Flexible and Implantable Polyimide Aptamer-Field-Effect Transistor Biosensors

Abstract

Monitoring neurochemical signaling across timescales is critical to understanding how brains encode and store information. Flexible (vs. stiff) devices have been shown to improve in vivo monitoring, particularly over longer times, by reducing tissue damage and immunological responses. Here, we report our initial steps toward developing flexible and implantable neuroprobes with aptamer-field-effect transistor (FET) biosensors for neurotransmitter monitoring. A high-throughput process was developed to fabricate thin, flexible polyimide probes using micro-electro-mechanical-system (MEMS) technologies, where 150 flexible probes were fabricated on each 4-inch Si wafer. Probes were 150-μm wide and 7-μm thick with two FETs per tip. The bending stiffness was 1.2 × 10⁻¹¹ N·m². Semiconductor thin films (3-nm In₂O₃) were functionalized with DNA aptamers for target recognition, which produces aptamer conformational rearrangements detected via changes in FET conductance. Flexible aptamer-FET neuroprobes detected serotonin at femtomolar concentrations in high-ionic strength artificial cerebrospinal fluid. A straightforward implantation process was developed, where micro-fabricated Si carrier devices assisted with implantation such that flexible neuroprobes detected physiological relevant serotonin in a tissue-hydrogel brain mimic.

Graphical Abstract

Implantable bioelectronics are powerful tools to monitor neural activity.^1–4 Implanted devices enable new discoveries associated with understanding healthy brain function and the biological basis and treatment mechanisms of neurological and neuropsychiatric disorders.^5–10 Devices for monitoring in vivo electrical signals have been extensively developed enabling chronic recordings lasting over six months with single neuron resolution.^11–20 In addition to electrical signals, chemical communication via neurotransmitters plays a central role in brain information processing. However, existing device technologies for neurochemical recordings are more limited.^5,21–23 A major challenge in neurochemical sensing is the development of biosensors that can detect arrays of neurotransmitters in vivo with high sensitivity and selectivity.

Recently, we developed aptamer-field-effect transistor (FET) biosensors for neurotransmitter detection.^4,23–28 Thin films of biocompatible In₂O₃ (~3 nm) for signal transduction were functionalized with single-stranded DNA (aptamers) for target recognition.^24–26 Aptamers selected for specific targets were covalently coupled to semiconductor surfaces.²⁴ Upon target capture, negatively charged oligonucleotide backbones undergo conformational changes near FET surfaces resulting in measurable changes in FET conductance and hence, target-related current. This sensing mechanism is universal since it is independent of the charge or electrochemical properties of the analytes themselves.²⁴ We have shown that aptamer-FETs can be used to detect biomarkers, including serotonin, dopamine, glucose, the membrane lipid S-1-P, phenylalanine, and cortisol in complex physiological environments.^24–27 Aptamer-FET biosensors show high selectivity to targets vs. structurally similar molecules. Recently, Zhang and coworkers developed graphene-based aptamer-FET biosensors using our aptamers for dopamine and serotonin sensing.^29,30

To monitor neurotransmitters in vivo, we developed Si-based implantable neuroprobes with miniaturized aptamer-FET biosensors. Probes were either 150 μm or 50 μm in width and thickness at the tips.⁴ The larger neuroprobes were implanted in awake, behaving mice where they detected electrically stimulated serotonin release with high sensitivity. While our previous study showed results for acute measurements (i.e., hours), challenges remain for stable, chronic in vivo recordings.^31,32

Conventional implantable electrodes, including the first-generation neuroprobes we developed, are typically fabricated using rigid substrates such as metals or Si, which are much stiffer than biological tissues.³³ For example, Si has a Young’s modulus of ~200 GPa, while brain tissue has a Young’s modulus of ~10 kPa. This Young’s modulus mismatch of >10⁷ induces immunological responses and subsequent glial scar formation. For biosensing applications, immunological responses lead to signal drift due to biofouling.^31,34 Scar formation decreases device recording accuracy and sensitivity over time.^31,34–38 Moreover, the formation of glial scars leads to biosensor failure due to limited access between the sensing surface and tissue interface.³⁴

Recent developments in soft and flexible electronics show potential for overcoming these limitations.^{14,20,22,39–45} Flexible bioelectronics with lower device Young’s moduli (i.e., increased elasticity) showed reduced immunological responses after brain implantation compared with rigid substrates.^{31,34,43,46,47} Flexible devices with the capability to detect neurotransmitters in real time are necessary to understand chemical neurotransmission, particularly via chronic neural recordings.^4,21,30,35 Along these lines, we recently developed a wearable smartwatch for sweat cortisol sensing using flexible polyimide-based aptamer FETs.²⁸ We also fabricated In₂O₃ nanoribbon aptamer-FET biosensor arrays on flexible poly(ethylene terephthalate) (PET) for the multiplexed detection of serotonin, dopamine, pH, and temperature in real time.²⁶

Many flexible substrates (e.g., PET and polyimide) still have Young’s moduli (typically on the order of GPa) that are orders of magnitude greater than brain tissue. One promising strategy to reduce the bending stiffness of a material is to reduce its thickness, as bending stiffness decreases cubically with thickness (equation shown in the Supporting Information).⁴⁸ Recently, polyimide devices with thicknesses <10 μm have been shown to have optimal bending stiffness and conformal contact with tissue.^20,40,42,49 However, ultrathin devices are difficult to process and handle.^30,40,50 Moreover, due to the low bending stiffness, thin flexible probes suffer from buckling during penetration into tissues, making them challenging to implant.

Here, we report on fabrication and implantation strategies for flexible and implantable polyimide (7-μm) aptamer-FET neuroprobes to monitor the small-molecule neurotransmitter serotonin. We designed a high-throughput process to fabricate polyimide neuroprobes using micro-electro-mechanical-system (MEMS) technologies, which are scalable, high throughput, and compatible with other microfabrication processes.⁴ Quasi-2D In₂O₃ FETs were fabricated on the probe tips. With aptamers functionalized on the In₂O₃ surfaces, flexible neuroprobes were used to detect serotonin at femtomolar concentrations in artificial cerebrospinal fluid (aCSF). We used an artificial brain tissue matrix to develop and to test a process to implant flexible probes using a rigid carrier as a temporary shuttle device with straightforward probe release. Flexible aptamer-FET probes were implanted into a brain-tissue hydrogel mimic and showed high sensitivities to serotonin.

Experimental Section

Materials

Oligonucleotides (aptamers) were obtained from Integrated DNA Technologies (Coralville, IA). The SYLGARD 184 used to make polydimethylsiloxane (PDMS) wells for recordings was from Dow Corning Corporation (Midland, MI). Deionized water (18.2 MΩ) was dispensed via a Milli-Q system (Millipore, Billerica, MA). Polyimide solution (#PI-2611) was provided by Dupont Teijin Films (Chester, VA). Prime quality 4-inch Si wafers (P/B, thickness 500 μm and 150 μm) were purchased from Silicon Valley Microelectronics, Inc. (Santa Clara, CA, USA). All other materials were purchased from Sigma-Aldrich Co. (St. Louis, MO) unless otherwise noted.

Neuroprobe fabrication

Polyimide solutions were spin-coated at 3000 rpm for 30 s onto 500 μm thick Si wafers to form a 7-μm thick film, as reported previously.²⁸ After coating, substrates were pre-heated at 90 °C for 90 s and then at 150 °C for 90 s on a hot plate. Afterward, the substrates were transferred to an oven and annealed at 350 °C for 30 min. Aqueous solutions (0.1 M) of indium(III) nitrate hydrate (99.999%) were then spin-coated at 3000 rpm for 30 s onto polyimide-coated Si wafers. After deposition of indium precursors, substrates were heated to 150 °C for 10 min, followed by 3 h of annealing at 350 °C.^23,51 Source and drain electrodes (10 nm Ti and 30 nm Au) were deposited by electron-beam evaporation (CHA Industries, Inc., Fremont, CA) and patterned via standard photolithography. A second photolithographic process was performed to define the outer perimeter of the probes. Reactive ion etching was used to etch through the polyimide substrates using a plasma etching system (STS-AOE, Plasmatreat, Hayward, CA) with the protection of the photoresist. Probes were released from carrier wafers using tweezers. The source and drain electrodes of the neuroprobes were connected with wiring using Ag epoxy (Ted Pella, Redding, CA).⁴

Implantation of flexible neuroprobes

Shuttle devices were fabricated from Si using previously published methods.⁴ Briefly, Si substrates (150 μm thick) were coated with photoresist to define probe profiles. Deep reactive ion etching using the Bosch process was used to etch through the unmasked regions of the silicon substrates using a Deep Silicon Etcher III (Plasma-Therm, Fremont, CA). To attach the flexible polyimide probes, Si shuttles were immersed into a physiological buffer solution to form a thin solution film on the Si surfaces. Since probes are destined for eventual brain implantation, we used aCSF as the buffer. Detailed aCSF preparation protocols are described in our previous work.²⁷ The polyimide probes were then attached to the Si shuttles via capillary forces. Alignment between polyimide probes and Si shuttle devices was achieved using tweezers under an optical microscope (Figure 3c).

Figure 3. Implantation of a flexible neuroprobe into a brain-mimic hydrogel.

(a) Schematic illustration of device assembly before implantation. A flexible probe was attached to a Si shuttle via surface tension and aligned under a microscope. (b) Schematic illustration of flexible probe implantation using a Si probe as the shuttle device. The assembled device penetrates the hydrogel without buckling. After insertion, the buffer layer between the flexible probe and the Si shuttle device diffuses between the hydrogel. The edge surface tension is relieved, enabling the flexible probe to be separated from the Si shuttle. The shuttle is removed, leaving the flexible probe implanted in the hydrogel. (c) Optical microscope image of an assembled device, where the flexible probe is partially aligned to the Si shuttle device. Photographs of (d) an assembled device implanted into the hydrogel, and (e, f) a freestanding flexible probe implanted in the brain-mimic hydrogel.

To mimic the properties of brain tissue for implantation, we used a 0.6% (w/v) agarose hydrogel, which was prepared by mixing agarose powder and aCSF. The mixture was microwaved until the agarose powder was dissolved (~1 min), poured into a container, and cooled to room temperature. Assembled probes and shuttles were implanted into the brain-mimic hydrogel using tweezers. After implantation for ~1 min, the Si shuttle devices were removed from the hydrogel leaving the polyimide probes in place (Movie S1).

In vitro experiments

To test device biosensing capabilities in vitro, we rinsed the FET probes in ethanol and dried with N₂ immediately before functionalization with thiolated serotonin aptamers. Mixtures of (3-aminopropyl)triethoxysilane and trimethoxy(propyl)silane (1:9, v/v) were thermally deposited on In₂O₃ at 40 °C for 1 h and annealed at 60 °C for 10 min.^4,23,24 Source and drain electrodes were insulated with a self-assembled monolayer of 1-dodecanethiol by immersing in 1 mM ethanolic solutions for 1 h. Probes were then immersed in 1 mM 3-maleimidobenzoic acid N-hydroxysuccinimide ester dissolved in 1:9 (v/v) mixture dimethyl sulfoxide and phosphate-buffered saline (PBS, Gibco, Fisher Scientific, Waltham, MA) for 30 min. To immobilize aptamers, probes were immersed in a 1 μM solution of thiolated DNA in PBS overnight. Probes were rinsed with deionized water and dried with N₂ before measurements. In vitro testing was performed by adding different concentrations of serotonin into aCSF in PDMS wells sealed to each device.

Ex vivo experiments

Brain tissue lacking serotonin was obtained from Tph2 knockout mice.⁵² Mice were exsanguinated by cardiac perfusion and brains were cleared of blood containing serotonin synthesized by peripheral Tph1. The brains were then shipped to UCLA on dry ice from the laboratory of Dr. Donald Kuhn (Wayne State University, Detroit, MI). Brain tissue was stored at -80 °C until use. Brain tissue collection procedures were approved by the Wayne State University Institutional Animal Care and Use Committee.

Brain tissue was homogenized in ice-cold aCSF (1:1 w/v) using a VirTis Virsonic600 ultrasonic cell disruptor (Gardiner, NY) with the microtip set at 4 and a 50% duty cycle for 30–40 pulses/s. Homogenized brain tissue solutions were mixed with agarose powder and aCSF to make 0.6% agarose gels. The mixture was microwaved just until the agarose powder dissolved (~1 min, melting point of ~65 °C). The solution was poured into a brain mold to mimic brain cortical morphology and immediately cooled. Flexible neuroprobes were implanted into the brain tissue hydrogel using the method described above. A Ag/AgCl gate electrode, constructed from a 0.010-inch diameter Ag wire freshly coated with AgCl (A-M Systems, Sequim, WA) by immersing in a bleach solution (Clorox, Oakland, California) for 5 min, was implanted into the brain tissue hydrogel next to the probe. A 2-μL aliquot of 50 μM serotonin was injected into the hydrogel (~0.2 cm away from the probe tip) for a final serotonin concentration of 100 nM (calculated based on 100-mL aCSF in the hydrogel).

All FET measurements were performed using a Keithley 4200A SCS semiconductor analyzer (Tektronix, Beaverton, OR). Source-drain current (I_DS) transfer curves were obtained by sweeping the gate voltage (V_GS) from 100 to 350 mV while maintaining the drain voltage (V_DS) at 10 mV, which took ~5 s for each scan. Calibrated responses were calculated at 300 mV to minimize device-to-device variation, as previously described.^24,53

Statistics

Data from FET calibrated responses are reported as means ± standard errors of the means and were analyzed using GraphPad Prism 7.04 (GraphPad Software Inc., San Diego, CA) via one-way analysis of variance followed by Tukey’s multiple comparisons post hoc tests with repeated measures.

RESULTS AND DISCUSSION

The process used to fabricate flexible In₂O₃ FET neuroprobes is described in detail in the Experimental Section and illustrated in Figure 1 and Figure S1. We leveraged our earlier high-throughput fabrication process for 150-μm and 50-μm Si neuroprobes using MEMS technologies.⁴ Polyimide (~7-μm) and In₂O₃ (~3-nm) films were fabricated sequentially on Si substrates using solution processing, followed by photolithography to define the Au and Ti source and drain electrodes. A layer of parylene can be added to the probe shafts to isulate source/drain interconnects, as demonstrated by us previously.⁴ Probe outlines were also defined photolithographically. We used In₂O₃ as the semiconductor due to its straightforward sol-gel fabrication. Others have shown that In₂O₃ is more stable than other metal oxides (e.g., indium-gallium-zinc oxide (IGZO)) in physiological electrolyte solutions.^23,54 The thickness of In₂O₃ on polyimide (3 nm) was determined by atomic force microscopy, as shown in our previous work.²⁸

Figure 1. Schematic of the flexible neuroprobe fabrication process.

A polyimide film was deposited on a Si substrate (1), followed by the deposition of In₂O₃ (2). Source and drain electrodes were patterned by photolithography followed by metal evaporation of Au and Ti (3). To define the outline of the neuroprobe, a second lithography process was used followed by oxygen-plasma etching of In₂O₃ (4) and polyimide (5). Each fabricated polyimide device was released from the Si substrate using tweezers (6) to obtain individual flexible neuroprobes (7).

This fabrication process is high throughput and scalable such that 150 probes were fabricated per 4” wafer (Figure 2a). Probes were delaminated from Si substrates using tweezers to produce free-standing devices. A delaminated probe is shown in Figure 2b, leaving an outline of the probe on the wafer. The freestanding polyimide probes are flexible and make conformal contact with the convoluted surface of an agarose hydrogel brain model (Figure 2c). Each probe was ~7 μm thick and ~150 μm wide with two FETs side-by-side on the tip (Figure 2d).

Figure 2. Serotonin sensing using aptamer-functionalized flexible neuroprobes.

(a) Photograph of 150 flexible probes fabricated on a 4” Si wafer. (b) Photograph of released probes on a Si substrate. As highlighted in the white box, the probe on the left has been partially released while the probe on the right has been completely released from the substrate. (c) Photograph of a flexible polyimide probe on a hydrogel matrix, which serves here as a brain mimic, showing the probe flexibility (tip width 150 μm). (d) Optical microscope image showing the tip of one neuroprobe with two field-effect transistors side-by-side. (e) Schematic of the measurement setup of a flexible neuroprobe, where a Ag/AgCl electrode was used to gate the FET through the buffer solution. (f) Schematic of the surface chemistry used to covalently functionalize aptamers onto the In₂O₃ surface. PTMS, trimethoxy(propyl)silane; APTMS, (3-aminopropyl)trimethoxysilane; MBS, 3-maleimidobenzoic acid N-hydroxysuccinimide ester. (g,h) Transfer and output curves, respectively, of a representative FET on a polyimide neuroprobe in phosphate buffered saline. PBS, phosphate buffered saline. (i) Sensing results for serotonin in artificial cerebrospinal fluid, where the highlighted box shows the physiological range of serotonin in the extracellular space. aCSF, artificial cerebrospinal fluid. Error bars are standard errors of the means of measurements from N=3 different probes.

The bending stiffness of these polyimide probes is 1.2 × 10⁻¹¹ N·m², which is seven orders of magnitude lower than that of the smallest optical fibers (230-μm-outer diameter polyethylene fibers with silica cores, 1 × 10⁻⁵ N·m², calculations shown in the Supporting Information).^30,55 The effective bending stiffness of the polyimide probes (bending stiffness per width) is ~8 × 10⁻⁸ N·m, which is less than that of commonly used soft elastomers, e.g., polydimethylsiloxane (2.2 × 10⁻⁷ N·m at 100 μm thickness).⁵⁶ Previous histological studies have shown that polyimide devices of similar thickness and bending stiffness induce minimal immune responses.^30,40 Note that the width of the probes and FETs can be scaled down in a straightforward manner to 50 μm using photolithography, as we have previously demonstrated using Si as the device substrate.⁴ Polyimide is thermally stable and compatible with most microfabrication processes up to 350 °C. Previously, we showed that the electronic and biosensing performance of In₂O₃ FETs on polyimide remains stable even after 100 bending cycles.²⁸ This straightforward and universal fabrication strategy opens opportunities to integrate polyimide neuroprobes with other neural recording and modulating modalities on a single flexible platform, e.g., electrophysiology and optical stimulation capabilities.

A Ag/AgCl electrode was used to gate thin-film In₂O₃ FETs on flexible neuroprobes through the buffer solution, where electrical double layers were used as the gate dielectric (Figure 2e). Source and drain electrodes were connected to the semiconductor analyzer via copper wiring to apply voltages to the interdigitated electrodes covered by a buffer solution.⁴ For serotonin sensing, the In₂O₃ surface was covalently functionalized with serotonin aptamers.²⁴ The detailed functionalization process is described in the Experimental Section and depicted in Figure 2f. Here, self-assembled monolayers on the Au source and drain electrodes combined with the electrical double layers in the buffer solution serve as capacitors for insulation.^24,27 Typical transistor performance was observed from the FETs in aCSF, with transfer and output characteristics shown in Figure 2g and 2h, respectively.

Serotonin was introduced to sensors to test responses. We used aCSF to mimic the ionic strength of the extracellular space in brain tissue. The FETs on flexible probes responded to serotonin in the pM to μM range covering the physiological range of serotonin in the brain extracellular space,^57–60 as highlighted in yellow in Figure 2i. The serotonin aptamer used here was previously demonstrated to be highly selective for serotonin vs. interfering molecules (i.e., L-5-hydroxytryptophan, 5-hydroxyindoleacetic acid, dopamine, L-tryptophan, uric acid, and ascorbic acid) when tested on FETs on Si.^4,24 We expect similar selectivity for the flexible polyimide neuroprobes since only the substrate was changed; we used the same aptamer to detect serotonin.

To monitor serotonin in vivo, neuroprobes need to be implanted into different areas of the brain where serotonin cell bodies and axons are found.⁵⁷ However, the softness of the polyimide probes leads to bunching during tissue penetration.^3,32 There have been several strategies developed to assist in the implantation of soft probes, including coating probes with a stiffening polymer, rapid freezing, or using liquid metal or temperature-adaptive Young’s moduli systems.^{42,47,50,61–63} However, current implantation strategies typically require re-engineering of device architectures or complicated assembly processes. Moreover, since biosensor surfaces are usually functionalized with receptors, e.g., antibodies or aptamers, before the implantation process, implantation strategies should not damage the functionalized layer.

Here, we developed a one-step shuttle-assisted process for flexible neuroprobe implantation using surface tension. As shown in Figure 3a, a Si shuttle was fabricated to have the same lateral dimensions as the flexible neuroprobes (150 μm). Fabrication was the same as described in our previous work⁴ and in the Experimental Section. The similar shapes of the Si shuttle device and the flexible polyimide neuroprobes minimized tissue damage from the shuttle device during the penetration process. The Si shuttle device was treated with O₂ plasma to make it hydrophilic before dipping into the buffer solution (i.e., aCSF). The thin layer of buffer coating the Si shuttle surface was used to affix a flexible neuroprobe via van der Waals forces. The neuroprobe and shuttle were aligned under an optical microscope using tweezers (Figure 3b).

After assembly, the shuttle+neuroprobe was implanted. To mimic brain tissue, we used a 0.6% agarose hydrogel with aCSF, which closely resembles several critical physical characteristics of brain tissue, including the penetration force.^47,64 After inserting the assembled device into the brain-mimic, the buffer layer between the flexible neuroprobe and shuttle device diffused into the hydrogel, reducing the surface tension holding the two parts together. The shuttle device was then retracted leaving the flexible device in place in the hydrogel (Figure 3b–f, Movie S1). During the removal of the shuttle device, we did not observe movement of the flexible probe, as shown in the movie (Movie S1). For future in vivo experiments, it would be necessary to investigate further the accuracy of positioning during the implantation processes.

Sensing in vivo entails detecting specific targets in complex biological matrices containing a wide range of neurotransmitters (nontargets), metabolites and other interferants, and biomacromolecules (biofouling). To investigate the sensing capabilities of the flexible neuroprobes in a biological matrix that more closely approximates an in vivo environment, we used brain tissue from Tph2-null mice.⁴ These mice do not express the rate-limiting serotonin synthetic enzyme in the central nervous system (CNS), and thus lack endogenous brain serotonin (Figure 4a).

Figure 4. Ex vivo sensing of serotonin in a brain phantom.

(a) Schematic illustration of brain phantom preparation. Brains from Tph2 null mice were removed and the tissue was homogenized in artificial cerebrospinal fluid (aCSF). The brain tissue homogenate was then mixed with 0.6% agarose gel in aCSF to make the brain-mimic for flexible probe implantation. (b,c) Photographs of an implanted flexible probe in the brain phantom, where a Ag/AgCl electrode was also implanted to apply a gate voltage. The source and drain electrodes were wired, sealed with epoxy, and connected to external measurement instrumentation. (d) Transfer characteristics of a representative field-effect transistor on a flexible probe after implantation in the brain phantom. (e) Output characteristics of a field-effect transistor on a flexible probe after implantation in the brain phantom. (f) Schematic illustration of serotonin injection using a syringe placed next to an implanted probe. (g) Five consecutive overlapping transfer sweeps of an implanted neuroprobe in the tissue-hydrogel brain-mimic before serotonin injection. (h) Calibrated responses of a representative neuroprobe after exposure to serotonin diffusing to the probe over time (100 nM final concentration). Error bars are standard errors of the means for N=5 sweeps. ***P<0.005 for 1 min vs. 5 and 10 min. (i) Calibrated responses from a different neuroprobe after exposure to 100 nM serotonin over time. Error bars are standard errors for N=5 sweeps. *P<0.05 for 1 min vs. 7 and 10 min.

Brain tissue homogenates from Tph2-null mice were incorporated into 0.6% agarose hydrogel in aCSF. As such, we produced an ex vivo model that mimicked the ionic strength of the extracellular space and approximated the chemical and physical properties of the in vivo brain environment. Using brain tissue that lacks serotonin prevented swamping the FET sensors with high concentrations of endogenous intracellular stores of serotonin liberated during tissue homogenization. In vivo, FET sensors would not encounter vesicular serotonin except briefly during implantation.

This type of brain phantom mainly mimics the physical characteristics of the brain (e.g., Young’s modulus and penetration force) and ionic strength of the extracellular space in the brain.^47,64 The brain tissue in the agarose gel introduces biomolecules found in the in vivo environment. Nonetheless, the preparation process may affect the molecular composition originally present in the brain. Previously, we tested similar biosensors on Si substrates in fresh brain tissue containing endogenous small molecules and macromolecules. In both cases, aptamer-FET sensors showed high target sensitivity and low biofouling.^4,24

Figure 4b and 4c, depict implantation of a representative flexible neuroprobe into a brain mimic using a Si shuttle, as described above. A Ag/AgCl electrode was then implanted to apply gate voltages through the hydrogel matrix (Figure 4b). We characterized the transistor performance of the flexible neuroprobes in this tissue-hydrogel phantom. As shown in Figure 4d and 4e, an implanted neuroprobe showed typical transfer and output characteristics with gate voltage and source-drain voltage modulation.

To simulate neurotransmitter release and diffusion in the brain, we used a syringe to inject 50 μM serotonin into the hydrogel. The recording sites were approximately 2 mm away from the implanted neuroprobes. The final concentration (100 nM) was selected as a physiologically relevant serotonin concentration.^57–60 Data were collected before the introduction of serotonin and at different time intervals after its introduction. As shown in Figure 4g, five consecutive transfer curves overlapped with each other in the tissue-hydrogel brain-mimic, illustrating that the flexible neuroprobes were stable in the matrix without detectable signal drift prior to injection.

We observed significant increases in the responses of two different neuroprobes as early as 5 min after serotonin addition (see Table S1 for statistics). These data demonstrate that the flexible aptamer-FET neuroprobes detected diffusion-related changes in serotonin levels. We previously found that the response times of our aptamer-FET sensors are on the order of seconds;²⁶ the 5-min delay in response time here is due to target diffusion over ~2 mm in the hydrogel matrix from the introduction location to the recording site, which is similar to what we observed in a previous study using Si neuroprobes.⁴ These findings illustrate that our flexible neuroprobes can be implanted into a solid matrix that mimics the rigidity and chemical composition of brain tissue to monitor physiologically relevant serotonin release.

CONCLUSIONS

We developed flexible and implantable neurochemical probes based on aptamer FET biosensors. A high-throughput fabrication strategy was developed to fabricate In₂O₃ thin-film (~3 nm) field-effect transistors on polyimide (7 μM) with up to 150 flexible neuroprobes on a 4-inch Si substrate. Flexible aptamer-FET neuroprobes detected serotonin down to femtomolar concentrations in artificial cerebrospinal fluid. To facilitate the implantation of these flexible devices, we developed a straightforward process where a rigid Si carrier device was used to assist in tissue penetration. To test the functionality of the flexible neuroprobes for future in vivo studies, we developed a brain phantom as an ex vivo model. We used agarose gel to mimic critical physical characteristics of the brain and aCSF to mimic the ionic strength of the brain’s extracellular space. Moreover, we combined agarose gel with mouse brain tissue containing interferants and biomacromolecules present in the brain. The brain tissue phantom served as a model to evaluate sensor performance towards in vivo applications for neurochemical sensing. We showed that the flexible neuroprobes were stable while implanted in the hydrogel matrices with reproducible transistor performance and without short-term biofouling. We were able to detect serotonin injected into the brain tissue phantom suggesting the potential to monitor changes in serotonin levels in vivo.

While soft neuroprobes remain an important goal for tissue-like implantable bioelectronics, a longstanding question remains as to how soft and flexible is soft and flexible enough such that immunological responses are minimized? This issue is a grand challenge to the bioelectronics community that needs to be addressed and requires further study. Future histological studies as a function of implantation duration involving flexible vs. stiff probes will provide comparative information on immunological responses. Another key question involves the impact of device dimensions. Smaller implanted devices appear to reduce immune responses.^31,32,34 However, understanding how immunological responses change systematically in response to miniaturization is still unclear. Along these lines, we have previously fabricated Si neuroprobes with 50-μm widths and thicknesses to compare with 150-μm neuroprobes in terms of tissue response. Scaling down the flexible neuroprobes developed in this work can be achieved using a similarly straightforward approach. Further histological studies comparing our devices of different sizes and stiffness are ongoing.

In addition to FETs, aptamers can be coupled with redox reporters for electrochemical sensing. For example, electrochemical aptamer-based (EAB) sensors have been used to track plasma drug concentrations in real-time.^65,66 Recently, a dopamine aptamer was coupled with carbon-fiber microelectrodes noncovalently or using an electrochemical conjugation strategy for dopamine sensing in vivo using cyclic voltammetry.^67,68 We envision that next-generation neuroprobes will need to be soft, multi-modal (i.e., sense multiple neurotransmitters and electrophysiological signals, produce optical or electrical stimulation, and drug delivery all on the same device), and have long-term recording capabilities in vivo. Soft bioelectronics with the capability to detect multiple neurotransmitters simultaneously in real-time are needed to investigate chemical information processing in the brain, particularly in the context of chronic neural recordings during behavior.²¹