Stretchable multichannel antennas in soft wireless optoelectronic implants for optogenetics

The device platform enables separate, wireless operation of a collection of injectable microscale inorganic light-emitting diodes (μ-ILEDs, based on unpackaged devices with dimensions of 220 µm width, 270 µm length, and 50 µm thickness) in designs that are also adaptable to other light-delivery platforms. A radio frequency energy harvester receives signals from a transmitter, rectifies them, and triples the resulting voltage to provide a direct-current output for the μ-ILEDs. An impedance matching circuit, designed to maximize the received power, uses a ceramic chip capacitor (3 pF; 0.20 mm width, 0.4 mm length, 0.22 mm thickness) and an inductor (2.7 nH; 0.20 mm width, 0.4 mm length, 0.22 mm thickness) connected in series. The rectifier incorporates miniaturized Schottky diodes (1.7 mm width, 1.5 mm length, 0.5 mm thickness) and ceramic chip capacitors (5 pF; 0.20 mm width, 0.4 mm length, 0.22 mm thickness). The multiplier involves three Schottky diodes, identical to those in the rectifier, for the purpose of boosting the voltages provided by the rectifier (∼0.9 V) to values sufficient to operate the μ-ILEDs (∼2.7 V). To accommodate anatomical shapes and natural motions inside the targeted biological environment, the devices exploit principles of stretchable electronics in the form of serpentine metal interconnects passivated by layers of polyimide (40 µm thickness), with the entire system encapsulated in a low modulus silicone elastomer (∼0.5 MPa, 100 µm thickness for superstrate and substrate) to yield soft mechanics at the system level (effective modulus, ∼1.7 MPa) (Fig. 1A) (7). The fabrication procedures, along with details of the components and the circuit layouts, appear in Methods and SI Appendix, Fig. S1 and Table S1.

The thin, miniaturized geometry (4.3 mm width, 8 mm length, and 0.7 mm thickness); ultra-lightweight construction (33 mg); and mechanical compliance of these devices facilitate full implantation and chronic use in the brain (Fig. 1B). The system includes two parts: a back-end power harvesting and control unit and an injectable needle-shaped substrate that allows delivery of the μ-ILEDs (Fig. 1C, Right Top and Right Bottom) to targeted structures in the deep brain. Both components offer soft mechanics to impose minimal constraints on natural motions of the surrounding tissue (Fig. 1 C, Left and D), with reliable operation under levels of deformation that significantly exceed those expected in freely moving animals (7). Accelerated testing indicates that the soft silicone encapsulation preserves device functionality for 6 d when immersed in PBS (7.4 pH) at 90 °C, and the devices show no degradation after more than 2 mo in similar solution at 37 °C (SI Appendix, Fig. S2). Arrhenius-based extrapolations of testing at 60 °C and 90 °C suggest lifetimes of ∼6 mo at 37 °C (SI Appendix, Note S1) (15). These soft, flexible components also reduce brain trauma upon insertion and minimize tissue damage after chronic implantation and activation (9, 10, 16). The high degree of compliance of the injectable part of the system can, however, confound mechanical penetration into the brain tissue. Previous approaches exploit releasable injection needles to overcome this challenge (9, 10). Here, a biodegradable needle made of poly–lactic-coglycolic acid (PLGA) obviates the need for extraction; this needle provides sufficient stiffness for injection into regions of the deep brain such as the LC and nucleus accumbens (NAc) but fully dissolves after exposure to biological fluids (SI Appendix, Fig. S3A) (17–19). In vitro testing shows that at 37 °C the needles lose half of their thickness (an eightfold reduction in their bending stiffness) in 3 d, with complete dissolution within a week (SI Appendix, Fig. S3 B and C). Heat generation is also a potential concern for long-term implantation and operation of electrical devices in animals. Experimental studies using implantable thermal sensors show that optical power densities of 10 mW/mm² (50% duty cycle; 10 Hz period; 50 ms pulse width) do not cause any significant, steady state temperature changes in the surrounding tissue (SI Appendix, Fig. S4). We note that the same concepts in multichannel antennas and wireless, battery-free operation can be applied to a broad range of other platforms for light delivery, including waveguide arrays and other technologies.

The key feature of these devices is that they support multichannel operation, thereby overcoming a critical limitation of the single-channel, single-wavelength capabilities of previously reported wired and wireless systems (7–11, 13). Here, an advanced, stretchable antenna structure integrates multiple capacitive coupling traces to yield nonoverlapping resonances for frequency-selective harvesting and control. Each two-channel antenna consists of three serpentine lines (Fig. 2A, blue, red, and black line), where each pair (blue and red line, red and black line) involves resonant capacitive coupling with its neighbor. In particular, coupling between the blue and red traces (channel 1, Fig. 2A) and between the red and black traces (channel 2, Fig. 2A) enables operation at 2.3 (channel 1) and 2.7 (channel 2) GHz, respectively. To illustrate this operation, computations of electric field magnitudes near the surfaces of the serpentine traces at a frequency of 2.3 GHz reveal enhanced capacitive coupling (20), and therefore resonant operation, associated with channel 1 but not channel 2, as featured in black and red dotted boxes (Fig. 2B, Right). Such coupling can be captured by the reflection coefficient (S₁₁) for incident electromagnetic waves (7, 20). Reductions in S₁₁ correspond to decreased reflections and therefore improved power transmission efficiencies. The results exhibit two separate resonances, corresponding to two independently addressable channels of operation (Fig. 2C). The circuit diagram in Fig. 2D highlights the nature of signal flows from the antenna to the corresponding DC voltage output nodes. Images of an operating device demonstrate that the frequency of the incident rf radiation can be adjusted to activate the channels independently (Movie S1). The optical intensity generated in the nontargeted channel as a function of that in the targeted channel appears in Fig. 2 E and F. As indicated by the dotted lines, the intensity can reach ∼12 mW/mm², which is more than sufficient to activate optogenetic proteins, before the onset of the operation in the nontargeted channel (21–23). The direct proximity of the light emitters in these injectable optoelectronic systems to the regions of interest in the brain enables activation of light-sensitive proteins with intensities of ∼10 mW/mm², much lower than those needed at the light source regions of systems that require delivery of light via use of waveguides (∼100 mW/mm²) (24). The designs of the stretchable antennas can be selected to achieve channel separations that can meet nearly any requirement in cross-coupling, as in SI Appendix, Note S2. Fig. 2G exhibits the maximum optical intensity at the targeted channel before the onset of the operation at the nontargeted channels as a function of channel separation from 0.4 to 0.8 GHz. Measurement results reveal that multichannel optoelectronic systems can drive optical intensities over 100 mW/mm², when the channel separation is above 0.7 GHz. In this way, cross-coupling between channels can be designed to negligible levels.

Advanced versions of the antenna design in Fig. 2 can support independent manipulation of three separate channels. The scattering parameters of such an antenna that consists of four serpentine lines (Fig. 3A, red, black, blue, and green line) where each pair (red and black line, black and blue line, and blue and green line) exhibits capacitive coupling to its neighbor at a distinct frequency indicate operation at 2.3 (channel 1), 2.7 (channel 2), and 3.2 (channel 3) GHz. As before, computations of the electric field magnitude at 2.3 GHz illustrate selective capacitive coupling associated with the pair of traces that defines channel 1. These findings are confirmed through a set of measurements and images of the devices (Fig. 3 C–G). As with the two-channel system, there is no significant cross-coupling for the conditions of interest (Fig. 3 D–F). For both two- and three-channel systems, the angular variations in power received at each channel can affect operation (4, 7, 20). Calculations show that the radiation patterns associated with each channel, in all cases, are similar, ensuring that each receives (or transmits) an equivalent amount of power at a given orientation (Fig. 3 H and I). In addition, the serpentine layouts of the antennas and the soft, elastomeric substrate and superstrate afford linear elastic response to strain (7). Tensile strains lead to shifting of the resonant frequencies to lower values, as expected from previously reported single-channel antennas (Fig. 3J) (7). This three-channel example highlights clearly the dramatic reduction in overall size compared with an otherwise equivalent arrangement of three separate, single-channel antennas (Fig. 3K). Advanced impedance matching techniques allow for further minimization of cross-coupling, thereby extending the number of channels that can be independently operated (SI Appendix, Fig. S6) (20). A practical limit for operation in the low gigahertz frequency ranges (0.7∼3.5 GHz, corresponding to a window of transparency for biological tissues) is ∼eight channels, based on a frequency range/channel margin of 2.8/0.4 (25, 26).

Multichannel operation relies on the capacitive coupling that occurs between adjacent serpentine lines. This coupling can be captured analytically with solutions to the 2D Laplace equation (Fig. 4 and SI Appendix, Note S3) for simple antenna structures with serpentine geometries approximated by sinusoidal curves (Fig. 4 A and B). Here the electromagnetic interactions can be captured as a sum of infinitesimally small capacitors with gaps defined by the separation between adjacent lines (Fig. 4B). As the gap increases, the capacitive coupling decreases, with an analytical dependence that is consistent with numerical simulations (Fig. 4 C and D). These studies reveal that the total capacitance of a stretchable antenna is approximately ∼1.5 pF. To achieve resonance in the low GHz range, inductances of 2∼3 nH are required, which is feasible within the dimensions of the antennas (20). Similar strategies can be extended to calculations of the inductances to yield the resonance frequencies. This analytical approach dramatically reduces the computational burden of full simulations of the actual antenna layouts by providing a starting point for the designs. Moreover, the results confirm that the operation of these types of antennas relies critically on capacitive effects.

The input impedance of the antenna is also important, as proper impedance matching can maximize the power transmission efficiency. The designs exploited here use an asymmetric geometry to increase the impedance over the very low values that are typically associated with antennas in this small size regime. Studies of input impedance as a function of the location of the input port provide insights into the role of this asymmetric geometry in input impedance and thus radiation efficiency. Moving the input from port 1 to port 4 breaks the symmetry, thereby increasing the impedance (Fig. 4 E and F) (20). Using these concepts to realize impedances of 50 ohms ensures maximum power transfer (20). The values of the reflection coefficient, S₁₁, are extremely sensitive to input impedance (Fig. 4G). The reflection values at each port, from 1 to 4, are 50%, 5%, 0.3%, and 25%, respectively (Fig. 4H).

Even with optimized impedance matching and antenna design, as the number of channels increases, corresponding increases in the total transmitted power could be necessary, depending on the nature of the optogenetic experiment. The upper limits could, with traditional systems, exceed exposure guidelines (25, 26). An actively managed array of antennas and a separately controlled transmission system that tracks the motions of the animals enable overall reductions in power by confining power delivery only to the relevant regions of an experimental assay chamber. Here, digital image processing on data collected with a monitoring camera (SI Appendix, Note S4 and Fig. S7A) locates the animal and then sends control signals to a multiplexer (MUX) that activates the antenna array in a way that directs power selectively to the corresponding area (SI Appendix, Fig. S7 B and C) (27). As such, transmission only occurs in the immediate environment of the targeted mouse, thereby greatly enhancing the efficiency of operation, to allow average power levels as low as 0.5 W (SI Appendix, Fig. S8) for a typical case (Movie S2). This type of selective activation also lends itself well to complex behavioral assays such as conditioned place preference or those that demand coverage over large areas, social interactions, home cage manipulations, or complex environments.

Multichannel operation that overcomes a critical limitation of the single-channel, single-wavelength capabilities introduced here demonstrates their utility for neural manipulation in native, homecage environments as well as standard behavioral assays with freely behaving mice. The LC has previously been implicated as a potent arousal center, such that activation of LC neurons can lead to sleep/wake transitions and speed emergence from anesthesia (28, 29). We set out to determine whether these subdermal devices can optogenetically drive naturalistic behaviors within the home cage by wireless photostimulation of LC-NEs. To do so, we used mice that express Cre recombinase under the promoter for galanin (Gal-Cre), which is expressed in a majority of noradrenergic LC neurons (LC-NEs) in combination with virally transduced Cre-dependent expression of ChR2 (14, 28–30). In these animals, the μ-ILED was placed directly adjacent to the LC to assess the ability of LC activation to drive locomotor activity during normal daily sleep periods in the mouse home cage (Fig. 5 A and B). Immunohistochemical analyses in Gal-Cre mice injected with AAV5–DIO–EF1α–ChR2 showed robust ChR2 expression (yellow) in the LC that overlaps with tyrosine hydroxylase (TH) expression (red), consistent with expression selectively in the galaninergic subset of LC-NE neurons (Fig. 5C) and previous studies (30). We singly housed Cre-expressing (Cre+) and Cre-negative (Cre–) littermates in home cages mounted on custom-built adapters that hold rf-emitting antennas in close proximity to the home cage (∼5 cm from the cage wall and floor) (SI Appendix, Fig. S7A and Fig. 5D). We followed a normal 12:12 h light:dark cycle and recorded video footage of the mice during a period of normal sleep and recorded for three successive 15-min periods. During the first and third periods, there was no activation of the implantable system (baseline and poststimulation), but during the second period, the implanted device was activated to drive LC neuronal activity by ChR2 (10 Hz, 50 ms pulse widths) (Fig. 5B) (14). Photostimulation of Galanin-expressing LC neurons drove increased locomotor activity in Cre+ animals compared with the pre- and postactivation period (Fig. 5E). In contrast, photostimulation of LC neurons in littermate Cre– mice did not alter the locomotor activity (Fig. 5F) (****P < 0.0001 two-way repeated-measures ANOVA with Bonferroni post hoc test). Taken together, these data demonstrate that selective photoactivation of galanergic LC neurons produces arousal and furthermore that these wireless devices can be easily used to perform behavioral experiments within the native home cage environment, which include studies of sleep or resting states that would otherwise be difficult with tethered cables.

Wireless subdermal optogenetic probes that offer multichannel integration with discrete neuronal subpopulations are needed to extend our knowledge of complex neural circuitry in more ethologically, naturalistic environments and behaviors. For example, multichannel capabilities are valuable in the context of experiments where multiple neuronal subpopulations need to be independently engaged. Recent efforts have shown that photostimulation of dynorphinergic neurons in the NAcSh is sufficient to induce both aversive and preference behaviors (12). Photostimulation of dynorphinergic cells in the ventral NAcSh (vNAcSh) shell through activation of the kappa opioid receptor (KOR) elicits aversive behavior. Activation of nearby dorsal NAcSh (dNAcSh) dynorphin cells, however, in another KOR-mediated process elicits preference behavior. Characterization of nearby neuronal circuits previously required separate animals, with separate fiber optic implants within each subregion for optogenetic stimulation. The multichannel devices introduced here can define and modulate this spatially compact circuit with a single group of animals. Multichannel devices implanted in preprodynorphin–IRES–cre-positive mice that were infected with an AAV5–DIO–ChR2–eYFP virus to selectively express ChR2 in dynorphinergic neurons in the NAcSh demonstrate this capability (Fig. 6 A–C) (12). Here, the vNAcSh targets one μ-ILED (channel 1) and the nearby dNAcSh targets the other μ-ILED (channel 2), separated by a distance of 1 mm (Fig. 6A). A custom-made unbiased, balanced two-compartment conditioning apparatus (52.5 × 25.5 × 25.5 cm³; Fig. 6D) allows assessment of preference or aversion behavior by comparing the amount of time spent in the μ-ILED–on (rf-on) side to the time spent in the μ-ILED–off side. Isolated photostimulation of the vNAcSh through activation of channel 1 produces aversive behavior (Fig. 6 E and F, ventral and SI Appendix, Fig. S9), whereas in the same animal illumination of the dNAcSh through activation of channel 2 produces a real-time preference behavior (Fig. 6 E and F, dorsal, and SI Appendix, Fig. S9; one-way ANOVA Bonferroni post hoc; ****P < 0.0001 rf-off vs. vNAcSh and dNAcSh vs. vNAcSh and **P < 0.01 dNAcSh vs. both and vNAcSh vs. both). Here we show the subdermal multichannel operation can photostimulate two discrete subregions in the NAcSh in a single animal using a fully subdermal implantable tether-free mode. These devices expand the range of complexity for in vivo optogenetic experiments that can be performed and provide a flexible solution for multichannel functionality for neural circuit dissection.

Fabrication was conducted on a clean glass slide (75 mm length, 50 mm width, and 1 mm thickness). Spin-casting polymethyl methacrylate (PMMA; 495 PMMA A8, Microchem) at 0.177 × g for 30 s and curing at 70 °C for 6 h formed a coating with thickness of 200 nm. Subsequently, spin-casting a precursor to polydimethylsiloxane (PDMS; Sylgard 184) at 1,000 rpm for 30 s and curing at 70 °C for 6 h yielded an overcoat with a thickness of 100 µm. An 18 µm-thick copper foil (Dupont) laminated onto the PDMS, patterned by photolithography (AZ 5214E, AZ Electronic Materials) and copper etching, defined the antenna and interconnects. Small amounts of lead-free solder paste (SMD290SNL250T5, Chipquik) applied to the surface mount devices and the µ-ILEDs (Cree Inc.) and baked at 285 °C in a vacuum oven for 10 min bonded the components to the patterned interconnects. An epoxy adhesive (200 nm), cured at 20 °C for 5 min, bonded this platform to the needle. Spin-casting PDMS at 1,000 rpm for 30 s and curing at 70 °C for 6 h formed a final overcoat to complete the fabrication.

Hot pressing the PLGA (65:35, Sigma-Aldrich Inc.) thin film at 100 °C for 5 min, followed by laser-cutting, formed thin PLGA (∼75 μm thick) needle-shaped substrates.

We used a commercially available finite element method tool, high frequency structural simulator (HFSS), to calculate the S₁₁ values, the normalized electric field magnitudes, and angular radiation patterns with a mouse mesh model where layers corresponding to biological tissues are modeled as Cole-cole relaxation models (25). Simulations revealed the following parameter values: 4.84 dBd antenna gain, 97% antenna efficiency, 95% rectifying efficiency, and 30% led efficiency.

We submerged multichannel optoelectronic systems in saline solution and monitored optical intensities as a function of time at various temperatures from 37 °C to 90 °C. Here, a hotplate was used to control the temperature of saline solution. For measurements of LED output, we used an optical measurement system (Extech LT300 Precision Digital Light Meter).

We used an optical measurement system (Extech LT300 Precision Digital Light Meter) to measure output of the LEDs. For accurate measurements of each individual channel, we prepared three types of samples, each with a different color LED, to facilitate spectrally separate measurements of each channel.

Of 15 devices implanted for animal studies, all remained functional, without significant change in operation, for more than 2 mo. Movie S2 shows a representative device during operation.

Adult (25∼35 g) male mice were group-housed, given access to food pellets and water ad libitum, and maintained on a 12:12-h light/dark cycle (lights on at 7:00 AM). The Animal Care and Use Committee of Washington University approved all procedures, which conformed to NIH guidelines.

For the arousal experiments, we used mice expressing Cre recombinase under the promoter for galanin (Gal-Cre) or littermate controls in combination with virally transduced Cre-dependent expression of ChR2 to target galaninergic neurons in the LC (33, 34). For surgery, we anesthetized the mice with isoflurane in an induction chamber and placed the animals in a stereotaxic frame that maintained anesthesia with 1∼2% isoflurane through a nose cone. We injected AAV5–DIO–ChR2 unilaterally into the LC [coordinates from bregma: –5.45 anterior–posterior (AP), ±1.25 medial–lateral (ML), and –.00 mm dorsal–ventral (DV)] in mice (Cre+ or Cre–) (30). For multichannel control of neuronal subpopulations experiments, we injected AAV5–DIO–ChR2–eYFP into the dNAcSh of preprodynorphin–IRES–cre-positive mice with targets of stereotaxic coordinates from bregma (+1.30 AP, ±0.5 ML, –4.25 mm DV) or vNAcSh stereotaxic coordinates from bregma (+1.30 AP, ±0.5 ML, –4.75 mm DV) (12). A minimum of 2 wk after the viral injections, we stereotactically implanted the devices in mice using the same coordinates as above for AP and DV but at ±0.3 ML. Then, we secured the device to the skull with tissue glue and sutured the skin.

For the home cage arousal experiments, we performed all behavioral experiments in a sound-attenuated room maintained at 23 °C at least 1 wk following the final surgery. At 10:00 AM, during a period of normal sleep and less activity, we recorded the activity of the mice for three successive 15-min periods by using Ethovision 8.5 (Noldus Information Technologies, RRID: rid_000100). During the first and last period, there was no activation of the implantable system. During the second period, the implanted system illuminated the LC under stimulation conditions (10 Hz, 50 ms pulse widths), which were previously shown to drive LC activation by ChR2 (14, 30).

For multichannel control of neuronal subpopulation experiments, we placed mice in a custom-made unbiased, balanced two-compartment conditioning apparatus (52.5 × 25.5 × 25.5 cm³) as described previously (12). During a 20-min trial, entry into one compartment led to activation of the devices and provided photostimulation to the NAcSh (rf-on) (10 Hz, 50 ms pulse widths), whereas the animal remained in the light-paired chamber and entry into the other chamber ended photostimulation (rf-off). Arrangement of the transmitting antennas targeting only one side of the real-time place preference assay led to the unilateral rf transmission. With this configuration, the active side of the assay always received rf transmission (rf-on). The rf signal generator power output ranged from –10 to 0 dBm at 2.3, 2.5, and 2.7 GHz for each LED device.

We showed all grouped data as mean ± SEM. Data were normally distributed, and independent Student’s two-tailed, unpaired or paired, t tests determined differences between two groups. One-way or two-way analysis of variances (ANOVAs) followed by post hoc Bonferroni or Dunnett’s multiple comparisons, if the main effect was significant at P < 0.05, determined differences between multiple groups. We took statistical significance as *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001 and conducted all analyses using Prism 5.0 (GraphPad).