Precision electronic medicine in the brain

Abstract

Periodically throughout history developments from adjacent fields of science and technology reach a tipping point where together they produce unparalleled advances, such as Allen Brain Atlas¹ and the Human Genome Project². Today, research focused at the interface between the nervous system and electronics is not only leading to advances in fundamental neuroscience, but also unlocking the potential of implants capable of cellular-level therapeutic targeting. Ultimately, these personalized electronic therapies will provide new treatment modalities for neurodegenerative and neuropsychiatric illness, powerful control of prosthetics for restorative function in degenerative diseases, trauma and amputation, and even augmentation of human cognition. Overall, we believe that emerging advances in tissue-like electronics will enable minimally invasive devices capable of establishing a stable long-term cellular neural interface and providing long-term treatment for chronic neurological conditions.

Neurotechnologies that interface directly with the human nervous system have reached a tipping point—one that could open new applications for electronic implants in neuroscience and medicine. Decades of research and clinical applications of therapeutic electrical stimulation^1–3, as well as the development of neural probes for neuroscientific exploration^4–6, provide a strong foundation for this future. However, despite this positive trajectory, we argue here that current neural interfaces are only a stop gap until basic structural, mechanical, and topological mismatches between electrical probes and the cellular networks comprising the brain are resolved⁷. In this Perspective, we highlight the need for truly stable and minimally invasive brain–electronic interfaces that mimic the natural properties of neural tissues and their constitutive cells. Approaches that allow stable mapping and modulation of the same individual neurons and neural circuits over extended periods of time promise to unlock new avenues for delivering personalized therapy to individuals with complex neurological and psychiatric disorders, as well as powerful control of prosthetics for restorative function in degenerative diseases, trauma and amputation— what we term here ‘precision electronic medicine’. The key components of precision electronic medicine are as follows: (1) stable recording and tracking of the same individual neurons that comprise neural circuits over the time, where most current technologies do not have this capability^8,9; (2) stable modulation of the individual neurons in neural circuits based on changes in recorded signals monitored in (1), where current technologies can only modulate regions of the brain comprising thousands of neurons¹; (3) closed-loop feedback and control based on the stable tracking and stable modulation of individual neurons in neural circuits; and ultimately, (4) monitoring and modulation at the level of specific neuron subtypes.

In this Perspective, we suggest that a central component for achieving these breakthroughs will require development and adoption of ‘tissue-like’ neural technologies capable of producing a stable interface at the cellular to subcellular level in the brain over extended periods of time. We first outline our vision of precision electronic medicine, essential pieces needed for implementation of this vision, and areas where it might impact in terms of basic science and therapeutics. Next, we step back to discuss briefly the current state-of-the-art in neural implant technologies both for medical and research applications. We highlight substantial advances made in front-end integration where implants connect to the brain and backend input/output connectivity and data processing, as well as highlight fundamental mechanical, structural, and biochemical mismatches between implants and cellular networks within neural tissues that ultimately limit the ability to have ‘precise communication’ with the same neurons over the life of an implant and more sophisticated biological functionality. We then describe how applying concepts of biomimicry have yielded ‘tissue- and neuron-like’ electronics with immune-privileged characteristics capable of stably integrating and recording from the brain over long periods of time. Lastly, we discuss developments that could produce a cell-type specific bidirectional electrical interface, modification of the tissue-like implants to enable cellular development for neural (or tissue) healing, as well as limitations that must be overcome to realize precision electronic medicine.

Trends in neural recording and neuromodulation

The three key components of neuromodulation and neuroprosthetic systems are sensing, control, and processing (Fig. 1). Among the diverse technologies used in these three areas, there are commonalities that can both help assess the advantages and disadvantages of existing and emerging neural devices and provide a framework to contextualize our vision for precision electronic medicine. When referring to sensing, we consider signals of activity recorded directly from the brain (e.g., surface or implanted electrode arrays), as well as from devices used, for instance, to detect external visual or audio signals. Signals that provide control are those that can be delivered to a part of the brain or peripheral nervous system via implanted electrodes, or for example, to a prosthetic limb. The last component, the processor, we define as the signals that are sent to a control device or nervous tissue. In many commercial implanted stimulators, the processor and control electrodes form an open loop—without direct sensory feedback— although the processor can be adjusted and/or subsequently optimized to maximize effectiveness based on observed patient response. Current and future trends point to closed-loop systems in which feedback signals, especially from the brain or nervous system, are used directly in the processor to optimize the control signals in real-time to maximize effectiveness^10,11. Ultimately, this would allow for more precise targeting and control of neural biomarkers directly related to symptom relief, thus improving therapeutic efficacy and reducing unwanted side-effects.

Figure 1:

Unidirectional and bidirectional neurostimulation approaches. To date, the vast majority of commercially available neurostimulation devices are unidirectional capable of singularly recording or stimulating. For example, unidirectional recording devices (red dashed-line) like the motor cortical prosthetics decode motor intention from motor cortical networks to actuate a robotic arm and restore movement^12,13. Similarly, unidirectional stimulation devices (blue dashed-line), such as retinal prosthetics map visual-spatial information from cameras to create visual percepts by stimulating retinal receptive fields^15,16. Bidirectional neurostimulation devices are capable of both sensing and stimulating in a real-time and adaptive manner, thus creating new opportunities leveraging closed-loop approaches.

To date, the vast majority of neural devices are unidirectional—capable of singularly recording or stimulating neural activity. A unidirectional recording device, such as a motor prosthetic, can decode motor intention from cortical neural activity and actuate a robotic arm in order to restore movement in a paralyzed individual^12,13. Similarly, unidirectional stimulation devices have yielded successes in modifying and augmenting brain function for therapy or support, including cochlear implants¹⁴, deep brain stimulation¹, and motor¹² and visual prosthetics^15,16. Technological and scientific constraints, including relatively small numbers of electrodes in multi-site neural stimulators, a lack of stable neural interfaces making difficult stable tracking of neural activity necessary for feedback, limitations in computational processing, as well as insufficient understanding of the underlying neural code, have limited progress in expanding beyond unidirectional prosthetics, though this is slowly changing^17–19.

Neural stimulation systems.

The first account on the clinical application of electrical brain stimulation can be traced to a Roman physician, Scribonius Largus in A.D. 46, which detailed the application of a bioelectric fish, Torpedo ocellata, to the cranial surface for the treatment of headache and gout²⁰. Today, though we have access to more modern technologies, the same open-loop electrical stimulation concept is used in devices implanted on the surface or deeper tissue of the brain (Figure 1). These approaches, which include deep brain stimulation (DBS), are being used for treatment of movement disorders and neurological and neuropsychiatric disorders, including Parkinson’s disease, obsessive-compulsive disorder, depression, epilepsy, and Alzheimer’s disease (for review, see ref. 1).

Specifically, these neural stimulation approaches are relatively ‘brute-force’ therapeutic interventions involving widespread modulation of neural activity through implantation of large low-impedance stimulating electrodes. For example, reduction of motor symptoms such as tremor, bradykinesia, and rigidity with bilateral subthalmic nucleus DBS treatment of Parkinson’s disease with implantable electrodes is well established, but is prone to the following limitations (e.g., see p.XXXBROWN)²¹: first, DBS electrode sizes and corresponding estimated stimulation volumes encompass large numbers of distinct types of neurons and different functional pathways^22–24, which has the potential for unwanted side-effects, and precludes therapeutic applications of higher precision; second, stimulation is typically applied without feedback, except for adjustments made by the neurologist to optimize effectiveness post-implantation through iterative and periodic patient observation, limiting the efficiency of therapy; and third, the continuous mode of operation and large implant designs limit the effective lifetime of implants in terms of the battery life^25–27 and adverse tissue immune response to the implants²⁸, respectively.

Several of these limitations are being addressed by efforts focused on improving clinical implants. To provide finer control of the effective stimulation volume commercial designs that segment and increase the number of addressable electrodes are being implemented^29,30. However, the typical sizes of these segmented electrodes remain large with respect to individual neurons and finite element models suggest they may have limited ability to steer therapeutic stimulation currents beyond the four radial electrodes of common DBS electrode design³¹. Demonstrations of closed-loop stimulation in research studies^17,32 where brain activity is monitored through local-field potentials (LFPs) have led to the implementation of upgraded DBS systems for clinical evaluation^33,34. Real-time recordings from DBS electrodes or tandemly placed brain surface electrodes can provide feedback-controlled neural stimulation, adjusting stimulation parameters such as voltage and timing through embedded algorithms³⁵. Although advances in closed-loop stimulation delivery have immediate applications for the improved treatment of diseases, such as Parkinson’s, Tourette’s Syndrome^36,37, and epilepsy³⁸, they do not overcome fundamental limitations of selective circuit-to-neuron group pairings necessary in precision electronic medicine. Ultimately, these approaches remain low resolution in terms of indiscriminately stimulating large numbers of different types of neurons and distinct neuronal pathways, are unable to provide detailed feedback information from neuronal spiking circuit activity, and do not address the intrinsic mismatch of implants with tissue that result in an immune response and ultimately limit electrode stability and lifetime.

The development of improved neural interfaces^39,40, including the engineering and manufacture of fully implantable neural recording systems capable of large-scale high-bandwidth recordings and stimulation as well as algorithms capable of real-time closed-loop therapy and prosthetic control^41,42 have also been an emphasis of large research programs in USA, Europe, and Asia^43–45,46. The primary objective of these programs is growth of neurotechnologies that bridge the spatial-temporal gap between the sub-millisecond functioning of networks with the micron-scale connectivity of neurons. For example, these programs have promoted high-density integration (discussed in the next section) and have led to substantial advances in integrated chips for processing between the ‘sensing’ and ‘control’ components (Fig. 1). Recently, a fully implanted closed-loop device was produced capable of recording LFPs, as well as performing on device single-unit detection on raw neural signals^43,45. The availability of custom chip design allows configurable recording and stimulation mosaics based on implant location, duration, and physiological function⁴³. Although these advances represent important milestones in demonstrating proof-of-principle in technological hardware capable of supporting bidirectional neural applications, these approaches are fundamentally limited by the use of non-precise and relatively rigid neural interfaces that preclude stable interfaces to cellular and sub-cellular neural elements, and correspondingly, prevent stable monitoring and modulation of the same neurons relevant to precise therapy and prosthetic control.

High-resolution neural interfaces.

In parallel with developments in neural stimulation systems, substantial effort has focused on increasing electrode density of neural recording probes (sensing in Figure 1) to record spiking activity from greater and greater numbers of individual neurons^5,47,48 (see p. AnikeevaXXX). These efforts have been motivated by prior contributions to fundamental neuroscience that identified the role of single neurons in behaviors, such as the discovery of simple cells in visual encoding⁴⁹, place/grid cells in spatial encoding^50,51, and motor population coding⁵². In addition, advances in the fabrication technology for electronics have allowed the production of devices capable of simultaneous recording on the order of 1000 neurons. Immediate opportunities to, for example, improve brain–machine interfaces for prosthetic control, are possible by increasing the number of simultaneously recorded neurons because larger numbers allow additional degrees of freedom for more natural control^53–55.

Emerging large-scale electrode technologies^56–58 have enabled the generation of neural recording datasets of unprecedented size across functionally connected networks in different regions of the brain, but not without limitation. Foremost, micromotion and the foreign body response elicited by rigid neural probes makes difficult the tracking of single neuronal activity over extended periods of times as would be necessary for a precision therapeutic implant⁶. Many factors contribute to chronic inflammatory response, including the physical size, mechanical properties and biochemical composition of the probe^48,59–61. For example, activated microglia and reactive astrocytes attempt to sequester the probe, which they recognize as a foreign body, eventually forming a multinucleated dense encapsulation layer between the probe and parenchyma. This barrier reduces probe signal-to-noise, inhibits local axonal growth, and results in neural atrophy⁴. The fundamental mismatch of the sizes and mechanical properties of probes and their impact on the foreign body response have been reviewed^48,62 (see Fig. 2 for a summary with respect to the brain and effective probe lifetime). Biomimicry and biocompatibility approaches have been explored with varying degrees of efficacy, including biomolecular coatings to promote neural density and to reduce immunoreactivity near the probe (for comprehensive reviews, see refs 4–6). Although these approaches can improve the performance of rigid neural probes, they do not address the fundamental size and mechanical mismatches with host cells and tissues.

Figure 2:

Challenges with current neural interfaces. Mismatches in structural, mechanical, and topological features between the brain and interface lead to micromotion and a prolonged chronic immune response limiting the longevity of conventional neural recording probes⁶. Similarly, factors including the physical, chemical, and mechanical composition of the electrode influence probe features such as diameter, shape, cross-sectional area, and size of recording surfaces governing the spatial resolution of the interface^5,84. Mesh electronics optimize the neural interface design for structural, mechanical, and topological similarity between the implant and neural substrate in order to create an interface that “looks” and “feels” like the cellular and sub-cellular networks comprising the brain⁶².

Neural probes tailored to the brain architecture

From the perspective of probe function, the different elements that make up nervous tissue span a wide range of sizes: synapses can vary from 20–40 nanometers, whereas neuron cell bodies and glia can span 4–100 microns⁶³. Microwire and conventional silicon array-based technologies remain larger than the scale of neural elements, particularly in high-density probes ⁶⁴. In terms of probe flexibility, neural tissue is soft, and the subcellular structures of neurons, such as axons, are even softer^65,66. Moreover, neural tissue undergoes periodic motion due to blood flow and periodic pressure changes associated with beating of the heart ⁶⁷ as well as motion of the brain within the skull during locomotion and head movements⁴⁷. Mechanically, conventional neural probes, such as silicon, carbon, or so-called flexible polyimide probes, have bending stiffnesses at least two-three orders-of-magnitude greater than that of brain tissue, which is between 10^–4 to 10^–1 nN m per unit width for a 20–100 µm thick brain slice⁶⁸, with the values for axons being several orders of magnitude smaller^65,66. The size and mechanical differences between the probe and brain are at the root of chronic inflammatory responses and gliosis, which result in scarring and degradation of the neural interface. Mechanical stiffness mismatch, which readily leads to micromotion, also makes it difficult to track the same individual neurons and neural circuits over time. Finally, the topology of standard probes is different from the interconnected open three-dimensional (3D) structure of the brain with neurons, astrocytes, and glia⁶⁹. Without adopting such a topology, probes not only exclude cells from their occupied volume, but also perhaps more importantly, preclude formation/re-formation of connections across the excluded space and inhibit the free diffusion of molecules that maintain homeostasis⁷⁰.

To develop probes that more closely resemble brain properties our group (C.M.L.)^8,9 has developed mesh electronics, which optimize the neural interface design for structural, mechanical, and topological similarity between the implant and neural substrate. The idea is to create an interface that resembles the cellular networks comprising the brain^9,48,62 (Fig. 2). The mesh is fabricated with cellular to sub-cellular sized components within a two-dimensional (2D) ultra-flexible scaffold with a bending stiffness comparable to neural tissue. This ultra-flexible macroporous interconnected arrangement of the mesh allows deep integration and interpenetration of neurons and glia without disruption to the local cytoarchitecture^9,71.

Mesh electronics can be introduced into the brain using a conventional syringe⁷² similar to many biological ‘therapeutics’, allowing ease of implantation for less conventional targets, such as the eye⁷³. The ultra-flexible 3D structure elicits only a minor foreign body response measured up to one year post-implantation in mice⁷⁴, enabling long-term stable recordings from a set of approximately 200 neurons in a single mouse⁷¹, stable tracking of the activity of the same neurons and local neural circuits for over 8 months⁷⁵. Despite these promising results in research there are areas that must be addressed with respect to human translation, including development of connection to interface cable or controller chip that is compatible with constraints of neurosurgery, demonstration of the stability and safety on time scales longer than year and approaching projected maximum timescale of patient treatment, and increasing substantially the number of addressable electrodes similar to the advances being made with high-density silicon probes.

Tissue-like implants enabling precision electronic medicine

New developments in neurotechnology promise to fundamentally shift proof-of-concept studies into applications in basic research and ultimately medical therapy (Fig. 3). For example, the functionalization of neural probes such that individual electrodes promote interactions with specific cell-surface protein markers, could allow precise measurement and control of direct and indirect pathway activity. In models of Parkinson’s disease, theoretical and computational models have postulated the role of neural subtypes such as basal ganglia D1 and D2 dopamine receptor subtypes⁸⁰; designing neural probes to interact with these cells may have a beneficial therapeutic effect by promoting selective targeting of underlying neural circuits such as the direct and indirect basal ganglia pathways. Development of neural devices to incorporate both electrophysiological recording and biochemical sensing (e.g., of dopamine and/or glutamate) through functionalized field-effect transistors^76,77 may also contribute towards the dynamic application of precision electronic medicine for Parkinson’s disease, in which neurotransmitters, such as dopamine, play an extensive role⁷⁸. For example, the dissociable roles of neural synchronization and dopamine release following electrical stimulation can be used to promote the therapeutic effects on motor symptoms while reducing any unwanted side-effects related to excessive dopamine release.

Figure 3.

Schematic representation of syringe-injectable mesh electronic implant into a human brain. Mesh electronics directly address the structural, mechanical, and topological mismatch between probe and host tissue. The mesh is fabricated with subcellular-sized nanowire field-effect transistor detectors which maintain measurement sensitivity and allow for highly localized sampling, formation of artificial synapses, and minimally invasive intracellular recordings and single-neuron stimulation (stimulation pad depicted by red circle)^76,85–87. The cellular and sub-cellular sized components are incorporated onto 3D ultra-flexible scaffold with a bending stiffness similar to the brain⁷¹, and the macroporous interconnected arrangement of the mesh allow deep integration and interpenetration of neurons and glia without disruption to the local cytoarchitecture⁷¹. Schematic representation of mesh electronics implanted in region of heterogenous neuronal sub-types (green and blue dashed-squares). Identifying unique cell-surface protein patterns may enable targeted cellular recording/stimulation interfaces through, for example, expression of complementary antibodies or aptamers on the mesh electronics electrode surfaces and/or polymer-encapsulated mesh structure.

Work from our laboratory (C.M.L.)⁸⁶ has shown that cells adhere and migrate along the electronically-active scaffold of mesh electronics. Along these lines, it may prove useful to modify neural surfaces to promote interactions with neural progenitor cells and tissue remodeling factors, as this would potentially create an opportunity to build interfaces that address both structural and functional components of neuromodulation. It could also provide a means of a unique active regenerative therapy to the hippocampus and adjacent cortical regions that are often sites of early neural neurodegeneration in diseases, such as Alzheimer’s disease⁷⁹. Additionally, emerging evidence suggests that electrical stimulation in the hippocampus induces neurogenesis in the dentate gyrus layer of the hippocampus⁸⁰. Multi-pronged therapeutic strategies in which temporally interleaved stimulation could induce neurogenesis, allow directed migration and differentiation of new nerve cells on the electronic scaffold, and functionally incorporate these cells into existing neural circuits through electrical stimulation^81,82. Such approaches would enable novel and exciting applications in precision electronic medicine.

The encouraging prospects for the creation of a long-term stable neural interface, we believe, will result in the enhancement of emerging approaches to neuromodulation therapies as well as open new approaches previously unconsidered. We envisage that future neurotechnologies will continue to blur the mechanical, structural, and biochemical dissimilarities between probe and tissue. However, developments are needed in two major areas to realize the full potential of the technology.

First, applying advances in semiconductor fabrication and microscale chip design for signal multiplexing, which have been demonstrated for high-density silicon probes⁵⁶, will allow increases in local recording density and volume over larger swaths of the brain and thus sould be capable of improved capture of the intrinsic network activity observed throughout the human brain over long periods of time. Additional benefits, such as the need for template-based spike sorting, will be obviated by the implementation of more robust signal triangulation methods following attainment of a critical density of recording contacts⁸³. Similarly, increases in sensor density will provide high-fidelity control over targeted neurons and circuits through directed exposure to electrical stimulation.

Second, advances in neural implants capable of handling high-density sensing and stimulation will be required to enable next-generation applications of implants that match the brain and are capable of stably recording from populations of neurons over extended periods of time. Challenges relating to mechanical and biocompatibility of high-density connectors, cables, and housing along with issues relating to battery longevity and recharge cycles will pose substantial barriers to the implementation of precision electronic medicine.

Conclusions

The field of electronic neural implants is poised to usher in a new era of basic research, therapeutic intervention, and other applications in neuroscience. This perspective outlines some of the constraints for realizing a vision of precision electronic medicine, and highlights the importance of a natural tissue-electronics interface for long-term implants that are immune-privileged and free of the foreign body response to enable precise recording and stimulation of the same individual neurons72 and neural circuits over extended time periods. Given the importance for both fundamental neuroscience research and human translation we expect to see efforts focused on fabrication of highly flexible probes such as mesh electronics, which do not exhibit an immune response, with high densities of recording and stimulation electrodes in the near term. Moreover, we expect that interfacing such ultraflexible implants with mature silicon-based processor chips will see increased emphasis as it could allow for efficient handling of expected large data streams as well as would being central to closed-loop controllers.

With these technology advances emerging, we envision that high-density immune privileged interfaces will rapidly expand our ability to study the brain unlike ever before, building the foundation of understanding necessary to unlock the potential for seamless neural-electronic systems. In the not too distant future, we posit that our level of technological and scientific insight will be sufficient to stably interface with the human brain in a manner that mirrors the organization of the brain itself. This will first arise in the form of enhanced therapeutic approaches to treating some of the most challenging neurodegenerative and neuropsychiatric disease, such as Alzheimer’s disease, depression, and obsessive-compulsive disorder. Subsequent applications will likely involve interfacing the brain in the absence of disease, to enhance or prevent the decline of cognitive capacities and expanding the exploration of what is possible with such a novel interface. Given that the brain is the very organ that makes us human, careful consideration of ethical issues will be required, although it is our opinion that the opportunity to develop powerful treatments for neurodegenerative and neuropsychiatric diseases as well as enhancing restorative function in trauma and amputation mandates that these efforts should proceed forward aggressively.