Neural Recording and Modulation Technologies

Abstract

Within the mammalian nervous system, billions of neurons connected by quadrillions of synapses exchange electrical, chemical and mechanical signals. Disruptions to this network manifest as neurological or psychiatric conditions. Despite decades of neuroscience research, our ability to treat or even to understand these conditions is limited by the tools capable of probing the signalling complexity of the nervous system. Although orders of magnitude smaller and computationally faster than neurons, conventional substrate-bound electronics do not address the chemical and mechanical properties of neural tissue. This mismatch results in a foreign-body response and the encapsulation of devices by glial scars, suggesting that the design of an interface between the nervous system and a synthetic sensor requires additional materials innovation. Advances in genetic tools for manipulating neural activity have fuelled the demand for devices capable of simultaneous recording and controlling individual neurons at unprecedented scales. Recently, flexible organic electronics and bio- and nanomaterials have been developed for multifunctional and minimally invasive probes for long-term interaction with the nervous system. In this Review, we discuss the design lessons from the quarter-century-old field of neural engineering, highlight recent materials-driven progress in neural probes, and look at emergent directions inspired by the principles of neural transduction.

(Web summary)

Understanding the dynamics and architecture of the nervous system requires tools for recording and modulating billions of neurons. This article reviews the history of neural engineering and the materials innovation at the interface between neural tissue and synthetic sensors.

Introduction

Understanding the information transfer and processing within the mammalian nervous system is one of the most urgent challenges faced by the biomedical community. Neurological, neurodegenerative, psychiatric, and neuromuscular conditions, for example, Parkinson’s disease, Alzheimer’s disease, major depressive disorder, and multiple sclerosis, respectively, affect an ever-increasing percentage of our aging population (Fig. 1a–b) [1–3], and traumatic injuries to the nervous system contribute 2.4 million patients each year to the disability burden in the US alone [4]. Furthermore, inflammatory conditions, such as hypertension and arthritis, have been linked to changes in peripheral nerve activity [5]. Although basic neuroscience research of the past century has greatly advanced our understanding of neuronal function, the ability to record and manipulate the dynamics of the nervous system remains insufficient to treat these psychiatric and neurological disorders or to restore function following nerve injury.

Figure 1│. Challenges of designing for the nervous system.

a│ The worldwide population of diagnosed Parkinson’s disease cases in 2005 (4.1 million) and the corresponding projection to 2030 (8.7 million) [1]. b│ Combined annual cost of multiple sclerosis [2], spinal cord injuries [214], Parkinson’s disease [215], and depression [3] in the US displayed in US$ billions. c│ Dimensions of the brain and spinal cord. d│ Brain tissue encompassing neurons, glia and blood vessels. e│ Circuit diagram for direct and indirect motor pathways in the brain [216]. f│ Physiological facts pertaining to brain function. g│ Spinal-cord cross-section. h│ Circuit diagram of motor pathway in the spinal cord [217, 218]. i│ Cross-sectional diagram of a peripheral nerve. Physiological facts pertaining to the spinal-cord and peripheral-nerve function [10]. SNc, substantial nigra pars compacta; GPi, internal external globus palidus; GPe, internal globus palidus; STN, subthalamic nucleus; D1 and D2, dopamine receptors.

Within the 1.3 litre volume of the human brain, billions of neurons, divided into thousands of genetically and structurally defined subtypes, communicate with each other through quadrillions of synapses (Fig. 1d–g) [6]. Notably, neurons represent only ~50% of the cells in the brain. The other half are electrically inactive glia that include astrocytes, oligodendrocytes and microglia,[7] the roles of which in brain function remains an active area of research [8, 9]. Similarly, the ~45-cm long, 1.5-cm thick spinal cord encompasses a variety of motor and interneurons, Schwann cells, and neuronal processes connecting the brain to the >150,000 km web of the peripheral nervous system [10] (FIG. 1h–j).

Neural activity is marked by millisecond-long 80–100 mV spikes in cell-membrane voltage called action potentials [6]. Aided by voltage-sensitive ion channels, action potentials propagate across neural membranes to release neurotransmitters into the synaptic cleft. Neurotransmitter uptake by the dendrites of synaptic partners results in signal transduction. In addition to chemical and electrical signals, neurons can respond to physical stimuli, such as pH [11], temperature [12, 13], pressure and tension [14] mediated by a variety of ion channels. Understanding neural function hence requires tools that can communicate with neurons across the diversity of their signaling capacities.

Driven by Moore’s law over the past 25 years, miniaturization of electronic devices down to the nanoscale has delivered unprecedented computational power. Although neuroprosthetic research has undoubtedly benefitted from these advances, additional design parameters need to be included for effective long-term operation and clinical translation of neural interfaces. The complexity of neural tissue requires further materials innovation beyond microfabricated semiconductor circuits. Consideration of the brain’s mechanical, electrical and chemical properties, and the diversity of neural signaling pathways may guide materials design to establish intimate long-term synthetic interfaces capable of faithful recording and stimulation of neural activity.

History of neural engineering

From neuroscience to neural engineering.

Motivation to answer fundamental questions in neuroscience has consistently inspired the invention of new technologies (FIG. 2). In the 1950s, wires were introduced to measure electrical activity in the nervous system [15], followed by patch-clamp electrophysiology in the 1970s [16] to elucidate synaptic transmission. In the 1980s to 1990s, silicon multielectrode arrays [17, 18], and stereotrodes and tetrodes [19, 20] have been developed to investigate communication between large groups of neurons within distinct neural circuits.

Figure 2│. Historical view of neural interface research.

Progress in genetic tools for imagining and manipulating neural circuits paralleled engineering trends in neural probe design. The development of neural interfaces, driven by the needs of brain–machine interface research, has been confounded by the failure of neural probes in chronic long-term experiments. Recently, there has been a materials-driven pursuit to reduce the foreign-body response, push the resolution of neural interfaces to sub-cellular dimensions and integrate the capabilities required for deployment of genetic tools.

These probe innovations relied on tailoring device geometry and electrical impedance to a particular neuroscience application. For extracellular recordings, the impedance is determined by the capacitive characteristics of the interface between the electrode and the cerebro-spinal fluid, which is a function of the electrode material and the tip area. High-impedance electrodes (5–10 MΩ) are typically used to record action potentials with high signal-to-noise ratios (SNR > 20) from neurons within a 10-μm radius [21]. Insulated steel, tungsten, gold and platinum microwires with dimensions of approximately 10–100 μm and impedances of less than 1 MΩ are commonly employed for extracellular recordings from moderate numbers of neurons [22]. Stereotrodes and tetrodes consist of 2 or 4 polymer-insulated 12-μm nickel–chromium (NiCr) microwires that are electrochemically coated with gold to lower the impedance from about 1–3 MΩ to 100–500 kΩ and to improve biochemical stability [19, 20]. The close proximity of multiple microwires enables deconvolution of overlapping population signals to identify action-potential shapes corresponding to specific cells via principal component analysis [23]. Neurotrophic electrodes integrate a microwire within a pipette carrying neurotrophic factors that stimulate neuronal ingrowth for high-fidelity recordings [24].

Microfabricated neural probes.

By the 1980s, advances in semiconductor microfabrication led to the development of silicon-based multielectrode arrays known as Utah arrays [17] and Michigan probes [18]. Utah arrays, which contain up to 128 sharp, metal-tipped electrodes with a pitch of 200–400 μm, are produced by a combination of micromachining and lithography from thick silicon wafers. Implanted Utah arrays ‘float’ on the surface of the brain and are connected to skull-mounted interface boards by flexible cables. Due to their relatively large area and electrode count, these devices have become crucial components in studies of cortical circuits in non-human primates and are the only neural probes approved by the US Food and Drug Administration (FDA) for chronic use in human patients [25]. As Utah arrays only permit recordings from the topmost 1–3 mm of the cortex with large footprints limiting their utility in small animal models, alternative strategies have been designed for monitoring neural activity at different spatial scales. Michigan probes are composed of lithographically defined patterns of metallic electrodes on thin silicon substrates for depth-defined recordings in subcortical structures [26]. Compatibility of these designs with modern CMOS (complementary metal-oxide semiconductor) processing enables straightforward integration of data acquisition and signal amplification capabilities while allowing for facile back-end connectorization [27, 28]. Both Utah arrays and Michigan probes employ the polymers polyimide and parylene C for insulation. Powered by advances in micro-electromechanical systems (MEMS) fabrication, the past three decades of neural engineering have delivered a multitude of sophisticated probes inspired by Utah and Michigan designs with increased resolution, reduced dimensions and expanded capabilities [29].

The emergence of the fields of neuroprosthetics and brain–machine interfaces, which seek to restore voluntary motor control to paralyzed patients, necessitated the development of high-resolution recording probes. By the late 1990s, decoding algorithms were developed to control robotic joints using rodent and primate brain activity in real time [30, 31]. The surge of scientific and clinical enthusiasm around brain–machine interfaces propelled this technology into clinical trials in 2004, producing encouraging demonstrations of tetraplegic patients controlling computer cursors [25] and robotic arms [32] with implanted electrode arrays. However, maintaining high SNR recordings from thousands of neurons proved challenging, and biocompatibility and reliability issues often resulted in probe failure only a few weeks after implantation [29, 33, 34].

Materials mismatch

Failure modes of neural probes can be broadly classified into those of engineering and biological origin [33–35] (FIG. 3). The former includes mechanical failure of interconnects between interface boards and implanted probes, degradation of electrical and environmental insulation, and delamination of the different layers constituting the device [29, 35]. To avoid premature failure, microfabricated devices have to be safeguarded from exposure to warm, aqueous and saline environments, and oxidative stress associated with the foreign-body response. Consequently, strategies for packaging and encapsulation have emerged as key aspects of neural-probe design. These include hermetic titanium casings typical of clinical devices, and polyimide, polyurethane, poly(dimethyl sulfoxide) (PDMS), SU-8 and parylene C coatings, which are common in research-grade tools [29].

Figure 3│. Mechanisms of neural-probe failure.

Failure modes of neural probes manifested as a loss of neural recording capability can be classified into those relating to device design [29, 35] (above the arrow) and foreign-body response [34, 36] (below the arrow). Design failure mechanisms include mechanical failure of interconnects [35], degradation and cracking of the insulation [219], electrode corrosion [220, 221] and delamination of probe layers [222]. Biological failure mechanisms include initial tissue damage during insertion [34, 37]; breach of the blood–brain barrier [41]; elastic mismatch and tissue micromotion [35, 38, 39]; disruption of glial networks [40]; formation of a glial scar; and neuronal death associated with the abovementioned factors, as well as with materials neurotoxicity [42] and chemical mismatch [34].

Foreign-body response.

Neuronal death and formation of glial scars (~100-μm thick) around implanted probes can lead to loss of recorded signals [34, 36]. A subject of several detailed reviews, the glial scar is composed of reactive astrocytes and activated microglia that form dense and electrically inactive tissue along the device surface [34, 36]. A number of factors have been suggested to contribute to this inflammatory response, including: initial tissue damage during device insertion [34, 37]; elastic mismatch between the neural probes and the neural tissue in the context of relative micromotion [35, 38, 39]; disruption of glial networks [40]; chronic breach of the blood-brain barrier [41]; materials neurotoxicity [42]; and chemical mismatch between the implant surface and the cell membranes and extracellular matrix [34].

Although the initial damage due to device insertion may have short-term effects on the recording performance, long-term probe performance is of particular concern to the brain–machine interface community. Introduction of a foreign object with dimensions exceeding 20 μm has been hypothesized to disrupt local communication between glia, which triggers the release of pro-inflammatory cytokines and recruitment of additional activated microglia and reactive astrocytes [34, 40]. The effects of size are challenging to decouple from those associated with elastic mismatch, as smaller device dimensions will also decrease bending stiffness. The elastic mismatch between the neural tissue (kilo- to megapascals) and the implanted probes (1–100 GPa), typically tethered to bone, causes repeated injury to the tissue every time the brain or nerve is displaced relative to the device [35, 38, 39]. This continuous impact has been linked to chronic breach of the blood–brain barrier [41], which activates glia and astrocytes to form a scar around the hemorrhage. In addition to the probe’s mechanical properties, surface chemistry and topography may also influence the extent of inflammation [43].

Three decades of neural engineering studies have focused on increasing the reliability and biocompatibility of microfabricated probes through continued miniaturization and addition of coatings to reduce inflammation [44] and modulus mismatch between the probes and neural tissue [45]. However, the innate materials properties of the implants can limit their biocompatibility, which motivated the recent emergence of materials-driven strategies to match the devices’ mechanics and geometry to that of neural tissue [46].

Strategies to improve compatibility

Making stiff electronics soft.

Compliant coatings can reduce the apparent elastic mismatch between stiff silicon- or metal-based probes and neural tissue [47]. Hydrogels based on synthetic, biological, or a hybrid of these materials can decrease the surface hardness of the probes, lower electrode impedance and diminish glial scarring. For example, poly(vinyl alcohol) matrix doped with the conductive polymer poly(3,4-ethylenedioxythiophene) (PEDOT) and copolymers of collagen and the organic semiconductor polypyrrole lowered the impedance of lithographically defined metallic electrodes on silicon or glass substrates while improving cell adhesion [48]. Incorporation of neuronal adhesion molecules (for example, L1) [45] or even live cells [49] into the coating may additionally promote local axonal growth.

Polymers have been used as alternative substrates to support microfabricated probes [29, 46] (Fig. 4a). Metal electrodes and interconnects can be lithographically defined or microcontact printed onto polyimide and parylene C substrates. The same polymers are also used for insulation and packaging. Using microcontact printing, integrated multiplexed circuits were integrated with flexible polyimide substrates to map cortical dynamics of epileptic seizures with 500-μm resolution and millisecond precision [50]. Conformal chemical vapor deposition of parylene C onto curved substrates was applied to create complex sheath and hemispherical cavities with internal metal contacts. These flexible neurotrophic electrode cavities were filled with growth factors (for example nerve growth factor and neurotrophin-3) to promote nerve ingrowth and to establish better interfacial contact for improved chronic recording performance [51, 52]. PDMS is also compatible with soft lithography and has been incorporated into devices intended to conform to the dynamic and flexible surfaces of the spinal cord [53] and peripheral nerves [54–56]. Regenerative sieves [56] and nerve-cuff [57] electrodes have also taken advantage of PDMS substrates, which can be rolled into cylindrical shapes tailored to the nerve dimensions to support nerve ingrowth.

Figure 4│. Examples of approaches intended to overcome foreign-body response and increase the resolution of neural interfaces.

a│ Use of flexible substrates and modulus-matching coatings to reduce elastic mismatch between the neural probes and neural tissues. b│ Contact-printed serpentine, wavy, and fractal interconnects. c│ Thermally drawn fibers with micron-scale electrodes and carbon-fiber microelectrodes. d│ Conductive polymers used as flexible alternatives to crystalline semiconductors and metals. e,f│ Conductive composites of flexible and stretchable polymers and carbon and metal micro- and nanoparticles.

To overcome the challenges associated with buckling of flexible probes during insertion into the tissue, polymers that undergo a dramatic reduction in modulus from giga- to megapascal in response to temperature (that is, thermoplastics) [58] and hydration (for example, poly(vinyl acetate)–cellulose composites inspired by sea-cucumber dermis) [59] have been explored as stimuli-responsive substrates for neural interfaces.

The material’s bending stiffness (Eq. 1) rather than the Young’s modulus determines the tissue interactions for devices tethered to the skull or vertebrae:

$$\frac{F}{d}=\frac{48EI}{L^3}$$ Equation (1)

The bending stiffness is defined as a force needed to achieve a certain deflection. Here F is the force, d is the deflection, E is the Young’s modulus, I is the moment of inertia, and L is the length of the device. For a probe with a rectangular cross-section, inputting the expression for the moment of inertia yields:

$$\frac{F}{d}=\frac{4Ew{t}^3}{L^3}$$ Equation (2)

where w and t are the width and thickness of the device, respectively.

Because the stiffness scales cubically with the material’s thickness (Eq. 2), reducing the dimensions of a high-modulus device can dramatically improve its flexibility [60, 61]. This scaling rule has been applied to fabricate compliant microscale wavy surfaces, meshes, serpentines and fibres composed of materials with Young’s moduli in the gigapascal range [50, 62–65] (Fig. 4b,c). Contact printed meshes and fractal patterns of silicon and metal ribbons with thicknesses of a few micrometres can act as flexible interconnects between microscale islands bearing recording electrodes as well as other sensors and processing electronics [50, 62]. Out-of-plane buckling of serpentine electrodes deposited onto pre-strained PDMS substrates accommodates stretching by up to ~100 %, which significantly exceeds the strains experienced by peripheral nerves [61] (Fig. 4b). Fibre-drawing can also be applied to reduce the dimensions of metallic electrodes embedded in polymer matrices down to 1–5 μm [65]. Fiber-probes with dimensions of human hair (~80 μm) can integrate 7–10 electrodes with impedance values of ~ 800 kΩ. These probes can record single-unit action potentials and produce negligible impact on the surrounding neural tissue (Fig. 4c). Similarly, carbon fibre can be manually separated [66, 67] or electrospun from carbon nanotube solution [68] to fabricate electrodes 5–10 μm in diameter with impedances of 100 kΩ – 4 MΩ (Fig. 4c). The miniature dimensions of the carbon electrodes also permit isolation of individual action potentials and improved tissue compatibility during chronic implantation [66–68].

Organic and hybrid electronics at neural interface.

Owing to their low moduli, organic conductors are promising platforms for devices intended to interface with biological systems [69] (Fig. 4d). Aromatic small molecules and conjugated polymers have been integrated into neural probes deployed in vitro [70]. However, with the exception of PEDOT, the transition of conjugated organics in vivo has been impeded by their poor environmental stability [71]. PEDOT mixed with poly(styrene sulfonate) (PEDOT:PSS) is a heavily-doped p-type semiconductor frequently used as a hole-injection and extraction layer in organic light-emitting devices (OLEDs) and photovoltaics, and as a channel in organic field-effect transistors (OFETs) or organic electrochemical transistors [72]. PEDOT:PSS coatings on standard metallic electrodes are known to increase the SNR and neuronal unit yield [73, 74]. Most recently, flexible PEDOT:PSS organic electrochemical transistors fabricated on 2-μm-thick parylene C substrates were shown to record local field potentials as well as isolated action potentials from the cortical surface [75, 76].

Blends of polymers and conductive micro- and nanoparticles (carbon [77–79], metals [80, 81], conjugated organics [82]) allow for independent control of both the electrode conductivity and bending stiffness [83]. A polymer matrix can be selected for its low modulus, and the conductivity, concentration and geometry of the micro- or nanoscale dopants determine the blend’s electronic properties [84] (Fig. 4e,f). Although the majority of nanocomposites are applied as coatings over metallic electrodes, fibre-drawn polyethylene electrodes doped with carbon back were used to record neural activity in the spinal cord of a mouse despite having a high impedance (> 1 MΩ) [78] (Fig. 4e). Doping of PDMS with carbon below the percolation threshold is a promising strategy in designing microstructured piezo-resistive touch sensors [79], and carbon-doped PDMS may eventually be integrated into neural recording probes. More recently, composites of silver nanowires and styrene-butadiene-styrene elastomer patterned into serpentine meshes with elastic moduli of 3–45 kPa (depending on the serpentine curvature) and conductivity ~10⁴ S m⁻¹ were employed to modulate cardiac firing in vivo [80], illustrating the future promise of this material platform for functional interfaces with other electrically active cells (Fig. 4f). As interest in flexible (opto)electronics continues to drive research into conductive polymer composites, a variety of designs initially intended for photovoltaics, OFETs and OLEDs [83, 85] may likely be integrated into future neural interfaces.

Bi-directional neural interfaces

Electrical stimulation.

Although recording electrodes provide insight into the electrophysiological activity within the nervous system, they do not allow for control over its dynamics. Electrical stimulation is clinically approved to treat Parkinson’s disease [86], neuropathic pain [87] and is a candidate for alleviating symptoms of major depression [88]. Chemically stable and biocompatible platinum and platinum–iridium electrodes with low impedance (<10 kΩ) are commonly used for electrical stimulation [89]. The relatively modest charge injection capacity (CIC ≈ 0.05–0.15 mC cm⁻²) of these electrodes can be improved by coatings of iridium oxide to (CIC ≈ 5 mC cm⁻²) or titanium nitride (CIC ≈ 0.9 mC cm⁻²). In addition to reducing electrochemical impedance, PEDOT, polypyrrole and their blends with conductive nanomaterials can improve charge injection (CIC ≈ 1–15 mC cm⁻²) of the stimulating electrodes [89]. The CIC can be further improved by increasing the electrode surface area through texturing. Akin to its deleterious effects on recording electrode performance, glial scarring diminishes the CIC of stimulating electrodes. Hence, strategies aimed to reduce the inflammatory response (for example, flexible substrates and modulus-matching coatings) may simultaneously improve stimulation fidelity [29, 46, 47, 55].

Engineering with optogenetics.

Owing to interference with electrophysiological recording and lack of cellular specificity, the basic mechanisms and therapeutic outcomes of electrical stimulation remain poorly understood [90]. Optogenetics can simultaneously perturb and probe the activity of genetically defined neuronal populations because the optical stimuli are decoupled from electrical recordings [91]. This method relies on genetic introduction of light-sensitive microbial ion channels and pumps, called opsins, into mammalian neurons [92] and other electrically active cells [93–95]. Opsins are membrane proteins harboring the chromophore retinal, which changes conformation in response to visible light to transport ions [96, 97]. These proteins can be used for both excitation and inhibition of neural activity. Excitatory opsins (for example, channelrhodopsin 2 (ChR2)) are cation channels that mediate membrane depolarization to fire action potentials in response to light. Inhibitory opsins are proton or chloride pumps (for example, halorhodopsin, NpHR, and archeorhodopsin, ArCh) that move ions against their concentration gradients to silence neurons through optically driven hyperpolarization [98]. Over the past decade, a palette of opsins with varied absorption spectra and photocycle kinetics have been discovered and engineered [99, 100].

Because visible light needed for opsin activation is scattered and absorbed by neural tissue, optogenetic neuromodulation relies on implanted devices to deliver optical stimuli. To take advantage of simultaneous electrophysiological recording and optogenetic control, early efforts outfitted mature neural recording technologies, such as Utah arrays (Fig. 5a), Michigan probes (Fig. 5b), tetrodes, and microwires (Fig. 5c), with commercially available silica fibres [91, 101–103]. Multimaterial semiconductor processing methods have recently been applied to monolithic integration of LEDs within silicon probes [104, 105]. For example, indium–gallium–nitride (InGaN) quantum wells were epitaxially grown on (111) silicon substrates to create arrays of up to twelve 10 μm × 15 μm blue-emitting micro-LEDs integrated with 32 metallic electrodes (Fig. 5d). Sharing geometry with Michigan probes, these devices enabled depth-defined recording and ChR2-mediated optogenetic stimulation of pyramidal neurons in the rat hippocampus [105]. An innovative alternative to silicon probes is a transparent electrode array composed of n-doped (resistivity, ρ ≈ 0.15 Ω·cm) wide-bandgap semiconductor zinc oxide (ZnO), which permits simultaneous electrical neural recording and delivery of optical stimuli with the same material (Fig. 5e)[106]. Another approach to fabricate transparent probes relies on integration of graphene electrodes and interconnects onto Parylene C substrates. The resulting transparent probes permit optical stimulation through the substrates during neural recording (Fig. 5f) [107].

Figure 5│. Probes for bi-directional communication with neural circuits.

a–c│ Integration of optical fibers into Utah arrays [101], Michigan probes [102] and tetrodes [91], respectively, intended for optogenetics research. d│ Gallium-nitride-based micro-LEDs monolithically integrated onto silicon substrates within Michigan-style probes [105]. e│ Transparent multi-electrode arrays composed of conductive ZnO [106]. f│ Transparent and flexible arrays of graphene surface electrodes [107]. g│ MEMS-processed tri-functional polymer probe integrating an SU-8 waveguide, metal electrodes and a microfluidic channel on a polyimide substrate [108]. h│ Multifunctional microcontact printed probe containing an electrode, micro LEDs, a photodetector and a thermistor [109]. i│ Multifunctional all-polymer fiber-probe integrating a waveguide, carbon composite electrodes and microfluidic channels [65]. LED, light-emitting device.

In addition to bi-directional interface capabilities, biocompatibility should be factored into the design of integrated optoelectronic probes based on semiconductors. As crystalline group III–V and II–VI semiconductors exhibit elastic moduli similar to those of silicon and metals (tens to hundreds of gigapascals), the tools based on these materials are likely to elicit inflammatory responses similar to those observed for mature electrode technologies. Consequently, soft- and hybrid material designs have been explored. Using MEMS processing methods, polymer waveguides composed of SU-8 photoresist were integrated with metal-electrode arrays on polyimide substrates. Although SU-8 can be cytotoxic [42] with limited transmission in the visible spectrum, these initial efforts laid the groundwork for multifunctional and flexible optoelectronic probes (Fig. 5g)[108]. Contact printing of metallic interconnects, electrodes, blue-emitting GaN micro-LEDs, microscale thermistors and silicon photodiodes onto PDMS produced highly compliant multifunctional probes suitable for optogenetic control of mouse behavior as well as neural recording with minimal damage to surrounding tissue (Fig. 5h)[109].

Alternatively, polycarbonate and cyclic olefin copolymer waveguides and carbon-black-polyethylene composite electrodes can be integrated within all-polymer probes via high-throughput fibre drawing. The flexible fibre probes exhibited bending stiffness of 10 N m⁻¹ and enabled simultaneous recording and optical neuromodulation in the mouse brain and spinal cord without triggering significant inflammatory reaction (Fig. 5i) [65, 78].

Delivery of chemical stimuli.

To perturb neural activity via traditional pharmacological [110] or chemogenetic approaches [111], microfluidic channels can be integrated into neural probes to deliver chemical and biological agents into the nervous system. Optogenetic stimulation augmented by delivery of a neuromodulator (for example, a synaptic blocker) may reduce the ambiguity in optical cell-type identification experiments, where opsin expression is used to mark a particular neuronal population [103, 104]. Furthermore, microfluidic channels permit injection of viral vectors needed for genetic experiments when transgenic animal models are not available [112]. Glial scar formation within the probe vicinity may block the microfluidic channels or change the diffusivity of the compounds of interest. Although integration of microfluidic features into flexible MEMS-processed [51, 108], microcontact printed [112] and fibre-drawn neural probes [65] increases the size of the device, these probes are significantly smaller than devices assembled from bulky cannulas, silica optical fibers and metallic electrodes, and hence may reduce the foreign-body response (Fig. 5g–i).

Wireless transmission.

Backend connections scale proportionally with the addition of functional features integrated into the implants. The increased device complexity and weight can affect the behavior of the experimental subjects, especially smaller animal models. Wireless power and data transmission can eliminate the need for multiple tethers and headstages [113], but may result in significant heat dissipation within the implanted hardware owing to high data rates and power draws. Furthermore, the physical dimensions of radiofrequency antennas may not be compatible with implantation into small rodents unless designed for a limited transmission range.

Recent implants powered by radiofrequency antennas enable wireless operation of semiconductor micro-LEDs to optically drive behaviour in mice expressing ChR2 in the brain, spinal cord and peripheral nerves [114, 115]. Both direct transmission [115] and evanescent coupling through the animal body [114] has been explored as a means to deliver the necessary power. The former relied on microcontact serpentine antennas with areas of ~3×3 mm², and the latter required a specialized radiofrequency cavity positioned directly underneath a behavioural enclosure. For applications requiring additional functionality, such as multiple optical sources or microfluidic drug delivery, larger antennas mounted on the skull were used [112].

Independently operating wireless microcircuits tens to hundreds of micrometres in size, called ‘neural dust’ motes, have been proposed as an alternative technology to traditional probes [116]. Initial millimeter-scale prototypes have been fabricated and are capable of recording and electrically evoking activity in peripheral nerves [117]. However, further miniaturization poses challenges to device placement and stability within the nervous system.

From microprobes to nanotransducers

Micro- and nanostructured electrodes permit intracellular recordings and modulation of sub-threshold events at finer temporal resolution and longer duration than conventional patch-clamp electrophysiology. Micrometre-sized gold mushrooms, functionalized with RGD peptides to promote cell adhesion, provided intracellular-like recordings of action potentials as well as subthreshold synaptic activity for days in vitro [118, 119] (Fig. 6a). However, translation of this technology in vivo is challenging because of its reliance on intimate contact between high impedance (~30 MΩ) mushroom electrodes and neurons. Monocrystalline gold nanowires ~100 nm in diameter also exhibit high impedances (~5 MΩ), but were found to be effective for acute recordings of single unit activity in the CA1 region of the mouse hippocampus [120]. Silicon-nanowire and gold-nanopillar electrodes were used for intracellular recordings following localized membrane electroporation, although intracellular access was lost within minutes owing to membrane resealing [121, 122]. To prolong intracellular access, hollow nanotube electrodes that form tight junctions with the cell membrane were synthesized [123]. Stealthy probes with amphiphilic coatings that mimic the chemical structure of cell membranes may also provide passive access to the cytosol through spontaneous fusion [124] (Fig. 6b).

Figure 6│. Micro- and nanoprobes for intracellular recordings.

a│ Endocytosed mushroom electrodes for high-SNR recordings [118, 119]. b│ Cell-penetrating nanopillar [122], nanowire [121], and nano-straw electrodes [123]. c│ Field-effect devices based on kinked semiconductor nanowires. [129–131]. SNR, signal-to-noise ratio.

In addition to capacitive voltage recordings, modulation of the gate voltage of semiconductor-nanowire FETs by controlling the local ion flux can also be used for in vitro extracellular neural recording (Fig. 6c) [125, 126]. The same study found that nanowire FET arrays are sensitive enough to capture the propagation of action potentials down individual axons [125]. To achieve high-resolution intracellular recordings in three dimensions, FET extension from planar substrates was enabled by the synthesis of kinked nanowires [127, 128]. Although electrode miniaturization and morphological control are promising strategies for long-term intracellular recording, the challenges in device connectorization and implantation currently limit these technologies primarily to in vitro models.

To harness the resolution of nanoscale electronics in vivo, macroporous meshes have been developed that have mechanical properties similar to neural tissue [129] [130]. These metal meshes supported by SU-8 substrates served to connect nanowire FETs and platinum nanoelectrodes to backend electronics and could be injected into the brain. These devices provided chronic single-unit recording while also acting as ‘neurophilic’ scaffolds promoting neuronal migration [129–131].

Despite the miniaturization and increased flexibility of neural probes, a device that eliminates connectorized hardware and tissue damage while communicating with the nervous system over arbitrarily long timescales remains to be demonstrated. By contrast, contactless stimuli coupled to nanoparticle transducers may facilitate localized readout or perturbation of neural activity at single-protein resolution [132]. Advances in synthetic methods have delivered nanomaterials with tunable size, shape and composition. Furthermore these nanomaterials can be decorated with functional groups for cell-specific targeting and improved biocompatibility [133, 134].

Contactless communication

Optical, acoustic and magnetic stimuli offer application-specific advantages for neural stimulation and recording [135] (Fig. 7). While optical stimulation and imaging approaches, such as infrared light [136], genetically encoded opsins [137, 138], and calcium and voltage indicators [139], offer superior spatial and temporal resolution, they are limited to tissue penetration depths of less than 1 mm owing to absorbance and scattering. Even three-photon excitation with a 1675-nm laser can only penetrate to a depth of 1.5 mm into brain structures [140].

Figure 7│. Tissue penetration by optical, ultrasonic and magnetic signals.

Electromagnetic waves in the visible and (near) infrared optical spectrum afford superior spatial and temporal resolution but limited penetration depth (~1–1.5 mm) [137, 140]. Ultrasound can access deeper brain regions (>50 mm), with a spatial resolution that is inversely proportional to the wavelength, which in turn scales inversely with the penetration depth (in general for ultrasound spatial resolution is > 1 mm³, temporal precision >10 ms) [143]. Alternating magnetic fields (AMFs) with low frequencies (<1 kHz) and high amplitudes (0.1–2 T) inductively couple to the upper 1–10 mm of tissue [147]. AMFs with amplitudes of ~1–100 mT and frequencies in the low radiofrequency range (0.1–1 MHz) travel through tissue unaffected [146]. Temporal precision of neural activity is dependent on the chosen magnetic scheme.

Acoustic waves can extend the penetration depth by one to two orders of magnitude in a frequency-dependent manner [141, 142]. Transcranial focused ultrasound at frequencies below 0.65 MHz and amplitude of 0.12 MPa was shown to modulate activity 3-cm deep in the human somatosensory cortex [143]. Although ultrasound signals with lower frequencies offer deeper tissue penetration without significant attenuation, the resolution is proportional to wavelength, which yields modulation volumes greater than 1 mm³ [142]. Functional ultrasound can map blood vessels in the neural tissue based on the characteristic Doppler frequency associated with the movement of blood cells through the imaging pixels [144]. Because neural activity within a given region is correlated with increased blood flow, this approach permits non-invasive and indirect measurement of neural dynamics. Without a contrast agent, a 15-MHz functional ultrasound probe enables hemodynamic imaging across the entire rat brain with a resolution 100 μm × 100 μm × 200 μm in 200 ms [144]. Although this method does not rely on implanted conduits, it necessitates mounting the functional ultrasound probe over an aqueous cranial window to avoid scattering from the skull.

Magnetic fields are the only modality that can access arbitrarily deep tissues with little attenuation owing to the low magnetic susceptibility of biological matter [145, 146]. Magnetic fields with frequencies below 1 kHz and amplitudes above 1 T can inductively evoke ionic currents to non-invasively modulate brain activity through the skull [147]. However, transcranial magnetic stimulation is limited by the coil geometry [148], and clinical applications are constrained primarily to cortical structures with centimetre spatial resolution. Functional magnetic resonance imaging (fMRI) exploits the differences between the magnetic properties of oxygenated and deoxygenated hemoglobin to measure a local increase in blood flow that accompanies neural activity [149]. Although non-invasive, this method has limited spatial and temporal resolution of ~0.5 mm × 0.5 mm voxel area and a sampling rate of less than 2 Hz.

Nanoscale transducers that convert optical, acoustic and magnetic stimuli into biological cues may extend the signal penetration depth and improve the spatial and temporal resolution of noninvasive neural recording and modulation approaches [150] (Fig. 8). Semiconductor quantum dots, gold nanoparticles and upconversion NPs can respectively convert light into voltage, heat, or light of a different wavelength (Fig. 8a–c). The deformation induced by ultrasound can be amplified by microbubbles [151] or be converted into electric fields by piezoelectric materials (Fig. 8d–e). Magnetic nanoparticles can cluster [152], transduce force [153] and apply torque in stationary field gradients, or dissipate heat through hysteresis in alternating magnetic fields [154] (Fig. 8g–i). Interfacing magnetostrictive and ferroelectric materials enables conversion of a magnetic field into an electrical potential [155]. This list, by no means exhaustive, represents a diverse and relatively unexplored materials toolkit for potential next-generation neurotechnologies [150].

Figure 8│. Nanomaterials as local transducers of external stimuli.

a│ Optoelectronic transitions in semiconductor quantum dots can be used to convert optical stimuli into electrical signals and vice versa, which can enable stimulation and recording of neural activity [156–159]. b│ Metallic nanoparticles couple to visible and near-infrared light through their plasmon resonance [134]. Plasmon energy is dissipated as heat, which can be used to modulate membrane capacitance or trigger heat-sensitive ion-channels in neurons [162, 163]. Plasmon resonance is also sensitive to electric fields, and can be used to detect changes in neuronal membrane voltage [166]. c│ Upconversion nanoparticles convert infrared to visible light via a two-photon absorption process. This effect may extend the penetration depth of optical neuromodulation techniques [172]. d│ Vibration, rotation and cavitation processes in hollow micro- and nanobubbles may induce action-potential firing in targeted neurons at ultrasound intensities otherwise insufficient to evoke excitation [173]. Microbubbles can also improve the resolution of functional ultrasound imaging by acting as contrast agents [151]. e│ Mechanical deformation by ultrasound can polarize piezoelectric nanomaterials for localized electrical stimulation [174]. f│ Genetically encoded gas vesicles function similarly to synthetic inert gas micro- and nanobubbles [207]. g│ Clustering of magnetic nanoparticles in time-constant magnetic-field gradients can be used to control membrane-protein complexes and influence cell fate [152]. h│ In slowly varying magnetic fields, magnetic nanodisks exert torque on the cell membrane [180] and magnetic nanoparticles exert tensile force. These mechanical cues can trigger the opening of mechanosensitive ion channels [175]. i│ Hysteretic heating of magnetic nanoparticles exposed to alternating magnetic fields triggers heat-sensitive ion channels causing Ca²⁺ influx and action-potential firing in neurons [183, 184].

Neural interrogation with nanomaterials

Optical nanomaterials.

Optically active inorganic nanomaterials have been explored as non-genetic tools for precise spatial and temporal control and recording of neural activity (Fig. 8a–c). Optical excitation of quantum dots embedded within the cell membrane can, in principle, generate sufficient localized electric fields to trigger action-potential firing [156] (Fig. 8a). However, to date, only quantum-dot films have been demonstrated to excite neurons upon light exposure [156, 157]. Field-induced photoluminescence quenching in quantum dots owing to ionization or the quantum-confined Stark effect may enable their use as voltage sensors [158, 159]. Technical challenges of using these materials in vivo include the cytotoxicity of quantum dots, which are composed primarily of heavy metals (Cd, Pb), and the difficulty in embedding them into the cell membrane [160]. Gold nanoparticles are significantly more biocompatible and can couple to light through surface plasmon resonance [161]. Plasmonic heating in gold nanoparticles can induce neural inhibition under constant illumination [162] or neural excitation with pulsed laser light at frequencies up to 40 Hz [163] (Fig. 8b). This photothermal approach allows operation at optical power densities lower than those needed to excite ChR2 while offering sub-cellular resolution [163–165]. Because surface plasmons are sensitive to local electric fields, they may be developed into voltage sensors [166], albeit with narrower dynamic range than quantum dots [159, 167]. Nanodiamonds harbouring nitrogen-vacancy colour centres with photoluminescence sensitive to temperature [168], electric and magnetic fields are emerging as biocompatible candidates for optical imaging of cellular activity [169, 170].

Upconversion nanoparticles absorb near-infrared light and emit shorter wavelengths via multiphoton processes (Fig. 8c). This effect can extend the penetration depth of optical stimuli, because tissue scattering is reduced in the near-infrared-light region [171]. For example, lanthanide-doped upconversion nanoparticles absorbing 980-nm light were used to control isomerization of the blue-light-sensitive LOV-domain to drive gene transcription [172].

Mechanoresponsive nanomaterials.

To improve the resolution of transcranial ultrasound stimulation while maintaining penetration depths greater than 1 cm [143], mechanoresponsive nanomaterials can be used to control targeted neurons by local amplification of low-intensity ultrasound. Perfluorohexane microbubbles undergoing physical deformation upon exposure to ultrasound (0.2–1 MPa, 2.55 MHz) have been shown to affect the behaviour of C. elegans [173] (Fig. 8d). This effect was linked to the mechanosensitive ion channel TRPV4 expressed in sensory neurons, which transduce the localized deformation during microbubble cavitation into cation influx and firing of action potentials [173]. To eliminate the reliance on mechanosensitive channels, piezoelectric barium titanate (BaTiO₃) nanoparticles have been engineered to convert ultrasound into an electric field [174] (Fig. 8e). Ca²⁺ influx in response to ultrasound was observed in neuron-like PC12 cells in close proximity to micrometre-sized BaTiO₃ aggregates [174]. These early examples demonstrate the potential of acoustic sensitizers to evoke neural activity when placed near the cell membrane. Future work in primary neurons and animal models should assess the efficacy of this neuromodulation technology. In addition to their use in neural stimulation, microbubbles and genetically encoded nano-vesicles (Fig. 8e, f) can improve the resolution of functional ultrasound imaging by two orders of magnitude, achieving hemodynamic brain mapping with a pixel size of 8 μm × 10 μm in a given illuminated plane [151].

Magnetic nanomaterials.

Magnetic nanoparticles (MNPs) can transduce a broad range of stimuli in the presence of magnetic fields and offer a versatile toolkit for contactless perturbation of cellular function (Fig. 8g–i). In a static field gradient, MNPs transduce force to actuate the opening of mechanosensitive ion channels or cluster to switch on aggregation-dependent cell pathways [175]. Relatively large forces in the piconewton range are required to trigger mechanosensitive ion channels [176], which necessitates the use of high field gradients (>5 T m⁻¹) and particles with large magnetic moments to evoke a cellular event [153, 175, 177, 178]. By contrast, forces of only a few femtonewton can drive lateral association of receptors embedded in lipid membranes (Fig. 8g). This mechanism was exploited to induce apoptosis through clustering of MNPs linked to death receptor 4 (DR4) in an applied field of 0.2 T [152]. Because the MNPs’ magnetic moment and targeting moieties can be varied independently via their size, composition, and surface functionalization, a variety of mechanical and chemical cues can be delivered with high spatial and temporal precision [179].

Alternating magnetic fields (AMFs) can also be applied to control cellular function. For example, micrometre-sized magnetizable ferrite-doped beads and permalloy discs can disrupt the cytoskeleton or cell membrane by exerting torque in a slowly varying field (10–50 Hz) [180, 181] (Fig. 8h). AMFs with frequencies up to 5 kHz coupled to superparamagnetic zinc-doped ferrite (Zn_0.4Fe_2.6O4) nanocubes can induce deflection and Ca²⁺ influx in mechanosensitive inner-ear hair cells [153] (Fig. 8h). In AMFs with low radiofrequencies (100 kHz–1 MHz) and amplitudes (10–100 mT), chemically inert ferrite MNPs (Me_xFe_3−xO₄, Me = Fe, Co, Mn, Zn) can undergo hysteresis and dissipate heat [182] (Fig. 8i). This phenomenon, which arises from irreversible work done during a magnetization cycle, has been investigated as a potential cancer treatment for nearly 5 decades. More recently, magnetic hyperthermia was applied to remote control of action-potential firing [183, 184] and gene transcription [185]. AMF-driven heat dissipation by MNPs can trigger the opening of the heat-sensitive transient potential receptor vanilloid channel 1 (TRPV1) leading to Ca²⁺ influx and action-potential firing in neurons [183, 184]. Magnetothermal neuromodulation can be formulated as a simple injection into a deep brain structure sensitized to heat via cell-specific TRPV1 expression to enable chronic stimulation [183].

Multiferroic composite nanoparticles consisting of magnetostrictive and piezoelectric phases enable remote magnetic control of the nanoparticle’s dipole moment. Magnetoelectric nanomaterials composed of magnetostrictive CoFe₂O₄ nanoparticles coated with piezoelectric BaTiO₃ shells can release electrostatically adsorbed drugs [186], and may be employed as transducers for wireless magnetoelectric neuromodulation [187].

MNPs can also be used as MRI (magnetic resonance imaging) probes of neural activity. At physiological temperatures, isolated superparamagnetic nanoparticles on average have zero magnetic moment. When clustered, the dipole–dipole interactions between the nanoparticles impart non-zero net magnetic moment onto the aggregates [188]. The change in magnetic moment can be recorded by MRI as a change in contrast, allowing for detection of events other than hemodynamics [189]. For example, MNPs conjugated to the Ca²⁺-binding protein calmodulin enable MRI measurement of Ca²⁺ concentration [190]. Although diffusion kinetics currently limits the efficacy of this method in vivo [191], improvement in clustering schemes may significantly enhance sensor sensitivity and the dynamic range [192]. With further refinement, these nanoparticle sensors may offer higher signal contrast at lower concentrations than current small molecule agents used for real-time MRI monitoring of neurotransmitter levels in the brain [193].

Prospects and challenges

Emerging applications of nanomaterials at the neural interface demonstrate the possibility of contactless neuromodulation and recording without fixed hardware and transgenes. Further gain in function can be envisioned through combinatorial strategies involving multiple nanomaterials and external stimuli. For example, independent neuromodulation of different cell populations may be possible by tailoring the nanomaterial properties, such as using MNPs that selectively dissipate heat in distinct AMF conditions or plasmonic gold nanorods that absorb light at different wavelengths [194, 195]. Materials optimization for a specific transduction mechanism can reduce stimulation latency [196] or enhance reporter sensitivity [159]. Finally, integration of probes that transduce orthogonal stimuli may further expand the experimental palette.

The delivery of nanomaterials in vivo is perhaps a bigger design challenge than optimizing the nanomaterials’ intrinsic properties [197]. The nanoparticles’ ability to target the site of interest and its residential lifetime at the neural interface depends on a number of factors including nanoparticles’ surface chemistry, size, shape, composition and concentration. Although antibody conjugation to NPs affords specificity in vitro, non-specific protein adsorption and poor colloidal stability may impede this targeting approach in vivo [198, 199]. Difficulty in traversing the blood–brain barrier [197] and transient lifetime on the cell membrane [200] render cell-specific and chronic neural interfaces with nanoparticles technically challenging. Non-surgical strategies to deliver nanoparticles to a particular brain region include functional-ultrasound-assisted temporary breach of the blood–brain barrier [201] and, for MNPs, spatial localization with magnetic-field gradients [202]. Improvement of these delivery methods may eventually allow nanoparticles to target a particular neural circuit by periodic intravenous or oral administration to overcome the transient nature of the nanomaterial–neuron interface. When invasive delivery is acceptable, nanoparticles can be directly injected into the region of interest. Amino-functionalized MNPs can persist up to a year in the human brain [203]. In the peripheral nervous system, nanoparticles can be injected into a targeted nerve encapsulated within a scaffolding material that delays nanoparticle clearance [204–206].

Biogenic nanomaterials can circumvent the difficulties in cell-specific nanoparticle targeting and their clearance through constant expression of the relevant protein complexes from the integrated transgenes. Gas-filled vesicles can be genetically encoded and engineered to improve temporal and spatial resolution of functional ultrasound imaging [207]. Magnetogenetic excitation and inhibition of neural activity in mice was recently reported with ion channels fused to ferritin [208, 209], a protein which can mineralize iron into weakly paramagnetic ferrihydrite [210]. Unlike synthetic MNPs composed of ferrimagnetic iron oxide and its alloys, ferritin possesses negligible magnetic moment, and the biophysical origins of the observed physiological response to applied magnetic fields thus remain unclear [211]. Magnetotactic bacteria are known to produce magnetite nanoparticles, which have significantly higher magnetic moments than ferritin [212]. However, transferring their genetic machinery into mammalian cells requires further optimization [213].

Conclusion and outlook

Multifunctional devices capable of forming intimate interfaces with neurons without provoking a severe foreign-body response would provide detailed maps of neural activity and serve as neuroprosthetic links with electronic circuits. Although nano- and microfabricated neural probes offer unprecedented resolution and temporal precision, the mismatch in mechanical and chemical properties of metals, oxides and semiconductors with neural tissue has frequently lead to the probe failure with time.

Fueled by advances in materials synthesis and flexible electronics, over the past decade, materials-centric designs have steadily reduced the foreign-body response at the brain–machine interface. Concomitantly, the development of opto- and chemogenetic tools has brought new opportunities for neuroscientists and inspired engineers to integrate optical and microfluidic features into neural probes. However, the reliance on fixed hardware for signal transduction, still limits the number of neurons that can be independently addressed across the nervous system.

Synthetic nanomaterials, with their diverse and tunable properties, can transduce optical, acoustic and magnetic stimuli into biological signals and vice versa. Although targeting of nanomaterials into specific locations of the nervous system presents a formidable challenge, the potential payoff is the ability to remotely control and record the activity of billions of neurons. Genetically encoded reporters and mediators of neural activity have improved our understanding of brain and nerve function, and have contributed insight into future therapies for neurological disorders. Their clinical translation, however, remains uncertain, and will require new technologies that will likely operate at the nanoscale to conform to the complexity of the nervous system.