Learning from the Brain’s Architecture: Bioinspired Strategies Towards Implantable Neural Interfaces

Abstract

While early neural interfaces consisted of rigid, monolithic probes, recent implantable technologies include meshes, gels, and threads that imitate various properties of the neural tissue itself. Such mimicry brings new capabilities to the traditional electrophysiology toolbox, with benefits for both neuroscience studies and clinical treatments. Specifically, by matching the multi-dimensional mechanical properties of the brain, neural implants can preserve the endogenous environment while functioning over chronic timescales. Further, topological mimicry of neural structures enables seamless integration into the tissue and provides proximal access to neurons for high-quality recordings. Ultimately, we envision that neuromorphic devices incorporating functional, mechanical, and topological mimicry of the brain may facilitate stable operation of advanced brain machine interfaces with minimal disruption of the native tissue.

Introduction

Since the days of Fritsch [1] and later Hubel [2], neural interfaces have illuminated the complex circuitry of the brain through both electrical stimulation and recording modalities [3,4]. Thanks to modern fabrication techniques, simple interfaces made of single microwires have since evolved to shanks that can simultaneously record from hundreds of neurons [5,6]. However, today’s established neural implants such as deep-brain stimulation (DBS) electrodes, Michigan arrays, and Utah arrays are mechanically rigid in nature, and it remains challenging for their topologies to conformally interface the complex neural network. Such mismatch in the mechanical and topological properties between implants and neural tissue can lead to device failure and tissue damage [7,8].

Nevertheless, taking inspiration from the brain’s own properties can improve the chronic outcomes of neural implants while also enhancing recording capabilities [5]. For example, reducing the bending stiffness of implants to approach the mechanical properties of brain tissue minimizes gliosis [9], enabling chronic stability, long-term recordings, high signal-to-noise ratio (SNR), and efficient neurochemical sensing [10]. Further, three-dimensional (3D) and macroporous device architectures inspired by the neural network increase the number of interfaced neurons while minimizing disruption to endogenous electrical and chemical signaling processes [5]. In this review, we will highlight the strategies for mechanical and topological mimicry of neural tissue to optimize recording capabilities (Fig. 1). By matching the properties of the endogenous tissue, neural implants can achieve seamless integration over chronic timescales, opening new possibilities for neuroscience study and brain computer interfaces.

Figure 1.

Mechanical, topological, and functional properties of the brain, along with devices that mimic these properties to achieve a seamless and chronic interface.

Mechanical Mimicry

Neural tissue is a mechanically complex system, originating from its multiple cell types and the extracellular matrix (ECM) surrounding neurons and glia [11] The complexity can be characterized as two-fold: multivariate representation of mechanical properties and the heterogeneity of neural tissue. Firstly, a series of physical parameters is required to mechanically model neural tissue [12]. Young’s modulus and bending stiffness are two common parameters representing the softness and the flexibility of neural tissues, respectively. Their definitions and corresponding applicable scenarios have been summarized in another review [13]. Moreover, evaluated by rheological study, the mechanical response of neural tissue is dependent on the applied strain rate. The viscoelastic properties thus provide a more complete characterization of neural tissues, including both the storage modulus and loss modulus [14]. Secondly, the anatomical classification provides a pragmatic perspective on the mechanical heterogeneity of neural tissue. For instance, the tissue located in the central nervous system (CNS) exhibits distinct mechanical properties from that in the peripheral nervous system (PNS), as a result of evolutionary adaptation. In the CNS, specifically, white matter, grey matter, and dura mater have different mechanical properties due to the heterogeneity of chemical compositions and cellular differences [15].

With the advent of flexible electronics, the updated neuroengineering toolbox has heralded a new era of mechanical mimicry for chronic and seamless implantable neural interfaces (Fig. 1) [15]. Firstly, inspired by the extremely low modulus of the brain tissue, materials with similar elastic moduli (on the kPa scale), such as polymers and hydrogels, have been actively applied and functionalized in implantable microelectronics for next-generation neural probes [16]. Multishank polyimide electrode arrays with up to 1,024 electrodes host high channel counts with flexibility that affords chronic single-unit recordings for over five months [17]. Secondly, structural designs to reduce the effective bending stiffness of implanted probes offer another strategy for mechanical mimicry. The probes with tissue-level bending stiffness (on the nN·m scale) minimize the excessive stress generated during local bending, thus preserving the natural environment of intact neural tissue during chronic recording [9,18]. Additionally, despite its various representations, effective bending stiffness is proportional to the third power of material thickness [13,19]. Therefore, high-performance but rigid materials can also be designed to be “bendable” like neural tissue by minimizing the device sizes [13], Thirdly, the strain-rate-dependent properties of neural tissue also call for multivariate mechanical mimicry for seamless dynamic interfacing. To match the viscoelasticity of neural tissue and its dynamic growth, stretchable and non-Newtonian probes act as a crucial platform to actively minimize the stress, which is represented by the morphing electronics based on viscoplastic electrodes [20].

Structural and Topological Mimicry

The Structure and Topology of the Nervous System

The nervous system comprises structures on multiple length scales that are all morphologically distinct. Axons can vary in length from 100 microns to over 2 meters in the central nervous system of some vertebrates, with diameters ranging from 0.2 to 20 microns [21]. The cell body of a neuron, also known as the soma, can range in diameter from 10 to 50 microns [22]. Presynaptic varicosities are much smaller, occupying a volume of approximately 1 femtoliter [23]. Potassium ion channels are approximately 45 angstroms long with diameters ranging from 10 angstroms at the widest to 7.5 angstroms at the selectivity filter [24]. Neural circuits are more difficult to define, spanning length scales from only several neurons to macroscopic regions of the brain anywhere from sub-millimeter to centimeters in scale [25]. Neuronal and glial networks have complex topologies resulting from the dense interconnection of the diverse family of these various nerve cell types. For instance, in the mouse cerebral cortex, the synapse density can be as high as one synapse for every 1.13 cubic microns [26]. Adding to the complexity is the fact that proximity is an insufficient metric for predicting synaptic connectivity [26]. Even similar functions in the nervous system are accomplished by multiple cell types, such as the case of synaptic sheathing, in which non-myelinating Schwann cells sheathe synapses in the PNS, while astrocytes sheathe synapses in the CNS [27]. The brain extracellular space (ECS), consisting of the fluid and extracellular matrix-filled interstices external to cell membranes, also extensively contributes to the topology of neuronal networks with a foam-like, porous structure that is only tens of nanometers wide [28].

Advantages of Neural Interfaces that Employ Structural and Topological Mimicry

Structural and topological mimicry can be used to great effect to enhance the neural integration and recording density of neural probes (Fig. 1). Neuron-like electronics (NeuE) are biomimetic neural probes that take inspiration from the interconnected structure of dendrite branches and approach neurite diameters in size [29]. By mimicking endogenous neural scaffolds composed of radial glia, astrocytes, and ECM, the NeuE ribbons can facilitate the migration and differentiation of neural progenitor cells and reduce the activation of astrocytes and microglia [29]. The fenestration of neural probes, as employed in mesh electronics, is another form of biomimicry used in neural probe design [30]. The macroporous structure of mesh electronics results in a probe that is 90% free space, thus enabling the interpenetration of axons and neuron somata with natural cell distributions [30]. Another benefit of this design is that perforations in these neural probes allow the transduction of cytokines and nutrients for enhanced neural integration [31]. Precise targeted delivery of NeuE or mesh electronics is achievable via stereotaxic injection by taking care to minimize relative motion between the device and brain upon injection [32]. Imaging is possible with current histological techniques such as tissue slicing and immunostaining.

Besides the 3D architecture, several biomimetic two-dimensional (2D) neural probes have been designed to match the topology of cortical and retinal surfaces. NeuroGrid is a 2D electrode array that can noninvasively record action potentials from the superficial cortical layers of rodents and humans [33]. In addition, Neural Matrix is a flexible, multiplexed 2D electrode array that samples a centimeter-scale cortical region with a high multiplexity of >1,000 channels and a chronic stability of >1 year [34]. Besides conformally interfacing with the cortex, e-dura is another 2D electrode array that can record electrospinograms (ESG) from the dorsal surface of the spinal cord [35,36]. Further, a subretinal prosthesis based on conjugated polymers conforms to the retinal surface with light sensitivity compatible with human vision [37]. Subsequent experiments with injected semiconducting polymer nanoparticles have also demonstrated a tight interface with bipolar cells in the retina [38,39]. Recent advances suggest the advantages of nanoparticles and even small molecules to replace larger implants as less invasive and more biocompatible neural interfaces [40–44]. Finally, a self-unfolding epiretinal prosthesis made of shape memory polymers transforms from a rolled tube to a 2D sheet upon implantation in the eye, minimizing the size of the surgical incision [45].

Along with structural and topological mimicry on the organ and circuit levels, emerging designs of the implantable neural probes are increasingly inspired by the size and topology of neuron somata and neurites. Specifically, by imitating the size of neuron somata and the topology of neurites, NeuE achieves seamless 3D integration with endogenous neuronal networks and minimal relative shear motion, thereby enabling stable tracking of single neuron activity over long time scales [29]. In addition, NeuroRoots mimic natural axonal bundles in a 3D structure, resulting in a larger sampled volume than rigid 2D interfaces, which are constrained in a plane [46]. Neurotassels and Neuralink devices are thread-like implants that offer similar 3D integration [47,48]. Aside from the stereotaxic injection approach for NeuE, a temporary engaging method using dissolvable polymers and a shuttle device ensures the precise targeting of NeuroRoots, Neurotassels and Neuralink devices. Additionally, inspired by the climbing abilities of some plants, twining electrodes made of shape memory polymers assume a 3D helical topology to form a tight interface with nerves for peripheral neuromodulation [49].

Other approaches to structural and topological mimicry

Nanowire field-effect transistors (nanoFETs) can be incorporated into implantable neural probes for interfacing individual ion channels and penetrating the neuronal membrane. Active semiconductor nanoFETs are sufficiently small to approach the scale of a single ion channel, enabling minimally invasive penetration of cell membranes, with the advantage of not having impedance scale inversely with probe size [50,51]. Besides nanoFETs, the nanometer precision of a flexible nanopipette can be used to patch clamp single ion channels and to scan and image the topography of the membrane surface [52]. The scale of the nanopipette allows for deeper, less-invasive penetration with greater reusability and reduced clogging [52].

Conclusion and Outlook

We argue that seamlessly integrated neural interfaces can be realized by designing probes with the mechanical and topological properties of the neural tissue in mind. The many examples above demonstrate the benefits of this strategy, with some implants remaining chronically integrated with the endogenous tissue for at least one year, and others tightly conforming to the curved geometries of the retina and cortex.

In addition to mimicking the mechanical and topological properties of the brain, next-generation brain machine interfaces may also benefit from the functional mimicry of the neural network (Fig. 1). While traditional computers are characterized by separate computational and memory units (Von Neumann architecture), neuromorphic electronics integrate processing and memory into the same unit, with short- and long-term plasticity between units modelled after synaptic plasticity between neurons. To date, proof-of-concept neuromorphic elements have only been demonstrated in vitro, namely by their dynamic responses to presynaptic dopamine signaling from PC-12 cells [53]. Bioactive coatings can also provide enhanced biocompatibility and neuroadhesion for electrodes [54]. Electrodes can be decorated with living cells and tissue as a means of promoting greater long-term stability [55]. Conductive hydrogel coatings can be used for their capacity for the controlled release of growth factor and other biomolecules [56]. Ultimately, bioinspired organic electrochemical devices will offer biocompatibility over extended timescales for potential restoration or augmentation of the nervous system in vivo [53,57,58].