Micelle-enabled self-assembly of porous and monolithic carbon membranes for bioelectronic interfaces.

Real-world bioelectronics applications, including drug delivery systems, biosensing and electrical modulation of tissues and organs, largely require biointerfaces at the macroscopic level. However, traditional macroscale bioelectronic electrodes usually exhibit invasive or power-inefficient architectures, inability to form uniform and subcellular interfaces, or faradaic reactions at electrode surfaces. Here, we develop a micelle-enabled self-assembly approach for a binder-free and carbon-based monolithic device, aimed at large-scale bioelectronic interfaces. The device incorporates a multi-scale porous material architecture, an interdigitated microelectrode layout and a supercapacitor-like performance. In cell training processes, we use the device to modulate the contraction rate of primary cardiomyocytes at the subcellular level to target frequency in vitro. We also achieve capacitive control of the electrophysiology in isolated hearts, retinal tissues and sciatic nerves, as well as bioelectronic cardiac sensing. Our results support the exploration of device platforms already used in energy research to identify new opportunities in bioelectronics.

Modern bioelectronics with multi-scale structures are used extensively for drug delivery, biosensing, and biological modulations[1-4]. Through modulation of biological activities, these bioelectronics have contributed to our improved understanding of biological dynamics and function. Moreover, they hold great therapeutic potential for treating many biological disorders, including Parkinson’s disease, congenital heart defects, and paralysis. However, macroscopic bioelectronic devices are usually rigid and mechanically invasive to cells and tissues. Their large feature sizes also make subcellular biointerfaces difficult to form. Many devices also display faradaic reactions at the electrode surfaces, which can generate toxic reactive species, corrode electrodes, and cause permanent damage to adjacent tissues[5].

Nanostructuring of the bioelectronics surfaces represents a promising way to improve the device performance[6]. Standard bioelectronic stimulation materials, such as platinum, iridium oxide, or titanium nitride, all display enhanced charge injection limits when their surfaces become nanostructured or porous[7]. Nanostructured conducting polymer-based bioelectronics, such as those based on poly(3,4-ethylenedioxythiophene) polystyrene sulfonate (PEDOT:PSS), are also widely used for biological interfaces, and they show great electronic/ionic conductivity, internal porosity and volumetric capacitance, and excellent mechanical properties[8,9]. Improvements are ongoing to make these polymer electrodes electrochemically more stable, especially in terms of repeated operation and swelling during long-term soaking in saline solution[10,11]. Besides, electrode surface coating with carbon nanomaterials, such as graphene and carbon nanotubes (CNTs), are widely used to reduce the impedance at the electrode/saline junctions and increase the charge transfer rate[12-14]. However, the polymer binding process usually associated with carbon coating procedures may result in ‘dead’ volumes or surfaces and increased total device thickness[15]. Moreover, the potential production of freestanding carbon nanostructures after dissociation of the coating layer may result in biological complications in vivo[16,17]. Carbon nanostructures have nevertheless been advancing modern electronics[12-17] in multiple areas, such as supercapacitor[18,19] or micro-supercapacitor[20]-based power elements, as well as stretchable or bendable microelectrodes for biointerfaces[3,7,21-23]. Therefore, there are still numerous opportunities to further advance carbon-based materials toward cheap, stable, minimally invasive, high performance and multifunctional bioelectronic devices.

In this article, we aim to integrate many key features of the advanced biointerface-forming electrodes into the carbon-based bioelectronic devices, such as the surface and internal porosity, mechanical compliance, and the use of nanomaterial building blocks for bottom-up construction. We rely on micelle-enabled self-assembly to prepare a binder-free (i.e., monolithic), carbon-based, and flexible micro-supercapacitor-like system for various types of bioelectronic interfaces. The monolithic carbon minimizes the biocompatibility complication associated with freestanding carbon nanostructures. The interdigitated electrode design shrinks the feature size of the otherwise bulky electrode for subcellular interfaces. The devices can operate under either micro-supercapacitor-like or a traditional monopolar electrode configuration under physiological conditions, yielding modulation of cardiomyocytes (CMs) in vitro, excitation of isolated heart and retinal tissues ex vivo, stimulation of sciatic nerves in vivo, as well as bioelectronic cardiac recording.

Carbon membrane synthesis and characterization

We adopted a bottom-up approach for the direct preparation of monolithic (i.e., binder-free) and hierarchical carbon membranes. Nanoscale micelles were prepared through biphasic interaction between triblock copolymer Pluronic F127 (the primary organic template) and resin (the primary carbon precursor) in ethanol. The micelle solution was spin-coated onto the thermal oxide silicon wafer. Following solvent evaporation-induced self-assembly (Fig. 1a(i)), the packed micelles were carbonized completely into mesoporous carbon membranes. The carbonization process leads to the crosslinking of carbon precursors and the formation of monolithic membrane structures. The process yields highly ordered mesostructures with uniform pore size of 7 nm, as shown in transmission electron microscopy (TEM) images (Fig. 1b; Supplementary Fig. 1). Nitrogen adsorption/desorption measurements further confirmed a narrow pore size distribution (Supplementary Fig. 2).

Figure 1.

Hierarchical porous carbon synthesis and characterization.

(a) Illustration of the preparation of hierarchical porous carbon. (i) Synthesis of mesoporous carbon; (ii) synthesis of hierarchical carbon containing both mesopores and macropores; (iii) layer-by-layer assembly of carbon layers creating porosity gradient. The bottom layer of the carbon membranes will become the side for biointerface formation after the device fabrication is completed (Fig. 2a, and Supplementary Fig. 6). (b) Transmission electron microscopy (TEM; left; scale bar, 100 nm), corresponding fast Fourier transform (FFT) diffraction pattern (left inset) and scanning electron microscopy (SEM) images (right; scale bar, 200 nm) of mesoporous materials showing the highly ordered mesostructures. (c) Cross-sectional view (upper panels) and associated top view (lower panels) of the hierarchical porous material. The hierarchical structures display two components: a bottom layer constructed from ordered mesoporous structure and layers of porous vesicles assembled into multiple layers. Scale bar, 200 nm. (d) Left: representative load versus displacement plots of the hierarchical porous thin film measured with a nano-indenter with various numbers of vesicle layers. The experiment was repeated four times. Right: Hardness and Young’s modulus calculated from the load versus displacement plots (mean ± s.e.m., n = 4 independent measurements at various locations on each kind of sample). As the number of layers increases, the Young’s modulus and hardness decrease, indicating that the film becomes softer. (e) Cross-sectional (false colour) SEM image of cardiomyocytes (CMs) cultured on the hierarchical porous film, implying that soft-hard hybrid interfaces can form between the film and cells. Scale bar, 100 nm. Results in b, c, and e are representative of five independent repeats.

To reduce the stiffness of carbon films for improved compliance with soft biological interfaces, we introduced an inorganic template for the creation of a macroporous layer. Addition of ~200–300 nm-sized silica (SiO2) spheres coated with dopamine into the Pluronic F127 and resin mixture during the membrane preparation (Fig. 1a(ii)) introduced macroporous (i.e., with pore size >50 nm) structures (Fig. 1c; Supplementary Figs. 3 and 4). The role of dopamine here is to stabilize dispersion of SiO2 spheres in the micelle solution and to provide an additional carbon source. The hierarchical porous carbon preparation scheme permitted layer-by-layer assembly (Fig. 1a(iii)) of mesoporous carbon by spin coating, where the thickness and porosity of individual layers can be controlled. Subsequent buffered hydrogen fluoride (HF) treatment removes the SiO2 templates and releases the membrane (Supplementary Figs. 5). While the mesoporous film had a Young’s modulus and hardness of 25.30 GPa and 3.82 GPa respectively, an addition of ~700 nm of the macroporous layer reduced these values to 4.20 GPa and 0.69 GPa (Fig. 1d). We hypothesized that electrochemical devices made from highly porous layers may deliver efficient capacitive currents that may be useful for biological modulation when biointerfaces are formed (Fig. 1e)[24].

We fabricated the device in a micro-supercapacitor-like design and over an SU-8 substrate. The SU-8 think film is chosen because it is photo-patternable, mechanically flexible, and chemically stable, and it has been commonly applied as a substrate or encapsulation material in flexible electronics[25,26]. Since SU-8 is incompatible with the carbonization conditions under high temperature, all photolithography processes have to be performed on the fully carbonized membrane. To avoid difficult substrate transfer steps, we followed an ‘upside-down’ fabrication method (Fig. 2a; Supplementary Fig. 6), where the last step releases the device from the substrate and exposes the side for potential biointerfaces (i.e., the original interface between the thermal oxide and the carbon, Fig. 1a (iii)). This fabrication technique requires that the softer layer (i.e., the layer that forms direct biointerfaces) be on the bottom of the as-made membrane, hence the layer-by-layer assembly has to start from the layer that yields macropores. We used patterned metals layer as the mask for carbon etching as well as the electrical conductor to the carbon-based porous biointerfaces. By shrinking the pad size for interdigitated patterns, we were able to produce multichannel micro-supercapacitor-like devices (Supplementary Fig. 5). Raman spectra mapping showed carbon peaks at 1330 and 1590 cm−1 only in the patterned region, confirming the presence of carbon membrane (Fig. 2c).

Figure 2.

Device fabrication and characterization.

(a) Overview of a flexible device fabrication workflow. (b) An optical image (left; scale bar, 1 mm) and a close-up view of the hierarchical porous carbon micro-supercapacitor device (right; scale bar, 200 µm). Photographs are representative of more than twenty devices that were fabricated. (c) Raman spectra for the micro-supercapacitor device at the indicated locations (left; scale bar, 10 µm). The spectra showing typical carbon peaks at around 1330 cm−1 and 1590 cm−1 are only found in the designed pattern region (right). (d) CV profiles for the micro-supercapacitor at different scan rates in cell culture medium DMEM. (e) Charge-discharge curves in DMEM medium of the micro-supercapacitor device at different current densities for the time window of 6 s (top) and 10 s (bottom). (f) Electrochemical stability of the device during storage at 37°C in a buffer solution (PBS) over a 1-month period. (g) Electrochemical stability over 1,000,000 CV cycles in the range −0.1 to 0.1 V at the frequency of 4 Hz at 37°C in a buffer solution (PBS).

Mechanical simulations for puncture deformation showed that compared to the uniformly covered multilayered device, the interdigitated pattern can reduce the maximum von Mises stress present in both the carbon and metal layers, suggesting a mechanical justification of the micro-supercapacitor-like device (Supplementary Fig. 7). We also performed electrostatics simulations to model the electric potential at the surface of the device and between the interdigitated electrodes (Supplementary Fig. 8). Comparing a 2-prong and a 10-prong electrode model with the same total electrode area, the 10-prong electrode geometry yields a more confined electrical potential distribution at the cross-section. This is reminiscent of the local return electrode designs for the photovoltaic devices used for retinal implants[27]. We speculate that this confined z-axis electrical potential may be beneficial for stimulation of cell monolayers in vitro or tissues with multilayered circuit organization (e.g., retina).

Next, we conducted electrochemical tests using several physiological electrolytes. The electrochemical performance of the porous carbon membranes was analysed by cyclic voltammetry (CV) and galvanostatic charge/discharge techniques (Figs. 2d and 2e; Supplementary Figs. 9-12). In Na2SO4 electrolyte solution, the device showed near-rectangular CV profiles up to a scan rate of 100 mV/s (Supplementary Fig. 9), confirming that the device acted as an electrostatic double-layer capacitor. Notably, the devices also showed a near-rectangular CV profile in Dulbecco’s Modified Eagle Medium (DMEM) up to a scan rate of 100 mV/s, demonstrating that the capacitor behaviour was maintained even in cell culture media (Fig. 2d). Areal specific capacitances calculated from the CV profiles (Supplementary Fig. 9) were comparable to those of other supercapacitors or micro-supercapacitors[28-30] (Supplementary Table 1). Electrochemical impedance spectroscopy tests revealed low equivalent resistances of ~ 680 Ω and 715 Ω of the device in phosphate-buffered saline (PBS) and DMEM solutions, respectively (Supplementary Fig. 13), suggesting good charge transport properties. Without optimization, the charge injection limit of the device is estimated to be at least ~ 120 µC/cm2 (Supplementary Fig. 14), which is comparable to Pt and Pt/Ir alloy electrodes used in tissue stimulations (Supplementary Table 2). Additionally, the electrode preserved a near-rectangular CV profile with a scan rates up to 4 V/s (Supplementary Fig. 15).

Lastly, the intrinsic device stability in physiological solution was tested by submerging the device in 37 °C PBS solution for one month, during which no obvious change in capacitance rate was observed (Fig. 2f). The device showed electrochemical and mechanical stabilities over the course of at least 1,000,000 cycles (Fig. 2g; Supplementary Figs. 15 and 16). Together, these results demonstrate that the device is promising for bioelectronic interfaces.

Biological training in vitro

We first confirmed the viability of CMs and primary rat cardiac fibroblasts (RCFs) cultured over the device. Live-dead assays showed that ~100% of CMs were still alive after a 3-day culture (Supplementary Fig. 17). RCFs nuclear staining on the device showed healthy nucleus morphology, and no obvious nuclear changes due to cell death or apoptosis were observed (Supplementary Fig. 18). To have a more sensitive evaluation of the device in vitro biocompatibility, we conducted the lactate dehydrogenase (LDH) cytotoxicity assay using RCFs. The LDH cytotoxicity percentages in SU-8 group on day 1 (mean value, 3.91%) and day 2 (mean value, 2.97%) indicate that SU-8 alone caused a very mild cytotoxicity in vitro, although SU-8 is widely recognized as a biocompatible material[25,26]. When all the device materials are present (i.e., SU-8, porous carbon and metals), the LDH percentages are 8.77 % (mean) and 4.93 % (mean) for day 1 and day 2, respectively. However, the LDH levels in both the SU-8 and the device groups decreased on day 2 versus on day 1, suggesting that the cellular tolerance for these synthetic materials increased over time.

Haematoxylin and eosin stain of mice skin tissues surrounding the implanted devices showed that subcutaneous implantation caused some mild histological changes in week 1 and 4 (Supplementary Fig. 19). No significant inflammation reactions and tissue damage were observed. The histological appearance of the skin tissue above the implanted device (i.e., the epidermis and the dermis) was normal in week 4, except for the mild thickening of subcutaneous connective tissue and mild recruitment of immune cells including macrophages and neutrophils. Additionally, we monitored mouse body weight and no weight loss was observed between control groups and device groups.

We analysed the formation of biointerface between the device and the cultured CMs by verification of expression of typical cardiac markers, such as cardiac troponin and connexin-43 (Fig. 3a). We additionally confirmed that cells could spread out on the device surface and form subcellular interfaces (Supplementary Figs. 20 and 21). To evaluate how the charging/discharging cycles from the micro-supercapacitor-like device affect the CMs electrophysiology, we applied the stimulation current to the CM-interfacing devices and performed simultaneous calcium imaging to monitor cellular electrical activity. Each side of the interdigitated electrodes acts as a lead and a square current waveform is applied between them (see Methods for details). Prior to stimulation, CMs were synchronized with a baseline rate of ~0.67 Hz. Upon application of an input electric current waveform, overdrive pacing was achieved, and the contraction rate immediately synchronized to the pacing frequency (Fig. 3b, large field of view Supplementary Fig. 23). It is noteworthy that the pacing rate doubles that of the stimulation rate (i.e., the applied current frequency) (0.5 Hz pacing rate in Supplementary Fig. 22, 1 Hz pacing rate in Fig. 3b and Supplementary Fig. 23). This is expected as both the anodic and the cathodic phases of the electrochemical stimulation (from the same finger electrode area) can depolarize the CMs, although the action potentials (APs) may be initiated at different subcellular locations. One advantage of the electrical stimulation from the interdigitated layout is that it can achieve direct cell modulation uniformly across the entire device area. Also, the confined electrical potential around the finger electrodes may help improve the efficiency of stimulation as CM cultures are typically monolayer or sub-monolayer, and it may be unnecessary to deliver an electric field far above the cell surface.

Figure 3.

In vitro biological training.

(a) Representative immunohistochemistry images for cardiac cells cultured on the micro-supercapacitor device. Cells were stained for cardiac troponin (CMs, green), connexin-43 (magenta), vimentin (fibroblasts, red) and DAPI (nuclei, blue). Dash lines, the edges of the interdigitated carbon electrodes. Scale bar, 10 µm. The experiment was repeated twice with similar results. (b) Upper left: representative image of CMs loaded with calcium sensitive dye; transmitted light shows the micro-supercapacitor (appeared in black). Scale bar, 10 µm. Upper right: device voltage (top) and input current density (bottom) during the stimulation. Bottom: cell contraction rate illustrated by plotting the intensity profile of the region of interest (ROI, highlighted in the image). Overdrive stimulation was reproduced with similar results in three independent experiments. (c) Representative image of CMs loaded with calcium sensitive dye recorded during subthreshold stimulation for 2000 s. Scale bar, 10 µm. Overlay shows approximate positions of CMs in the field of view and their cluster assignment was color-coded in blue (cluster 1) and red (cluster 2). (d) Representative traces of fluorescence intensities at the region of interest (ROI dots in panel c) at the beginning of experiment and at 1100, 1500 and 1900 s of stimulation. Microphotographs (c) and cell trainings (d) in subthreshold pacing are representative of three independent experiments. (e) Frequency analysis of cell contraction frequency during subthreshold training. Top panel shows the voltage (left) and the input current density (right) waveforms during the training process. Bottom: frequency analysis shows fluctuating but gradual increase in the contraction rate, which eventually reached the target training frequency of 1 Hz.

Although this ability to perform overdrive pacing of CMs may have promising therapeutic applications, we also wanted to demonstrate the utility of the device for in vitro cellular manipulation that would allow for basic mechanistic investigations. To this end, we applied current stimulation with subthreshold amplitude, which did not elicit a direct overdrive pacing response. Figure 3c shows a field of view in which 5 CMs synchronized in 2 clusters were identified with very low spontaneous activity (~0.2 Hz and 0 Hz). When a low amplitude current waveform was applied (5 times lower than for overdrive pacing), the CMs did not show an immediate response. However, when the stimulation was applied for a longer duration, a gradual increase in contraction was observed. After ~1900 s of stimulation, the cells were evidently ‘upregulated’ by the current stimulation and their contraction rate increased to the target stimulation rate of 1 Hz. Figure 3d shows representative calcium imaging traces from two regions of interest (ROI), before and during the training of the two cell clusters, and Figure 3e shows gradual increase in contraction frequency over time. These results resemble the gradual increase in the rate of electrical activity previously observed in CMs upon subthreshold optical stimulation[31]. While the cells within each cluster are synchronized throughout the training, the two clusters display negligible synchronization to each other before they reach the targeted frequency (Supplementary Fig. 24, Supplementary Video 1, Supplementary Video 2). This fact suggests that the subthreshold training displays cellular level heterogeneity and the stochastic events may be involved in this process. Although a detailed understanding of the precise underlying mechanism here requires further rigorous investigation, we postulate that repetitive stimulation alters the resting membrane potential of the CMs to the point that they are sufficiently depolarized to elicit APs. Possible mechanism might relate to the ‘memory effect’, which refers to the way supra-threshold stimulation alters the excitability of cells and their resulting resting frequency[32]. Because the porous carbon-based devices operate via capacitive charge injection, we were able to pace or train the CMs in a biocompatible way without generating faradaic reactions which, also could cause electrode hydrolysis, increase the resistance over time, and in severe cases, cause device failure.

Biological modulation at the tissue and organ levels

Intact neural tissues represent highly crowded environments with non-neuronal factors, such as the extracellular matrix and glial interactions, that can influence neuronal activation. To determine whether the micro-supercapacitor-like device can modulate activity in intact neural circuits, we performed stimulation experiments on isolated mouse retinas. The laminated organization of the retina and its diverse neural circuit motifs make the retina an accessible model system to study brain circuit functions. Additionally, the use of retinal tissue allows us to stimulate physiologically relevant “input neurons,” i.e., the photoreceptors, and record activity from the “output neurons,” i.e., the retinal ganglion cells (RGCs) of a well-defined sensory circuit. In these experiments, we utilized retinas isolated from transgenic mice (Slc17a6/Gt(ROSA)26Sor) expressing the calcium indicator GCaMP6 in all RGCs[33,34]. This enabled simultaneous monitoring of neural activity via multiple neurons. Micro-supercapacitor-like devices and interconnects were fixed on glass coverslips using glue and PDMS and used as bottoms for the retina perfusion chamber. Dissected retinas were then positioned with the photoreceptor layer facing the micro-supercapacitor-like device and the RGC layer facing upward (Fig. 4a). Using two-photon laser scanning microscopy, we recorded calcium transients from RGCs while stimulating the photoreceptor layer with micro-supercapacitor modulation. We used square current waveforms applied across the interdigitated electrodes within the device to modulate the neural activity of the ganglion cell layer. Current flow direction was periodically switched by alternating between the anodic and cathodic phases every 3.5 s. Throughout the stimulation process, large periodic transients could be observed across several RGCs. During the anodic phase, RGCs exhibited large calcium transients, while during the cathodic phase, the calcium levels in RGCs returned to resting levels (Fig. 4b). To corroborate that the micro-supercapacitor stimulation observed was due to activation of the glutamatergic pathway from photoreceptors to bipolar cells to RGCs, we applied a cocktail of glutamate receptor antagonists, including AP-5, L-AP4, and DNQX, to silence glutamate transmission in the retina. Upon glutamate receptor blockade, the large calcium transients could no longer be observed in RGCs during micro-supercapacitor stimulation (Fig. 4c). These results demonstrated that the charging and discharging stimulation at the supercapacitor-photoreceptor interface evoked changes in glutamate release from photoreceptor neurons and could therefore activate the retinal network.

Figure 4.

Biological modulation at the tissue and organ level.

(a) Left: schematic of the experimental setup for the retinal stimulation. Right: a max-intensity projection of RGCs expressing GCaMP6. Scale bar, 25 µm. (b) Top: a retinal calcium image showing activated RGCs upon the stimulation. Scale bar, 25 µm. Middle: representative calcium traces from individual RGCs (numbered in the upper image). Bottom: the input current density during the stimulation. Confocal images in a and b are representative field of views from four repeats. (c) Top: calcium transient traces represented as average ± standard error (s.e.m.) in both control and glutamate antagonist conditions. Bottom: the quantification of response amplitudes of calcium transients (mean ± s.e.m., n = 6 RGCs across various experimental conditions). Calcium transients from each RGC were averaged across five repeated measurements, and statistical analyses were performed using pairwise one-way analysis of variance (ANOVA) with Bonferroni correction. Adjusted P values were indicated in the plot; n.s., not significant. (d) Schematic of the Langendorff perfusion system. ECG and LVP were monitored to show the effect of stimulation on the heart rate. (e) Images of a micro-supercapacitor-like device conforming around a cylindrical holder (left; scale bar, 2 mm) and around the curvilinear and contractile cardiac tissue (upper right; scale bar, 5 mm) with a close-up view (lower right; scale bar, 500 µm). Images in e are representative of more than ten independent experiments. (f) Representative LVP profiles and ECG recordings of the isolated heart stimulated at a frequency of 1 Hz. Input current density was synchronized to the corresponding stimulated portions of the LVP and ECG recordings. Dashed boxes and arrows are spontaneous APs (black) and APs that follow positive (green) and negative (pink) artifacts. (g) A closer look of the APs in the ECG recording with corresponding colours. (h) Representative LVP and ECG profiles of the isolated heart stimulated by 2 Hz square current waveform (left), 5 ms alternating pulses (middle), 1 ms biphasic pulses (right) to achieve pacing at 4 Hz. Presented results from the isolated heart stimulation are representative of three independent experiments using independent hearts and electrodes.

Next, we evaluated the bioelectronic stimulation in tissues and organs that required the traditional charge injection configuration. The rat heart contained a ~75 µm thick epicardium separating the CMs and the device[35]. Therefore, a single micro-supercapacitor-like device was not sufficient to accumulate the depolarizing charge. In these experiments, one porous carbon-based device (i.e., working electrode) was placed on the left ventricular (LV) wall of a rat heart to apply different current waveforms, while the other device) was placed on the right ventricular (RV) wall (Fig. 4d). We found that the two porous-carbon device configuration is highly efficient as both flexible devices form good contacts with the heart surface (Fig. 4e). We first applied anodic/cathodic current square waves with different frequencies as this can yield similar charging/discharging processes from a supercapacitor device (1 Hz, ± 0.5 mA/cm2, Fig. 4f; 1.67 Hz, ± 0.4 mA/cm2, Supplementary Fig. 25; 2 Hz, ± 0.37 mA/cm2, Fig. 4h; and 5 Hz, Supplementary Fig. 26). Upon stimulation, the heart immediately contracted at double the stimulation rate (Figs. 4f and 4h; Supplementary Video 3). This is similar to the observed in vitro pacing with a micro-supercapacitor device configuration, suggesting that both the anodic and cathodic stimulations from the working electrode can achieve a pacing effect. Due to the capacitive nature of our device and lack of faradaic charge injection, we hypothesize that the mechanism governing the electrical stimulation is similar to field coupling[36,37]. We observed positive/negative ECG artifacts during the stimulation (Figs. 4f and 4g), while no artifact was associated with the spontaneous contraction. As the ECG was recorded from the aorta and the LV, the observed artifact may be attributed to the working electrode positioned on the LV. It should be noted that the artifact shapes and the positions are dependent on the device/ECG configurations, so they vary in different settings. As the elicited heart potentials were initiated near the onset of the anodic/cathodic phases of the square waves (Fig 4h left), we next shrank the duration for each phase while keeping the spacing of adjacent phases the same (i.e., 250 ms, Supplementary Fig. 27). The results from the alternating anodic/cathodic pulses (5 ms, Fig. 4h, middle) show effective overdrive pacing with 3.7 µC/cm2 charge injection per phase/pulse. Finally, we attempted the electrical pacing with conventional charge-balanced biphasic pulses and achieved efficient overdrive pacing at 0.7 µC/cm2 charge injection in 1 ms pulse width (Fig. 4h, right).

Finally, to demonstrate the utility of the porous carbon-based device for in vivo neuro-modulation applications, we interfaced it with sciatic nerves. We used an acute setting in which the device was interfaced with the exposed nerve. When one device was interfaced with the sciatic nerve and the other device was interfaced with the rat’s body, we observed that the associated limb was clearly moving with every cathodic phase of the current injection (Supplementary Video 4). This was further validated by electromyography (EMG) recordings from the rat limb, which showed large potential spikes that were synchronized to the cathodic phase (Supplementary Fig. 28).

Conclusion

In summary, we have developed a hierarchical carbon-based micro-supercapacitor-like device that is capable of biological modulation. The porous structures and the interdigitated designs improve the device mechanics and allow control over the current flow through capacitive charging/discharging. The device was successfully applied to in vitro, acute ex vivo, and acute in vivo modulations with the capacitive charging/discharging cycles or traditional current pulses. The device also shows promise with respect to long-term stability and reasonable biocompatibility. Future device improvements include increasing the charge injection limit of the materials to match that of the best charge injection electrodes such as titanium nitride or iridium oxide[7]. Other functions of this micelle-based and self-assembled bioelectronics device may incorporate the flexible electrical sensing, as shown preliminarily in the ECG recording from isolated hearts (Supplementary Fig. 29). Finally, while supercapacitors or micro-supercapacitors have been demonstrated in the past as implantable power elements, our findings suggest new utilities of these designs for future bioelectric therapeutics.

Methods:

Hierarchical porous carbon film preparation.

Resol was prepared by crosslinking of phenol and formaldehyde, using methods from the literature[38]. Pluronic block copolymer F127 (template) and phenolic resol (carbon source) were mixed at ratio of 1:2 into an ethanol solution. The precursor solution was stirred for 1 hour before use. Vesicle structures were constructed using silica nanosphere templates. 200 nm-diameter silica nanospheres (NanoComposix, 10 mg/ml) were surface modified with a dopamine layer, according to a literature method[39]. The modified nanospheres were then added into precursor solution and mixed until a homogeneous mixture formed. Silicon wafer (Nova Electronic Materials, p-type, 600 nm thermal oxide SiO2) was cut into small pieces with suitable sizes (e.g. 2 cm × 4 cm) and cleaned by O2 plasma (Plasma Etcher, PE100) at 100 W for 2 minutes. The precursor solution with silica nanospheres was spin-coated (Laurell, WS-650 spin coater) onto the surface-cleaned silicon at 1500 rpm for 45 s. Multiple layers were formed by leaving the silicon wafer at room temperature for 10 minutes and then repeating the spin-coating. Then pure precursor solution was spin-coated onto the substrate at 3000 rpm for 45 s. The wafers were kept immobile for 4–6 h at 25°C, then baked in an oven for 24 h at 100°C. After baking, the wafers were transferred into an inert argon atmosphere and heated at 700°C (temperature rise rate at 5°C/min) for 30 min. The thin film was etched off the silicon wafer and nanospheres were etched by submerging the wafers in buffered hydrofluoric acid (HF) for 8 h. The thin film was rinsed with deionized (DI) water 6 times and dried prior to further characterizations.

Device design.

CAD was performed in AutoCAD software and schematics of comb-like devices are shown in Supplementary Figure 6. The pattern of interdigitated electrodes was consistent in each of the fabricated devices. Electrodes had a width of 15 µm with 10 µm spacing between them and 300 µm distance to the edge of the comb. The single channel device was comprised of 150 pairs of electrodes connected to 2 large pads, allowing for manual connection to jumper wires. The multichannel electrode had 8 sites arranged in 4 columns and 2 rows, and each site had 40 pairs of electrodes. Traces coming from the electrode sites were connected to 16 pads (2.25 mm length, 0.3 mm width, 0.5 mm pitch) matching a ZIF connector (FH12A-16S-0.5SH(55), Hirose Electric Co Ltd.). SU-8 patterns were designed with perforations in the area outside of the electrodes to increase the surface area available to buffered HF and to reduce the time required for SiO2 etching. However, we did not notice significant difference in fabrication and performance of devices with and without perforations.

Device fabrication.

A hierarchical porous carbon film obtained on p-type wet thermal oxide silicon wafer was used as a substrate for fabrication. A standard photolithography procedure was applied to make the desired pattern on the substrate. A schematic of the microfabrication protocol is shown in Supplementary Figure 6. Metal layers act as a hard mask for reactive-ion etching, and as a conductive and mechanical support for the carbon films. Chromium was used to improve adhesion between material layers. 5 nm Cr, 100 nm Au and 5 nm Cr layers were evaporated on the patterned surface using an e-beam evaporator (EvoVac, Angstrom Engineering), and extra carbon was removed by reactive-ion etching (Oxford Instruments PlasmaPro, NGP80). The supporting SU-8 layer had a thickness of approximately 10 µm. Patterns were exposed using a direct writer (MLA150, Heidelberg). The pattern on the SU-8 layer was etched from the substrate with buffered HF and transferred into DI water, afterwards washing the device with DI water 6 times. For free standing devices, electrodes were transferred onto polyimide film and allowed to dry. For single site devices, jumper wires were connected to the pads using conducting silver paste (Ted Pella) and after 24 h of drying, the connections and excess of connecting traces were insulated using silicon glue (Kwik-Sil, World Precision Instruments). For multisite devices, excess of traces was partially insulated using silicon glue. The connector header was strengthened with polyimide tape, cut to appropriate dimensions, and inserted into the ZIF connector. For cell and tissue culture, the devices were directly transferred onto glass microscope slides or into glass-bottom culture dishes. After ensuring even placement, the devices were allowed to dry and partially adhere to the glass. Next, jumper wires were connected using silver paste and left to dry for 24 h, after which the device was fixed to the dish and connections insulated with silicon glue. Devices were sterilized by oxygen plasma (100 W for 1 min) and ultraviolet light irradiation before cell or tissue culture.

Raman spectroscopy.

Raman spectra were recorded using a LabRAM HR evolution system (Horiba, Japan), and the mapping was done in the ultralow frequency module using a 532 nm laser.

Electrochemical measurements.

Cyclic voltammetry (CV) was performed over a wide range of scan rates at room temperature or 37 °C in various electrolytes. A potentiostat (SP-200, BioLogic) controlled with EC-Lab software with the three-electrode cell was used. A platinum wire was used as the counter electrode, a Ag/AgCl electrode (1M KCl) as the reference electrode, and a micro-supercapacitor device as the working electrode. Areal cell capacitance (C) was calculated from the cathodic phase of CV by , where I is the current, s is the scan rate, V is the potential window, and A is area of the electrode. Galvanic charge/discharge measurements were tested using two symmetric devices in a two-electrode configuration using source meter (Keithley 2636A, Tektronix, Inc.) controlled by a LabVIEW program (National Instruments). The voltage shown was the potential between the two symmetric devices.

Voltage transients.

Voltage transients were measured using a potentiostat (SP-200, BioLogic) controlled with EC-Lab Express. A platinum wire was used as a counter electrode and a Ag/AgCl (1 M) electrode was used as a reference. For the measurement, the carbon electrode was stabilized at 0 V versus Ag/AgCl for 1 ms in PBS at room temperature. 700 µs long cathodic and anodic current pulses were delivered to the electrode with the interpulse delay of 28 µs. The most negative (Emc) and positive (Eac) polarization potentials were assigned 28 µs after the maximum of a respective peak to account for access voltage and instrument delay. The charge injection limit was taken as a maximum charge injected without exceeding the −0.6 – 0.6 V potential window.

Electron microscopy.

Transmission electron microscopy (TEM, Tecnai F30, FEI) and scanning electron microscopy (SEM, Carl Zeiss, Merlin) were used to characterize the hierarchical porous materials and material/cell interfaces. CMs on the materials were fixed in 5% glutaraldehyde PBS solution for 30 min, washed in DI water, and then dehydrated with an increasing ethanol gradient from 30% to 98%. The samples were dried in a critical point dryer (Leica EM CPD300) and observed on the same SEM after coating with an 8 nm Pt/Pd metal layer on a sputter coater (Ted Pella, Inc.). The SEM was operated at a 2-kV accelerating voltage. Images were analysed using ImageJ.

Mechanical test.

Indentation modulus and hardness were measured by performing nanoindentation using a Hysitron 950 TriboIndenter in ambient conditions with a Berkovich indenter (three-sided pyramid-shaped diamond tip, tip radius ∼100 nm). All measurements were kept at a constant displacement of 200 nm. The data were analysed using standard Oliver and Pharr (1) analysis to extract the reduced modulus (E) and hardness by selecting upper fit at 95% and lower fit at 20%. The Young’s modulus E of the samples was extracted based on Eq. (1)[40], a general relation that applies to any axisymmetric indenter. The diamond tip has a Young’s modulus E = 1141 GPa and a Poisson’s ratio v = 0.07[41]. Here, we assume the Poisson’s ratio of the samples v = 0.25, which has been widely used for amorphous carbon[42].

Finite element simulations.

Finite element analysis was performed using COMSOL Multiphysics software. For details on modelling of electric potential and von Misses stress distribution please refer to Supplementary Note 1 and Supplementary Note 2, respectively.

Cell culture.

Hearts were excised from P1–3 neonatal rats into ice cold HBSS (without Ca2+ or Mg2+). The hearts were cut to small <1 mm piece, and then rinsed with HBSS to remove blood. A Pierce™ primary cardiomyocyte isolation kit (Thermo Fisher Scientific) was used for digesting the tissue according to manufacturer protocol. After isolation, the suspended cells were pre-plated for 2 h, allowing fibroblasts to adhere to the tissue culture plate. Then the enriched CMs population was seeded onto the micro-supercapacitor device pre-treated with fibronectin (Sigma). The cells were allowed to sit in culture media (DMEM high glucose + 10% FBS, 1% Glutamax and 1% penicillin–streptomycin) for 24 h, then the media was changed to CM-specific media (DMEM high glucose + 10% FBS, 1% penicillin–streptomycin, and 0.1% growth supplement from isolation kit).

LIVE/DEAD Viability.

CMs were cultured on the device for 3 days and then stained with green fluorescent calcein-AM (Life Technologies), red fluorescent ethidium homodimer-1 (Life Technologies) for 30 min. DAPI was included for nuclear counterstaining in dead cells. 100% EtOH treated cells for 20 min were included as a positive control. The as-stained cells were imaged using a Leica SP5 confocal microscope.

Nuclear staining assay.

Primary rodent fibroblasts harvested from rat pup hearts (postnatal day 1–3) were seeded and cultured on top of sterilized carbon micro-supercapacitors with fibroblast culture medium (high glucose DMEM + 10% FBS) for 36 hours in an environment of 37°C and 5% CO2. The seeded device was washed with PBS and then fixed with buffered 4% paraformaldehyde at room temperature. After permeabilization with 0.25% Triton X-100 in PBS, cells were counterstained with Hoechst 33342 (0.1 µg/ml) and imaged on Olympus inverted fluorescent microscope and Leica Confocal SP5 system. Positive control samples were prepared by treating fibroblasts with 1mM hydrogen peroxide to induce apoptosis.

Lactate dehydrogenase (LDH)-cytotoxicity assay.

Primary rodent fibroblasts harvested from rat pup hearts (P1-P3) were seeded and cultured on top of sterilized carbon micro-supercapacitors with fibroblast culture medium (high glucose DMEM + 10% FBS) in an environment of 37°C and 5% CO2. The culture medium supernatant was harvested at 24 (day 1, D1) and 48 hours (day 2, D2). LDH activity in the medium supernatant was measured using Pierce LDH Cytotoxicity Assay Kit (Thermo, US). Five independent measurements were conducted in each condition. Cycloheximide (CHX) at the concentration of 10 μg/mL for 24 hours was used to induce cell death as a positive control for this assay. LDH cytotoxicity percentage was calculated by , where is the corresponding LDH release from the CHX-, device-, and SU-8-treated samples, is the spontaneous LDH release activity, and is the maximum LDH activity detected in whole cell lysates.

Evaluation of in vivo biocompatibility in mice.

A piece of carbon micro-supercapacitor (0.5 cm × 0.5 cm) was surgically implanted under the dorsal skin of 6 to 8-week-old C57BL/6J mice through incisions, which were later closed with sutures. Mouse body weights were recorded on a weekly basis. At 1 or 4 weeks, mice were euthanized, and skin tissue surrounding the implant site was excised and fixed with buffered formalin for histology. The excised tissues were processed and stained with hematoxylin and eosin (H&E) for histopathological evaluation at university human tissue resource centre.

Device connections for the bioelectrical stimulations.

Devices were generally connected in a two-electrode (i.e., working electrode and counter electrode) configuration, where the counter electrode was connected to the ground. Square current waveforms (Figs. 3b, 3e, 4b, 4f, and 4h (left); Supplementary Figs. 25, 27, and 28) were delivered using a source meter (Keithley 2636A) controlled using a LabVIEW program. Current pulses (Fig. 4h, middle and right; Supplementary Fig. 27) were delivered using a potentiostat (SP-200, BioLogic) using EC-Lab Express software. The electric potential was reported as the potential between the two electrodes. In the in vitro cardiac training (Figs. 3b-e, Supplementary Figs. 22-24), the square current waveforms were applied to the two interdigitated electrodes within the same carbon micro-supercapacitor-like device (i.e., one interdigitated electrode as the working electrode and the other interdigitated electrode as the grounded counter electrode). In the ex vivo retinal experiments (Fig. 4b), the stimulation of isolated hearts (Figs. 4f and 4h; Supplementary Figs. 25 and 26) and sciatic nerves (Supplementary Fig. 28), we used two carbon micro-supercapacitor-like devices (i.e., with one device as the working electrode, and the other device as the grounded counter electrode). For the traces reported in this article, the square current waveforms or the current pulses were applied the two macroscopic devices.

Calcium imaging.

CMs were treated with calcium sensitive dye (2 µM Fluo-4, AM, cell permeant, Thermo Fisher Scientific) for 30 min at 37 °C. Cells were rinsed and incubated for 30 min to allow complete de-esterification. The treated cells were then visualized with a Leica SP5, STED-CW Super-resolution Laser Scanning Confocal. Details of time-series processing from the subthreshold stimulation experiments are described in Supplementary Note 3. Representative videos of at the beginning (t=0–31.395 s) and at the end of stimulation (t=1900–1931.395 s) are attached as Supplementary Video 1 and Supplementary Video 2.

Immunocytochemistry.

Primary cardiac cells were cultured onto devices for 3 days and were fixed (4% paraformaldehyde), permeabilized (0.2% Triton-X), and then blocked (2% BSA solution) for an hour to prevent nonspecific binding. Cells were incubated overnight at 4 °C with rabbit anti-cardiac troponin I antibody (1:400 in 2% BSA), chicken anti-vimentin antibody (1:500 in 2% BSA), and mouse anti-Connexin 43 antibody (1:100 in 2% BSA). Cells were rinsed and subsequently stained for 1 h with Alexa Fluor 488 anti-rabbit, Alexa Fluor 594 anti-chicken, and Alexa Fluor 647 anti-mouse secondary antibodies (1:250). DAPI (Invitrogen, P36931) was used to label the nuclei. Stained cells were imaged using the Leica SP5 Confocal and analysed using ImageJ.

Retina slice.

Vglut2-IRES-Cre mice (016963-Slc17a6/J) and floxed Ai95(RCL-GCaMP6f)-D mice (028865-Gt(ROSA)26Sor/J) were acquired from The Jackson Laboratory and were crossed to each other in the laboratory of Dr. Wei to obtain hybrid Vglut2-IRES-Cre/GCaMP6f transgenic mice. Mice of both sexes (postnatal days 21–35) were used for retinal calcium imaging experiments. Mice were anesthetized with isoflurane and decapitated after dark adaptation. Under infrared illumination, retinas were isolated from the pigment epithelium in oxygenated Ames’ medium (Sigma-Aldrich, A1420), cut into dorsal and ventral halves, and mounted on filter papers as described in the literature[43]. Retinas were kept in the dark at room temperature in Ames’ medium bubbled with 95% O2/5% CO2 until use (0–7 h). Glass coverslips with adhered micro-supercapacitor devices were used as the bottoms of the imaging chamber. During imaging, retinas were placed on top of micro-supercapacitor devices and gently pressed against the surface using a platinum (Pt) wire weight and perfused with oxygenated Ames at 32–33°C. Cells were visualized with infrared light (>900 nm) and an IR-sensitive video camera (Watec). Calcium transients from GCaMP6f expressing retinal ganglion cells were recorded by a custom-built two-photon microscope (Bruker) equipped with a Ti:sapphire laser (Chameleon Ultra II; Coherent Technologies) tuned to 920 nm[44]. Data were acquired using PrairieView software from a 100 μm × 100 μm field of view with an acquisition rate of ∼15 Hz. ROIs were manually drawn in ImageJ to enclose the soma of each GCaMP6f-expressing cell and a background region where there was no detectable GCaMP6s expression. Using custom MATLAB scripts, we calculated the average intensity over time for all ROIs and subtracted the background trace from light-responsive somatic traces to remove noise. The calcium traces were resampled to 75 Hz and smoothed using a moving average sliding window of 25 data points (~333 ms). For each stimulation protocol, the average was calculated for every cell.

Isolated heart.

An adult rat was heparinized (1,000 IU/kg IP) and anesthetized using open-drop exposure of isoflurane in a bell jar configuration. The heart was removed and placed in ice cold HBSS buffer, and the aorta was cannulated in preparation for use in a Langendorff setup. Oxygenated HEPES-buffered Tyrode’s solution (containing, in mM, NaCl 126, KCl 5.4, Glucose 10, HEPES 10, MgCl2 1, CaCl2 2, MgSO4 1.2, NaH2PO4 0.39; bubbled with 99.5% O2; pH 7.3) was perfused through the cannulated aorta. The perfusion was passed through a heating coil and bubble trap (Radnoti), and the heart was placed in a water-jacketed beaker (Fisher Scientific) to maintain a temperature of 37°C. The perfusion pressure was maintained at 80–100 mmHg by adjusting the height of the IV bag containing the perfusion buffer. The sinoatrial node along with the atria were removed, which resulted in a slow atrioventricular node pace (~1 Hz). The perfusion and left ventricular pressures (LVP) were monitored using a BP-100 probe (iWorx) connected to the perfusion line and a water filled balloon inserted to the LV, respectively. For ECG recordings, needle electrodes were positioned on the LV wall and aorta and connected to a C-ISO-256 preamplifier (iWorx). All signals (perfusion, LVP and ECG) were amplified using an IA-400D amplifier (iWorx) and interfaced with a PC using a DigiData 1550 digitizer with Clampex software (Molecular Devices). Two devices were placed on the heart; one below the heart’s apex, and the other covering the LV wall. Recordings from supercapacitor devices (Supplementary Fig. 29) were achieved with C-ISO-256 preamplifier (iWorx) as in a regular ECG setup, but with devices connected as input electrodes.

In vivo rat nerve stimulation.

Adult rats were deeply anesthetized with isoflurane (3–4%). The fur was removed from the hindquarters using a surgical clippers and hair removal cream. A semi-circular incision across the midline was made in the skin, and the fascial plane was opened between the gluteus maximus and the anterior head of the biceps femoris, thereby exposing the sciatic nerve. In this setting, two devices were used. One device was placed under the sciatic nerve and the other under the rat’s skin.

Animal subjects.

Mice and rats were housed in the animal facility of the University of Chicago. The animal room was maintained at a humidity of 40–60 % and a temperature of 18–23 °C under a 12-h light/12-h dark cycle. The animals were allowed free access to food and water. All animal procedures were approved by the Institutional Animal Care and Use Committees (IACUC) of the University of Chicago.

General data processing.

Analysis of numerical data was performed using Microsoft Excel, OriginPro or Python scripts. Statistical analyses were performed using GraphPad Prism. Plotting was performed using OriginPro, Adobe Illustrator or Python scripts using matplotlib library.