Skip to main content
NCBI home page
Cold Spring Harbor Perspectives in Medicine logoLink to Cold Spring Harbor Perspectives in Medicine
. 2019 Nov;9(11):a034314. doi: 10.1101/cshperspect.a034314

Closed-Loop Neuromodulation in Physiological and Translational Research

Stavros Zanos 1
PMCID: PMC6824403  PMID: 30559253

Abstract

Neuromodulation, the focused delivery of energy to neural tissue to affect neural or physiological processes, is a common method to study the physiology of the nervous system. It is also successfully used as treatment for disorders in which the nervous system is affected or implicated. Typically, neurostimulation is delivered in open-loop mode (i.e., according to a predetermined schedule and independently of the state of the organ or physiological system whose function is sought to be modulated). However, the physiology of the nervous system or the modulated organ can be dynamic, and the same stimulus may have different effects depending on the underlying state. As a result, open-loop stimulation may fail to restore the desired function or cause side effects. In such cases, a neuromodulation intervention may be preferable to be administered in closed-loop mode. In a closed-loop neuromodulation (CLN) system, stimulation is delivered when certain physiological states or conditions are met (responsive neurostimulation); the stimulation parameters can also be adjusted dynamically to optimize the effect of stimulation in real time (adaptive neurostimulation). In this review, the reasons and the conditions for using CLN are discussed, the basic components of a CLN system are described, and examples of CLN systems used in physiological and translational research are presented.

Typically, in a physiology experiment, the subject is exposed to a set of controlled conditions and interventions, while the investigator takes functional measurements. These measurements are recorded during the experiment to be analyzed at a later time, offline (Fig. 1A). However, in some cases, the timing and other aspects of the intervention need to be linked to one or more physiological events and physiological parameters and tightly controlled, especially when those parameters change rapidly. In such cases, the intervention is delivered on the occurrence of certain physiological states, defined a priori by the investigator and inferred by an automated system that analyzes measurements taken simultaneously, in real time (Fig. 1B). This experimental model can be useful in the study of the nervous system owing to the inherently dynamic nature of neural activity because the same stimulus delivered against a different physiological state may have completely different physiological effects. It can also be used as a method for controlling dynamic neural processes, as well as other physiological functions that are themselves modulated by the nervous system, in a responsive and adaptive manner. In recent years, the term neuromodulation has been adapted to refer to these neural control systems, although in the more traditional usage, neuromodulation is the physiological process by which a neuron uses chemicals to regulate the activity of large, often distant, populations of neurons.

Figure 1.

Figure 1.

Open in a new tab

Principles of open-loop and closed-loop experiments or interventions. (A) In an open-loop context, a predetermined intervention is applied to an animal according to a hypothesis, a set of measurements are taken to characterize the response of the animal to that intervention, and those measurements are analyzed at a later time by the investigator. (B) In a closed-loop context, the investigator starts by defining a set of rules that will determine the conditions at which an intervention will be applied to the animal. An automated system (CPU) observes a set of measurements taken from the animal at regular intervals and delivers the intervention according to the defined rules, in real time.

Here we will discuss the most common reasons for using a closed-loop neuromodulation (CLN) approach and describe the basic components of closed-loop systems. Examples of CLN systems in the context of basic and translational physiological research will be presented. Finally, future directions of this line of research will be discussed.

REASONS AND CONDITIONS FOR THE USE OF CLOSED-LOOP NEUROMODULATION

There are two main reasons why one would want to use a closed-loop approach in a neuromodulation setting.

Need for Responsive Neuromodulation

The requirement for responsive interaction with the nervous system arises when neural processes that depend on precise timing between a physiological event or state and an intervention are studied, or when an intervention needs to happen during a certain physiological state for it to be successful.

In principle, the state dependency of the effects of neurostimulation on a dynamic neural or physiological process could be studied in either of two ways in a given experiment: (1) open-loop delivery of stimuli across different physiological states and registration of physiological effects separately for each state, and (2) closed-loop delivery of stimuli in response to a specific state and registration of the effects for that state only. In cases when the effects of neurostimulation are nonstationary and, more importantly, when they are affected by the history of neurostimulation itself, a closed-loop approach will more accurately address state-dependent effects.

One successful use of responsive CLN systems in physiological research has been in in vivo studies of synaptic plasticity. For spike timing-dependent synaptic potentiation to be induced, the detection of a spontaneous presynaptic action potential needs to be followed by electrical or sensory stimulation that elicits postsynaptic depolarization within a short window of time, typically, <50 msec (Dan and Poo 2004; Jackson et al. 2006a; Nishimura et al. 2013a). There are several examples of CLN systems that, in real time, monitor neural or physiological activity, detect relevant signal signatures in it (e.g., spikes, field potentials or muscle activity), and deliver neurostimulation to successfully induce neural plasticity (Rebesco et al. 2010; Guggenmos et al. 2013; Ethier et al. 2015; Fetz 2015; Oweiss and Badreldin 2015). Such systems have been used to investigate plasticity mechanisms in vivo (e.g., Jackson et al. 2006a; Carrillo-Reid et al. 2016), as well as to facilitate adaptive plasticity after neural injury (Edwardson et al. 2013; Nudo 2014; McPherson et al. 2015). Responsive CLN systems have also been used in experimental efforts to restore disrupted communication between brain regions in the context of a cognitive prosthesis (e.g., Deadwyler et al. 2017), between the brain and the spinal cord (Nishimura et al. 2013b; Zimmermann and Jackson 2014; Capogrosso et al. 2016) or the peripheral nervous system (Moritz et al. 2008; Bouton et al. 2016) in the context of restoration of motor movement is patients with paraplegia or quadriplegia.

On a more translational front, “on-demand” delivery of a neurostimulation-based therapy during certain physiological states, inferred by physiological and other biomarkers, has two main advantages over open-loop delivery: (1) higher probability of attaining desirable, state-specific effects, while minimizing the chance of undesirable side effects, and (2) more efficient operation of the stimulus generator because stimulation happens only when it is needed. For example, brain stimulation delivered through subdural electrodes in response to detection of abnormal brain activity can suppress the onset of epileptic seizures (Ramgopal et al. 2014; Geller et al. 2017). Closed-loop left cervical vagus nerve stimulation (VNS) triggered from seizure-related increases in heart rate (HR) reduces the frequency and severity of seizures more effectively than open-loop VNS (Fisher et al. 2016; Hamilton et al. 2018). Deep brain stimulation (DBS) delivered in response to pathologic brain activity or to the onset of hand tremor is at least as effective as and at least as safe as open-loop DBS, while consuming less power, hence significantly extending the battery life of the implantable generator (Gilat 2018; Kuo et al. 2018). Finally, a number of preliminary human studies suggest that closed-loop stimulation of the auricular branch of the vagus nerve (VN), triggered from the expiratory phase of the respiratory rhythm, which is known to strongly modulate vagal tone, can effectively induce analgesia in individuals with pelvic pain and reduce blood pressure (BP) in hypertensive patients (Napadow et al. 2012; Sclocco et al. 2017).

Need for Adaptive Neuromodulation

The requirement for adaptive neuromodulation arises when a neuromodulation intervention leads to physiological or clinical effects that are not entirely predictable and that need to be monitored for the parameters of the intervention to be optimized with regard to those effects.

In physiological research, an important application of CLN systems is determining stimulus-response characteristics of a sensory neural circuit by the iso-response method (Gollisch and Herz 2012). Iso-response curves are trajectories in the stimulus parameter space that elicit similar neural responses. To explore the parameter space while no significant changes in the neural response occur, a CLN system records and quantifies neural activity in real time, then selects the next stimuli so that the neural response remains on the iso-response curve. This adaptive selection of iso-response stimuli can significantly reduce experiment time (Benda et al. 2007) and reveal nonlinearities in the stimulus–response function, would be missed had the experiment been performed in an open-loop manner (Gollisch and Herz 2012). A proof-of-concept design of an adaptive CLN system has recently been explored in the context of VNS for HR control; in this system, VNS parameters are adjusted in a way that minimizes the difference between an observed physiological variable (i.e., HR) and a desired target value of that variable (Romero-Ugalde et al. 2017).

A more translational example of an adaptive CLN system is closed-loop spinal stimulation that aims to restore locomotion in paralyzed animals by activating sensory and motor spinal circuits; this system is dynamically adapted to maximize the precision or fluidity of the resulting motor movement by monitoring and analyzing the resulting movements themselves, in real time (Wenger et al. 2014). Closed-loop spinal stimulation is also used to relieve pain and associated symptoms in patients with back or leg pain; in this case, the recruitment of fibers in the dorsal column via monitoring of stimulus-evoked compound actional potentials is used to adjust stimulation parameters and maintain stimulation within an individualized therapeutic range (Russo et al. 2018).

In translational closed-loop applications, the choice of biomarkers used in the adaptation process warrants special consideration. When the entire range of physiological, desirable, and undesirable effects is well characterized, then the choice of biomarkers that “reward” and “penalize” a given set of stimulation parameters during the adaptation process is straightforward. However, this is not always the case, as the nonobvious effects of most neuromodulation therapies are incompletely understood, especially because they involve multiple organ systems and at different timescales. This situation is not unlike the unpredictable effects of drug therapies, often recognized years after the therapies are introduced clinically. More comprehensive characterization of the varied effects of neuromodulation therapies in relevant animal models and data collection from as many sensors and clinical and laboratory tests as it is feasible during their clinical application are the best ways to build more confidence in the selection of biomarkers for therapy optimization. Moreover, in clinical applications, the process of adaptation itself ought to be more conservative, because of the potential for undesirable, even catastrophic, events (e.g., Ali et al. 2004). A blind trial-and-error strategy to discover the directions in parameter space that minimize the discrepancy between current and desirable effects may not be ideal in terms of safety; however, it is one of the few options when no robust models relating stimulation parameters and physiological effects exist. In such cases, even small increases in the magnitude of unwanted effects in response to a new set of stimulation parameters could be heavily penalized and the change in stimulation parameters reversed immediately.

The potential clinical application of such adaptive approaches is significant, as it allows subject- and state-specific therapy to be “prescribed” without a priori knowledge of the variable, complex, and inherently dynamic effects, both desired and undesired, of neurostimulation on the target organs.

Conditions for the Use of a CLN System

Three conditions need to be met for a closed-loop approach to be meaningful and successful, at least in principle.

First, the physiology of the target organ and the mechanism of action of the intervention need to be relatively fast. A targeted physiological process that is inherently slow (e.g., the application of electrical fields for accelerating bone fracture or wound healing) is unlikely to benefit from a closed-loop approach that emphasizes rapid action and feedback, unless that is an inherently slow process that relies on fast physiology (e.g., fast synaptic plasticity that underlies learning). For that reason, physiological processes that benefit from CLN interventions include those that are under the modulatory control of the central or peripheral nervous system, such as restoration of motor movement in paralysis (Nishimura et al. 2013b; Wenger et al. 2014; Alam et al. 2016; Ganzer et al. 2018), alleviation of chronic pain (Russo et al. 2018), suppression of epileptic seizures (Ramgopal et al. 2014; Parastarfeizabadi and Kouzani 2017; Thomas and Jobst 2018), augmentation of brain plasticity after neural injury (Hays et al. 2013; Pruitt et al. 2016), improvement of movement deficits in Parkinson's disease ([PD]; Hebb et al. 2014; Meidahl et al. 2017; Parastarfeizabadi and Kouzani 2017), and even psychiatric disease (Lo and Widge 2017).

Second, the feedback signals that inform the CLN system of the relevant aspects of the dynamic state of the target organ need to be representative of that state. For example, although arm accelerometry may be an excellent feedback signal for the detection of the onset or the monitoring of an ongoing epileptic seizure, it may not be as useful for the prediction of an upcoming seizure with the intent of suppressing it before it becomes clinically evident (Ramgopal et al. 2014). In that regard, it is important to define appropriate signals as biomarkers that are reliably quantifiable, track the targeted physiological process in a timescale congruent with its dynamics, and correlate well with the clinical manifestations and the treatment results (Hebb et al. 2014; Thomas and Jobst 2018).

Finally, from a translational perspective, a CLN system ought to be used instead of an open-loop system only when the latter, because of its nonresponsive nature, cannot attain the desired effect or causes unwanted effects that closed-loop stimulation would minimize. Pharmacotherapy of human diseases is a good illustration of this dichotomy. For many diseases, we have a working model of the pathophysiology and its dependency on physiological state; we also have a good understanding of the time course and magnitude of the drug's physiological effects. Consequently, we can come up with a standardized, open-loop, daily delivery schedule that works well when it is adhered to, as is the case with antibiotics in infectious diseases. However, there are numerous diseases in which an intervention is successful only when it is delivered under the right circumstances and potentially deleterious when delivered outside of them: glycemic control in diabetes, heart rhythm control in arrhythmias, control of vascular resistance in hypertension, control of airway resistance in asthma, etc. Given that all these are examples of diseases in which pathogenesis involves the nervous system, closed-loop peripheral neuromodulation would be a meaningful therapeutic approach (Sharma and Weber 2018).

BASIC COMPONENTS OF A CLOSED-LOOP NEUROMODULATION SYSTEM

A CLN system comprises a few basic components: sensors, acquisition system, processing unit, and output device. When the system is chronically implanted, it also includes a case, a power source, and, in some cases, wireless transmission.

Sensors

The set of sensors are needed to obtain physiological measurements from the nervous system or other organ systems. These sensors need to have a relatively fast response time and be able to take repeated measurements, to provide an adequate representation of the dynamic biological system that is being monitored. The outputs of these sensors comprise the physiological signals that the closed-loop system uses to infer the status of the organ or the organism. Many of these sensors are invasive, meaning they require a surgical procedure to be implanted. Typically, these sensors need to be implanted chronically, and they need to be appropriately interfaced with the acquisition system and the rest of the closed-loop system. The surgical techniques and challenges, as well as the special engineering demands associated with chronic, invasive sensors, depend on the type and the anatomical location of the sensor and the connected device and are outside the scope of this work (Arle 2011). Sensors that measure electrical activity of neurons and other excitable cells are inexpensive, readily available, and they can be interfaced with a variety of amplification and acquisition systems. For all those reasons, they represent the first choice for sensing in CLN systems.

Sensors used in neuromodulation systems include:

  • Sensors for electrical neural activity. These are typically conductive elements (microelectrodes, microwires, pads, etc.) placed near the source of activity. Noninvasive sensors are those that are placed on the skin or on the scalp surface (Lopez-Gordo et al. 2014). However, most sensors are invasive as they are implanted subdermally (Young et al. 2006), subdurally (Schalk and Leuthardt 2011), intracortically (Gunasekera et al. 2015), in deep brain structures (Lewis et al. 2016), on the surface of the spine or intraspinally (Tator et al. 2012), or on peripheral nerves (Famm et al. 2013; Rijnbeek et al. 2018). Measuring electrical brain activity has the overall advantage of high temporal resolution (down to a submillisecond scale, if needed); spatial resolution can also be high, albeit only with invasive, high-channel count implants comprising microscale sensors. Invasive sensors generally give rise to better signal-to-noise ratio signals, as they typically lie closer to the source of the electrical signals and tend to pick up less ambient noise (Fig. 2).

  • Sensors for electrical activity of nonneuronal excitable cells. These include sensors for electrocardiography (ECG), electromyography (EMG), electrooculography (EOG), and different forms of electrodermal activity, including the galvanic skin response. These sensors can be noninvasive, typically placed on the skin surface in predetermined locations, or invasive (e.g., subcutaneous or epicardial ECG, intramuscular EMG).

  • Invasive sensors for other physiological measurements. These include pressure sensors placed in vessels or body cavities (e.g., ventricles of the brain), blood flow sensors, temperature sensors, biochemical sensors measuring blood glucose, pH, blood CO2, etc. Although the physical process involved in such measurements is different for different sensors, the output of these sensors is generally an electrical potential that can be registered in real time by a standard acquisition system, just like with neural signals. For that reason, these are also excellent sensors for CLN systems.

  • Noninvasive, wearable, and ambient sensors. This heterogeneous group of sensors includes wearable systems that capture acceleration of the torso, head or limbs, respiration, temperature, oxygen saturation, etc. (Patel et al. 2012). There are also ambient sensor systems that use light sensing, motion sensing, and video to monitor a space in which a patient operates daily. Signals from these sensors require special digital acquisition systems and, therefore, individualized CLN system design.

Figure 2.

Figure 2.

Open in a new tab

Digital signal processing (DSP) and event detection examples in a closed-loop neuromodulation (CLN) system. (A) In the top panel, a short snippet of neuronal activity is shown, acquired from an intracortical microelectrode, after high-pass filtering of the raw electrical signal to isolate fast spiking activity. Red arrows denote spiking events detected by a double time-window discriminator DSP chain operating on the filtered signal in real time, using the Neurochip CLN system (Jackson et al. 2006b). The graphical representation of that DSP chain is shown in the bottom panel: The filtered signal has to cross a voltage threshold, then go through both of the two voltage windows that are set in such a way to detect spikes with a certain waveform (black traces) and ignore threshold crossings with non-spikelike waveforms (gray traces). (B) In the top panel, a snippet of electrical activity from another intracortical microelectrode is shown, this time with no filtering applied; slow oscillatory activity (representing the local field potential [LFP]), as well as fast spiking events riding on top of it are shown. Arrows denote spiking events, just like in panel A; red arrows denote spiking events that occurred during a depolarizing (negative) phase of the oscillatory field potential, and black arrows denote the remaining spiking events. (Middle panel) The discrimination between the two populations of spikes happened through implementation of a second DSP chain, in addition to the spike detection, that gates the acceptance of spikes on negative values of the low-frequency band-filtered field potential (shown in the red trace). (Bottom panel) Average raw signal during the operation of the CLN system, in which single-pulse electrical stimuli were triggered from the accepted (red) spiking events. The triggered stimulus artifact is blanked by the gray vertical line. The average spike waveform that lead to triggering of neurostimulation events is shown just before the artifact (open arrow); that spike is riding on the trough of a slow, oscillatory field potential (filled arrow). This detection system was implemented on the Neurochip-2 CLN system (Zanos et al. 2011).

Acquisition System

The acquisition system amplifies, if needed, and digitizes the output of the sensors, and makes the digitized signals available to the processing unit. In some cases, the acquisition system is embedded in the sensors (active sensors) (Raducanu et al. 2017). The acquisition system may include wireless transmission if the processing unit is physically separated from the acquisition system (Won et al. 2018).

The acquisition system may include analog components (bioamplifiers), when electrical physiological activity that needs to be amplified is monitored (all neural signals, ECG, EMG, etc.). A detailed discussion of the properties of bioamplifiers and how those relate to the different sensor/signal/implant scenarios is beyond the scope of this review (Holleman 2016). Analog circuits for signal preconditioning are sometimes deployed (e.g., ac coupling, notch filtering of 60 Hz noise, low-pass or high-pass filtering, etc.) to ensure signals are within the requirements for digitization.

The analog-to-digital converter (ADC) operates on the analog signals (amplified or not) and converts them to digital signals of appropriate sampling rate, accuracy, and bitrate resolution. Different input signals have different ADC requirements. For example, neuronal spiking activity (Fig. 2) requires a much higher sampling rate than ECG and that, in turn, requires a higher sampling rate than BP signals.

Finally, the acquisition system may be used to suppress stimulation artifacts in neural or physiological recordings, arising from the operation of the stimulation device. Artifacts are orders of magnitude larger than physiological signals and introduce epochs during which no meaningful data can be recorded. Various analog- and digital-based methods have been developed for artifact suppression; however, this issue has not been fully resolved (Erez et al. 2010).

Processing Unit

The digitized signals are streamed, in individual samples or in a packet of more than one sample, to the processing unit, which is essentially a computer. The computer performs, in real time, several functions.

Function 1

It implements digital signal processing (DSP) functions (blue box labeled “A,” Fig. 3). In many cases, the digitized signals need to be further processed for relevant features to be extracted. For example, a voltage threshold crossing followed by a comparison with two consecutive voltage windows is a typical DSP chain that is used to detect neuronal spike waveforms in a signal from a single intracortical microelectrode (Fig. 2A). When the signal meets all these conditions, a spike is detected; the time stamps of occurrence of individual spikes, or the frequency of spiking are common features used in closed-loop systems (Franke et al. 2010). Similar DSP chains can be applied to ECG to extract normal or abnormal QRS complexes (Maheshwari et al. 2013) to arterial BP signals to measure systolic pressure, etc.

Figure 3.

Figure 3.

Open in a new tab

Functions of the processing unit in the context of a closed-loop neuromodulation (CLN) system. (A) The first function of the processing unit is to perform digital signal processing (DSP) and feature extraction on the input signals. These features could be something as simple as the average amplitude of an electromyography signal over a few seconds, or as complex as the exact timing of occurrence of a predetermined abnormally wide QRS complex from an electrocardiogram. (B) The second function is to estimate the physiological state, by using the features extracted from (A), and to compare that state with a set of “intervention rules” to make a decision about the delivery of the intervention. The parameters of the intervention are also programed and can be either fixed or adaptive. (C) The third function is to adapt the intervention parameters in a manner that compensates the deviation of the physiological response to the intervention from a “desired” response. That is performed by comparing the actual response to past interventions, estimated again from input signals and extracted features with the desired response and computing a “response error.” The system, by means of trial and error, learns how to alter the intervention parameters in a way to continuously minimize the response error.

Function 2

The computer combines different features from one or more input signals, estimates the physiological state of the system, and compares that state with a number of preprogramed physiological conditions, which should generate a certain output (the so-called “intervention rules” in the red box labeled “B,” Fig. 3). For example, in Figure 2B, an acceptance trigger is generated when a neuronal spike is detected at an intracortical electrode, while the low-frequency oscillatory component of the field potential is at a depolarizing (negative) voltage. In this experiment for induction of cortical plasticity, only spikes that occur during a depolarizing cortical potential lead to stimulation pulses, a powerful method that allows the in vivo study of the effect of postsynaptic polarization level on synaptic plasticity. Both these functions of the processing unit are related to the first of the two reasons for using a CLN system, namely, the real-time interaction with the nervous system.

Function 3

This relates to optimizing an intervention based on the outcome of preceding interventions (green box labeled “C,” Fig. 3). In this case, the processing unit compares the actual response to a neurostimulation event, as inferred from input signals, with a preprogramed desired response and calculates a so-called “response error.” The system then adjusts the intervention parameters in a direction that is likely to minimize that error in the next neurostimulation event. In essence, this represents a control system that optimizes the neurostimulation parameters with regard to the physiological or clinical effects of neurostimulation (Fig. 3).

Finally, the processing unit may include a local memory bank for storing signals and features, or a wireless communication system for transmitting information to a remote computer for further processing (Gutruf and Rogers 2018). The hardware and software aspects of implementing the functions of the processing unit depend on the number and nature of input signals, the complexity of the implemented DSP chain, whether the state estimation process and intervention rules are fixed or adaptive, the required frequency of computing the response error, the algorithm for minimizing the response error, and the complexity of the intervention parameter space (Denison and Litt 2014).

Output Device

The output device in a CLN system delivers the intervention and it is typically a programmable and triggerable neurostimulator or drug delivery system.

Neurostimulators are devices that deliver targeted energy to neural tissue by means of electrical current (van Dongen and Serdijn 2016), magnetic fields (Malmivuo and Plonsey 1995), ultrasound waves (Bystritsky et al. 2011), or light (Bolus et al. 2018). That energy can either excite or suppress activity of neural tissue, with concomitant effects on physiology. Neural tissues that are typically the target of neurostimulation include the cerebral cortex, deep brain regions, the spinal cord, and peripheral nerves. The energy is delivered through appropriate stimulation probes, placed in the proximity of the neural tissue of interest either invasive or noninvasively. Similar anatomical and surgical principles apply to stimulation probes as with signal sensors and, in fact, many of the sensors discussed above also serve as stimulation probes (Cogan 2008). The closer to neural tissue a probe lies, the smaller the energy required to excite (or inhibit) the tissue; also, the smaller the contact area between the probe and the tissue is, the more focused and, therefore, the more physiologically specific the modulation effect is (McCreery et al. 1986). Two additional considerations for invasive neurostimulation probes are the thermal and electrochemical effects of stimulation, which need to observe strict safety requirements (Merrill et al. 2005). Electromechanical microinfusion pumps and microfluidic probes are chemical delivery systems that can be electrically or remotely triggered or programmed and can locally administer neuroactive agents to neural tissue via an implantable channel (Sim et al. 2017).

EXAMPLES OF CLOSED-LOOP NEUROMODULATION SYSTEMS

In this section, we will discuss in some more detail a few examples of currently used CLN systems, from basic and translational physiology to more mature clinical systems, to showcase some practical implementations of this technology and its applications.

Induction of Neuroplasticity via Closed-Loop Cortical Stimulation

One of the first bidirectional CLN systems that allowed bidirectional interaction with the central and the peripheral nervous system was the Neurochip brain–computer interface (BCI) platform, developed by Eberhard Fetz's group at the University of Washington (Jackson et al. 2006b; Zanos et al. 2011). It features sensing of a variety of neural and behavioral signals, customizable DSP and feature extraction modules, programmable logic for the delivery of neurostimulation, and a multichannel neurostimulation output. The two main applications for the Neurochip BCI are the in vivo study of activity-dependent neural plasticity mechanisms and the restoration of transmission of motor signals across an interrupted neural pathway (Fetz 2015). Both these applications depend on real-time detection of specific signatures of neural activity and delivery of contingent electrical stimulation, continuously during unrestrained behavioral conditions, hence the need for an implantable CLN system.

The Neurochip BCI has been successfully used to induce spike-timing-dependent plasticity (STDP) between motor cortical sites (Jackson et al. 2006a) and between the motor cortex and the spinal cord (Nishimura et al. 2013a) in freely behaving monkeys, using penetrating wire implants, with important implications for motor recovery after stroke or spinal cord injury. Because of the challenges for maintaining stable recordings of spiking activity with penetrating wires over long periods of time, surface cortical and spinal probes have recently attracted attention among basic and translational researchers (Schalk and Leuthardt 2011). In a series of studies, we explored the potential for using surface electrocorticography (ECoG) arrays for recording from and stimulating the cortex of monkeys (Fig. 4A; Zanos 2009, 2013; Zanos et al. 2011, 2018; Rembado et al. 2017). In one of these studies (Zanos et al. 2018), β oscillations (12–25 Hz) in the ECoG were used as a population-level signature of cortical neuronal activity. During these oscillations, cells tend to fire at higher rates at the depolarizing (surface-negative) oscillatory phase, and at lower rates during the hyperpolarizing (surface-positive) phase (Fig. 4B). Using the Neurochip BCI, the depolarizing phase at one cortical site triggered electrical stimulation at a second site, creating the conditions for spike-timing-dependent synaptic potentiation. That led to an increase in the strength of the synaptic projection between the two sites (Fig. 4C). In separate experiments, the hyperpolarizing oscillatory phase of one site triggered stimulation at a second site, and that led to a decrease in the strength of the synaptic projection (Fig. 4C). These findings indicate that induction of bidirectional cortical synaptic plasticity is possible through the operation of a CLN system, using oscillatory signals recorded through minimally invasive neural probes. Although these plasticity effects last only for a few seconds, they do represent activity-dependent synaptic changes that may have a range of implications for the role of cortical oscillations in short-term plasticity, attention and learning, and for their role in movement brain disorders like PD (for a more detailed discussion, see Zanos et al. 2018).

Figure 4.

Figure 4.

Open in a new tab

A closed-loop neuromodulation (CLN) system for induction of cortical synaptic plasticity in awake primates. (A) Schematic diagram of the left hemisphere of a nonhuman primate, with the locations of the recording and stimulation probes chronically implanted epidurally via small burr holes in the skull. Two cortical sites that are synaptically connected are chosen for each experiment. The presence of a synaptic connection between cortical sites is revealed by electrically stimulating one site and recording an elicited neural response at a different site (cortically evoked potential [CEP]), as shown in the lower left inset; the CEP (blue trace) is the average of many individual responses (gray traces). In this case, stimulating the CSTIM site elicited a CEP at the CTRIG site, suggesting a CSTIM→CTRIG synaptic connection. The CLN paradigm aimed at inducing plasticity at the CSTIM→CTRIG synaptic projection by way of recording oscillatory potentials at CTRIG, selecting an oscillatory phase, either depolarizing (negative) or hyperpolarizing (positive), and triggering stimulation at CSTIM at the occurrence of that oscillatory phase in the ongoing signal from CTRIG. In some of these experiments, the Neurochip-2 brain–computer interface (BCI), an implantable CLN device, was used (as shown in top left inset) (Zanos et al. 2011). (B) Example of β-range (15–25 Hz) oscillatory potentials recorded at the cortical sites shown in A, with the corresponding colors. Four cycle-triggered (CT) stimuli were triggered from the depolarizing (negative) phase of the oscillations. Test stimuli (T) were delivered outside of the oscillations, both before and after the burst of CT stimuli, to elicit CEPs and measure the change in strength of the CSTIM→CTRIG synaptic projection that is caused by closed-loop stimulation. (C) (Left panel) When CT stimuli were triggered from the depolarizing phase of oscillations (increased neuronal activity at CTRIG), the size of the CEP after the burst (blue) was larger than before the burst (orange), an indication for synaptic potentiation. (Right panel) When CT stimuli were triggered from the hyperpolarizing phase of oscillations (corresponding to decreased neuronal activity at CTRIG), the size of the CEP after the burst (blue) was smaller than before the burst (orange), an indication for synaptic depression. (From Zanos et al. 2018; reproduced, with permission, from Elsevier © 2018.)

Treatment of Parkinson's Disease via Closed-Loop Deep Brain Stimulation

PD is caused by depletion of the dopamine neurons in the nigrostriatal pathway, resulting in dysregulation of the glutamatergic projection from the striatum to the motor cortex and an abnormal level of oscillatory activity in the reciprocal connections between the motor cortex, the thalamus, and the striatum (Caligiore et al. 2016). Patients with PD experience slowness of motor movement and tremor, among other symptoms, both of which have been correlated with abnormal neuronal activity in those circuits (Stein and Bar-Gad 2013). DBS delivers high-frequency electrical stimulation to the subthalamic nucleus, believed to create a reversible, functional suppression of the circuit, thereby stopping aberrant neuronal activity and alleviating symptoms (Little and Brown 2014; Tinkhauser et al. 2017).

DBS is delivered in an open-loop mode (i.e., in preprogramed “on” and “off” periods), irrespective of the level of neural dysfunction or symptoms. This results in neurological side effects arising from disruption of neuronal communication between other affected circuits. Delivering DBS in closed-loop mode (or adaptively), only when it is needed or when it is maximally efficient, would increase the therapeutic window and reduce the power drain on the battery of the pulse generator (Hebb et al. 2014; Meidahl et al. 2017).

Two types of biomarkers, related to the severity and time course of PD symptoms, can be used to optimize the timing of stimulation in adaptive DBS: brain activity and peripheral motor signals. Brain activity related to PD symptoms can be recorded invasively through the DBS electrode or ECoG electrodes implanted during the procedures, or noninvasively through electroencephalography (EEG) electrodes on the scalp (Morishita and Inoue 2017; Swann et al. 2018). Of these signals, invasively recorded local field potentials (LFPs) are the most widely used in closed-loop DBS systems, typically focusing on the amplitude and phase of oscillatory activity in the subthalamic nucleus or the motor cortex (Parastarfeizabadi and Kouzani 2017; Swann et al. 2018). Peripheral motor activity, captured through EMG electrodes or arm-mounted accelerometers, can be used alone or in combination with brain signals to infer the presence and severity of motor symptoms for adaptive DBS (Parastarfeizabadi and Kouzani 2017). In terms of effectiveness, compared with open-loop DBS, adaptive DBS has been shown to provide an additional clinical improvement of ∑25%–30%, a 20% reduction in side effects and a 40%–55% reduction in stimulation time (Meidahl et al. 2017). Two commercial systems are currently Food and Drug Administration (FDA) approved and can provide closed-loop DBS: the responsive neurostimulation (RNS) System by NeuroPace, which is also used to treat epilepsy through cortical stimulation, and the Activa PC+S system by Medtronic. They both respond with appropriately timed neural stimulation pulses to either brain or peripheral signals indicative of motor symptoms in PD.

Control of Hemodynamic Function via Closed-Loop Vagus Nerve Stimulation

The cardiovascular (CV) system is physiologically regulated by the autonomic nervous system in a responsive, dynamic fashion. In addition, in a variety of conditions, it shows complex time-dependent pathophysiology, with the autonomic nervous system implicated in it. Therefore, as is the case with the brain, it is appropriate to consider treating CV disorders within a CLN framework. Although closed-loop control of heart rhythm using pacing technology has been a mainstay of clinical cardiology, neuromodulation-based control of CV physiology and treatment of CV disorders have only recently received attention. Autonomic, sympathetic, or parasympathetic nerve stimulation has been successfully used in experimental animals to control systemic BP (Plachta et al. 2014), HR (Ardell et al. 2017), atrial fibrillation (Choi et al. 2017), ventricular arrhythmias (Bruegmann et al. 2016), heart failure (Premchand et al. 2014), etc. The main advantage of neuromodulation-based over traditional drug-based therapies in CV diseases is the potential for highly selective modulation of different hemodynamic parameters in a responsive and adaptive manner. However, no CLN system responding to changes in CV physiologic parameters has been implemented yet beyond some theoretical designs (Romero-Ugalde et al. 2017, 2018).

In a proof-of-concept experiment, we used responsive, closed-loop cervical VNS to control systemic BP in rats. Rats anesthetized with isoflurane were instrumented with ECG sensors, a nasal flow sensor, and a BP sensor in the femoral artery, which allowed us to monitor their HR, breathing rate (BR), and arterial BP (Fig. 5). After exposing the carotid sheath at the neck and separating the VN from its vascular elements, we placed a flexible cuff around the left VN and connected the cuff to a rack-mounted stimulator. In constant current mode, trains of monophasic rectangular pulses (100 µsec pulse width, 30 Hz, 300 pulses) were delivered to the VN, in increasing amplitudes, until a physiological response, defined as a stimulus-elicited change in HR, BR or BP, was noted; that amplitude was considered the physiological threshold. Delivery of similar VNS trains at supra-threshold amplitudes was associated with a rapid decrease in HR and in systolic arterial pressure (SAP); a similarly swift return of these parameters to prestimulation levels occurred once VNS was discontinued (Fig. 5A). Once all physiological parameters had stabilized, continuous intravenous administration of norepinephrine, an agent that causes vasoconstriction, was initiated and a gradual increase in SAP was noted (Fig. 5B). At the same time, the stimulator was programed to deliver continuous VNS (constant current, monophasic rectangular pulses, 100 µsec pulse width, 30 Hz, at current amplitude equal to 1.5 times the physiological threshold) while SAP exceeded 150 mmHg; it was also programed to discontinue VNS when SAP decreased below 115 mmHg. While the closed-loop VNS system was in operation, and despite the continuous delivery of norepinephrine, “normal” levels of SAP, within the 115–150 mmHg range, were maintained, without any occurrences of hypotension (Fig. 5B).

Figure 5.

Figure 5.

Open in a new tab

A proof-of-concept, rack-mounted closed-loop neuromodulation (CLN) system for the control of arterial blood pressure via closed-loop vagus nerve stimulation (VNS) in anesthetized rats. (A) Raw physiological measurements before, during, and after a short train of VNS. A rat was anesthetized with isoflurane and instrumented with a nasal temperature sensor to register nasal air flow (dark green) and calculate breathing rate ([BR]; magenta), a skin patch mounted on the chest to register electrocardiography (ECG) (blue) and calculate heart rate ([HR]; light green), and an intravascular pressure sensor in the femoral artery to register systemic arterial pressure (AP; yellow). The trunk of the left vagus nerve (VN) was surgically exposed at the level of the neck and a bipolar cuff electrode was placed on it. A 10-sec-long train of VNS, represented by the rectangular purple trace, was delivered: 300 monophasic square pulses of 100 µsec pulse width and 150 µA intensity, at 30 Hz pulsing frequency. VNS produced a decrease in HR and in arterial, both systolic and diastolic, pressure. All physiologic parameters quickly returned to prestimulation levels after the end of VNS. (B) Example of the system operating in closed-loop mode. A gradual increase in systolic arterial pressure (SAP) was accomplished by intravenous infusion of norepinephrine (NE), a vasoconstrictive agent. Continuous VNS delivery (100 µsec pulse width, 150 µA amplitude) was gated by an increase in SAP >150 mmHg and stopped when SAP decreased <115 mmHg. Once SAP rose beyond 150 mmHg, VNS was initiated, resulting in a quick decrease in SAP. Once SAP decreased <115 mmHg, VNS was turned off and SAP started increasing again. (C) Conceptual architecture of a CLN system for controlling BP in a closed-loop manner, based on the diagram in Figure 3. The system continuously monitors HR, SAP, and BR, by recording ECG, AP, and nasal air flow. The three parameters are used to calculate a physiological score (S), that is proportional to increases in SAP beyond a certain level (SAP0) and inversely proportional to decreases in HR and BR below a “safe” level (HR0 and BR0, respectively). When the score S increases beyond a predefined level, VNS is turned on; when it decreases below that level, VNS is turned off. This is the conceptual basis of a responsive, adaptive treatment of increased SAP that is sensitive to undesirable effects of the neuromodulation treatment (decreases in HR and BR).

Although there is evidence that VNS causes direct vasodilation in different vascular beds (McMahon et al. 1992; Feliciano and Henning 1998), which would explain the elicited decrease in SAP, the mechanism of action in this case likely also involves a decrease in HR, reduction in cardiac contractility, inhibition of sympathetic tone, reflexive changes in the central autonomic drive as a result of afferent vagal activation, and, finally, extraneural hemodynamic effects from the spread of current beyond the VN to surrounding muscles and vessels. Taken a step further, a CLN system that also takes into account additional undesirable effects of VNS (e.g., significant decreases in HR or in BR) could control AP in a safer manner (Fig. 5C).

CHALLENGES AND OPPORTUNITIES FOR CLOSED-LOOP NEUROMODULATION

There are several challenges for the wider use of CLN systems in basic and translational physiology and, more importantly, in clinical medicine. First, there is urgent need for understanding the anatomy and physiology of the central and peripheral circuits involved in physiological processes and disorders; that is, a prerequisite for selecting the correct biomarkers, control algorithms, and targets for neurostimulation. Several Brain Research through Advancing Innovative Neurotechnologies (BRAIN) initiative and National Institutes of Health Stimulating Peripheral Activity to Relieve Conditions (NIH-SPARC) funding opportunities have been addressing this gap for the past few years in the United States. Second, additional progress is needed in the fabrication of tissue-friendly sensors, stimulation probes, and implantable generators as well as in the design and implementation of energy-efficient and computationally powerful processors that able to handle more complex detection and optimization algorithms. Third, the specialized physician training requirements and the regulatory steps involved in bringing such devices into the market are more complex and expensive to navigate compared with open-loop stimulators, which make commercialization efforts riskier (Meidahl et al. 2017). However, as scientists, engineers, and physicians continue to define the principles, methods, and applications for CLN systems in physiological and translational research, in animal models, and in human subjects, it is expected that CLN systems will comprise a significant portion of the growth forecasted for the neuromodulation market, from approximately $2.8 billion in 2016 to more than $7 billion in 2025 (Accuray Research 2018).

Footnotes

Editors: Valentin A. Pavlov and Kevin J. Tracey

Additional Perspectives on Bioelectronic Medicine available at www.perspectivesinmedicine.org

REFERENCES

  1. Accuray Research. 2018. Global neuromodulation devices market analysis and trends—Industry forecast to 2025. Accuray Research, Noida, Uttar Pradesh, India. [Google Scholar]
  2. Alam M, Rodrigues W, Pham BN, Thakor NV. 2016. Brain–machine interface facilitated neurorehabilitation via spinal stimulation after spinal cord injury: Recent progress and future perspectives. Brain Res 1646: 25–33. 10.1016/j.brainres.2016.05.039 [DOI] [PubMed] [Google Scholar]
  3. Ali II, Pirzada NA, Kanjwal Y, Wannamaker B, Medhkour A, Koltz MT, Vaughn BV. 2004. Complete heart block with ventricular asystole during left vagus nerve stimulation for epilepsy. Epilepsy Behav 5: 768–771. 10.1016/j.yebeh.2004.05.008 [DOI] [PubMed] [Google Scholar]
  4. Ardell JL, Nier H, Hammer M, Southerland EM, Ardell CL, Beaumont E, KenKnight BH, Armour JA. 2017. Defining the neural fulcrum for chronic vagus nerve stimulation: Implications for integrated cardiac control. J Physiol 595: 6887–6903. 10.1113/JP274678 [DOI] [PMC free article] [PubMed] [Google Scholar]
  5. Arle J. 2011. Surgical techniques in neuromodulation. In Essential neuromodulation (ed. Arle J, Shils JL), pp. 269–282. Elsevier, Amsterdam. [Google Scholar]
  6. Benda J, Gollisch T, Machens CK, Herz AVM. 2007. From response to stimulus: Adaptive sampling in sensory physiology. Curr Opin Neurobiol 17: 430–436. 10.1016/j.conb.2007.07.009 [DOI] [PubMed] [Google Scholar]
  7. Bolus MF, Willats AA, Whitmire CJ, Rozell CJ, Stanley GB. 2018. Design strategies for dynamic closed-loop optogenetic neurocontrol in vivo. J Neural Eng 15: 026011 10.1088/1741-2552/aaa506 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Bouton CE, Shaikhouni A, Annetta NV, Bockbrader MA, Friedenberg DA, Nielson DM, Sharma G, Sederberg PB, Glenn BC, Mysiw WJ, et al. 2016. Restoring cortical control of functional movement in a human with quadriplegia. Nature 533: 247–250. 10.1038/nature17435 [DOI] [PubMed] [Google Scholar]
  9. Bruegmann T, Boyle PM, Vogt CC, Karathanos TV, Arevalo HJ, Fleischmann BK, Trayanova NA, Sasse P. 2016. Optogenetic defibrillation terminates ventricular arrhythmia in mouse hearts and human simulations. J Clin Invest 126: 3894–3904. 10.1172/JCI88950 [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Bystritsky A, Korb AS, Douglas PK, Cohen MS, Melega WP, Mulgaonkar AP, DeSalles A, Min BK, Yoo SS. 2011. A review of low-intensity focused ultrasound pulsation. Brain Stimul 4: 125–136. 10.1016/j.brs.2011.03.007 [DOI] [PubMed] [Google Scholar]
  11. Caligiore D, Helmich RC, Hallett M, Moustafa AA, Timmermann L, Toni I, Baldassarre G. 2016. Parkinson's disease as a system-level disorder. NPJ Parkinsons Dis 2: 16025 10.1038/npjparkd.2016.25 [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Capogrosso M, Milekovic T, Borton D, Wagner F, Moraud EM, Mignardot JB, Buse N, Gandar J, Barraud Q, Xing D, et al. 2016. A brain–spine interface alleviating gait deficits after spinal cord injury in primates. Nature 539: 284–288. 10.1038/nature20118 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Carrillo-Reid L, Yang W, Bando Y, Peterka DS, Yuste R. 2016. Imprinting and recalling cortical ensembles. Science 353: 691–694. 10.1126/science.aaf7560 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Choi EK, Zhao Y, Everett TH, Chen PS. 2017. Ganglionated plexi as neuromodulation targets for atrial fibrillation. J Cardiovasc Electrophysiol 28: 1485–1491. 10.1111/jce.13319 [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Cogan SF. 2008. Neural stimulation and recording electrodes. Annu Rev Biomed Eng 10: 275–309. 10.1146/annurev.bioeng.10.061807.160518 [DOI] [PubMed] [Google Scholar]
  16. Dan Y, Poo MM. 2004. Spike timing-dependent plasticity of neural circuits. Neuron 44: 23–30. 10.1016/j.neuron.2004.09.007 [DOI] [PubMed] [Google Scholar]
  17. Deadwyler SA, Hampson RE, Song D, Opris I, Gerhardt GA, Marmarelis VZ, Berger TW. 2017. A cognitive prosthesis for memory facilitation by closed-loop functional ensemble stimulation of hippocampal neurons in primate brain. Exp Neurol 287: 452–460. 10.1016/j.expneurol.2016.05.031 [DOI] [PMC free article] [PubMed] [Google Scholar]
  18. Denison T, Litt B. 2014. Advancing neuromodulation through control systems: A general framework and case study in posture-responsive stimulation. Neuromodulation 17 (Suppl. 1): 48–57. 10.1111/ner.12170 [DOI] [PubMed] [Google Scholar]
  19. Edwardson MA, Lucas TH, Carey JR, Fetz EE. 2013. New modalities of brain stimulation for stroke rehabilitation. Exp Brain Res 224: 335–358. 10.1007/s00221-012-3315-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  20. Erez Y, Tischler H, Moran A, Bar-Gad I. 2010. Generalized framework for stimulus artifact removal. J Neurosci Methods 191: 45–59. 10.1016/j.jneumeth.2010.06.005 [DOI] [PubMed] [Google Scholar]
  21. Ethier C, Gallego JA, Miller LE. 2015. Brain-controlled neuromuscular stimulation to drive neural plasticity and functional recovery. Curr Opin Neurobiol 33: 95–102. 10.1016/j.conb.2015.03.007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Famm K, Litt B, Tracey KJ, Boyden ES, Slaoui M. 2013. Drug discovery: A jump-start for electroceuticals. Nature 496: 159–161. 10.1038/496159a [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Feliciano L, Henning RJ. 1998. Vagal nerve stimulation releases vasoactive intestinal peptide which significantly increases coronary artery blood flow. Cardiovasc Res 40: 45–55. 10.1016/S0008-6363(98)00122-9 [DOI] [PubMed] [Google Scholar]
  24. Fetz EE. 2015. Restoring motor function with bidirectional neural interfaces. Prog Brain Res 218: 241–252. 10.1016/bs.pbr.2015.01.001 [DOI] [PubMed] [Google Scholar]
  25. Fisher RS, Afra P, Macken M, Minecan DN, Bagic A, Benbadis SR, Helmers SL, Sinha SR, Slater J, Treiman D, et al. 2016. Automatic vagus nerve stimulation triggered by ictal tachycardia: Clinical outcomes and device performance—The U.S. E-37 Trial. Neuromodulation 19: 188–195. 10.1111/ner.12376 [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Franke F, Natora M, Boucsein C, Munk MHJ, Obermayer K. 2010. An online spike detection and spike classification algorithm capable of instantaneous resolution of overlapping spikes. J Comput Neurosci 29: 127–148. 10.1007/s10827-009-0163-5 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Ganzer PD, Darrow MJ, Meyers EC, Solorzano BR, Ruiz AD, Robertson NM, Adcock KS, James JT, Jeong HS, Becker AM, et al. 2018. Closed-loop neuromodulation restores network connectivity and motor control after spinal cord injury. eLife 7: e32058 10.7554/eLife.32058 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Geller EB, Skarpaas TL, Gross RE, Goodman RR, Barkley GL, Bazil CW, Berg MJ, Bergey GK, Cash SS, Cole AJ, et al. 2017. Brain-responsive neurostimulation in patients with medically intractable mesial temporal lobe epilepsy. Epilepsia 58: 994–1004. 10.1111/epi.13740 [DOI] [PubMed] [Google Scholar]
  29. Gilat M. 2018. Towards closed-loop deep brain stimulation for freezing of gait in Parkinson's disease. Clin Neurophysiol 129: 2448–2450. 10.1016/j.clinph.2018.08.012 [DOI] [PubMed] [Google Scholar]
  30. Gollisch T, Herz AVM. 2012. The iso-response method: Measuring neuronal stimulus integration with closed-loop experiments. Front Neural Circuits 6: 104 10.3389/fncir.2012.00104 [DOI] [PMC free article] [PubMed] [Google Scholar]
  31. Guggenmos DJ, Azin M, Barbay S, Mahnken JD, Dunham C, Mohseni P, Nudo RJ. 2013. Restoration of function after brain damage using a neural prosthesis. Proc Natl Acad Sci 110: 21177–21182. 10.1073/pnas.1316885110 [DOI] [PMC free article] [PubMed] [Google Scholar]
  32. Gunasekera B, Saxena T, Bellamkonda R, Karumbaiah L. 2015. Intracortical recording interfaces: Current challenges to chronic recording function. ACS Chem Neurosci 6: 68–83. 10.1021/cn5002864 [DOI] [PubMed] [Google Scholar]
  33. Gutruf P, Rogers JA. 2018. Implantable, wireless device platforms for neuroscience research. Curr Opin Neurobiol 50: 42–49. 10.1016/j.conb.2017.12.007 [DOI] [PubMed] [Google Scholar]
  34. Hamilton P, Soryal I, Dhahri P, Wimalachandra W, Leat A, Hughes D, Toghill N, Hodson J, Sawlani V, Hayton T, et al. 2018. Clinical outcomes of VNS therapy with AspireSR (including cardiac-based seizure detection) at a large complex epilepsy and surgery centre. Seizure 58: 120–126. 10.1016/j.seizure.2018.03.022 [DOI] [PubMed] [Google Scholar]
  35. Hays SA, Rennaker RL, Kilgard MP. 2013. Targeting plasticity with vagus nerve stimulation to treat neurological disease. Prog Brain Res 207: 275–299. 10.1016/B978-0-444-63327-9.00010-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  36. Hebb AO, Zhang JJ, Mahoor MH, Tsiokos C, Matlack C, Chizeck HJ, Pouratian N. 2014. Creating the feedback loop: Closed-loop neurostimulation. Neurosurg Clin N Am 25: 187–204. 10.1016/j.nec.2013.08.006 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Holleman J. 2016. Design considerations for neural amplifiers. Conf Proc IEEE Eng Med Biol Soc 2016: 6331–6334. [DOI] [PubMed] [Google Scholar]
  38. Jackson A, Mavoori J, Fetz EE. 2006a. Long-term motor cortex plasticity induced by an electronic neural implant. Nature 444: 56–60. 10.1038/nature05226 [DOI] [PubMed] [Google Scholar]
  39. Jackson A, Moritz CT, Mavoori J, Lucas TH, Fetz EE. 2006b. The neurochip BCI: Towards a neural prosthesis for upper limb function. IEEE Trans Neural Syst Rehabil Eng 14: 187–190. 10.1109/TNSRE.2006.875547 [DOI] [PubMed] [Google Scholar]
  40. Kuo CH, White-Dzuro GA, Ko AL. 2018. Approaches to closed-loop deep brain stimulation for movement disorders. Neurosurg Focus 45: E2. [DOI] [PubMed] [Google Scholar]
  41. Lewis PM, Thomson RH, Rosenfeld JV, Fitzgerald PB. 2016. Brain neuromodulation techniques: A review. Neuroscientist 22: 406–421. 10.1177/1073858416646707 [DOI] [PubMed] [Google Scholar]
  42. Little S, Brown P. 2014. The functional role of β oscillations in Parkinson's disease. Parkinsonism Relat Disord 20 (Suppl. 1): S44–S48. 10.1016/S1353-8020(13)70013-0 [DOI] [PubMed] [Google Scholar]
  43. Lo MC, Widge AS. 2017. Closed-loop neuromodulation systems: Next-generation treatments for psychiatric illness. Int Rev Psychiatry 29: 191–204. 10.1080/09540261.2017.1282438 [DOI] [PMC free article] [PubMed] [Google Scholar]
  44. Lopez-Gordo MA, Sanchez-Morillo D, Pelayo Valle F. 2014. Dry EEG electrodes. Sensors (Basel) 14: 12847–12870. 10.3390/s140712847 [DOI] [PMC free article] [PubMed] [Google Scholar]
  45. Maheshwari S, Acharyya A, Puddu PE, Mazomenos EB, Leekha G, Maharatna K, Schiariti M. 2013. An automated algorithm for online detection of fragmented QRS and identification of its various morphologies. J R Soc Interface 10: 20130761 10.1098/rsif.2013.0761 [DOI] [PMC free article] [PubMed] [Google Scholar]
  46. Malmivuo J, Plonsey R. 1995. Magnetic stimulation of neural tissue. In Bioelectromagnetism—Principles and applications of bioelectric and biomagnetic fields, pp. 375–381. Oxford University Press, New York. [Google Scholar]
  47. McCreery DB, Bullara LA, Agnew WF. 1986. Neuronal activity evoked by chronically implanted intracortical microelectrodes. Exp Neurol 92: 147–161. 10.1016/0014-4886(86)90131-7 [DOI] [PubMed] [Google Scholar]
  48. McMahon TJ, Hood JS, Kadowitz PJ. 1992. Pulmonary vasodilator response to vagal stimulation is blocked by Nω-nitro-l-arginine methyl ester in the cat. Circ Res 70: 364–369. 10.1161/01.RES.70.2.364 [DOI] [PubMed] [Google Scholar]
  49. McPherson JG, Miller RR, Perlmutter SI. 2015. Targeted, activity-dependent spinal stimulation produces long-lasting motor recovery in chronic cervical spinal cord injury. Proc Natl Acad Sci 112: 12193–12198. 10.1073/pnas.1505383112 [DOI] [PMC free article] [PubMed] [Google Scholar]
  50. Meidahl AC, Tinkhauser G, Herz DM, Cagnan H, Debarros J, Brown P. 2017. Adaptive deep brain stimulation for movement disorders: The long road to clinical therapy. Mov Disord 32: 810–819. 10.1002/mds.27022 [DOI] [PMC free article] [PubMed] [Google Scholar]
  51. Merrill DR, Bikson M, Jefferys JG. 2005. Electrical stimulation of excitable tissue: Design of efficacious and safe protocols. J Neurosci Methods 141: 171–198. 10.1016/j.jneumeth.2004.10.020 [DOI] [PubMed] [Google Scholar]
  52. Morishita T, Inoue T. 2017. Need for multiple biomarkers to adjust parameters of closed-loop deep brain stimulation for Parkinson's disease. Neural Regen Res 12: 747–748. 10.4103/1673-5374.206642 [DOI] [PMC free article] [PubMed] [Google Scholar]
  53. Moritz CT, Perlmutter SI, Fetz EE. 2008. Direct control of paralysed muscles by cortical neurons. Nature 456: 639–642. 10.1038/nature07418 [DOI] [PMC free article] [PubMed] [Google Scholar]
  54. Napadow V, Edwards RR, Cahalan CM, Mensing G, Greenbaum S, Valovska A, Li A, Kim J, Maeda Y, Park K, et al. 2012. Evoked pain analgesia in chronic pelvic pain patients using respiratory-gated auricular vagal afferent nerve stimulation. Pain Med 13: 777–789. 10.1111/j.1526-4637.2012.01385.x [DOI] [PMC free article] [PubMed] [Google Scholar]
  55. Nishimura Y, Perlmutter SI, Eaton RW, Fetz EE. 2013a. Spike-timing-dependent plasticity in primate corticospinal connections induced during free behavior. Neuron 80: 1301–1309. 10.1016/j.neuron.2013.08.028 [DOI] [PMC free article] [PubMed] [Google Scholar]
  56. Nishimura Y, Perlmutter SI, Fetz EE. 2013b. Restoration of upper limb movement via artificial corticospinal and musculospinal connections in a monkey with spinal cord injury. Front Neural Circuits 7: 57 10.3389/fncir.2013.00057 [DOI] [PMC free article] [PubMed] [Google Scholar]
  57. Nudo R. 2014. Plasticity of cerebral motor functions: Implications for repair and rehabilitation. In Textbook of neural repair and rehabilitation, 2nd ed. (ed. Selzer ME, et al. ), pp. 99–113. Cambridge University Press, Cambridge. [Google Scholar]
  58. Oweiss KG, Badreldin IS. 2015. Neuroplasticity subserving the operation of brain–machine interfaces. Neurobiol Dis 83: 161–171. 10.1016/j.nbd.2015.05.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
  59. Parastarfeizabadi M, Kouzani AZ. 2017. Advances in closed-loop deep brain stimulation devices. J Neuroeng Rehabil 14: 79 10.1186/s12984-017-0295-1 [DOI] [PMC free article] [PubMed] [Google Scholar]
  60. Patel S, Park H, Bonato P, Chan L, Rodgers M. 2012. A review of wearable sensors and systems with application in rehabilitation. J Neuroeng Rehabil 9: 21 10.1186/1743-0003-9-21 [DOI] [PMC free article] [PubMed] [Google Scholar]
  61. Plachta DT, Gierthmuehlen M, Cota O, Espinosa N, Boeser F, Herrera TC, Stieglitz T, Zentner J. 2014. Blood pressure control with selective vagal nerve stimulation and minimal side effects. J Neural Eng 11: 036011 10.1088/1741-2560/11/3/036011 [DOI] [PubMed] [Google Scholar]
  62. Premchand RK, Sharma K, Mittal S, Monteiro R, Dixit S, Libbus I, DiCarlo LA, Ardell JL, Rector TS, Amurthur B, et al. 2014. Autonomic regulation therapy via left or right cervical vagus nerve stimulation in patients with chronic heart failure: Results of the ANTHEM-HF trial. J Card Fail 20: 808–816. 10.1016/j.cardfail.2014.08.009 [DOI] [PubMed] [Google Scholar]
  63. Pruitt DT, Schmid AN, Kim LJ, Abe CM, Trieu JL, Choua C, Hays SA, Kilgard MP, Rennaker RL. 2016. Vagus nerve stimulation delivered with motor training enhances recovery of function after traumatic brain injury. J Neurotrauma 33: 871–879. 10.1089/neu.2015.3972 [DOI] [PMC free article] [PubMed] [Google Scholar]
  64. Raducanu BC, Yazicioglu RF, Lopez CM, Ballini M, Putzeys J, Wang S, Andrei A, Rochus V, Welkenhuysen M, van Helleputte N, et al. 2017. Time multiplexed active neural probe with 1356 parallel recording sites. Sensors (Basel) 17: E23, 88 10.3390/s17102388 [DOI] [PMC free article] [PubMed] [Google Scholar]
  65. Ramgopal S, Thome-Souza S, Jackson M, Kadish NE, Sánchez Fernández I, Klehm J, Bosl W, Reinsberger C, Schachter S, Loddenkemper T. 2014. Seizure detection, seizure prediction, and closed-loop warning systems in epilepsy. Epilepsy Behav 37: 291–307. 10.1016/j.yebeh.2014.06.023 [DOI] [PubMed] [Google Scholar]
  66. Rebesco JM, Stevenson IH, Kording KP, Solla SA, Miller LE. 2010. Rewiring neural interactions by micro-stimulation. Front Syst Neurosci 4: 39 10.3389/fnsys.2010.00039 [DOI] [PMC free article] [PubMed] [Google Scholar]
  67. Rembado I, Zanos S, Fetz EE. 2017. Cycle-triggered cortical stimulation during slow wave sleep facilitates learning a BMI task: A case report in a non-human primate. Front Behav Neurosci 11: 59 10.3389/fnbeh.2017.00059 [DOI] [PMC free article] [PubMed] [Google Scholar]
  68. Rijnbeek EH, Eleveld N, Olthuis W. 2018. Update on peripheral nerve electrodes for closed-loop neuroprosthetics. Front Neurosci 12: 350 10.3389/fnins.2018.00350 [DOI] [PMC free article] [PubMed] [Google Scholar]
  69. Romero-Ugalde HM, Le Rolle V, Bonnet JL, Henry C, Bel A, Mabo P, Carrault G, Hernández AI. 2017. A novel controller based on state-transition models for closed-loop vagus nerve stimulation: Application to heart rate regulation. PLoS ONE 12: e0186068. [DOI] [PMC free article] [PubMed] [Google Scholar]
  70. Romero-Ugalde HM, Le Rolle V, Bonnet JL, Henry C, Mabo P, Carrault G, Hernández AI. 2018. Closed-loop vagus nerve stimulation based on state transition models. IEEE Trans Biomed Eng 65: 1630–1638. 10.1109/TBME.2017.2759667 [DOI] [PubMed] [Google Scholar]
  71. Russo M, Cousins MJ, Brooker C, Taylor N, Boesel T, Sullivan R, Poree L, Shariati NH, Hanson E, Parker J. 2018. Effective relief of pain and associated symptoms with closed-loop spinal cord stimulation system: Preliminary results of the Avalon Study. Neuromodulation 21: 38–47. 10.1111/ner.12684 [DOI] [PubMed] [Google Scholar]
  72. Schalk G, Leuthardt EC. 2011. Brain–computer interfaces using electrocorticographic signals. IEEE Rev Biomed Eng 4: 140–154. 10.1109/RBME.2011.2172408 [DOI] [PubMed] [Google Scholar]
  73. Sclocco R, Garcia RG, Gabriel A, Kettner NW, Napadow V, Barbieri R. 2017. Respiratory-gated auricular vagal afferent nerve stimulation (RAVANS) effects on autonomic outflow in hypertension. Conf Proc IEEE Eng Med Biol Soc 2017: 3130–3133. [DOI] [PubMed] [Google Scholar]
  74. Sharma G, Weber D. 2018. Closing the loop for bioelectronic medicine. MedDevice Online, www.meddeviceonline.com.
  75. Sim JY, Haney MP, Park SI, McCall JG, Jeong JW. 2017. Microfluidic neural probes: In vivo tools for advancing neuroscience. Lab Chip 17: 1406–1435. 10.1039/C7LC00103G [DOI] [PubMed] [Google Scholar]
  76. Stein E, Bar-Gad I. 2013. β oscillations in the cortico-basal ganglia loop during parkinsonism. Exp Neurol 245: 52–59. 10.1016/j.expneurol.2012.07.023 [DOI] [PubMed] [Google Scholar]
  77. Swann NC, de Hemptinne C, Thompson MC, Miocinovic S, Miller AM, Gilron R, Ostrem JL, Chizeck HJ, Starr PA. 2018. Adaptive deep brain stimulation for Parkinson's disease using motor cortex sensing. J Neural Eng 15: 046006 10.1088/1741-2552/aabc9b [DOI] [PMC free article] [PubMed] [Google Scholar]
  78. Tator CH, Minassian K, Mushahwar VK. 2012. Spinal cord stimulation: Therapeutic benefits and movement generation after spinal cord injury. Handb Clin Neurol 109: 283–296. 10.1016/B978-0-444-52137-8.00018-8 [DOI] [PubMed] [Google Scholar]
  79. Thomas GP, Jobst BC. 2018. Feedback-sensitive and closed-loop solutions. In Innovative neuromodulation (ed. Arle J, Shils J), pp. 41–59. Elsevier, Amsterdam. [Google Scholar]
  80. Tinkhauser G, Pogosyan A, Little S, Beudel M, Herz DM, Tan H, Brown P. 2017. The modulatory effect of adaptive deep brain stimulation on β bursts in Parkinson's disease. Brain 140: 1053–1067. 10.1093/brain/awx010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  81. van Dongen M, Serdijn W. 2016. Design of efficient and safe neural stimulators: A multidisciplinary approach. Springer, New York. [Google Scholar]
  82. Wenger N, Moraud EM, Raspopovic S, Bonizzato M, DiGiovanna J, Musienko P, Morari M, Micera S, Courtine G. 2014. Closed-loop neuromodulation of spinal sensorimotor circuits controls refined locomotion after complete spinal cord injury. Sci Transl Med 6: 255ra133 10.1126/scitranslmed.3008325 [DOI] [PubMed] [Google Scholar]
  83. Won SM, Song E, Zhao J, Li J, Rivnay J, Rogers JA. 2018. Recent advances in materials, devices, and systems for neural interfaces. Adv Mater 30: e1800534 10.1002/adma.201800534 [DOI] [PubMed] [Google Scholar]
  84. Young GB, Ives JR, Chapman MG, Mirsattari SM. 2006. A comparison of subdermal wire electrodes with collodion-applied disk electrodes in long-term EEG recordings in ICU. Clin Neurophysiol 117: 1376–1379. 10.1016/j.clinph.2006.02.006 [DOI] [PubMed] [Google Scholar]
  85. Zanos S. 2009. Neural correlates of high-frequency intracortical and epicortical field potentials. J Neurosci 29: 3673–3675. 10.1523/jneurosci.0009-09.2009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  86. Zanos S. 2013. “Cortical surface recurrent brain-computer interfaces.” PhD thesis, University of Washington, Seattle. [Google Scholar]
  87. Zanos S, Richardson AG, Shupe L, Miles FP, Fetz EE. 2011. The Neurochip-2: An autonomous head-fixed computer for recording and stimulating in freely behaving monkeys. IEEE Trans Neural Syst Rehabil Eng 19: 427–435. 10.1109/tnsre.2011.2158007 [DOI] [PMC free article] [PubMed] [Google Scholar]
  88. Zanos S, Rembado I, Chen D, Fetz EE. 2018. Phase-locked stimulation during cortical β oscillations produces bidirectional synaptic plasticity in awake monkeys. Curr Biol 28: 2515–2526.e4. 10.1016/j.cub.2018.07.009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  89. Zimmermann JB, Jackson A. 2014. Closed-loop control of spinal cord stimulation to restore hand function after paralysis. Front Neurosci 8: 87 10.3389/fnins.2014.00087 [DOI] [PMC free article] [PubMed] [Google Scholar]

Articles from Cold Spring Harbor Perspectives in Medicine are provided here courtesy of Cold Spring Harbor Laboratory Press

RESOURCES