Main

Electronic devices implanted in the body provide powerful tools for diagnosis, therapeutics and research1,2,3,4. For example, bioelectronic implants for brain stimulation have provided biological insights and have proved effective for treating many brain diseases1. However, placing a medical implant inside the brain typically requires invasive intracranial surgery, with associated pain and tissue damage along with risks of infection, ischemia, psychological distress, morbidity and mortality5. Even endovascular electrodes6,7,8, although not requiring intracranial access, still need endovascular surgery with its associated risks and complications. Moreover, they cannot achieve submillimeter spatial targeting precision and are unable to access most brain regions6,7,8. While attempts have been made to explore intravenous (i.v.) injection routes9, these have led to nonspecific stimulation of large brain regions without focality. On the other hand, existing noninvasive brain stimulation technologies, such as transcranial magnetic stimulation and transcranial direct current stimulation, lack the necessary spatio-temporal resolution10.

Here we have developed bioelectronic devices that, after i.v. injection, are trafficked through the circulatory system and implant autonomously in brain regions of inflammation. We also demonstrate that they enable wirelessly controlled focal stimulation of deep brain regions such as ventrolateral thalamic nucleus in the rodent brain providing a nonsurgical brain implant for focal neuromodulation that takes advantage of immune cells’ natural trafficking to sites of inflammation. We name electronics that circulate through the vasculature ‘Circulatronics’ (Fig. 1). Realization of the Circulatronics brain stimulator requires overcoming several hurdles: (1) development of efficient wireless free-floating electronic devices that are miniaturized to fit inside the vasculature, (2) circulation of these devices without being eliminated from the bloodstream and (3) recognition of and self-implantation in desired brain regions. To overcome these challenges, we built wireless optical energy harvesting electronic devices that are subcellular sized and self-standing with high efficiency (to achieve point (1)) and created hybrids with living immune cells (to achieve points (2) and (3)). We demonstrate this technology for brain regions of inflammation, an important therapeutic target for many neurologic diseases, including Alzheimer’s disease11,12, multiple sclerosis13, ischemic stroke14, brain tumor15, neuropathic pain16,17,18, spinal cord19 and peripheral nerve injury20, whose treatment may benefit from electrical modulation targeted at the inflamed region11,12,13,14,15,16,17,18,19,20. We describe the electronic device design and fabrication, the creation of cell–electronics hybrids, the nonsurgical focal brain stimulation and biocompatibility studies.

Fig. 1: Circulatronics.

Schematic diagram illustrating the concept of Circulatronics. Credit: Pablo Penso; human anatomy image from Shutterstock.

Results

Subcellular-sized free-floating wireless electronic devices

To fit and freely move inside the vasculature without clogging, the size of Circulatronics devices must be similar to or smaller than that of the circulating cells (for reference, a circulating cell such as monocyte has a diameter of 12–18 µm). Hence, we set out to develop subcellular-sized wireless electronic devices (SWEDs) that are free-floating and that can convert extracorporeally applied fields to electrical energy to enable electrical neuromodulation. While there are different modalities for wireless energy harvesting (such as radio frequency, electromagnetic, optical or acoustic21) each with its unique characteristics, we decided to use the photovoltaic principle, which involves wireless powering via optical fields. This is because optical modalities provide high spatio-temporal resolution, penetration depth of several centimeters in the human head with intact skull and have already been used for clinical studies (Supplementary Note). Moreover, photovoltaic devices generate d.c. potentials22 eliminating the need for any rectifying circuits (saving on-chip area and avoiding circuit complexity). Other modalities can also be employed in future for Circulatronics technology, based on user-defined requirements. Although photovoltaics have been applied previously for neuromodulation23, this study investigates subcellular-sized free-floating photovoltaic devices compatible with circulation through bloodstream for in vivo electrical brain stimulation. Moreover, none of the previous photovoltaic devices or implants with other modalities (optical, electrical, radio frequency, magnetic or acoustic) have been demonstrated for brain stimulation with high spatial resolution without surgery. We used organic semiconductors24,25 to leverage the photovoltaic effect, as they have unique advantages such as narrow bandwidth for enabling multiplexing, high optical absorption coefficients, mechanical flexibility allowing good interface with soft biological systems and biocompatibility. They also provide ease of fabrication and compatibility with complementary metal-oxide-semiconductor back-end-of-line processing26,27 creating opportunities for integrating advanced functionalities in the future (Supplementary Note). The equivalent electrical circuit of these photovoltaic devices consists of a current source (representing optical intensity dependent polaron generation), three diodes and several resistors (Extended Data Fig. 1a). Our devices consist of a three-layer structure: anode, binary blend of semiconducting organic polymers (acceptor and donor material forming the active layer where the excitons are generated) and cathode (Fig. 2a). By customizing the organic polymeric materials, these devices can be tuned to different optical wavelengths that will allow their independent control, enabling multiplexing. To achieve this, we used two different donor materials, poly(3-hexylthiophene) (P3HT) and poly(2,6-(4,4-bis-(2-ethylhexyl)-4H-cyclopenta(2,1-b;3,4-b′)dithiophene)-alt-4,7(2,1,3-benzothiadiazole)) (PCPDTBT), as their absorption spectra are complimentary to each other (Extended Data Fig. 1b). (6,6)-phenyl-C61-butyric acid methyl ester (PCBM) was used as the acceptor polymer in both cases. Further, poly(3,4-ethylenedioxythiophene):polystyrene sulfonate (PEDOT:PSS) and titanium (Ti) are used as the anode and cathode, respectively, and are chosen based on their work function (Extended Data Fig. 1c) and biocompatibility. To investigate the scaling behavior and gauge the feasibility of subcellular-sized devices, we developed a fabrication process (Methods section ‘Device fabrication’) to create devices with nanoscale thickness (about 200 nm) and different lateral length scales, ranging from diameters of 200 µm (>10 times the diameter of a monocyte) to 5 µm (subcellular size). The nanoscale thickness is critical not only to increase the mechanical flexibility for improved biological interfacing, but also for achieving optimal device performance due to the competing effects of increase in exciton generation as well as recombination effects with the increase in device active layer thickness28. These devices can be mass produced at the wafer scale (Supplementary Fig. 1). Figure 2b shows the scanning electron microscopy (SEM) images of the fabricated devices on a 4-inch silicon wafer. The inset shows the zoomed-in image of a single device. The setup for characterizing our devices (Methods section ‘Device characterization setup’) is shown in Fig. 2c. The scaling performance of the devices is illustrated in Fig. 2d. It is seen that these devices, even when scaled down to diameters ≤10 µm achieving subcellular sizes (and less than 0.01× volume of a cell with 12 µm diameter) can generate nanowatts of power. Figure 2e shows the current–voltage characteristics of a SWED (10 µm in diameter) illustrating the increase in short-circuit current (ISC) generation with the incident optical intensity. There is no major change in the open-circuit voltage (VOC) even when operating the devices at low intensities, which corroborates the ability of photovoltaics to generate near-constant potential22 in the open-circuit condition. Figure 2d,e correspond to P3HT-based devices and the characteristics of the PCPDTBT-based devices are shown in Extended Data Fig. 2. Even at this ultra-small dimension (subcellular sizes), SWEDs generate open-circuit voltages, VOC = 0.2 ± 0.008 V (P3HT), VOC = 0.17 ± 0.01 V (PCPDTBT) and short-circuit currents, ISC = 12.8 ± 2.15 nA (P3HT), ISC = 18.2 ± 2.56 nA (PCPDTBT) at incident optical intensity of 10 mW mm−2. Next, to achieve free-floating SWEDs required for Circulatronics, we developed a process flow to release the structures from the fabrication substrate (silicon wafer) through tetramethylammonium hydroxide (TMAH)-based etching of sacrificial aluminum layer and retrieve and collect them (Methods section ‘Device releasing and collection’ and Extended Data Fig. 3a,b). Our characterizations confirmed that even after the substrate-release process, the devices retain good performance (Extended Data Fig. 3c).

Fig. 2: Characterization of subcellular-sized electronics with optical remote control.

a, Schematic energy diagram illustrating the working principle of the photovoltaic devices (three-layered device based on organic polymers) and the final device structure developed. OSC, organic semiconductor. b, SEM images of the SWEDs fabricated at wafer scale. SWED structure, PEDOT:PSS|P3HT:PCBM|Ti. Scale bar, 10 µm (inset, 1 µm), n = 5. c, Schematic diagram of the setup used for in-air and ex vivo measurements. The devices were illuminated from the bottom with a laser. For ex vivo measurements, brain tissue or whole brain with the skull was placed on a no. 1.5 cover slip (while keeping the tissue wet in PBS) and in close contact with the devices. The microscope was used to assist with alignment while probing the devices from the top using the three-stage micromanipulators. The probes were connected to a potentiostat and the laser was operated in continuous-wave mode to obtain the current–voltage (I–V) curves. In-air measurements were as follows: d, Representative plot showing the power generated by different-sized devices as a function of applied voltage. Device structure, PEDOT:PSS|P3HT:PCBM|Ti, Intensity 10 mW mm−2 incident on the devices, n > 5 devices. Legend, device diameter. e, Representative I–V characteristics of a SWED (10 µm in diameter) for varied light intensities incident on the SWED. SWED structure, PEDOT:PSS|P3HT:PCBM|Ti, n > 5 devices. A 520-nm wavelength laser was used as the light source for d and e. Ex vivo measurements: f, Representative I–V plot for the SWEDs (10 µm in diameter) with light passing through different thicknesses of brain slices or whole brain (6.1 ± 0.3 mm) or whole brain with skull (6.5 ± 0.3 mm). The transmittance of NIR penetration through the brain tissue was measured and is presented in Supplementary Fig. 9 (Methods section ‘NIR light transmittance measurements’). SWED structure, PEDOT:PSS|PCPDTBT:PCBM|Ti, n = 3 devices. Laser, 792 nm; intensity, 24.6 mW mm−2 incident on the bottom surface of the brain. g, Maximum power generated by the SWEDs (10 µm in diameter) with light passing through the whole brain without the skull (and the whole brain with an intact skull) at different light intensities (792 nm) incident on the bottom surface of the brain. SWED structure, PEDOT:PSS|PCPDTBT:PCBM|Ti, values represent median ± standard deviation (s.d.) (n = 5 devices). h, Box plot showing the multiplexing effect using the two SWED structures (10 µm in diameter) we have developed: PEDOT:PSS|PCPDTBT:PCBM|Ti and PEDOT:PSS|P3HT:PCBM|Ti. The plot shows the mean (circle within the box), median (horizontal line within the box), lower and upper quartile (delineated by the box), whiskers extending to the most extreme data points within 1.5× IQR from the quartiles, with minima and maxima beyond this range shown as individual outlier points (n = 3 devices per condition). The inset shows the absorption coefficients for the two SWED structures indicating their orthogonal absorption coefficients at the operating wavelengths. Labels, α (absorption coefficient, ×105 cm−1) and λ (wavelength, nm). The measurements were done with light passing through a 0.5-mm-thick brain tissue slice. Intensities used for this measurement were 10 mW mm−2 (520 nm) and 2 mW mm−2 (785 nm) incident on the bottom surface of the brain slice.

Source data

To gain insights on the SWEDs operation in their free-floating form in the extracellular environment, we carried out simulation program with integrated circuit emphasis (SPICE) simulations (Methods section ‘SPICE simulations’). From our SPICE simulations, we estimated that the SWEDs operating point stabilizes at approximately 400 pA and 147 mV (Supplementary Fig. 2). To further characterize the energy harvested by the SWEDs when implanted in the brain, we employed our ex vivo measurement setup allowing probing of the devices and measurement of the generated power, with light penetrating through different brain tissue thicknesses (Fig. 2c). The current–voltage characteristics are plotted in Fig. 2f for various tissue thicknesses at an incident intensity of 24.6 mW mm−2 of near-infrared (NIR) light, showing the successful remote operation of the SWEDs (10 µm in diameter) even for the entire mouse brain with the intact skull. The power generated by the SWEDs is plotted as a function of optical intensity for the whole brain without and with the skull in Fig. 2g and as expected the power increases with increase in the incident optical intensity. It is seen that these SWEDs can generate 0.545 ± 0.058 nW and 0.482 ± 0.019 nW of power (for the whole brain without and with the skull, respectively) at an incident optical intensity of 46.06 mW mm−2 (note that continuous laser illumination up to 100 mW mm−2 is demonstrated as safe in vivo, while higher intensities can be used when using pulsed illumination29). The ability to independently control the SWEDs using different optical wavelengths when operated through brain tissue is demonstrated in Fig. 2h.

Autonomous implantation in inflamed brain regions

We then investigated whether the SWEDs can be transported by cells through the circulatory system and autonomously implant in the brain. We attached SWEDs onto the surface of immune cells (Fig. 3a,b), specifically monocytes, as they target the region of inflammation30 (Methods section ‘Creation of cell–electronics hybrids’) and can cross the blood–brain barrier31. We functionalized the surface of monocytes with azide groups, leveraging the available amines on the cell membrane proteins. On the other hand, we functionalized surface of the PEDOT:PSS layer on our SWEDs with dibenzocyclooctyne (DBCO) groups, to attach them to the cells using Click chemistry32. Fluorescence-activated cell sorting (FACS) was performed to isolate cell–electronics hybrids with a purity of 92.4% ± 5.2% (Methods section ‘Creation of cell–electronics hybrids’, Fig. 3c and Extended Data Fig. 4). The three-dimensionally (3D) reconstructed focused ion beam-SEM image of the hybrid of a SWED with a monocyte is shown in Fig. 3a while Fig. 3b shows the confocal z-stack projection views of the hybrid.

Fig. 3: Autonomous implantation of wireless bioelectronics in the brain.

a,b, 3D reconstructed focused ion beam-SEM image (a) and confocal z-stack image (b) (with cross-sections from various viewing planes) showing a SWED (10 µm in diameter) attached to the immune cell (monocyte, wehi-265.1). Depth resolutions, 20 nm (a) and 500 nm (b). Scale bars, 5 µm. c, Histogram showing the population of cell–electronics hybrids and only cells in the suspension prepared for i.v. injections before and after FACS. A double gating strategy was used to isolate the cell–electronics hybrids from the suspension with a purity of 92.4% ± 5.2%. Values represent mean ± s.d. (n = 3 independent experiments). d, Representative confocal images of mouse brain slice for the experimental group (n > 5 mice) in which cell–electronics hybrids were i.v. introduced in the mice with the target (inflamed) brain region (ventrolateral thalamic nucleus (VL), induced by LPS). Image illustrating (i) Hoechst staining, (ii) target region, (iii) self-implanted cell–electronics hybrids and (iv) overlay image showing localized cell–electronics hybrids self-implanted at the target region. e, Representative confocal images of mouse brain slice for the control group (n > 5 mice) with same conditions as d except that cell–electronics hybrids were replaced by SWEDs alone (without the cells). Image illustrating (i) Hoechst staining, (ii) target region, (iii) SWED channel and (iv) overlay image showing no discernible self-implanted SWEDs. f, Representative confocal images of mouse brain slice for the control group (n > 5 mice) with same conditions as d except that LPS was replaced by PBS. Image showing (i) Hoechst staining, (ii) injected PBS, (iii) cell–electronics hybrids channel and (iv) overlay image showing no discernible self-implanted cell–electronics hybrids. Scale bar, 100 µm for (i)–(iv) in d–f. g, Titanium content in the brain for quantification of the self-implanted SWEDs that contain a titanium layer (after subtracting the intrinsic baseline titanium content in mouse brain) measured by inductively coupled plasma-mass spectrometry when 2 × 106 number of SWEDs only or cell–electronics hybrids were injected into the animal i.v. for the experimental and control groups described in d–f. A significant increase in titanium content compared to the baseline was observed only for the experimental group while for both the control groups, the changes compared to baseline were not significant, values represent mean ± s.d. (n = 3 mice), ***P < 0.001 (one-way analysis of variance (ANOVA) test). h, The coefficient (β1) from the logistic regression model (n = 12) suggesting an influence of target region on cell–electronics hybrids localization in the actual experimental data (β1 = 2.2 ± 0.94), compared to the shuffled control (β1 = 2.7 × 10−4 ± 2.9 × 10−4). Values represent mean ± s.d., **P < 0.01 (unpaired two-tailed t-test). i, The ACC values indicate a better model fit for the experiment (n = 12) (ACC = 0.82 ± 0.05) compared to the shuffled control (ACC = 0.50 ± 0.01), supporting the hypothesis of cell–electronics hybrids localization at the target region. Values represent mean ± s.d., ****P < 0.0001 (unpaired two-tailed t-test). D, dorsal; L, lateral; M, medial; V, ventral.

Source data

To assess the stability and the transmigration capabilities of the hybrids, we performed transmigration assays (Methods section ‘Transmigration assay’) to characterize the rate of transmigration of various populations (cells, SWEDs and hybrids) and stability of hybrids during the process. The population distribution of the sample for the transmigration can be written as:

$$\begin{array}{l}{{{N}}}_{{\rm{PASS}}}({{t}})=\left((1+{{D}})\times,{{{n}}}_{{\rm{CELL}}}({{{t}}}_{0})\times,{{{r}}}_{{\rm{CELL}}}+(1-{{D}})\right.\\qquad\quad\left.\times,{{{n}}}_{{\rm{HYBRID}}}({{{t}}}_{0})\times,{{{r}}}_{{\rm{HYBRID}}}+(1+{\rm{D}})\times,{{{n}}}_{{\rm{SWED}}}({{{t}}}_{0})\times,{{{r}}}_{{\rm{SWED}}}\right)\times,{{t}}\end{array}$$

where NPASS(t) is total transmigration population at a given time t, D is the fraction of the hybrids that dissociate during the transmigration process, rCELL is the rate of transmigration of monocytes (14.14 ± 6.4 mm−2 h−1, measured experimentally), rHYBRID is the rate of transmigration of hybrids, rSWED is the rate of transmigration of SWEDs (0.44 ± 0.27 mm−2 h−1, measured experimentally), nCELL(t0) is the number of monocytes, nHYBRID(t0) is the number of hybrids and nSWED(t0) is the number of SWEDs at t = 0 (before transmigration).

By counting the proportion of various populations before (at time 0) and posttransmigration (at time t), and equating the coefficients using the above equation, we found that a substantial population (86.9% ± 0.9%) of the hybrids remained stable and did not dissociate during the various stages of the transmigration process (Extended Data Fig. 5a–c) and transmigrated at an average rate of 5.24 ± 0.98 mm−2 h−1.

To investigate whether these hybrids can be self-implanted in the inflamed brain region, we used a classical inflammation model in a deep brain region (ventrolateral thalamic nucleus) of mouse (Balb/C, Methods section ‘Subjects used for animal experiments’) through stereotactic injection of lipo-polysaccharide (LPS) (Methods section ‘Stereotactic injection for the inflammation model’). The dye allowed us to precisely locate the region of LPS injection later during postmortem imaging. The stereotactic surgery is done to emulate the inflammation, and application of our technology does not require any surgery. Next we administered the high purity suspension of cell–electronics hybrids (Fig. 3c and Extended Data Fig. 4b) through i.v. injection (Methods section ‘i.v. injection’). Then 72 h after the i.v. injection, the mice were transcardially perfused, their brains were harvested, fixed and brain slices were prepared for imaging (Methods section ‘Perfusion and imaging’). Our data show that the hybrids self-implanted in the brain (Fig. 3d,g and Extended Data Fig. 5d). In one control group, where only SWEDs were i.v. injected without attachment to immune cells with all other conditions remaining the same as the experimental animals, SWEDs were not observed in the brain images (Fig. 3e), demonstrating the role of immune cells for effective self-implantation. Another control group was studied under the same conditions as the experimental animals except that LPS was replaced with phosphate-buffered saline (PBS), and no hybrids were detected in the brain images (Fig. 3f) corroborating that the cell–electronics hybrids selectively enter the inflamed brain region. Quantification of the number of SWEDs (10 µm) implanted in the brain (Fig. 3g and Methods section ‘Quantifying cell–electronics hybrids in tissue’) found ~14,029 ± 4,154 (n = 3 mice) SWEDs installed in the brain of experimental animals. In the two control groups, the number was negligible (Fig. 3g). Further, to evaluate the colocalization of self-implanted cell–electronics hybrids in the region, we used logistic regression analysis (Methods section ‘Assessment of cell–electronics hybrid localization’), which showed that the target region is a determining factor in predicting the self-implantation of cell–electronics hybrids (Extended Data Fig. 6). The high coefficient (β1, actual 2.2 ± 0.94; β1, shuffled 2.7 × 10−4 ± 2.9 × 10−4; P < 0.01) (Fig. 3h) and a high accuracy (ACC) value (ACCactual = 0.82 ± 0.05; ACCshuffled = 0.50 ± 0.01; P = 0.000009) (Fig. 3i) of the actual experimental data compared to the shuffled control further confirmed the prediction.

Brain region-specific stimulation

Next we explored the capabilities of SWEDs to stimulate specific regions of brain. First, to characterize the neuromodulation capabilities of SWEDs, we conducted in vitro patch-clamp experiments by drop-casting SWEDs in neuronal cultures and applied optical illumination (Extended Data Fig. 7a and Methods section ‘In vitro patch-clamp experiments’). These studies revealed reliable light-triggered neuromodulation capabilities of our SWEDs. Our patch-clamp recordings demonstrate that a ten-pulse optical train consistently induces robust neuronal firing patterns, with each stimulation pulse reliably triggering multiple action potentials. The generated action potentials show precise temporal correlation to optical pulse offset (Extended Data Fig. 7b,c). Our control experiments (Extended Data Fig. 7d) confirm that optical illumination alone (without SWEDs) showed minimal membrane perturbations and does not elicit neuronal responses, validating that the observed effects are specifically due to SWED activation rather than nonspecific heating due to light or other optical artifacts.

Building on these in vitro findings, we next characterized brain stimulation in vivo using c-Fos immunohistochemistry (IHC), which is widely used to delineate cell activity33,34 and can provide information on spatial distribution of the stimulated region. For this, 18 mice were divided into experimental and control groups to assess SWED-mediated neuromodulation specificity (Fig. 4a). Nine mice received i.v. injections of cell–electronics hybrids (SWED structure PEDOT:PSS|PCPDTBT:PCBM|Ti|TiN, 10-μm diameter) for brain self-implantation, while nine received monocytes alone. After 72 h, mice were organized into 4 groups: hybrids with wireless actuation (n = 4), hybrids without actuation (n = 5), monocytes with actuation (n = 5) and monocytes without actuation (n = 4). NIR light actuation (792 nm, 15 mW mm−2, 10 ms pulses, 20 Hz) was applied for 20 minutes. Figure 4b–e shows representative confocal images of c-Fos activation in the brain region of experimental and control mice. Quantification of the brain tissue, using a pipeline optimized for counting c-Fos positive cells in tissues (Methods section ‘In-vivo c-Fos modulation and analysis’), revealed a statistically significant number of c-Fos positive cells (317.8 ± 80.96 cells per mm2, n = 4 mice) in the brain region of the experimental mice (hybrids + wireless actuation) compared to the control groups: hybrids + no wireless actuation (107.9 ± 40.57 cells per mm2, n = 5 mice), cells (monocytes) + wireless actuation (76.2 ± 64.34 cells per mm2, n = 5 mice) and cells (monocytes) + no wireless actuation (73.38 ± 44.14 cells per mm2, n = 4 mice) (Fig. 4f). These results collectively demonstrate that the activation of self-implanted cell–electronics hybrids by NIR light specifically induces neural activity, as evidenced by increased c-Fos expression. On the other hand, none of the presence of the monocytes only, the hybrids without NIR light or the NIR light alone is sufficient to elicit these neural changes. In addition, the spatial distribution of c-Fos positive cells showed a distinct pattern of clustering around the target area in mice where hybrids were self-implanted and had received wireless stimulation. Radial distribution analysis revealed a decrease in the density of c-Fos positive cells as the distance from the target locus increased and the c-Fos positive cell density returned to the baseline within a few tens of micrometers outside the target region boundary. This was in contrast with the control animals (which did not receive wireless stimulation), where the activity was uniformly dispersed, as shown in Fig. 4g. Note that in our experiments, light is not required to be localized at the target and is illuminated over a wider region.

Fig. 4: Nonsurgical targeted focal brain stimulation.

a–e, Schematic diagram showing the timeline and the wireless stimulation scheme (a). Representative images showing the (i) cell nuclei, (ii) target region, (iii) c-Fos activity and (iv) overlay confocal images for mice with self-implanted cell–electronics hybrids for the experimental group where optical actuation was applied (n = 4 mice) (b); and control groups where self-implanted cell–electronics hybrids were present but no optical actuation was applied (n = 5 mice) (c), where only cells were self-implanted and optical actuation was applied (n = 5 mice) (d) and where only cells were self-implanted and no optical actuation was applied (n = 4 mice) (e). The optical pulse sequence applied in b and d: 792 nm, 15 mW mm−2, 10-ms pulse width, 20 Hz, 20-min duration. Scale bars, 25 µm for (i)–(iv) for b–e; SWED structure used for b and c PEDOT:PSS|PCPDTBT:PCBM|Ti|TiN. f, Scatter plot showing the number of c-Fos positive cells in the target region for the experimental (self-implanted hybrids with wireless optical actuation) and control groups (self-implanted hybrids without optical actuation, self-implanted cells with wireless optical actuation and self-implanted cells without wireless optical actuation). Longer (shorter) horizontal lines represent the median (standard deviation) of the data; n = 4 mice (hybrids + wireless actuation, and cells + no wireless actuation), n = 5 mice (hybrids + no wireless actuation, cells + wireless actuation), **P < 0.01, ***P < 0.001; one-way ANOVA test; NS, not significant. g, Bar plot showing the distribution of c-Fos positive cells as a function of radial distance from the target region boundary (represented by 0 on the x axis; positive and negative values on the x axis correspond to regions outside and inside the target, respectively) for the experimental (cell–electronics hybrids + wireless actuation) and control (cell–electronics hybrids + no wireless actuation) groups, n = 4 mice (experimental group), n = 5 mice (control group), values represent mean ± standard error of the mean (s.e.m.). h, (i) Principal component analysis (PCA) cluster, (ii) spike waveform, (iii) representative raw trace from a single trial, (iv) super-imposed raw-traces from multiple trials, (v) raster-plot displaying spike timing across multiple trials, (vi) peri-stimulus time histogram and (vii) z-score of the representative recorded unit for the experiment when hybrids were self-implanted and NIR light was applied. The vertical shaded line ((iii)–(vii)) shows time at 0–100 ms for the optical pulse. Horizontal dashed line (in z-score plot): z = 2.33; bin, 100 ms, Optical pulse: 100 ms, 15 mW mm−2, 792 nm. i, Normalized ensemble activities (z-scores) of active units in the experimental dataset (LPS + hybrids + NIR light) (n = 14, 5 mice). Data are presented as mean ± s.e.m., vertical shaded line shows time at 0–100 ms for the optical pulse. i.c., intracranial. Illustrations in a created using BioRender.com.

Source data

To further characterize the neuronal stimulation in vivo, we carried out single-unit recordings to electrically measure the modulated neural activity in the presence of cell–electronics hybrids during NIR light stimulation (Methods section ‘Single-unit recording’). The recorded units (n = 14 out of 64 units, 5 mice) exhibited a statistically significant increase in neural activity with optical pulse stimulation (Fig. 4h and Supplementary Fig. 3). Comprehensive control studies in the ventrolateral thalamic nucleus region with NIR light alone (n = 58 units, 5 mice) and in LPS-inflamed ventrolateral thalamic nucleus with self-implanted monocytes (without SWEDs) plus NIR light (n = 61 units, 5 mice) confirmed that neuronal activation was specifically attributable to SWED stimulation rather than confounding factors (Extended Data Fig. 8 and Supplementary Fig. 4). Further, statistical analysis of spike timing (Methods section ‘Statistical analysis of spike timing and time locking’ and Extended Data Fig. 9) revealed that experimental group responses exhibited both temporal alignment to optical pulses and consistency exceeding all control cohorts. Compared to NIR light only (45.58 ± 4.84 percentile rank) and cells only + NIR light (51.24 ± 6.15) controls, experimental first-spike latencies relative to pulse offset ranked at the 99.18 ± 0.43 percentile. Temporal consistency showed even greater distinction, with median absolute deviations (MAD) in experimental responses ranking at the 99.94 ± 0.04 percentile versus 59.02 ± 3.36 (NIR only) and 51.47 ± 5.85 (cells only) controls. These quantitative measures provide statistical evidence that the neuronal responses exhibited temporal consistency associated with the optical pulse offset, with both the speed and deviation of responses being notably different from the spontaneous neural activity. Lastly, the pooled z-score ensemble data from the recorded active units (Fig. 4i) further demonstrate rapid, transient excitatory neuronal responses triggered only by the activation of cell–electronics hybrids in the presence of NIR light, highlighting the capability of Circulatronics to wirelessly stimulate neurons.

Biocompatibility studies

Both in vitro and in vivo biocompatibility studies were performed. The biocompatibility of SWEDs was first assessed in vitro against monocytes and cultured primary neurons. The metabolic activity assays (3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide, MTT) for cytotoxicity in monocytes (Methods section ‘Cytotoxicity assay for monocytes’ and Supplementary Fig. 5) and neurons (Methods section ‘Cytotoxicity assay for cultured neurons’ and Supplementary Fig. 6) indicate that the SWEDs do not impair the health or viability of these cells. Following this, the in vivo biocompatibility of our technology was evaluated for both potential short-term and long-term effects (Methods). Animals with the LPS injection in the brain were randomly divided into two groups: the experimental group received an i.v. injection of cell–electronics hybrids, while the control group did not. We performed the complete blood count analysis (Supplementary Table 1) and comprehensive blood serum chemistry analysis (Supplementary Table 2) at 3 days post LPS injection, which revealed no adverse effects from cell–electronics hybrids in the experimental group compared to the controls (Methods section ‘Blood count and blood serum chemistry analysis and histology’). Behavioral tests were conducted to evaluate the impact of cell–electronics hybrids on the animals’ locomotor and cognitive functions by open fiel