Neural Interfaces

Warren M Grill. American Scientist. Volume 98, Issua 1. Jan/Feb 2010.

A neural interface is an engineered device designed to exchange information with the nervous system. Signals can be introduced by localized electrical stimulation of neurons, and information can be captured by recording the electrical activity of neurons over time. Injuries to peripheral nerves or the spinal cord often result in catastrophic loss of function—limbs and organs become disconnected by the equivalent of cut wires. A possibility that energizes researchers in the field is that the connections and physiological functions can be restored by the development of devices that supply a bridge for the interrupted signal. A computer listens to a keyboard and communicates to a printer through cables. At the ends of the cables are specific adapters. Neural interfaces are adapters that can collect neural signals, port them through feedback and control circuits, and deliver them to neural circuits downstream from the cause of disruption in neural traffic. Neural routes that had fallen silent after injury can again become conductors of physiological signals, with potentially life-changing restoration of function.

Basic neural interfaces already have important medical roles, but there is far greater potential on the horizon if an array of fundamental biophysical and biomédical challenges can be overcome. The challenges fall into the following categories: selectivity—small numbers of neurons must be monitored and addressed uniquely in a noise-ridden environment; stability—optimally, surgically implanted interfaces will interact with specific neurons over a time frame of years; resolution versus invasiveness—a balance must be achieved between the selectivity of the signal and the degree to which delicate physiological structures will tolerate the intrusion of the device; and management of host-interface responses—reducing tissue reactions such as inflammation and scarring, which plague implanted devices today, will be critical in achieving selectivity, stability and resolution in long-term implants.

Approaches to overcoming these challenges include empirically derived advances in electrode design, building on ongoing experiments in material selection and geometry; biochemical manipulation of the injury response and healing; and in the case of electrical activation of the nervous system, experimental analysis of the response to changes in the timing, amplitude, and polarity of stimulation waveforms. On the horizon, research is being done on single-cell electronic connections—silicon-neuron hybrid circuits—and alternatives to stimulation by electrical current, including genetic modifications that produce neurons that can be activated by light.

The Wiring of the Body Electric

Communication through the nervous system takes the form of action potentials, which consist of rapid, transient changes in the polarization state of nerve cells. The number, rate and pattern of action potentials can be recorded to acquire information from the nervous system, and action potentials can be induced by tiny currents that cause the depolarization of nearby nerve cells and nerve bundles. Depolarization beyond a threshold amount is the trigger for an action potential.

The peripheral nervous system (PNS) relays information to and from the central nervous system (CNS), comprising the brain and spinal cord. Motor instructions and sensory feedback are conducted in opposite directions through nerve cell axons that lie side by side in the same nerve bundles.

The nerves of the PNS consist of both bare fibers and fibers wrapped in myelin. The myelin sheath insulates the nerve fibers, and it contributes to speedier transmission of action potentials by acting as a spacer between clusters of ion channels on the axon surface that open in response to voltage changes across the membrane. The depolarization that occurs during an action potential arises when positive ions flow into the nerve fiber, which is polarized (negative inside) in its resting state. Propagation of the action potential occurs as depolarization in a region of the nerve fiber spreads to cause voltage-gated ion channels in adjacent regions to open. The depolarization is communicated at much higher speed along segments sheathed in myelin, leaping between the gaps between myelinated segments. In discussing interfacing with the PNS, we are mainly interested in myelinated nerve fibers, which carry messages rapidly over long distances.

Nerve fibers are organized in bundles called fascicles, which are themselves bound in nerve cables that branch from peripheral nerve trunks. Helpfully, the topological arrangement of nerve bundles is related to the eventual anatomical targets of the neural signals. Classic anatomical studies by Sir Sydney Sunderland in the 1960s and 70s, confirmed and expanded by recent work, have shown that in the more peripheral parts of a nerve trunk, motor axons are grouped in fascicles that innervate single muscles or small groups of synergistic muscles. Careful staining experiments have shown that nerves innervating the fingertips of monkeys form a discrete geometrical group through the carpal tunnel of the wrist and all the way through the upper arm. If this were not the case, it would be much more difficult to discover where to introduce targeted signals intended to initiate specific motor responses. Geometry also assists because nerves that will branch at upcoming junctures are at the surface of nerve bundles—nerves peel off the central trunk from the outside layers.

Nevertheless, there are difficult problems posed by the geometry of the nervous system. Nerve cables are three-dimensional bundles, with most axons buried within the bundle. And here we run into a fundamental challenge of neural interfacing. To deliver or record signals effectively and selectively, an electrode must be implanted very close to the targeted nerve fiber, while avoiding injury arising from the implantation.

Links with Cuff Electrodes

The challenge of selectivity turns on two fundamental properties of nerve fiber stimulation: The current required to stimulate a neuron with an extracellular electrode depends on the distance between the electrode and the nerve fiber—the current-distance problem; it also depends on the diameter of the nerve fiber—the current-diameter problem. Due to biophysical characteristics of nerve fiber polarization, nerve fibers of larger diameter require less stimulating current.

Nerves are delicate anatomical structures. Simply piercing nervous tissue with an electrode is a direct and effective technique suitable for laboratory experiments on cultured cells, but the accompanying trauma is undesirable for therapeutic interfacing with the structures of the peripheral nervous system. A contemporary solution is cuff electrodes, which generally include several independently addressable conductive electrodes embedded in an insulating sheath (cuff). The cuff is surgically positioned with the electrodes directly contacting the surface of peripheral nerve trunks. The cuff exploits the current-distance relationship to enable selective and graded stimulation of an individual nerve fascicle lying closest to the electrode contact, ideally with minimal stimulation of fascicles more distant from the contact point.

One of the advantages of an encircling cuff of multiple, independently addressable electrodes is that the cuff can be installed without precise foreknowledge of the specific nerves being contacted. Once in place, a feedback device can then learn which electrodes to address by monitoring for functional outcomes in response to stimulation. In an early experimental use of the nerve cuff electrode, a multi-electrode cuff was placed around the optic nerve of an individual who had total loss of vision due to photoreceptor loss. Stimulation through the individual electrodes in the cuff generated visual sensations (phosphenes), notably in different portions of visual space, which is reflective of how the retina, optic nerve and visual cortex are organized.

Early animal studies revealed the challenges of such interfaces, including variability in the effectiveness of stimulation and recording. The classic response to foreign materials in tissue is a covering of fibrotic connective tissue. It had been hoped that tissue encapsulation accompanying healing would stabilize the cuff and the electrodeneuron interface, but in some instances, variability in input-output properties remained. Furthermore, there was evidence of mechanically induced neural injury, although the injury seemed to have minimal effect on the electrophysiological performance of the installation. The connection between performance and longer-term acquisition of injury-related damage remained uncertain as research moved to studies of humans with neurological diseases or injuries.

In recent work, cuffs have been installed in humans to attempt the restoration of upper-extremity motor function. Initial studies were conducted during unrelated surgical procedures during which nerves were exposed. Measurements were made on stimulation thresholds and selectivities that extended what had been learned from pre-clinical animal studies. Running through the choice of electrode contacts, researchers were able to activate certain target muscles from about 30 percent to nearly 100 percent activation before the activation spread to other muscles. This encouraging result was followed by the implantation of four nerve-cuff electrodes in an individual with tetraplegia (paralysis of all four limbs). The permanently implanted electrodes exhibited long-term stable thresholds while selective stimulation conferred functional movements such as lifting of the arm (shoulder abduction), elbow flexion and extension, and extension of the wrist and fingers. While the level of control achieved was coarse, it was a significant leap beyond no control at all and opened vistas on what will be medically possible.

In some therapeutic interventions, a nonspecific aggregate signal readily achieved by an electrode cuff is valuable in restoring function. Foot drop is a malady in which neurological damage renders an individual unable to turn the ankle and toes upwards, resulting in an awkward foot-slapping gait. A nerve cuff collecting an aggregate signal has been used to detect heel strike on the ground, triggering electrical stimulation of muscles that compensates for foot drop. Similar signal detection can be used to recognize the onset of an object slipping from the fingers of an individual with compromised grasp, triggering a stimulating signal to restore grip. In these applications, the recorded signal is an aggregate of many nerve fibers within a cuff. The selectivity of the recording is coarse but sufficient to drive a feedback circuit that delivers a marked functional improvement.

A FINEr Cuff

Advances in cuff design attempt to counter the challenging threedimensional geometry of the fascicle by changing the geometry. The Flat Interface Nerve Electrode (FINE) flattens the round cross-section of nerve bundles, which increases the perimeter of the nerve, bringing more nerve bundles into reach, and bringing nerves in the center of the nerve bundle closer to the encircling electrodes. Flat interface electrodes have proven to be well tolerated, and they can be selective enough to enable stimulation even of subregions of individual fascicles.

For both flat and round electrode cuffs, the direct contact between cuff and nerve can result in nerve damage, and pressure from the cuff may cause demyelination, which impedes neural conduction. Recent work suggests that these problems may be avoidable by implanting at regions of minimal nerve such as toward the center of long bones. Cuff electrodes can also be customized, with a various sizes available for cuffs that destined for specific nerves. A of softer and more readily tolmaterials is being explored for the insulating cuff, and progress is being made on determining the snugness of fit that gives the best tradeoff of fidelity versus tissue damage and fibrosis.

Intraneural Electrodes

A very high degree of selectivity can be achieved by threading electrodes inside the length of an axon bundle. Electrodes of very fine platinumiridium wire implanted in the sciatic nerve of the cat have been shown to stimulate so selectively that only portions of targeted fascicles were activated by the stimulating current, resulting in activation of single muscles—very high fidelity indeed. An additional advantage is very low stimulation thresholds—an order of magnitude lower than those required for the extracellular stimulation provided by cuff electrodes.

A drawback of intrafascicular electrodes is the invasiveness of the implant. Insertion of the electrode necessarily violates the protective tissue around the nerve bundle, breaching the blood-nerve barrier and disrupting the chemical and mechanical microenvironment on which continued nerve function depends. Edema, fibrosis and loss of nerve fibers are common results. As with other interface techniques, advances are being made in techniques for installing the devices and managing the healing response.

Regenerative Electrodes

A particularly selective and stable nerve-electrode contact can be made if individual nerves are coaxed to grow into contact with engineered receptacle electrodes. Axon-guidance techniques use structural cues to guide the nerve, such as aligned poly PAN-MA (acrylonitrile-methacrylate) nanofibers that form conduits for growth. Paths for growth may also be constructed by creating tracks of growth factors, proteins of the extracellular matrix and other biological molecules that influence the growth of the fiber as part of their biological function. In the regenerative sieve electrode, a perforated disk is apposed to the• end of a severed nerve, allowing cell processes—wiry extensions of the cell—to grow through the holes in the disk. The lining on the inside wall of each perforation serves as an individually addressable electrode, and the sieve electrode is seized to the end of the nerve by the processes that grow through the perforations, providing a measure of stability. However, the resulting fibers tend to be small—small fibers require higher stimulating currents—and they may have diminished or absent myelin sheaths. Current work suggests that sieve electrode holes larger than 40 micrometers may contribute to adequate nerve-end regeneration. As regenerative techniques develop, the stability of implants will likely be enhanced, and the signal-tonoise ratio will be improved by maximizing the proximity of the electrode to a specific nerve.

Signal to Noise

The small amplitude of neural signals and the ratio of signal to noise pose significant challenges for neural interfacing. The average resting potential of mammalian neurons is approximately -60 millivolts, which can reach positive values during the passage of an action potential. Electrodes that penetrate the neuron may encounter this level of signal, but the signals recorded from extracellular electrodes are much lower. For example, in regenerative electrodes, in which the nerve processes that grow toward an electrode may be very fine, the signal that is encountered may be just a few hundred microvolts—amidst a cloud of noise.

When attempting to use captured signals to direct motor functions, the difficulty is compounded by a basic physiological fact: Muscle is noisy. Neural signals near contracting muscles can be swamped by the much larger-amplitude electrical signals created by muscle activation (the electromyogram or EMG). In addition, if recording and stimulation are being done simultaneously, the electrical stimulation signal may swamp the recording with large electrical artifacts.

These problems are addressed in part by the properties of cuff electrodes. All action currents are corraled within the volume of the cuff, and the recording electrodes are to some degree shielded from muscle electromyograms or stimulation artifacts by the insulating properties of the cuff.

A more comprehensive solution is to move the neural interface upstream along the axon to the cell body of the neuron in the dorsal root ganglion of the spinal cord. (Neuron cell bodies reside in the dorsal root and project axons to the periphery, in some cases as much as a meter away.) Silicon-based implants in the dorsal ganglion offer the promise of stability, and the combination of simultaneous recordings from just 8 to 10 single axonal units can yield acceptably high-fidelity interpretations of limb position and motion, supplying the foundation for a therapeutically useful feedback and control circuit.

Connections in High Places

Interfacing to the central nervous system itself—the brain and spinal cord—is now a major focus of neuroengineering. Examples of therapies under intense study include the use of deep brain stimulation to regulate mood disorders, epilepsy, and Parkinson’s symptoms. Much work is also being done on developing the ability to control devices with a brain-computer interface.

Overall, the brain can be divided into several functional areas. At the base is the brain stem, center for the control of cardiac and respiratory functions. Above the brain stem are deep structures—the thalamus, hippocampus, basal ganglia and others—that have a range of functions, from temperature regulation to emotional response to stimuli to memory formation. The cerebellum, at the lower back of the skull, receives sensory and motor input and appears to be responsible for posture and motor coordination.

The higher brain functions reside in the cerebral cortex on the brain’s surface. The cortex consists of six layers, containing different cells or processes of cells that carry different kinds of information. The cortical surface is further divided into regions of functional specificity—visual cortex, auditory cortex, motor cortex and so on, each with specialized chores related to the initiation, execution or processing of neurological events.

In the spinal cord, the dorsal (back) side carries sensory information, and the ventral (front) side carries motor information. Along with neurons, many other cell types inhabit the CNS, including an array of glial cells (non-neuron support cells) that includes astrocytes, microglia, oligodendrocytes, Schwann cells and neural precursor cells. The interactions among these cells figure importantly in the response to implanted CNS interfaces.

Host Response in the CNS

When a device is implanted in the brain or spinal cord for extended periods, an unforgiving inflammatory response is initiated. Initial electrode insertion is likely to damage multiple structures, including capillaries, extracellular matrix and support cells such as astrocytes. Macrophages, the debris scavengers of the immune system, may enter through severed vessels. Microglia, which are debris scavengers specific to the brain, migrate to the site of the injury; staining shows them clustered around electrode interfaces. Activation of microglia is followed by encapsulation of the electrode by astrocytes, forming a dense cellular sheath that probably diminishes the ability of the electrode to record and stimulate by increasing tissue impedance. The chemical responses of the glial cells, including production of signaling molecules and reactive oxygen species, may contribute to a decrease in the number of neurons in proximity to the electrode over time.

In attempts to circumvent the host response, electrodes have been fabricated with different physical dimensions and geometries. Tip design can affect the acute (initial) response, but within a few weeks, the inflammatory response looks similar in all cases. Coating the electrodes with anti-inflammatory agents seems to inhibit glial activation, but it is not yet known whether this strategy will improve the longevity of cortical recording. Overall, it is likely that the more the glial response can be inhibited, the better the prognosis for long-term implants.

As in the peripheral nervous system, dealing with host response in the CNS requires a tradeoff between resolution and invasiveness. Penetrating electrodes are not always necessary. For example, electrocorticogram (ECoG) electrodes are placed inside the skull but above the brain. However, the signals accessible from the brain surface are the summation of a vast number of neural signals that deliver imprecise spatial information. The ECoG is the standard tool for defining zones linked to epileptic seizure, but for many applications a higher degree of resolution may be required.

Between the summed signal of an ECoG and the fidelity of a single cell, it may be desirable to address a specific localized region of the cortex. Penetrating microwire arrays such as the Utah arrays can address single units or small regions of the brain. More physical parts protruding in the brain is inevitably accompanied by more tissue damage, but such devices have been used with success in a variety of nonhuman primate and human studies.


The neuron-packed brain and spinal cord are complex volume conductors. Signals in the CNS are typically small in amplitude and deterioriate rapidly over space. Stimulating electrodes are placed in proximity with cells, axons and dendrites (neuron processes that receive signals from other neurons and conduct the messages back to the cell body of the neuron). It is unclear which elements of which neurons are most responsive to stimulation. Current may activate or inactivate neurons and axons depending on morphology, distance from the electrode, orientation and other stimulus parameters. Stimulation of axons carrying longdistance messages through the brain (called fibers of passage) results in action potentials propagating in both directions, the normal direction toward the axon (orthodromic) and the reverse direction toward the cell body (antidromic). Cells and fibers of passage have similar thresholds for activation, and stimulation may activate both pre- and post-synaptic elements. (Communication between neurons takes place at intimate gaps between cells called synapses.) Furthermore, brain function is localized but some of the functions are coordinated over considerable physiological distances, certainly very large distances compared to implant size. In sum, the structure of the brain presents a difficult challenge when the goal is a selective neurological response.

In some applications, alteration of the stimulation waveform and polarity can increase selectivity by stimulating some neurological elements preferentially. For example positive-phase waveforms selectively stimulate local cells, whereas negative-phase waveforms preferentially stimulates axons of passage. The range of stimulation patterns, including asymmetric waveforms and assorted polarity manipulations, are being evaluated to increase selectivity in both the central and peripheral nervous system.

Emerging Opportunities

The intimacy and density of neural interfaces promises to be dramatically increased by connections between microcircuits and single neurons or even single ion channels. In 2004, R. Alexander Kaul at the Max Planck Institute for Biochemistry and coworkers developed a silicon-neuron hybrid by culturing pre- and postsynaptic nerve cells on a silicon chip. The presynaptic cell was stimulated by a capacitor on the chip, leading to stimulation of the postsynaptic cell, which was recorded by a transistor on the chip. Repetitive stimulation by the capacitor strengthened the synaptic communication between the two cells, demonstrating neuronal memory on the silicon chip. The direct integration of a neuronal circuit and a semiconductor chip will likely prove to be an important step in the integration of neuroelectronic devices.

A feature of the intimate contact between neuron and device on the silicon chip is that it enables capacitative stimulation by generating field effects through the accumulation of charge without the actual transfer of charge. The miniaturization of electrodes is limited by the ability to pass charge from the metal electrode to ionic carriers in tissue, which is related to the surface area of the electrode. Excess charge per unit area damages the electrode via corrosion and damages tissue via chemical reactions at the electrodetissue interface. With capacitative currents, stimulation occurs without electron transfer or chemical reactions at the electrode surface. Efforts to move the capabilities of siliconneuron hybrids from in vitro to in vivo are generating great excitement. Neurons cultured in wells on silicon-based electrodes have been shown to grow extensions away from the probe, suggesting the possibility of in vivo circuitry. An ongoing challenge is maintaining the stability of the circuit, a task that includes preventing the neurons from migrating away from the probe.

A different and particularly arresting approach is the use of light to replace electrical stimulation. In classic work in the early 1970s, Richard L. Fork at Bell Laboratories demonstrated that laser illumination could produce excitation of molluscan neurons through a mechanism that was unknown at the time, and remains murky to this day. Later experiments suggest that photochemical reactions at the surfaces of neurons may generate reactive oxygen species that lead to depolarization, possibly coupled with photothermal injury that transiently perforates the membrane, again leading to depolarization. The intensity of light required by Fork was very high; subsequent work on focusing of the optical stimulation has brought that intensity threshold closer to biologically tolerable limits. Recently, laser illumination was used to directly excite the rat sciatic nerve, during which it was possible to grade the amplitude of the evoked muscle response by the intensity of the light.

An alternative to direct optical stimulation brings in genetic engineering to create light-responsive neurons. Among the approaches being explored are the transformation of normally non-light-responsive cells by genetically linking lightsensitive molecular substituents to ion channels expressed in the cell, and engineering chemical “cages” that in response to illumination release neurotransmitters that excite neurons possessing receptors for the transmitter. These optogenetic tools, possibly linked to a fiber-optic-based system for localized ilumination, offer the promise of highly selective stimulation of target cells and may form the basis for a new generation of neurotechnological solutions for research and medicine.

Cultivating light responsiveness in neurons has a sound biological heritage. The retina of the eye is composed of specialized light-sensitive neurons that can become activated by as little as one photon. In 2002, Boris Zemelman and colleagues used genetic manipulation to make neurons responsive to illumination by expressing genes in cultured hippocampal neurons that coded for elements of the invertebrate retina. The expressed retinal elements produced a light-controlled excitatory current in the affected cells, just as they would in the native retina, although slowly, with latencies of one second to tens of seconds. Firing frequency increased with increases in the light intensity.

This approach was extended by Matthew Banghart at the University of California and colleagues when they developed a potassium channel that opened and closed in response to different wavelengths of light. The channel was joined with a photosensitive molecule (azobenzene) whose conformation determined whether an attached molecule could reach the channel and block it. After expressing the engineered channels in neurons from rat hippocampus, ultraviolet light of 390 nanometers in wavelength caused a conformation change in the azobenzene that removed the channel blocker and allowed the outflow of potassium. Visible green light at 500 nanometers altered the shape of the azobenzene, and blockage of the channel was restored. A photoswitch like this could someday be used to restore photoreceptor-like function in cases of retinitis pigmentosa, a group of conditions in which disorders of the retinal photoreceptors (the light-sensitive rod and cone cells of the retina) lead to blindness. This would be an alternative to recent, highly publicized work in the development of visual prosthetics that capture scenes like a camera and directly stimulate retinal or cortical neurons to replace the signal normally received from photoreceptors.

The power of genetic approaches was seen in 2006 when Anding Bi at Wayne State University and colleagues achieved long-term expression of an algal light-sensitive protein in the inner retinal neurons (nonphotoreceptor cells) of the rat. The transformed neurons responded to illumination as photoreceptors would, with membrane depolarization and light-induced action potentials. The researchers later demonstrated that light stimuli evoked potentials in the visual cortex comparable in amplitude to those evoked in control animals.

These results demonstrate the feasibility of genetically transformed cells as a technique for neural interfacing and potentially for medical rehabilitation. The potential advantages of optical stimulation include selectively activating only targeted neurons, eliminating electrical artifacts that complicate recording of neuronal activity in the presence of electrical stimulation, and eliminating the electrochemical reactions at the interface of electrode tip and tissue that lead to electrode decay and tissue damage. But the challenges are many. The ability to adequately express inserted proteins over the long term is still developing, and doing so in specific subsets of neurons is a daunting task. A fiber-optic -based system for selective neural stimulation holds great appeal, although it will retain some of the problems of invasiveness of electrodes and will be constrained by the same intensity-distance relationship of conventional electrical stimulation, with the robustness of neuronal activation droppping off quickly with the distance from the light source.

Integration of external electronics and the nervous system has immense potential to advance our understanding of the nervous system and the brain and will certainly be transformative for the lives of individuals with compromised neural function due to disease or injury. In the short term, individuals are already regaining some measure of control over lost motor functions and benefiting from better management of maladies like Parkinson’s disease and epilepsy. In the long term, we can envision exquisitely sensitive research and diagnostic tools and neural-electronic hybrid prosthetics that rival the endowments of a sound body.