplitude optical artifacts coinciding with pulse onset and offset were systematically removed using automated detection and filtering protocols (Supplementary Fig. 12). Thereafter, single units (signal-to-noise ratio ≥4) were identified by principal component-based spike sorting, where a transient excitatory response was counted if the unit activity exceeded the 99% confidence interval (Z value > 2.33) in 2 consecutive bins (bin size, 100 ms) in the experimental condition (0–2 s after stimulus onset) in the normalized peri-event histogram and, when compared with baseline activity, the maximum firing rate increased in more than 50% of the trials. Optical stimulation was delivered using a NIR laser pulse (792 nm wavelength, 100 ms duration, 15 mW mm−2) every 10 s during recordings. These optical parameters induced minimal tissue temperature changes, as demonstrated by heat diffusion simulation studies (Supplementary Fig. 13).
For our methodology, we found that 14 out of 64 units showed temporally consistent activation patterns following optical stimulation. These active units exhibited temporally consistent responses correlated with optical pulse timing (within hundreds of milliseconds after optical pulse offset (Fig. 4h and Supplementary Fig. 3)) as seen from the pooled z-score plot (Fig. 4i), while nonactive units showed no temporal relationship with stimulation parameters (Supplementary Fig. 14). This 22% sampling efficiency compares favorably with FDA-approved neuromodulation technologies such as repetitive transcranial magnetic stimulation46, which demonstrates 28% efficiency in single-unit activation studies.
Statistical analysis of spike timing and time locking
To evaluate the likelihood that the short first spike timing from the optical pulse offset observed in our data occurred by chance, we generated 10,000 control datasets, in which the t = 0 event of each trial was randomly reassigned to a time point within the baseline period, that is, t = −3 s to t = 0 s of that trial. Note that t = 0 in the original trial marks the onset of the laser pulse. This randomization approach creates a null distribution that helps determine whether the observed neuronal responses are truly linked to the optical stimulation rather than occurring by random chance.
We determined the percentile rank of the observed median timing of the first spike timing from the optical pulse offset within the distribution of median timings from the control datasets. This percentile rank provided a statistical measure of how likely the observed shorter spike timing could have occurred by chance. Similarly, to evaluate the temporal consistency, we determined the percentile rank of the observed MAD of the first spike timing from the optical pulse offset within the distribution of MADs of the timings from the control datasets. The use of MAD provided a robust measure of variability that is less sensitive to outliers than the standard deviation, allowing us to quantify how consistently neurons respond to the optical stimulation.
Imaging for electrode mapping with hybrid localization
To assess the spatial and functional relationship between electrodes, recorded single units and self-implanted cell–electronics hybrids, we used a combination of electrode prestaining and postexperiment IHC. Electrodes were stained with (1,1′-dioctadecyl-3,3,3′,3′-tetramethylindocarbocyanine perchlorate) (DiI dye, 2% w/v in DMSO) before insertion. Following single-unit recording experiments, brains were extracted and postfixed in 4% PFA overnight. Axial brain sections (100 µm) were prepared using a vibratome (Leica VT1000 S) and stored in PBS-1×. The sliced brain tissues were then analyzed to correlate electrode proximity to cell–electronics hybrids with recorded neuronal activity. The histological analysis combined with electrophysiological recordings (Supplementary Fig. 15) demonstrated that only neurons adjacent to cell–electronics hybrids responded to optical stimulation, while neurons recorded from electrodes positioned farther away remained unresponsive, confirming spatial precision of neuromodulation.
Cytotoxicity assay for monocytes
Colorimetric MTT assays were performed in 96-well plates where SWEDs at various concentrations (10–10,000 SWEDs per µl) were added to each well plated with the cells (2–4 million per ml). After each time interval (Supplementary Fig. 5), 10 μl of MTT was added to the medium and incubated for 4 h. Then 200 μl of DMSO was added and mixed in each well. The absorbance signal was measured using a spectrophotometer (Spark, Tecan) at 570 nm and the background (measured at 630 nm) was subtracted from the signal to obtain the normalized values.
Cytotoxicity assay for cultured neurons
E18 Sprague Dawley rat dissociated hippocampal neurons were purchased from Brainbits. Neurons were cultured in the 96-well plates. The well plates were coated with 50 µl of poly-d-lysine (100 µg ml−1) to promote cell adhesion. Neurons were plated at a concentration of 104 per well and allowed to grow for 7–10 days. Then, 50 µl of SWEDs at various concentrations (105 to 107 ml−1) were incubated for the studied time intervals (Supplementary Fig. 6). Thereafter, 10 μl of MTT was added to the medium and incubated for 4 h. Then 200 μl of DMSO was added and mixed in each well. The absorbance signal was measured using a spectrophotometer (Spark, Tecan) at 570 nm and the background (measured at 630 nm) was subtracted from the signal to obtain the normalized values.
Blood count and blood serum chemistry analysis and histology
Blood samples were collected at two distinct time points (day 3 and day 12 after LPS injection). For the complete blood count analysis, 6 drops of blood from the facial vein pricked by a 21-G needle were collected in an EDTA-lined tube (20.1278.100, Sarstedt) and mixed by inverting back and forth on a rocker and submitted immediately for analysis. Before transcardial perfusion, 400 µl of blood was collected via cardiac puncture and stored in serum separator tubes (BDAM367985, VWR) for serum analysis. The blood was allowed to coagulate at room temperature for 15 min and spun down at 2,000 rpm for 10 min. The serum was collected and stored at −20 °C until analysis.
After transcardial perfusion, tissue samples of the major organs were collected for histological examination. The tissue samples were fixed, paraffinized, sectioned, H&E stained and scanned using a digital whole slide scanner (Aperio).
OFT
The OFT was conducted (day 2 and day 11 after LPS injection) to assess the locomotor activity of the mice. After habituation for 10 min individually in the home cage in the behavior testing room, each mouse was individually placed in the center of an OF arena of size 400 mm × 400 mm × 300 mm and allowed to explore freely for 10 min. A 10-min behavior recording was started right after the mouse was place in the OF arena. Locomotion was recorded two-dimensionally at 10 Hz from top-view with a CCD video camera installed above the center of the OF arena. Right after recording, the mouse was placed back in its home cage, and returned to the holding room. Video image data were processed using a custom script written in MATLAB R2023b (MathWorks).
NORT
The NORT was performed (day 3 and day 12 after LPS injection) to evaluate the recognition memory of the mice. The test consisted of a 10-min training phase, where the mice were exposed to 2 identical objects, and a 5-min testing phase, where 1 of the familiar objects was replaced with a novel object, with a 1-h interval between the 2 phases. Right after recording, the mouse was placed back in its home cage, and returned to the holding room. The time spent exploring the novel object versus the familiar object was analyzed using custom-written MATLAB scripts. The discrimination index was defined as tnew/(tnew + told), where tnew and told are the times spent exploring the new and old objects, respectively.
Testing immunoreaction to SWEDs
To examine potential immunoreaction to SWEDs, eliminating the immune reactions stemming from LPS (which is used to create the inflammation model), SWEDs were directly injected intracranially into the brain (instead of the self-implantation procedure). Intracranial injection of PBS under identical conditions was used as a control. SWEDs (at a concentration of 10 million per ml) or PBS (control) were unilaterally injected (2 μl) using a glass pipette into the mouse brain at the following coordinates: anterior–posterior −1.6 mm, medial–lateral 1 mm and dorsal–ventral −2.5 mm. Mice were euthanized at 1 day, 3 days and 7 days postinjection (n = 3 mice for each time point). Brains were extracted, coronally sectioned (50 μm) and immunohistochemically (IHC) stained for astrocyte marker GFAP (1:500, Thermo Fisher) and microglia marker-ionized calcium-binding adapter molecule 1 (Iba-1) (1:500, SYSY) following the manufacturer’s protocol. GFAP and Iba-1 IHC used goat anti-rat Alexa Fluor 488 (1:500, Biotium) and donkey anti-guinea pig Alexa Fluor 633 (1:500, Biotium), respectively, for secondary antibody staining. Hoechst (1:10,000, Thermo Fisher) was used to counterstain labeled cell nuclei.
Confocal imaging of stained sections was done using a Nikon Ti microscope (CSU-X1 confocal module). The SWEDs or PBS injection sites were delineated using a MATLAB script. Cell Profiler was used to quantify GFAP and Iba-1 expression based on fluorescence intensity within the defined injection region. Measurements were averaged across three sections per animal. SWEDs and PBS groups were statistically compared at each time point.
Clearance studies of i.v. injected cell–electronics hybrids
The clearance kinetics of the injected cell–electronics hybrids were investigated using the IVIS Spectrum imaging equipment (PerkinElmer). The mice with the LPS injection in the brain were i.v. injected with the fluorescently labeled cell–electronics hybrids. At various time points after injection, the mice were anesthetized with isoflurane (3% in air) and the fluorescence intensity in the animal body was examined under the IVIS system to track the clearance of the cell-electronic hybrids. Once clearance of the cell–electronics hybrids was confirmed (fluorescence intensity returning to baseline) under in vivo imaging, the mice were euthanized by transcardial perfusion and major organs (kidneys, spleen, heart, brain, liver and lungs) were harvested. These organs were subsequently imaged ex vivo under the IVIS system to provide additional confirmation regarding the clearance.
Continuous health monitoring
Throughout the study, the health of the animals was continuously monitored. This included regular checks on their body condition score, a widely accepted method to assess the overall health status of an animal. The body condition score provides a visual assessment of an animal’s muscle and fat, which can indicate whether the animal is underweight, overweight or at an ideal weight.
In addition to body condition scoring, the water intake and the body weight of the animals was also tracked. Changes in water consumption can be an early indicator of health issues. Similarly, changes in body weight can signal potential health problems.
H&E studies of brain tissues
To understand whether the SWEDs cause any adverse effect on brain tissue (eliminating any influence of LPS, which is used to create the inflammation model as well as to study chronic effects), SWEDs (10 µm diameter, 5 µl of solution at a concentration of 10 million SWEDs per ml) were directly injected stereotactically into the hippocampal region (anterior–posterior −2 mm, medial–lateral 1 mm, dorsal–ventral −2 mm). The animals were euthanized via transcardial perfusion in 4% v/v of PFA in PBS-1× at 1 day, 1 week, 1 month and 6 months postinjection. The brain was isolated and stored in 4% PFA (in PBS-1×) overnight before slicing it using a vibratome (Leica VT1000 S). Automated stainer (Tissue-Tek Prisma Plus) was used for staining the samples with H&E and slides were imaged using a digital whole slide scanner (Aperio) at ×40.
Biphasic pulse generator
Simulations were performed using Cadence Virtuoso and Spectre. Experimentally measured current–voltage characteristics of 10-μm diameter P3HT (D1) and PCPDTBT (D2) SWEDs when embedded in the brain tissue at a depth of 0.5 mm and wirelessly controlled with 520-nm and 785-nm illumination (Fig. 2h), were fitted using third-order polynomials. The polynomial fits were used to model the operation of D1 and D2 in Supplementary Fig. 16 using a Verilog-A script. Metal-oxide-semiconductor field effect transistors M1 and M2 were modeled using BSIMSOI v.4.7.1 with gate length L = 45 nm, gate width W = 450 nm, buried oxide thickness TBOX = 25 nm, top silicon thickness TSi = 10 nm and gate dielectric thickness Tox = 2 nm, as fully depleted silicon on insulator field effect transistors. The complementary metal-oxide-semiconductor footprint for the designed circuit was <1 μm2 (Supplementary Fig. 17). The load was modeled based on the electrode–electrolyte impedance of 2-μm diameter raised Pt-electrodes in PBS-1× (ref. 47), with series parasitic resistance Rp = 1.02 MΩ and pseudo-capacitance CP = 74.4 pF per electrode. Temperature was set at 37 °C to account for operation of the circuit in vivo.
The operating points of metal-oxide-semiconductor field effect transistors M1 and M2 were set self-consistently in the subthreshold regime based on the operating points of the driving SWEDs. When actuated by a 785-nm (520 nm) laser pulse, D2 (D1) developed a larger voltage than D1 (D2), which appeared as the gate-source voltage (VGS) for M2 (M1). For M1 and M2 operating in the subthreshold regime, a relatively small difference in VGS resulted in a large difference between their drive currents. Thus, M2 (M1) was selectively turned on and current flowed from D2 (D1) to the load and back through M2 (M1). We created a negative pulse of 65 μs (ref. 48), a gap (no illumination) of 50 μs followed by a charge balancing, 190-μs positive pulse by remotely controlling the illumination timings. The frequency of operation was chosen to be 200 Hz.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
