Challenges and opportunities of acquiring cortical recordings for chronic adaptive deep brain stimulation

Abstract

Deep brain stimulation (DBS), a proven treatment for movement disorders, also holds promise for the treatment of psychiatric and cognitive conditions. However, for DBS to be clinically effective, it may require DBS technology that can alter or trigger stimulation in response to changes in biomarkers sensed from the patient’s brain. A growing body of evidence suggests that such adaptive DBS is feasible, it might achieve clinical effects that are not possible with standard continuous DBS and that some of the best biomarkers are signals from the cerebral cortex. Yet capturing those markers requires the placement of cortex-optimized electrodes in addition to standard electrodes for DBS. In this Perspective we argue that the need for cortical biomarkers in adaptive DBS and the unfortunate convergence of regulatory and financial factors underpinning the unavailability of cortical electrodes for chronic uses threatens to slow down or stall research on adaptive DBS and propose public–private partnerships as a potential solution to such a critical technological gap.

Deep brain stimulation (DBS) is an effective and safe treatment for many brain disorders. It involves high-frequency electrical stimulation of structures near the centre of the brain through permanent indwelling electrodes implanted under precision neurosurgical guidance^1,2. In the United States DBS was approved by the Food and Drug Administration (FDA) for the treatment of essential tremor (with regulatory approval granted in 1997), Parkinson’s disease (approved in 2002), dystonia (2003; approved under humanitarian device exemption), epilepsy (approved in 2012) and obsessive–compulsive disorder (OCD; approved in 2009 under humanitarian device exemption).

Continuous DBS

Deep brain stimulation is also under investigation for Tourette syndrome, major depressive disorder, chronic pain, obesity, anorexia nervosa, tinnitus, addiction, Alzheimer’s disease, post-traumatic stress disorder and anxiety^3–6. Collectively, these disorders represent some of the biggest public health challenges of our era and are leading causes of disability worldwide⁷. Treatments for many of these disorders are limited. For instance, more than 30% of patients with major depressive disorder and up to 50% with chronic pain do not achieve relief from available therapies^8,9. For other disorders, such as anorexia or post-traumatic stress disorder, there are no specific medications and the psychotherapies available at present are often unacceptable to patients or not effective^10,11. All approved clinical indications use open-loop or continuous DBS (cDBS)—a pre-established set of stimulation parameters (such as frequency, amplitude, pulse width, stimulation contacts and duty cycle)—delivered continuously throughout the day. Continuous DBS is generally safe and is a commercial success in the approved indications. In experimental indications, however, it often proves challenging to optimize stimulation parameters to relieve symptoms¹². Furthermore, cDBS can cause side effects that can limit its efficacy^13,14. The specific side effects depend on the brain target and diagnosis but often result from undesired activation of structures neighbouring the therapeutic target. For instance, cDBS for essential tremor, Parkinson’s disease and dystonia can cause speech and gait problems through the stimulation of passing fibre tracts to the brainstem and the cerebellum. Those side effects and the overall invasiveness of DBS may be one reason the DBS market has remained relatively stagnant in terms of the number of implants performed per year. Side effects also motivate the development of additional non-invasive neuromodulation approaches such as vibrotactile and temporal interference stimulation^15,16.

Adaptive DBS

In contrast to the continuous delivery of therapy in current systems, most brain disorders involve symptoms that fluctuate throughout the day, and the response to stimulation can be highly dependent on the current brain state^12,17–19. To more effectively engage target brain circuits and account for state dependence, there is a growing interest in adaptive closed-loop DBS (aDBS)^12,20–24. In aDBS, stimulation is adjusted in real time in response to a physiologic signal (‘biomarker’) that correlates strongly with the disease state of the patient^22,25,26. Adaptive DBS remains investigational and may not be needed in all indications. However, it has been suggested that, in specific contexts, aDBS may provide effective therapy in indications where open-loop DBS has failed or it may mitigate some limiting side effects of stimulation. This is particularly true in cases where symptoms present episodically and briefly, for example, during movement in essential tremor or in response to tic urges in Tourette syndrome. In these cases, aDBS might improve the therapeutic window by delivering stimulation only when it is needed, thus reducing side effects.

There are two critical challenges for each aDBS application: the identification of one or more relevant biomarkers and the design of an algorithm that maps changes in these biomarkers to stimulation adjustments. The biomarkers might be derived from brain, muscle or neurochemical signals, or even be recorded by a peripheral sensor such as a kinematic device^27–30. The majority of studies involving aDBS have focused on the local field potential (LFP) as a signal that can be efficiently recorded with existing implants (such as Percept PC/RC, Activa PC+S/RC+S and NeuroPace RNS)^31–34 and contains rich information about disease states. The algorithms for mapping the fluctuations in stimulation adjustment come with new free parameters and they potentially increase the patient-specific programming difficulty beyond what clinicians can accomplish in practice given the current status of automation developments. With modern device electronics and system design, LFP sensing can be added for as little as 5% of the total power budget or battery capacity of a device³⁵ and the power drain of signal processing can be more than compensated by the reduction in stimulation amplitude or duty cycle given that stimulation is far and away the primary source of battery drain.

Biomarkers for use in aDBS

All approved and proposed DBS indications share the feature of being network diseases that involve interactions between cortical and subcortical structures^26,36–39. This raises the critical question of where the biomarkers for aDBS should be sensed. Deep brain targets are critical hubs in disease networks and can hence be effective stimulation targets^38,40,41. Yet they present substantial potential disadvantages as biomarker sources when compared with cortical biomarkers (Fig. 1).

Fig. 1 |. Movement decoding and aDBS in Parkinson’s disease is more reliable using cortical signals.

a, Using STN DBS for Parkinson’s disease as an example, LFPs recorded through DBS leads can be used for feedback control in the adaptation of DBS parameters. Implantation of an additional ECoG strip electrode extends the recordable signals to the premotor cortex, the primary motor cortex as well as the sensory and parietal cortices. b, Movement-aligned analyses illustrate typical patterns of readiness potentials during periods of high grip force (black line below the raw data) in ECoG but not subthalamic LFPs, which have up to 10× lower raw signal amplitudes. Moreover, in the time–frequency domain, the typical low-frequency desynchronization and gamma-band activation is observable in both deep brain LFPs and cortical ECoG. The heat map shows the percentage deviations from the −2 s to the −1 s baseline before movement onset. The ECoG shows much stronger activations, especially in higher frequency ranges typically associated with local action-potential firing. c, Consequently, gradient-boosted decision-tree models trained on ECoG outperform STN LFP-based models. R² is the coefficient of determination. ***P < 0.001. Adapted from ref. 53, CC BY 4.0.

First, LFPs from deep brain structures have a lower signal amplitude and a narrower spectral range than cortical LFPs (also referred to as electrocorticography signals; ECoGs). Hence, cortical LFPs provide a richer signal space that can be mined for biomarkers. However, that richness may bring in confounders; for example, ECoGs will often reflect circadian variation, such as sleep. Still, such confounders may actually be clinically useful: for instance, delivering DBS only when patients are awake would avoid needless battery consumption and avoid inadvertent disruptions of sleep architecture. Although this would require more complex and possibly multiple decoders operating in parallel, it could provide clinical advantages.

Second, because deep targets (for example, in the motor thalamus) map onto much larger cortical regions, the signals of interest are spatially separated in the cortex, allowing for more precise decoding from standard electrode arrays (Fig. 1).

Third, because DBS by definition involves active stimulation of a subcortical target, recordings from those structures will be contaminated by stimulation artefacts (usually single-digit volts rather than tens to hundreds of microvolts for typical LFPs). As a related point, stimulation will probably change the content of subcortical signals over months to years of DBS exposure, whereas cortical signals seem to be more stable. Although cortical signals seem to be important for successful aDBS, they are increasingly difficult to capture and study in patients, solely because of a lack of suitable chronic recording electrodes for human use (that is, of chronic recording electrodes that are technologically feasible, ethically viable and regulatorily approved).

In the following sections we review the applicability of aDBS to the treatment of essential tremor, Parkinson’s disease and psychiatric disorders.

Essential tremor

One of the more apparent use cases for aDBS is essential tremor, a movement disorder where intentional movement causes uncontrollable tremors. Because symptoms are limited to attempts to use the affected limb, there is a clear delineation between when patients need stimulation and when they do not. Continuous stimulation of the ventral intermedial (VIM) nucleus of the thalamus (the most common essential tremor target for DBS) drains the device battery and often causes side effects, including dysarthria (the difficulty to pronounce words) and paraesthesia⁴² (feelings of tingling or numbness; that is, of ‘pins and needles’). The early proofs of principle for aDBS in essential tremor used wearable sensors to trigger stimulation based on movement, with effective symptom suppression^43–45. Ultimately, however, wearable systems add complexity (such as a wireless data link in the critical control loop, which may be subject to data and control breaches), security risks (such as risks to patient privacy and possibly risks of interference with the control loop) and patient inconvenience (pertaining to the charging and maintenance of the wearable). All of these points make wearables inferior to direct sensing and control through a single implanted device. Yet, even if the addition of sensing electrodes adds complexity, it does not necessarily affect other practical considerations; for instance, there is no need for additional battery charging (that is, beyond what is needed for the primary DBS system).

As for aDBS based on LFPs, cortical signal amplitudes are notably larger and contain richer spectral content with respect to VIM LFPs (Fig. 2a–c), which enables greater discrimination of movement from rest (Fig. 2d). Although it is possible to identify volitional movement or tremor using the thalamic electrode either through externalized leads or via implanted systems⁴⁶, when the stimulation engine is ‘on’, it creates artefacts that dominate thalamic recordings at extremely low stimulation amplitudes, making it impossible to capture the different features (Fig. 2e). This does not occur in the cortex, where sensing from a strip of electrodes chronically implanted over the primary motor cortex allowed for the first demonstration of a self-contained aDBS that effectively controlled tremor^46,47 and for evidence of cortical signals that were superior to thalamic LFPs owing to higher signal amplitudes and distance from the stimulation site. Cortical recordings also yield more stable biomarkers; in fact, long-term primate studies established that the stability of motor cortical ECoG can last for years^48,49, whereas in humans, with appropriate algorithms, stable decoder performance can be maintained for months^50,51. In contrast, in VIM, the spectral power separation between baseline and tremor or movement was specifically unstable across months⁵², making aDBS for essential tremor clinically infeasible if limited solely to sensing from the VIM stimulation target. It is possible that cortical signals may change over long time scales—for instance, through network remodelling by the disease or via long-term neurostimulation³⁷. That slow drift, however, could be easily compensated through small and infrequent adjustment of parameters of the aDBS algorithm.

Fig. 2 |. Movement decoding and aDBS in essential tremor is more reliable using cortical signals.

a, ECoG strip placement over the primary motor cortex (M1) of the hand. i–iv denote the individual contacts of an ECoG recording strip. b, Implanted depth lead in the VIM thalamus. Voa, ventralis oralis anterior nucleus of the thalamus; vop, ventralis oralis posterior nucleus of the thalamus. c, Raw time-domain signals of a single hand movement in response to a ‘go’ command recorded from the M1 via an ECoG strip (top) and from a depth lead implanted in the VIM thalamus (bottom). The ECoG signal amplitude is substantially larger. d, Normalized power separation between baseline activity and movement or tremor, expressed as changes in normalized power between conditions, with VIM DBS off. Low-frequency oscillations (LFOs) and high-frequency activity (HFA) were defined as 1–30 Hz and >70 Hz, respectively. Cortical recordings show greater differences in both bands. Statistically significant differences from 0 dB were determined using a Wilcoxon signed rank test; *P < 0.01; diagonal line indicates no significant difference. Boxes show the median, first and third percentiles, and the whiskers represent the minimum and maximum range with outliers (indicated by red crosses) excluded. e, Spectrograms of M1 (top) and VIM (bottom) signals in the presence of short therapeutic stimulation blocks (middle; timing of stimulation indicated by grey boxes). Stimulation produces artefacts at a frequency of 130 Hz (outlined in orange) and at its subharmonics in the VIM but not in M1. These artefacts in the VIM jam the ability of detecting LFO modulation with movement. Turning the stimulation engine on, even with a stimulation amplitude of 0 mV, generates artefacts. Adapted with permission from ref. 46, AAAS.

Parkinson’s disease

Similar examples have emerged in DBS for the treatment of Parkinson’s disease. ECoG strip recordings were directly compared with subthalamic nucleus (STN) LFPs for decoding grip force as a representative example of the type of movements that DBS is meant to augment⁵³. Machine-learning models trained on a single channel of motor-cortex ECoG outperformed models trained on STN, even when the STN models were allowed to use multiple channels. The result was robust to changes in the underlying model. Motor-cortex ECoG also provided more robust signals for rapid closed-loop control to mitigate dyskinesia owing to overstimulation^54,55. In four of five patients undergoing long-term cortical and subthalamic recording at home, motor ECoG provided superior decoding of the ‘on’ versus ‘off’ states⁵⁴ (that is, low versus high symptoms). ECoG could also be combined with basal ganglia LFPs to provide superior motor-state decoding than what was achieved using signals from either site alone. Note that cortical-driven aDBS in Parkinson’s disease is not focused on treating tremor. Continuous DBS can control Parkinsonian tremor but does less well with akinesia or rigidity, where patients often experience fluctuations between bradykinetic (understimulated) and dyskinetic (overstimulated) states; instead, aDBS can reduce the occurrence of dyskinetic side effects and in the process reduce battery consumption⁵⁶. Thus, although some studies have reported improved tremor control with aDBS⁵⁴, a pivotal clinical trial meant to make the case for aDBS in Parkinson’s disease being specifically focused on reducing dyskinesia⁵⁷. This represents a difference from the essential tremor use case, because Parkinsonian tremor is more rest-associated, whereas essential tremor is more movement-associated. In Parkinson’s disease, decoding of motor intent becomes less important; instead, decoding pathological beta-band activity (a biomarker that is specific to Parkinson’s disease yet irrelevant to essential tremor) is more suitable.

Tourette syndrome

Tourette syndrome is a chronic neurodevelopmental disorder characterized by involuntary tics that can substantially diminish the quality of life of affected individuals⁵⁸. It is similar to essential tremor in the sense that Tourette syndrome symptoms involve specific, brief and episodic movements, and that tics are often preceded by a subjective urge (as with movement intentions in essential tremor and Parkinson’s disease). Hence, intermittent stimulation may be sufficient to control tics. Deep brain stimulation is indeed a promising therapy for carefully selected individuals with severe treatment-refractory Tourette syndrome⁵⁹. For instance, intermittent scheduled electrical stimulation of the centromedian thalamus suppressed tics with a mean of 2.3 h of DBS per day (with a s.e.m. of 0.9 h per day)⁶⁰. Closed-loop stimulation based on ECoG achieved a similar improvement in symptoms⁶¹, thus suggesting that aDBS may be an effective approach to treating Tourette syndrome. Because DBS for Tourette syndrome requires relatively high amplitudes and pulse widths, aDBS may be superior to cDBS because it will lead to less battery drain (in fact, a pilot study involving four patients reported that aDBS may be superior to cDBS for symptom control but the study was not statistically powered; in this aDBS application, ECoG recordings were required because, as with essential tremor, stimulation artefacts corrupted centromedian thalamus recordings and prevented their real-time use in tic detection).

Psychiatric disorders

In view of the extreme prevalence of medication-refractory psychiatric illnesses and their high economic and social costs⁶², these illnesses may ultimately be one of the highest value applications for DBS. Unfortunately, standard DBS has not met its endpoint in any multi-centre trial for any psychiatric disorder despite small or single-centre studies showing dramatic improvement in otherwise-refractory individuals^63,64. There are difficult challenges in correctly engaging target circuits and a major barrier to DBS optimization in psychiatry is the lack of an objective readout of target engagement^12,23.

However, there is a strong interest in aDBS as a path forward^12,38,65 because it might overcome these challenges via the use of biomarker-based programming algorithms. At present, psychiatric DBS is titrated on the basis of patient self-reporting, which is easily distorted by cognitive biases (for example, by recency bias) and does not distinguish well between true disease relapses and transient emotional responses to life stresses. Adaptive DBS and the related biomarkers might better discriminate those states⁶⁶ and enable programming to drive the brain to a specific physiologic state^12,67 or to activate ‘detect and treat’ systems that directly respond to symptom flares¹⁸. In all of these paradigms, the three relevant factors (signal granularity, signal amplitude and artefact robustness) that make cortical signals attractive aDBS biomarkers are particularly relevant. First, brain states related to subjective distress and cognition are strongly encoded in the cortex^68–73. Second, the best-studied DBS targets for psychiatric indications are in white matter^40,74–78 but LFPs recorded from white matter have much lower amplitude than LFPs from grey matter targets and are notoriously difficult to interpret, in part because the origin of these signals is unknown⁷⁹. Third, artefact concerns are even stronger in psychiatric applications because stimulation amplitudes are often greater than those in movement disorders, creating larger artefacts^80–83. These principles have been demonstrated in two psychiatric disorders: depression and OCD.

In depression, neuroimaging suggests that symptoms arise from a distributed network of both subcortical and cortical brain regions^84,85, suggesting that cortical recordings might be critical for the robust decoding of depression symptoms. Recent work with distributed multi-site LFP and ECoG recordings of patients with epilepsy showed that cortical regions are needed for the decoding of mood states related to depression and anxiety and, when personalized mood decoders were trained separately for each individual patient, that the decoders consistently selected the orbitofrontal cortex (OFC) as an essential mood-predictive region in most of the patients⁷¹. Similarly, when exploring the decoding of chronic pain levels from LFP and ECoG recordings, decoding relied on the OFC in all four patients⁶. Because of the strong overlap between depression and pain, these results support the value of cortical recordings as biomarker sources for psychiatric disorders.

The most common target for DBS in the treatment of OCD is the ventral internal capsule/ventral striatum (VC/VS) as it may normalize pathological frontostriatal network activity^86,87. Because OCD pathophysiology is strongly believed to involve cortico-striatal networks^40,88,89, symptom states requiring adjustments to DBS (such as increases in OCD-related distress or the emergence of DBS-driven hypomania^13,90,91) might best be detected from a combination of cortical and subcortical recordings. Cortical decoding might be necessary because relevant signals such as OCD symptom burden are encoded primarily in the cortex, as demonstrated when chronic recording electrodes were implanted over the supplementary motor area to complement VC/VS recordings¹⁷. Of the top five features predicting symptoms in a machine-learning model, three of them were cross-structure interactions and two were cortical power signals. It is important to note that LFPs from the VC/VS were not selected in this data-driven analysis and that another reason for why cortical decoding might be necessary is that LFPs are less affected by DBS artefacts. An ongoing clinical trial (NCT04806516) seeks to develop aDBS using chronic ECoG recordings from the OFC in OCD⁶⁵ (Fig. 3a,b). In the first patient, OFC recordings showed smaller stimulation artefacts (Fig. 3c–e) and more spectral peaks, which may be biomarkers of the pathology of OCD and the responses to it. Measurements of the same OFC area could also be performed via a standard cylindrical depth lead but that placement involves trajectory and burr-hole planning to avoid ventricles and blood vessels, making it much more challenging and potentially lengthening the intra-operative time owing to the need for mapping electrode depth. Placement of a cortical strip (Fig. 3) is a much faster procedure that is less prone to interference from, for instance, brain shifts caused by the previous placement of a VC/VS lead.

Fig. 3 |. Stimulation artefacts and signals from chronic recordings in OCD.

a,b, Frontal (a) and inferior (b) views of the reconstructed cortical surface, subcortical structures, anterior commissure and DBS leads in the (VC/VS) as well as ECoG electrodes over the OFC. c, Power-spectral-density plot showing artefacts in the VC/VS (left), medial OFC (middle) and lateral OFC (right) at two stimulation settings (DBS on) as well as the absence of artefacts during DBS off. As the stimulation amplitude and pulse width increase, the 150.6 Hz peak widens in the VC/VS recordings but not in the OFC recordings. Likewise, lower-frequency artefacts in the 60–100 Hz range⁸⁰ appear during the 5.5 mA condition in VC/VS but not in the OFC recordings. The grey boxes correspond to the frequency range (0–50 Hz) in d. d, Power-spectral-density plots (formatted identically to those in c) of OFC recordings show the presence of neural activity above the characteristic 1/f in the 0–20 Hz range, whereas VC/VS recordings do not.

The technology gap

Cortical LFPs and ECoGs are the most promising source of biomarkers (and in some cases their only proven source), enabling aDBS in the preliminary work carried out to date across brain disorders. However, cortical LFPs and ECoGs also carry additional risks that need to be balanced against the potential benefits (Table 1). Although none of the studies discussed here reported specific adverse events related to the placement of a second cortical lead, every additional piece of hardware in the body adds to the risks of infection, failure and tissue injury, and lengthens the surgical time needed to place the extra cortical leads, which always increases risk. Furthermore, aDBS in any form is more complex than cDBS, adding the risk of inadvertent clinician error or an unanticipated adverse effect of the specific control policy chosen. That complexity may also create a burden for patients in that their therapy and its effects may be perceived as less predictable or less controlled⁹². Our contention is that these risks are relatively small and they are outweighed by the potential clinical benefits identified in the pilot studies discussed above (however, those results will need to be proven and validated in larger studies).

Table 1 |.

Weighing the cost–benefit trade-offs of cortical electrodes for aDBS

Advantages (putative benefits: new biomarkers that improve therapy outcomes)	Disadvantages (putative costs: additional material and time expenses and patient risks)
The distance between sensing and stimulation sites eliminates or minimizes stimulation artefacts.	Additional hardware, adding risk (not yet quantified) of haemorrhage, infection or foreign-body responses.
Coverage of large brain areas across the functional organization of the cortex, potentially using multiple high-density electrodes.	Additional surgical complexity, requiring tools and techniques that are not yet standard outside expert centres, that would bring up risks of prolonged surgery.
The electrodes can be specifically optimized for high-fidelity sensing as opposed to deep brain leads optimized for charge delivery.	Optimal targeting for specific applications yet to be defined.
High signal-to-noise ratios reduce the detrimental effects of physiological artefacts such as movement or cardioballistic activity.	Additional complexities associated with methodological procedures and signal processing, and the hardware increases the risk of operator errors.
Therapeutic electrodes in the DBS targets do not need to be ‘sacrificed’ for sensing; aLL can be used for stimulation.

Unfortunately, larger studies have essentially become impossible to carry out owing to a lack of hardware. Cortical recordings are indeed generally performed with strip- or paddle-style electrodes placed in the subdural space (that is, in direct contact with the putative signal source). Because these electrodes are insulated on one surface (they are directional), they isolate the signal of interest and are better able to reject extracranial noise. Those recording paddles are then connected to an implantable pulse generator (IPG), usually in the chest. Almost all published work in this space has used one specific technology: a repurposed spinal-cord-stimulation paddle connected to a Medtronic IPG with the capacity for chronic brain sensing. Three generations of those IPGs^31–33 have enabled a range of studies in the clinic, the laboratory and at home^{17,54,93–96}.

However, at present, there are no electrodes (for spinal-cord stimulation or otherwise) that are both approved by the FDA for chronic subdural use and that can connect to these sensing IPGs. For a brief period of roughly five years, a four-contact epidural spinal-cord stimulation paddle was available from Medtronic, on a limited basis, to support investigator-initiated studies as part of a public–private partnership for the Brain Research Through Advancing Innovative Neurotechnologies (BRAIN) Initiative. A design-history file with the FDA was supplemented with additional information to support subdural use in regulated registered clinical trials. This file included reports of additional testing, such as endotoxin levels and expanded biocompatibility data. Those tests were necessary because the FDA and other regulatory bodies apply stricter requirements for devices exposed to cerebrospinal fluid compared with devices confined to epidural use. The chosen lead was already in the process of being discontinued because it had only four active contacts and clinical trends in spinal-cord stimulation had moved to higher contact counts to enable fine-grained control. A limited remaining number of leads were kept available as research products and, similarly, there was a limited manufacturing run to support planned clinical cases. Once that supply was exhausted, there was no commercial reason to manufacture more because no aDBS application had arisen that clearly supported further investment and no replacement technology exists at this time. Several manufacturers create grid- or paddle-style arrays for acute or short-term surgical monitoring (less than 30 days) but none are qualified for chronic use. In theory it is possible to qualify specific batches for that use but the only known example involved the hand-building of specialized connectors⁵⁰. That is not a scalable strategy across institutions. Furthermore, the longevity of such electrodes is unknown. Paddle or grid leads are generally manufactured by bringing the component materials (insulating polymer substrate or patterned conductors) together as a ‘sandwich’, with bonding enhanced by the use of various adhesives and surface treatments. In addition, the adhesive and material choices may be substantially different if the final construct needs to maintain identical performance for years to decades.

The development of aDBS requires understanding of how the performance of a system and the relevant biomarkers may vary across days to months, making chronic implantability essential. In psychiatric applications, biomarker decoding requires data containing a range of symptoms and generally DBS drives changes in symptoms over months^20,23,38. Moreover, sensing signals from the deep stimulation lead cannot act as a substitute because of the artefact concerns that have been specifically validated in recent analyses^97,98. In other words, owing to this technological gap, meaningful aDBS advances may become essentially impossible in the short term.

Closing the technology gap

Although repurposed spinal-cord leads were a key enabling technology for first-generation aDBS studies, they still presented major limitations. Common configurations consisted of large-diameter electrodes (2–3 mm) that were sparsely spaced (10 mm) and embedded in thick and minimally flexible silicone (thicknesses of about 1.5–2.0 mm; Fig. 4)^99,100. They did not conform well to the brain, were not easy to position over the targeted gyrus and required a sizeable craniotomy that added to patient morbidity. A lack of conformity to brain surfaces and bulkiness have been directly linked to an increased number of complications¹⁰¹. Stiff electrodes can also compress brain tissue and cortical veins, resulting in cerebral oedema and potentially causing increased intracranial pressure, bleeding and microinfarcts^102,103.

Fig. 4 |. Dimensions of available paddle-style recording electrodes.

Dimensions of chronic (top) and acute (less than 30 days; bottom) paddle-style recording electrodes. These include the Medtronic Resume 09130 lead, the NeuroPace RNS cortical lead (PMA100026)¹¹⁷ and ECoG strips manufactured by Ad-Tech (K053363)⁹⁹ and NeuroOne (K192764)¹¹⁴. Orange boxes show the relative thickness of the electrodes from a side view. Different leads have different tail diameter and contact spacing, which creates fundamental incompatibilities for electrical connections. For instance, the Medtronic Resume 09130 lead uses a ring/spring connection from Bal Seal, whereas the NeuroPace DL-344–3.5 uses a proprietary connector with a custom-manufactured cover. These parts cannot be substituted or readily adapted to each other. N/A, not applicable.

New electrode materials and fabrication technologies are providing solutions that begin to address these problems, potentially reducing paddle thickness (to less than 100 μm) and allowing for different sizes and geometrical configurations of contacts, and for increasing flexibility and conformability. These properties have great potential to reduce the surgical invasiveness of the implantation procedure, mitigating some risks of cortical sensing. Several research laboratories have produced high-density thin-film electrodes using substrates of polyimide, liquid-crystal polymer or parylene-C^104–111. Commercially available solutions include thin-film polyimide (manufactured by NeuroOne^112–114; at present, it cannot connect to an IPG but has been FDA-cleared for clinical use for less than 30 days) and silicone electrodes (manufactured by CorTec)¹¹⁵ that can be coupled to an investigational brain-interchange platform that provides closed-loop capability but that is yet to be validated for use in clinical trials (and hence it is not FDA-approved). Iris Biomedical¹¹⁶ (Ripple spin-out) is also working on fully implantable electronics (the Athena platform) with recording and stimulation capabilities as well as wireless data transmission. However, the prototype device only provides externalized headers to connect to commercially available electrodes made by other manufacturers, which do not provide cortical leads for chronic use. The only chronic cortical lead that is commercially available is part of the NeuroPace RNS system¹¹⁷, it is not sold separately and is not compatible with non-NeuroPace devices. Figure 4 illustrates the different connectors used in subacute or monitoring electrodes versus those used in chronic leads. Connectors for subacute leads are not designed for use in continued contact with body fluids and updating those connectors for compatibility with an implantable DBS system is not trivial (and, from a regulatory perspective, it implies redesigning the whole lead). A subacute grid was used for investigational use⁵⁰ but it required manual fabrication and the testing of a small batch of repurposed electrodes (R. Franklin, Blackrock Neurotechnology, personal communication). This is not a scalable approach to aDBS investigation and the results would be difficult to translate to a product suitable for pivotal clinical trials.

Therefore, although there are multiple options that could develop into cortical-sensing solutions with the correct investment, there are major technical and regulatory barriers to using them for next-generation aDBS systems. Available research results are preliminary, obtained in a few patients by expert academic investigators and none have reached the level of robustness needed for a critical trial that would support cortical sensing as a standard clinical procedure. Without strong trial results and the corresponding prospect of substantial clinical revenue, there are no strong financial reasons for any manufacturer to develop a chronic cortical electrode. Qualifying a device for chronic use in direct contact with the brain requires extensive and expensive testing, even for small-scale experimental applications. The testing would include numerous benchtop experiments (mechanical and material-based, electrical, of hermeticism, of accelerated lifetime and of corrosion; as per ISO-14708). Manufacturers must also prove biocompatibility (as per ISO-10993), which consists of a battery of tests performed both in vivo and in vitro (including cytotoxicity, sensitization, genotoxicity, neurotoxicity, chronic toxicity, carcinogenicity and a six-month brain implant in a large-animal model). Furthermore, such testing must be done on prototypes that closely resemble the final design, materials and manufacturing process. Hence, testing can only be done towards the end of a project, not iteratively in parallel with design activities. As an example of the expense, a standard biocompatibility testing required for FDA submission, which includes a 180-day large-animal study, costs approximately US$1.5 million and can take up to one year of planning; furthermore, even after a device is approved, there are ongoing costs for inspection, testing and documentation for each production lot to ensure that manufacturing standards are maintained.

All these costs are extremely difficult to justify when aDBS remains investigational and cannot produce a marketable and revenue-generating clinical application for chronic cortical recording leads for at least the next several years. There is an important contrast to the early days of cDBS: although the concept was investigational, there was evidence of substantial and immediate benefit within seconds of starting stimulation, which greatly reduced the perceived risk. In many aDBS applications, particularly in psychiatric disorders, the benefits are much less visible and the symptoms themselves are ill-defined¹², which, together with multiple unsuccessful pivotal trials, greatly raises the perceived risk and leaves little-to-no incentive for the private sector to invest the necessary time and resources. At the same time, unless those investments occur, it may be impossible to move aDBS towards clinical viability.

There is therefore a canonical impasse between biomedical research and technology development. Academic researchers have the capabilities and deep knowledge of needs that could be addressed, with long-term potential for new marketable devices. Industry has knowledge on how to manufacture and test technologies and to obtain and maintain the necessary regulatory approvals but cannot focus on long-term opportunities with uncertain time horizons. Hence, neither can proceed. Canonical problems, however, have canonical solutions. In other biomedical domains, similar impasses have been addressed by a public–private partnership where government and sometimes non-profit entities provide ‘risk capital’ that allows industry to enter agreements that would otherwise be economically unviable, which in turn enables academics to conduct innovative and otherwise impossible research^118–121 (see also https://www.nih.gov/research-training/accelerating-medicines-partnership-amp). Such partnerships have broken logjams in drug development and biomarker development for major diseases^122,123, most recently in coronavirus disease 2019-vaccine development¹²⁴. Most of the aDBS research that we have described was conducted under a public–private partnership through the BRAIN Initiative¹²⁵, which has been highlighted as a major success in its mid-initiative report (https://braininitiative.nih.gov/vision/nih-brain-initiative-reports/brain-20-report-cells-circuits-toward-cures). The time is thus ripe for a new public–private partnership focused specifically on enabling technologies for aDBS and for device manufacturers to be interested in supporting academic research. The US National Institutes of Health (NIH) are interested in supporting new recording technologies and next-generation clinical trials, particularly of closed-loop neurotechnologies (such as the research for applications (RFAs) RFA-NS-21–023, RFA-NS-21–024 and RFA-NS-21–026) through the BRAIN and Blueprint MedTech Initiatives. One or more manufacturers could be funded by either the NIH or the recently awarded Blueprint MedTech Incubator Hubs to perform initial development and testing (biocompatibility, implantability and large-animal chronic safety) of one or more next-generation cortical designs. Those electrodes would need to be designed to capture and decode from the motor cortex (a single gyrus), or from the prefrontal or OFC (multiple gyri with signals of interest), accommodating their use across a range of potential disease applications; furthermore, their development should incorporate feedback from potential academic users (both surgical and non-surgical). The new electrodes could be designed to be compatible not just with the systems used for research but with connector architectures in common use by multiple large strategic manufacturers (existing DBS devices use a common in-line design, with the only difference being the contact spacing; Fig. 4). This option may not even require cooperation from the strategic manufacturers because, in theory, their products could be purchased ‘off the shelf’ and explicitly put through testing with the novel electrodes (for example, to prove that signal integrity is maintained over time).

The output of this process would be a design-history file (DHF; a formal record of every step in the design of a medical device, from initial concept to translation of the concept into formal requirements, including the results of testing against those requirements). For a new cortical lead, those tests might include verification that manufacturing is consistent in the final geometry (within tolerances), ISO-10993 biocompatibility and verification of signal quality and integrity, covering a reasonable set of partner devices and use cases. If developed with public funds, one would expect the DHF to be made publicly available, as opposed to the DHFs for commercial systems that are deposited with the FDA as proprietary information. Academic users would then apply for specific investigational-device exemptions with reference to that DHF. Because the testing would probably not be exhaustive, incorporation of the cortical leads in the DHF would impose additional restrictions on some studies. For instance, most modern DBS systems are magnetic resonance imaging-compatible but when paired with a third-party cortical lead the magnetic resonance safety would need to be re-verified. This testing is probably too expensive for a limited public–private partnership and thus, the aDBS system would be considered magnetic resonance imaging-unsafe within these investigational uses. These restrictions, however, also applied to the previous generation of tools under the BRAIN public–private partnership.

In this scenario, electrode fabrication would be performed by an original equipment manufacturer that would continue to maintain the DHF and that would provide cortical electrodes to researchers (which might include spin-out companies commercializing novel aDBS approaches). The costs of that regulatory maintenance and small-batch fabrication would be bundled into the unit price of these cortical leads, which in turn would be included in grant proposals of clinical trials, making the costs dependent on demand (and thus, amortizing the cost of a full-time product manager over all electrodes ordered in a year). There are many fine details that would need to be resolved (such as how the designs might survive bankruptcy of a manufacturer or change of business and how to avoid excessive profit at public expense) but the basic concept should be feasible. We believe there is broad investigator support for that concept (as the authorship list of this article suggests). It also highlights a policy challenge: tools such as cortical recording do not ‘belong’ to any one disease or researcher and yet most medical research is funded by individual grants to individual investigators for single and specific medical conditions. It is hard to fit tool development into that rubric, especially when the tool to be developed is so simple yet so necessary.

Outlook

Deep brain stimulation, particularly aDBS, offers the hope of treating a wide range of brain disorders that are both disabling and lacking in specific and effective treatments. A major barrier agreed on by experts across a wide range of disorders and disciplines is that effective aDBS is likely to require the ability to sense biomarkers from the cerebral cortex while stimulating the deep and subcortical brain. That premise has been proven in innovative studies from multiple laboratories but those studies are unable to continue for lack of a single part: a chronically implantable and sensing-compatible cortical electrode. The problem is regulatory and economic rather than technical. New medical technologies only emerge when several criteria are met: a favourable balance of risks and benefits, viable economics (benefits more than offset incremental costs), efficient clinical workflows and the commercial freedom to operate. Cortical sensing for aDBS can only meet these criteria through extensive further investigations that would map promising biomarkers into robust clinical benefits in controlled studies. The use of those markers and the surgical placement of cortical electrodes would need to become simple enough for widespread clinical adoption. There is a large gap between the needs of early adopters and the pragmatic concerns of future users, and the tools needed to bridge the gap are not economically viable for researchers.

Therefore, despite the promise of aDBS and the robust commercial success of traditional DBS for movement disorders, misaligned economic incentives and appropriately stringent regulatory requirements imply that there is no manufacturer willing to develop and market a chronic cortical electrode until the science is already complete and the intellectual property is secured. If investigators, industry and funders can find a new way to work creatively and collaboratively to develop cortical recording as a technology, millions of patients stand to benefit.