The Emerging Role of Biomarkers in Adaptive Modulation of Clinical Brain Stimulation

Abstract

Therapeutic brain stimulation has proven efficacious for treatment of nervous system diseases, exerting widespread influence via disease-specific neural networks. Activation or suppression of neural networks could theoretically be assessed by either clinical symptom modification (ie, tremor, rigidity, seizures) or development of specific biomarkers linked to treatment of symptomatic disease states. For example, biomarkers indicative of disease state could aid improved intraoperative localization of electrode position, optimize device efficacy or efficiency through dynamic control, and eventually serve to guide automatic adjustment of stimulation settings. Biomarkers to control either extracranial or intracranial stimulation span from continuous physiological brain activity, intermittent pathological activity, and triggered local phenomena or potentials, to wearable devices, blood flow, biochemical or cardiac signals, temperature perturbations, optical or magnetic resonance imaging changes, or optogenetic signals. The goal of this review is to update new approaches to implement control of stimulation through relevant biomarkers. Critical questions include whether adaptive systems adjusted through biomarkers can optimize efficiency and eventually efficacy, serve as inputs for stimulation adjustment, and consequently broaden our fundamental understanding of abnormal neural networks in pathologic states. Neurosurgeons are at the forefront of translating and developing biomarkers embedded within improved brain stimulation systems. Thus, criteria for developing and validating biomarkers for clinical use are important for the adaptation of device approaches into clinical practice.

ABBREVIATIONS

AD

Alzheimer's disease

aDBS

adaptive DBS

BCI/BMI

brain computer/machine interface

DBS

Deep brain stimulation

ECoG

electrocorticography

EEG

electroencephalogram

EMG

electromyogram

ECAP

evoked compound action potential

FOG

freezing of gait

GPi

globus pallidus internus

IPG

internal pulse generator

LFP

local field potential

PD

Parkinson disease

PAC

phase amplitude coupling

SCS

spinal cord stimulation

SCC

subcallosal cingulate

STN

subthalamic nucleus

TMS

transcranial magnetic stimulation

TBI

traumatic brain injury

TRD

treatment resistant depression

VNS

vagal nerve stimulatorstimulation

Vim

ventral intermediate nucleus of the thalamus

Both extracranial and intracranial brain stimulation can exert influence over multiple brain areas through modulation of disease- and patient-specific neural networks.^1-4 Placement of intracranial brain stimulation and recording electrodes is currently executed via preoperative recordings and detailed imaging. In contrast, extracranial stimulation can be either diffusely applied to the skull or targeted to engage specific brain regions, such as in transcranial alternating current and transcranial magnetic stimulation (TMS),^5-7 or lie outside of the cranium, such as vagal nerve stimulation (VNS)⁸ and spinal cord stimulation (SCS). Engagement of appropriate circuitry and treatment efficacy in all these forms of brain stimulation is typically verified by clinical assessment and symptom reduction as well as direct electrophysiological recordings with stimulation.

Clinical biomarkers or endpoints may also provide objective endpoints to therapy or elucidate why patients with optimal electrode position still have such variable clinical outcomes, including development of dose-response functions.⁹ The current clinical approach to programming stimulation parameters requires significant time and effort from both patient and clinician and can be subjective Use of biomarkers may facilitate objective and/or automated automatic adjustments,¹⁰ improved power efficiency, and fewer side effects by linking both underlying physiologic network activity and symptom relief. This may be achieved through intermittent or continuous, dynamic adjustment based on amplitude- or time-dependent changes.^2,11

Neurosurgeons have been at the forefront of concept and device testing in clinical research in brain stimulation. The number of clinical and feasibility studies, usually in small pilot groups of patients for brief time frames (hours), indicate that this is an area moving rapidly from the laboratory to the bedside and eventual standard clinical applicability.^12-16 A number of clinical trials for newly marketed and prototypical responsive, scheduled, and adaptive systems are ongoing at academic institutions. Here, we describe direct neurosurgery translation of intra- and extracranial feedback modulation of brain stimulation, updating an earlier, more comprehensive review.²

BIOMARKERS BY DISEASE

Characteristics of an ideal surrogate biomarker are noted in Table 1. The complexity of potential biomarkers and their barriers to implementation highlight the need for neural circuitry modeling. There are many avenues to analyze these biomarkers in initial clinical trials, ranging from intraoperative testing to verify appropriate electrode position, using percutaneous wires after surgery for short-term testing, or using one of the implanted sensing/recording devices for longer term data collection. Clinical implementation of biomarkers ranges from FDA approved devices to human trials with research devices to experimental biomarker identification pending clinical application. Table 2 provides a broad overview.

TABLE 1.

General Characteristics of a Desirable Biomarker

Desired biomarker characteristic	Description
Directly correlates to clinical symptoms	Relevant in time course and extent to a measurable symptom
Constantly/dynamically tracks disease state	Dynamically changes as stimulation alters neural circuits to reflect symptom improvement
Minimal sampling error	Ability to differentiate location where best, representative signal obtained
Signal stability over time	Signal faithfully and consistently tracks disease state across multiple conditions, such as walking, sleeping and activity
Signal can be differentiated from background “noise”	The signal can be distinguished from ongoing spontaneous activity to be distinct and measurable
Confirmation of desired signal	Distinguish pathological signal from overlapping normal cortical signals
Adjustable	Ability to fine tune for patient to patient variability and across dynamic states

A biomarker must do more than simply match clinical symptoms in order to be effectively implemented in a future adaptive or closed-loop device. This table highlights keep qualities necessary to make controls implementation and device design efficacious.

TABLE 2.

Possible Targets, Potential Surrogates, and Current Developmental Hurdles

Disease	Stimulation targets	Surrogate/biomarker	Biomarker development challenges
Epilepsy	Anterior thalamic nucleus, CM thalamus, localized seizure focus, vagal nerve	Intracranial: abnormal synchrony and excitability noted on EEG, ECoG and depth electrodes	Ability to sense pre-ictal event, validate clinical efficacy in human trials
		Extracranial: heart rate (VNS)
Parkinson (rigidity and bradykinesia)	STN, GPi	Beta hyper synchrony (Beta band oscillations)	Direct correlation with clinical symptom improvement
		Phase amplitude coupling	Signal differentiation
Parksinon (dyskinesia)	STN + Gpi (dual electrode)	Gamma oscillations
Parkinson (FOG)	Pedunculopontine nucleus	DBS: Increased beta frequency or cholinergic neuron action potentials
		SCS: Spinal cord evoked recordings or secondary sensory evoked recordings
Tourette	Centromedian nucleus of thalamus and GPi	Low frequency thalamic oscillations, cortical oscillations	Identify best control system mode (continuous versus adaptive versus responsive)
Essential tremor	Vim (of thalamus)	Internal: ECAP External: EMG, accelerometry	Signal stability over time; signal variability between patients; signal to noise ratio
Alzheimer's disease	Fornix, entorhinal cortex, hippocampus, cingulate, precuneous, frontal cortex	Electrophysiological: hippocampal evoked potentials chemical: cholinergic activity	Very theoretical, stimulation target/biomarker identification
Depression	Subcallosal cingulate (SCC) and	Imaging based: tractography	Leading biomarker identification
	Area 25 (medial forebrain bundle),	intersection of 3 fiber bundles near
	nucleus accumbens	SCC, frontal lobe evoked potentials

These sites summarize the disease-based discussion in the text. (CM – centromedian nucleus (of the thalamus); PET – positron emission tomography; SPECT – single photon emission computed tomography; STN – substantia nigra; EEG – electroencephalogram; ECoG – electrocorticography.)

EPILEPSY

Intracranial Stimulation

Intracranial neurostimulation for medication resistant epilepsy includes the responsive NeuroPace system (NeuroPace Inc, Mountain View, California) or intermittent, scheduled stimulation via anterior thalamic (ANT) DBS (both now approved in the US – NeuroPace for responsive stimulation and ANT DBS for intermittent, open loop stimulation).¹⁷ ANT DBS, placed via microelectrode recording and frontal lobe scalp electroencephalogram (EEG), may modulate general frontal lobe networks to reduce seizure susceptibility.¹⁸ Scalp EEG in this setting could also be used as an intraoperative or permanent surrogate biomarker for both localization and to initiate stimulation in relation to onset of an epileptic event.¹⁹

The NeuroPace system is intermittently triggered by ictal events (or their precursor signatures) using detectable depth electrode or electrocorticography (ECoG) cortical ictal activity as the biomarker. Regions based on preoperative localization are stimulated to prevent seizure onset and propagation.^20,21 NeuroPace's clinical limitations include accurate localization of both of both inputs and outputs, detection of pre-ictal events sufficiently far enough in advance of a seizure so that the patients does not detect clinical symptoms, and determining exact stimulation algorithms.

Extracranial Stimulation

Extracranial stimulation includes cervical VNS, likely affecting diffuse brain networks.^8,22 Recent studies have indicated that direct electrocardiographic detection of tachycardia in the pre-ictal state may provide a key feedback signal to trigger scheduled intermittent stimulation. This signal is available since typically the internal pulse generator (IPG) for VNS is placed in the left chest region. Additionally, focused extracranial alternating current stimulation may also be activated (using intracranial sensing electrodes) to increase seizure threshold and/or treat ictal events.⁷ This is a promising new approach that may be able to take advantage of temporal interference to both spatial and temporal focusing of extracranial stimulation.⁵

PARKINSON DISEASE

Although ∼75% of Parkinson disease (PD) patients obtain some symptom improvement with DBS placement in the subthalamic nucleus (STN) or globus pallidus internus (GPi),^23,24 there remains considerable variation in outcomes as the pathways and pathophysiology of the disease are not currently fully understood.²⁵ It is possible that different biomarkers will likely be required for different PD symptoms, rather than attempting to treat the various disease phenotypes with one treatment biomarker.

Beta Band Oscillations

Spontaneous beta band oscillatory activity at 13 to 30 Hz spreads through the cortico-basal network upon synchronization of the cortex, basal ganglia, and thalamus.²⁵ This beta band activity thought to be a marker of PD state in animal models and humans and a possible surrogate for treatment effect for bradykinesia (not tremor).²⁶ Hypersynchrony lessens after therapeutic doses of dopaminergic medication and clinical levele DBS stimulation.²⁷ Such synchrony can be recorded from motor cortex (ie, ECoG)²⁸ or directly from STN or GPi DBS contacts.^29-31 Whitmer et al²⁵ recorded subdural cortical ECoG and spontaneous STN local field potential (LFPs) (in a clinical study of 13 humans). Several clinical studies have now shown proof of principle that adaptive adjustment of beta band oscillations may be equi-efficacious to continuous DBS and more efficient.^12,32

Phase Amplitude Coupling

Based on primary motor cortex ECoG, de Hemptinne et al^28,33 demonstrated that DBS reduces phase amplitude coupling (PAC) between beta oscillations and higher frequency superimposed oscillations characteristic of PD. PACs are thought to cooridinate timing between various cortical areas for execution of tasks.³⁴ While they are seen in normal cortical activity, the increased PAC in PD possibly reflects neurons restricted to an inflexible pattern by PD, leading to the hallmark symptoms of rigidity and bradykinesia.³³ Adaptive DBS (aDBS) using motor cortex sensing appears to be equally efficacious to continuous DBS in a few long-term patients.^12,16

Evoked Potentials and Oscillations

In some cases, more than one DBS electrode may be needed for clinical efficacy, and an additional electrode at another anatomical site (ie, STN + GPi together) may prove to be beneficial. Instability and freezing of gait (FOG) are significant causes of morbidity in Parkinson patients.³⁵ There is evidence of this in a few closed-loop systems described in the literature. For example, one study utilized a closed-loop system to alter GPi stimulation based on pedunculopontine nucleus potentials (correlated with FOG and instability) as the outputs.³⁶ Dyskinesias after STN DBS are another application for dual electrode recordings by targeting STN together with GPi.^37,38 Dual electrodes can allow for stimulation on one electrode and recording on another, although stimulation evoked potentials can also be recorded from STN DBS electrodes alone,³⁹ showing an interesting resonant neural activity. Lastly, spontaneous gamma band oscillations (ie, 60-90 Hz) may also indicate hyperkinetic PD symptoms, (eg, dyskinesias) as another biomarker beyond beta band oscillations as discussed earlier.⁴⁰

Spinal Cord Stimulation

SCS may improve gait in particular, and freezing episodes, possibly through direct activation of lower extremity circuits and/or secondarily through indirect activation of intracranial circuits.^41,42 Biomarkers derived from SCS evoked recordings (ie, from stimulating one contact and recording on other contacts) or secondary sensory stimulation evoked recordings (ie, stimulating on a SCS contact and recording from sensory cortex) may be useful to titrate the level of SCS required (ie, above or below the level of paresthesia). This is an area of early clinical investigation.

TOURETTE SYNDROME

Tourette syndrome is an idiopathic neuropsychiatric disorder defined by motor and phonic tics^43,44 and intraoperative thalamic recordings of the centromedian parafasicular complex suggest that low frequency bursts correlate with the clinical phenotype. Maling et al⁴⁴ studied 5 Tourette patients implanted with the Neuropace device localized using cortical ECoG strips, CT-MRI fusion and intraoperative microelectrode recordings to delineate their anatomic CM target. Best symptom relief from stimulation correlated with increased gamma activity and eventual return to higher thalamic frequencies. Alternatively, low frequency oscillations in the GPi that precede electromyogram (EMG)-proven tic recordings by 50 to 2000 ms may be a possible anticipatory biomarker.^13,45 Okun et al⁴⁶ demonstrated long-term efficacy of scheduled stimulation in Tourette Syndrome as compared to purely continuous stimulation. They have also reported proof of concept for responsive DBS (where the spontaneous biomarker triggers onset of stimulation) for Tourette's pathology with improved battery life.¹⁵

ESSENTIAL TREMOR

Ventral intermediate nucleus of the thalamus (Vim) stimulation can help to treat essential tremor.⁴⁷ Kent et al⁴⁸ investigated the electrical stimulation-induced evoked compound action potential (ECAP) intraoperatively arising from neural elements (likely axons) near the lead.³⁹ Neural activation appeared to correlate with programmatic stimulation adjustment and clinical tremor as measured by accelormeter.⁴⁹ Specifically, low frequencies near 10 Hz worsened the tremor, but high frequencies closer to 130 Hz improved it. Unfortunately, ECAP signal has significant patient-to-patient variability, as significant as an order of magnitude between subjects.^39,49 ECAP signal may also be altered over time by glial scarring, potentially hampering long-term applications.⁴⁹ One idea to improve the signal to noise ratio during Vim DBS would be to record the electrically evoked field potential with a second electrode placed anteriorly in VOP to reduce stimulation artifact.^43,48,50

Indeed, accelerometers for dynamic tremor measurement are examples of external biomarkers from wearable devices that can potentially provide closed-loop control.⁵¹ Cagnan et al⁵² developed a prototype device coupling the measured phase of tremor to the DBS IPG for intermittent, phase - specifica stimulation. Alternatively, Basu et al⁵³ utilized surface EMG and accelerometry in an on/off adaptive control system incorporating a tremor-predictive algorithm.⁵³ Another highly innovative tremor treatment approach utilizes the typical clinical observation that patients with essential tremor have minimal to no tremor at rest. By recording and interpreting a motor cortical ECoG signal indicating pending arm motor activity, Vim DBS may dynamically control tremor prior to and during activity with less power use.¹³

ALZHEIMER’S DISEASE

It is unclear the degree to which the abnormal circuits in Alzheimer's disease (AD) may be treated or affected by stimulation.^54,55 Novel techniques such as DBS of the fornix in clinical trials have thus far shown limited efficacy.^56,57 Large hippocampal evoked potentials could potentially become biomarkers both to adjust fornix stimulation amplitude (currently empirically determined) and to estimate plasticity following fornix stimulation. Another alternative site is the nucleus basalis of Meynert,⁵⁸ although a cholinergic biomarker and associated dose-response curve might be required to titrate effectively such widespread and diffuse cholinergic enhancement.

Extracranial stimulation has also been implemented to improve^59-61 and potentially to forestall worsening in AD. Extracranial stimulation may demonstrate the advantage of widespread circuitry activation, matching the nearly global pathology in more advanced AD. Intracranial markers may help adjust and define a dose-response curve for extracranial stimulation.⁶²

DEPRESSION

DBS for treatment resistant depression (TRD) with numerous potential sites, including but not limited to subcallosal cingulate (SCC) white matter, Brodmann area 25 gray matter, and ventral capsule/ventral striatum.^63,64 While small clinical trials have demonstrated even long-term efficacy,⁶⁵ larger industry sponsored trials such as BROADEN (St. Jude) have not.⁶⁶ Intraoperative testing is challenging as patient response is highly variable, personal, and may be delayed over weeks to months. TRD is a significant example of a disease state where nonelectrophysiologic biomarkers are the leading candidates. For example, when utilizing tractography to guide lead placement, autonomic effects (tachycardia and increases in skin conductance) correlate with intraoperative testing.⁶⁷ SCC DBS lead placement for research studies was initially anatomically guided⁶⁵ but therapeutic responses were widely variable.⁶⁸ Prospective trials instead targeting white matter tract intersections near the SCC and interrogation of these individual white matter tracts with ECoG monitoring may be promising to better guide electrode placement and long term efficacy.^69-71

Both VNS and TMS of the left frontal lobe have also been FDA approved for treatment of depression.⁷² However, a clear dose-response curve based on symptoms and intracranial responses will be critical to establish a circuitry basis for stimulation.

STIMULATION AND CONTROL MODES

Time constants are a measure of the system's response time to an input.⁷³ These can be significantly different. For example, tremor responses to thalamic DBS may stabilize within 10 to 20 s⁷⁴ whereas subthalamic DBS for bradykinesia may initiate within seconds but require more than 30 min to stabilize.⁷⁵ There are 5 major subtypes of systems control policy as related to stimulation (Figure and Table 3). The clinician's input decreases and system autonomy increases through the spectrum. In order of increasing complexity, the five subtypes include:

1. simple continuous (Figure A) with clinical adjustment;

2. intermittent or scheduled using a fixed schedule and fixed amplitude (Figure B), such as the 30 sec on/5 min off commonly used cycling in VNS;

3. responsive with a preset amplitude and width but requires a trigger for initiation (Figure C) from threshold changes in a few channels of input signal;

4. adaptive with flexibility of variable durations of stimulation in response to a single biomarker input threshold, or a variable amplitude (Figure D) typically set to minutes for on/off cycling;

5. closed-loop within the brain-machine interface context, this isa multidimensional input, rapid outputs from continuousprocessing of inputs (ie, 10 to 20 Hz), and appropriate feedback signal (Figure E).

FIGURE.

DBS stimulation control models show the input below each pulse sequence, reflecting continuous changes in the biomarker, and the upper reflects the actual system output in response to the input. The green color indicates stimulation on and the red color indicates stimulation off. The various modes include: A, continuous stimulation; B, scheduled intermittent stimulation; C, responsive stimulation. Each stimulation is the constant but requires a threshold to be met; D, adaptive stimulation. The number and amplitude of pulses may vary to achieve a control point value; E, closed-loop stimulation. A dynamic continuous output informs a continuous dynamic input. F, Depending on disease pathology, input may be from the DBS lead or secondary DBS, EEG, or ECoG electrodes. Output stimulation from the IPG is primarily though the DBS/parenchymal lead but future subdural or scalp hardware options are possible. Reprinted from: Hoang et al² Copyright © 2017 Hoang, Cassar, Grill and Turner. CC BY.

TABLE 3.

Control Systems Description

Control mode	Feedback type	Description of feedback	Time constant of activation
Continuous	Clinician observation	Clinician manual adjustment	Monthly or frequency of clinic visits
Scheduled Intermittent	None	Preset stimulation amplitude turned on or off at preset timing	Preset timing determined by system physiology or empirically
Responsive	Triggered by threshold event	Preset stimulation amplitude turned on or off by trigger (with defined lockouts)	0.5-5 s, can be repeated
Adaptive	Single biomarker input, continuous monitoring	Stimulation output can be turned on or off, or scaled, for continuous adjustment	Tremor ∼ 10 s Rigidity, Gait ∼ 60-90 s
Closed-Loop	Multiple channels of input biomarkers for continuous analysis	Continuous prediction of brain intent for action	20-50 ms for information update

The use of biomarkers can be described in various approaches, including continuous and intermittent. Responsive and adaptive show progressively more flexibility in when to perform stimulation (ie, triggered by an event or threshold) and adaptive has inherently further flexibility in prolonged stimulation and levels of stimulation when on. Closed loop can apply to any scheme where a feedback signal is used to alter stimulation, but commonly is used in a brain-machine context. The chart gives the type of feedback which can be used, the nature of the feedback and time constants to be considered in delivering the feedback.

In these various modes the control policy can be defined internally (ie, a state transition map embedded within the device based on contingencies) or externally.²² Each state transition can then lead to a predefined stimulation change. Table 4 categorizes the devices and biomarkers previously described by the control subtypes discussed here.

TABLE 4.

Biomarkers Classified by Control Subtypes

Control mode/class	Specific examples of biomarkers/devices	Description
Simple continuous	Open Loop DBS	All adjustments via clinician/direct patient observation
Intermittent	ANT DBS	Scheduled on preset timing, therefore no biomarkers used (exception: VNS and heart rate as biomarker preceding seizure event)
	VNS
Responsive/triggered	Neuropace	Predetermined output in response to threshold
	DBS for tourette	Conditional output in response to motor cortex patterned responses
	DBS for essential tremor	Output in response to motor cortex movement initiation signal
Adaptive	DBS for Parkinson disease (eg, PC + S and RC + S)	Currently external; response tailored to continuous input signal.
		Includes motor cortex/STN coherence (eg, PAC), spontaneous beta band, stimulation-evoked responses.
Closed Loop	Braingate system	Multiple inputs (∼100 channels) process to form contingent output
	Other custom systems

SCHEDULED INTERMITTENT STIMULATION

Epilepsy and Parkinson Disease

Rather than continuous stimulation (ie, Figure A), ANT DBS and VNS for epilepsy implement an intermittent, scheduled stimulation at a preset level¹⁷ (Figure B, Table 3). Scheduled intermittent stimulation or additional on-demand stimulation can reduce the amount of unnecessary stimulation and may preserve network sensitivity which can otherwise adapt or fade with constant stimulation. Phase-dependent stimulation intermittently dampens the network may critically dampen the system if the response is sufficiently rapid.⁵² There are criteria for a stimulation trigger with a “lockout” or safety mechanism preventing overstimulation. The pulse repetition frequency of DBS for epilepsy is critical as EEG synchrony caused by alternative frequencies can amplify seizures.²¹ The SANTE trial targeting the anterior nucleus of the thalamus (ANT) demonstrated reduction in seizure frequency, even over the long term, using scheduled, intermittent programming.¹⁸ Specifically, SANTE randomized participants to stimulation at either at 5 V or 0 V (no stimulation) with set parameters of 90 microseconds, 145 Hz, and a 1 min “on” and 5 min “off”.

Newer Medtronic DBS devices (ie, Medtronic PC + S and RC + S; Medtronic Inc, Dublin, Ireland) also feature lock-out, phase-in timing and adjustable response based on the specific input (Figure C).^76-78 The RC + S has been implemented with more advanced features than the PC + S, including built in logic for contingent, incremental stimulation changes in response to a recorded signal, as well as a distributed network which can encompass both internal and external devices.

RESPONSIVE CONTROL

Epilepsy

Certain disease processes, like epilepsy, may have long asymptomatic or unpredictable periods between events where constant stimulation programming may not be ideal.²⁰ The key difference from the scheduled intermittent control is the recording of network signals to initiate the stimulation. For the previously mentioned Neuropace system, stimulation at a preset level is delivered in a binary on/off fashion when upper or lower thresholds are reached, much as a thermostat controls a furnace.

ADAPTIVE CONTROL

Parkinson Disease

Arguably, the most advanced control systems concepts regarding brain stimulation are the last 2 types – aDBS and closed loop (Figure D and Figure E). Adaptive stimulation has adjustable or contingent stimulation in response to the external or internal biomarker, as compared to the fixed output of a responsive system. There is a growing body of evidence that aDBS is clinically feasible and safe.^12,16,32 Little et al^79,79 utilized STN aDBS via an externalized system on/off system linked to processing of LFPs (beta frequency amplitude). Notably the stimulation was dynamic with variable widths. They also compared continuous, scheduled intermittent, and adaptive systems directly and found statistically significant improvement with the adaptive system in motor scores, decreased speech side effects, reduction in stimulation time, and decreased energy utilization.⁸⁰ However, longer-term data showed that the adaptive system demonstrated much improved efficiency but only a modest improvement in efficacy.⁸¹ Most recently, Arlotti et al¹² studied neurophysiologic and clinical responses of LFP-based aDBS for advanced Parkinson for 8 h after implantation in 2 patients. Beta-band power correlated with clinical manifestations and aDBS automatically decreased DBS amplitude during “on” states effectively while preventing dyskinesia.^12,82

A theoretical variation called a “scalar adaptive system” utilizes inputs of varying amplitudes to approach a desired set point.^11,73 The difference between the desired set-point and the current value is noted and a larger difference generates a correspondingly larger change in stimulation, hence the term “scaled output”(Figure D). This slightly more sophisticated variation embodies classic control system principles by minimizing the amount of output oscillation and time to achieve a steady state. It also mitigates large or complete on/off changes which may cause uncomfortable side effects for the patient. A scalar response may also adapt to varying needs during task performance to dynamically control system response.

CLOSED LOOP

Motor and Psychiatric Disease

At the most complex end of the control spectrum, brain computer/machine interfaces (BCI/BMI) require constant or near-constant sensing, feedback parameter, and output for motor control.⁸³ Whereas adaptive stimulation may have only a single setpoint, closed-loop stimulation can utilize dynamic or multiple setpoints, adjusted with the information from the rapidly updated feedback parameter.⁸⁴

The initial work in BCI/BMI has largely focused on motor disorders. The previously mentioned Medtronic “Activa PC + S” has been used in a number of pertinent examples such a BCI in a locked-in patient with amyotrophic lateral sclerosis.⁸⁵ More recently Widge and Sahay⁸⁶ have also proposed extensions of closed-loop theory to psychiatric applications. Early work suggests that closed-loop feedback can uniquely modify neural network firing patterns with continued exposure to BCI training. This may have unique applicability to psychiatric disease which can fluctuate or clinically change over time.⁸⁶ Additionally, a preliminary low frequency signal has been recorded in nucleus accumbens, which may be useful for measurement of impulsivity.⁸⁷

Adaptive Stimulation with Extracranial Stimulation

There are a number of strategies to improve and focus extracranial stimulation and to develop contingent stimulation dependent upon intracranial biomarkers.^7,8 VNS may also be contingent upon either direct patient triggering or electrocardiographic signals, such as tachycardia which may precede or accompany an aura or seizure onset.⁸ In this case, the VNS may be rapidly triggered to attempt to abort a seizure at the onset or in a precursor stage.

CONCLUSION

Identifying and determining efficacy of stimulation biomarkers will require considerable additional development, mainly in collaboration with neurosurgeons who can facilitate development, clinical testing and implantation of new systems. Electrophysiological measures, neurochemical or other markers may also be reasonable options beyond current imaging and electrophysiological measures. While treatment of disease can obviously be optimized, the potential to better understand the underlying circuitry on many pathologies is a more fundamental goal of this study. It is critical at this stage to identify the most promising biomarkers into proof of concept translational work, and well-designed clinical trials to demonstrate efficacy, safety, or noninferiority in comparison to conventional devices. Neurosurgeons are the driving force in many instances to implement and improve adaptive modulation systems as well as understand the underlying clinical diseases.

Disclosures

This work was supported by NIH R01 NS079312, NIH R37 NS040984, and NIH UH3 NS103468. The authors have no personal, financial, or institutional interest in any of the drugs, materials, or devices described in this article.