Introduction

Dyspnoea, defined as “a subjective experience of breathing discomfort” that is prevalent across multiple conditions [1], severely impacts quality of life [2], which may reflect a sparsity of safe and effective treatments. A comprehensive understanding of the central neurophysiology of dyspnoea will help to discover targeted therapies. This approach is facilitated by several advances: 1) distinct components of breathlessness, encapsulated by “air hunger”, “sense of breathing effort” and “chest tightness”, have been characterised that can vary independently [3]; 2) different neural mechanisms have been postulated for the different components [4]; and 3) methods have been established to induce specific components in experimental settings [5]. Air hunger, defined as an “uncomfortable urge to breathe”, can be induced by raising inspired carbon dioxide (CO2) while constraining ventilation, thereby providing a reliable experimental model of a particularly unpleasant component of pathological breathlessness [6].

Cerebral mechanisms have been studied primarily using brain-imaging of experimentally induced breathlessness in healthy individuals. Experimental air hunger, and breathlessness associated with resistive loading, have both elicited strong activation of the insular cortex [7–12] with consistent activation of other regions such as anterior cingulate, orbitofrontal cortex, thalamus, amygdala and basal ganglia also implicated in these studies. How these different areas function as a network for dyspnoea perception is yet to be unravelled.

Deep brain stimulation (DBS), involving implanted electrodes providing constant electrical stimulation of specific brain regions, is a therapy for various neurological conditions including movement disorders, and intractable pain [13]. Several of the DBS sites coincidentally overlap with areas identified in brain-imaging studies of dyspnoea [7] thus offering an alternative approach to investigate cerebral mechanisms.

We previously reported air hunger relief during DBS of the motor thalamus (ventral intermediate nucleus (VIM)), in an individual with post-stroke tremor who coincidentally had pre-existing breathlessness from COPD [14]. The thalamus mirrors phrenic nerve firing, representing the drive to breathe once a certain threshold is reached [15]. One hypothesis that follows is that the air hunger signal, generated by the mismatch between brainstem respiratory corollary discharge and vagal afferents from the lungs, projects to the thalamus, whereby a dyspnoea signal is distributed to cortical sensory areas. Here, we hypothesised that DBS of the VIM would mitigate experimentally induced air hunger, raising the possibility that this region could be a target for relief of intractable dyspnoea by neuromodulation.

MethodsParticipants

16 patients who underwent DBS of the bilateral VIM to treat chronic tremor were recruited consecutively from a single centre at John Radcliffe Hospital (Oxford, UK). All participants provided written informed consent. Ethical approval was provided by South Central Oxford REC (11/SC/0229). The trial was registered at Clinicaltrials.gov (identifier NCT04058457). Eligibility criteria included individuals aged >18 years who have DBS of the VIM. Exclusion criteria included pregnant females, subjects participating in a clinical investigation that includes an active treatment arm which may affect the respiratory system, and indication of acute respiratory problems at the time of the experimental session.

Sample size

Previous studies involving experimentally induced air hunger rated on a visual analogue scale (VAS) by healthy volunteers showed a linear increase in VAS ratings of air hunger with a slope of 6.7%VAS for every 1 mmHg rise in end-tidal CO2 tension (PETCO2) above normocapnia (40 mmHg). The standard deviation of this response slope was 2.4%VAS·mmHg−1 [16]. From this data we determined that an increase in PETCO2 to 47.5 mmHg would produce a mean air hunger rating of 50%VAS with a standard deviation of ±19%VAS. The minimal clinically important difference (MCID) for VAS ratings of air hunger is estimated to be between 10 and 20 mmVAS [17, 18]. We chose 15%, as this lies in the middle of this range to determine the number of participants we would need as a result of a change of this magnitude to be definitive. Assuming a true difference of ±15%VAS in the mean VAS rating of air hunger at this level of hypercapnia between DBS “OFF” versus “ON”, which is above the minimal clinically important difference of ±10%VAS for breathlessness ratings using VAS, we determined that we would need to study 16 patients to be able to reject the null hypothesis with 85% power and a Type I error probability of 0.05 (PS v3; https://cqsclinical.app.vumc.org/ps/).

Experimentally induced air hunger

Participants sat semi-reclined in a comfortable chair. They breathed through a mouthpiece connected via a bacterial filter to a pneumotachograph. The airflow signal was electronically integrated to provide online tidal volume (FV156 respiratory flow integrator; Validyne Engineering Corp, CA, USA). A fast-responding gas analyser (ML206; AD Instruments, Oxford, UK) was used to measure breath-by-breath expired CO2via a sample line inserted into the mouthpiece. A second sample line inserted in the mouthpiece was connected to a differential pressure transducer (±50cmH2O; DP45, Validyne Engineering Corp, CA, USA) for continuous measurement of airway pressure. One-way breathing valves (Hans Rudolph, KS, USA) separated inspiration from expiration. A 3-L anaesthetic bag provided the inspiratory reservoir.

A fixed flow of heated and humidified air (HC150 humidifier; Fisher & Paykel Healthcare, New Zealand) was fed into this bag. Participants breathed to a metronome with a beep-rate set to match the participant's resting spontaneous breathing frequency. To induce air hunger, up to 7%CO2 was added to the inspiratory reserve using an air–oxygen blender (Inspiration Health, Croydon, UK) to which gas cylinders containing 10% CO2 in air, and medical air were connected. Flow of fresh gas to the inspiratory reserve was kept constant and set to match the participants’ spontaneous resting ventilation. Participants rated their air hunger using a slider to operate an electronic 100-mm VAS. Ratings were cued by a light-emitting diode that lit every 15 s (figure 1a). Arterial oxygen saturation was measured using a finger-pulse oximeter. Blood pressure was measured every 2–3 min using the oscillatory cuff method and ECG using six-lead cutaneous silver chloride electrodes.

FIGURE 1

Experimental setup and protocol. a) Experimental setup: participants breathed via a mouthpiece from a 3-L anaesthetic bag into which the flow of fresh gas was set to the participants' baseline minute ventilation (V′E). A metronome was used to set breathing frequency (fR) to the participants' spontaneous rate at baseline. b) Protocol: during the first ramp test (ramp 1), 1-min increments in inspired carbon dioxide (CO2) were implemented using a gas blender which mixed medical gases from compressed gas cylinders, while participants rated any breathing discomfort on a 100-mm visual analogue scale (VAS). The standard debrief afterwards ensured that participants recognised air hunger (AH) as a dominant component of their respiratory discomfort and that they had used the VAS correctly. During steady-state tests, a constant level of inspired CO2 was imposed, targeting 50% full scale of the VAS, which was determined from the initial practice ramp test. O2: oxygen; Bal: balance; N2: nitrogen; VT: tidal volume; PETCO2: end-tidal CO2 tension; Paw: airway pressure; DBS: deep brain stimulation; RD: respiratory discomfort.

Protocol

Participants completed three practice “ramp” tests involving 1-min increments in inspired CO2. For the first ramp, participants rated “any breathing discomfort”. Subsequently, a debrief questionnaire [19] involving volunteered comments followed by patient selection of respiratory and nonrespiratory descriptors from pre-set lists, was used to ensure participants could differentiate air hunger from other sensations. Participants were then instructed to solely focus on, and rate air hunger, during subsequent testing. Two steady-state air hunger tests were then completed which involved a sustained increase in inspired CO2 for 5-min at a level targeting the PETCO2 associated with air hunger ratings approximating 50%VAS during initial ramp tests. The order of ON and OFF DBS was randomised between steady-state tests (figure 1b). End-point was when tolerance was reached, participants came off the mouthpiece or PETCO2 reached 60 mmHg.

Data processing and analysis

Analogue signals were digitised (Micro1401; Cambridge Electronic Design, Cambridge, UK) at a sample rate of 20 Hz and stored for offline analysis using Spike2 software (v10; Cambridge Electronic Design). VAS ratings of air hunger and breath-by-breath PETCO2 were derived by peak-detection (Spike2).

Shapiro–Wilks test was used to check if the data were normally distributed. Given that this was the case, a two-tailed paired t-test was performed to compare average air hunger ratings in the last minute of steady-state between DBS ON and OFF conditions. This region of interest took place 15 min after switching ON or OFF DBS to allow for stabilisation of the patient's tremor. Figure 2 shows a sample physiological trace of the practice ramp (figure 2a) and steady-state tests (figure 2b). The green box represents the region of interest where data were averaged and analysed.

FIGURE 2

Physiological recordings during ramp and steady-state hypercapnic air hunger tests. a) Raw physiological traces for air hunger, end-tidal carbon dioxide tension (PETCO2) airway pressure (Paw) and tidal volume (VT) during the practice hypercapnic ramp with constrained ventilation in both “OFF” and “ON” deep brain stimulation (DBS) conditions. b) Raw physiological traces during the hypercapnic steady-state air hunger tests. Variables which lie within the last minute of the test (green box) were processed and compared between ON and OFF DBS conditions. VAS: visual analogue scale.

Brain imaging

Lead-DBS V3 [20], an electrophysiologically validated processing and analysis pipeline, was used to localise and visualise electrodes. One patient dataset was excluded (011) as the subject had unilateral electrodes and this process requires bilateral electrodes. Pre-operative T1 magnetic resonance imaging (MRI) and post-operative computed tomography (CT) scans were co-registered using a two-stage linear registration (rigid followed by affine) as implemented in advanced normalisation tools (ANTs) [21]. Electrode localisations were corrected for brainshift in post-operative acquisitions by applying a refined affine transform calculated between pre- and post-operative acquisitions that were restricted to a subcortical area of interest. Pre- and post-operative acquisitions were spatially normalised into MNI152NLin2009Asym space (MNI152) [22] using symmetric diffeomorphic image registration implemented in ANTs.

Electrode models were selected and automatically pre-localised in native and template spaces using the PaCER algorithm [23]. If these failed to accurately localise electrodes, tips and trajectories were manually processed within a user interface in Lead-DBS. Orientation of directional DBS leads was determined using the algorithm published by Dembek et al. [24].

Electrodes were then manually localised based on post-operative acquisitions using a tool specifically designed for this task, rendered in template space (MNI152) using a template to define regions of interest, in this case the DISTAL-medium atlas defining subdivisions of the thalamus [25]. Post-operative CT scans were also checked against the electrode positioning in template space. Lead-group [26] was then used to group electrode localisations in template space (figure 3a). Amplitudes were inputted for each electrode in each hemisphere, and active contacts selected (figure 3b). Air hunger responses were then correlated with active contact positionings (figure 3c).

FIGURE 3

Deep brain stimulation (DBS) electrodes and their active contacts in relation to the ventral intermediate nucleus (VIM) for 15 participants visualised in standard MNI space using Lead-DBS V3 software. a) Electrode positionings (solid grey electrodes) within MNI space [25] with the VIM visualised using the DISTAL-medium atlas [29]. b) Transparent electrodes and their active contacts. c) Point-cloud visualisation of active contacts, with their colour correlated to extent of air hunger relief from “OFF” to “ON” DBS. Colour of dots represent extent of relief with blue (most relief) to red (least relief/heightening).

To verify within-subject and MNI space registration accuracy of the LeadDBS model, individual electrode reconstructions were performed in subjects’ native space in a parallel, confirmatory analysis. Post-operative CT images were registered to subjects’ T1-weighted pre-operative MRI series using FMRIB's Linear Image Registration Tool (FLIRT) [27, 28] as implemented in the FMRIB Software Library (FSL) version 6.0.7.10 [29]. Active contacts were reconstructed from known electrode geometry and CT artefacts in subjects’ native space. The FSL FIRST toolbox [30] was utilised to provide individual model-based segmentation of each subject's thalamus, applying recommended boundary-correction settings; grey–white matter segmentation using FSL FAST (FMRIBS Automated Segmentation Tool).

ResultsParticipants

36 patients with DBS of the VIM were approached to take part in this study. 30 patients were eligible, with nine declining participation. Five patients were unable to complete the air hunger test due to their tremor severity during OFF DBS. 16 patients (three female) with essential (n=11), dystonic (n=2), both essential and dystonic (n=1) and Parkinsonian tremor (n=2), were studied between 12 September 2019 (date first patient was studied) and 27 June 2023 (date last patient was studied). Mean±sd age, height and weight were 66±10 years, 174±8 cm and 182±26 lb, respectively (table 1). Electrodes were implanted bilaterally in the VIM in 15 patients (figure 3), and unilaterally on the left in one (014). One of the 15 patients with bilateral electrodes only had left-sided stimulation (006). One patient also had bilateral electrodes in the globus pallidus internus which were OFF at the time of testing (008). Mean±sd time since clinical diagnosis was 25±20 years. Median time from surgery to testing was 23 months (range 1–97 months).

TABLE 1

Demographics, clinical characteristics and deep brain stimulation (DBS) parameters

PatientAge yearsSexHeight cmWeight lbDBSTremorDisease duration monthsMedicationsTime since DBS monthsComorbiditiesAmplitudePulse width μsLeftRightLeftRightVmAVmA169Male170212Bilateral VIMDT14Candesartan, dipyridamole, duloxetine, simvastatin1Previous stroke2.93.58070275Male171187Bilateral VIMET15None1Irregular heartbeat229090375Male164165Bilateral VIMET12None35OCD338070467Male177170Bilateral VIMET10None36None4.151.218060540Female169165Bilateral VIMET51Propanolol 40 mg6None2.42.57060663Male182143Bilateral VIM left side ONET57Clindamycin12None3.5110753Female173154Bilateral VIMDT2None3None1.52.29090873Female163185Bilateral VIMET+DT features40Atenolol, candesartan, pravastatin, fusidic acid25High blood pressure3.23120120972Male182220Bilateral VIMET7Sodium valproate, felodipine, bisoprolol, aspirin, atorvastatin20Acute STEMI, smoker, osteoarthritis, BCC, epilepsy1.92.05110901074Male180177Bilateral VIMET70Apixaban, digoxin, clonazepam, salbutamol, salmeterol inhaler, Calcichew,4Subependymal lesion, asthma2.32.280801175Male180169Bilateral VIMET12None50None3370701275Male162162Bilateral VIMET14None97None2.12.160801369Male176209Bilateral VIMET11Candesartan, amlodipine, lansoprazole, tolterodine37Hypertension, cataract left eye, acute ulcerative colitis, stomach ulcer, urinary urgency1.50.770701471Male190227Left-side VIMPD27Sinemet, pramipexole, opicapone55None2.5701559Male178182Bilateral VIMPD18Stalevo, sinemet, quetiapine91None5.33.51001001659Male174209Bilateral VIMET33None12Smoker, overactive bladder0.7536070Steady-state air hunger test

VIM DBS was observed to have modulatory effects causing a relief of air hunger in 13 patients, an increase in two and no change in one. Test levels of hypercapnia, ventilation, tidal volume and respiratory frequency were well-matched for ON and OFF conditions (mean±sd PETCO2 42.7±4.2 and 42.8±4.4 mmHg; ventilation 13.7±5.6 and 13.4±4.7 L·min−1; tidal volume 0.9±0.5 and 0.9±0.4 L; respiratory frequency 16±5.3 and 15.3±3.2 breaths·min−1). Overall mean steady-state air hunger was significantly lower in ON compared to OFF (52.1±27.8%VAS versus 66.5±20.3%VAS; figure 4a), with a significant mean reduction of −14.4±15.5%VAS (p=0.002), which exceeds the published MCID of 10% for VAS ratings of air hunger [17, 18]. Individual changes in air hunger responses with ON condition are displayed in figure 4b.

FIGURE 4

Effect of deep brain stimulation (DBS) of the ventral intermediate nucleus (VIM) on hypercapnic air hunger. a) Box and whisker plot showing the median (horizontal solid line) and mean (horizontal dashed line), interquartile range (shaded boxes) and upper/lower extremes (whiskers) for ratings of air hunger on a 100-mm visual analogue scale (VAS) during experimentally induced steady-state hypercapnic air hunger with constrained ventilation with DBS electrodes in the VIM with DBS switched off (OFF) and switched on (ON) in 16 tremor patients. b) The change in %VAS air hunger responses when DBS of the VIM is switched ON. The dotted line represents the minimal clinically important difference (MCID) of 10%VAS for VAS ratings of air hunger [17, 18].

14 participants completed the standard debrief after the initial practice ramp to interrogate the respiratory sensations felt during the practice test. Figure 5 depicts the frequency of descriptors rated according to clusters of air hunger, work and effort and “other” components, showing that patients were able to distinguish air hunger from the other clusters. Patients commonly confused the mental work associated with the test, with physical work of respiratory muscles; this may account for the high frequency of selecting “breathing required more work” (figure 5a).

FIGURE 5

Respiratory descriptors associated with the initial practice ramp. The frequency of choosing air hunger, work and effort and the “other” cluster of descriptors as one of the top three sensations experienced at the peak of the initial practice hypercapnic ramp test. Participants were instructed to rate any breathing discomfort during this test. a) Frequency of each descriptor; b) sum of each cluster of descriptors according to their category.

Brain imaging

Supplementary figure S4 shows native space thalamic segmentations, confirming appropriate segmentation accuracy. Supplementary figure S5 demonstrates three-dimensional renderings of each subject's active contacts in native space with individual thalamic segmentations (right column). This was compared with normalised MNI-space electrode reconstructions performed in LeadDBS, with thalamus and VIM estimations from the DISTAL atlas [25] shown for each subject for comparison (left column). This comparison confirms appropriate registration and standard-space normalisation accuracy of the LeadDBS method by an independently Bayesian model-based (FSL/FIRST) approach in subjects’ native space.

Discussion

We have systematically studied the effect of motor thalamic DBS on experimentally induced air hunger in 16 tremor patients. During steady-state tests, participants gave significantly lower air hunger ratings when DBS of the VIM was ON (p=0.002). The extent of air hunger relief (−14.4±15%VAS) exceeded the published MCID [17, 18].

The ascending air hunger signal via the thalamus

The ascending air hunger signal is generated by corollary discharge of respiratory drive from the brainstem tempered by vagal afferents from pulmonary stretch receptors reporting prevailing ventilation; Any mismatch modulates the air hunger signal. This is supported by a variety of evidence, as follows. Grogono et al. [31] demonstrated that inhaled furosemide sensitised pulmonary stretch receptors relieving hypercapnic induced air hunger. Fowler [32] showed that rebreathing after breath hold acutely relieved air hunger in healthy individuals. Flume et al. [33] showed more rapid onset of air hunger during breath-hold and lesser air hunger relief during rebreathe in lung transplant patients who had fewer pulmonary stretch receptors compared to healthy controls.

There remains speculation about the site at which the “mismatch” comparison occurs. The thalamus has been proposed to gate, and subsequently distribute, the ascending air hunger signal to cortical sensory areas where air hunger is consciously perceived. Electrophysiological evidence from studies in cats provides direct evidence for the thalamus representing an intermediary site for the ascending breathlessness signal [15]. In mechanically ventilated paralysed cats, activity of the phrenic nerve, whose firing represents the drive to breathe, was mirrored within thalamic neurons during increasing hypercapnic stimulus once a threshold was reached. It would be interesting to see if this could be confirmed in humans, potentially with the use of intracranial electroencephalography. Human studies also report distinct structural and functional subdivisions of the thalamus being involved in respiratory control receiving respiratory afferents [34].

Role of the thalamus

Functional brain imaging studies report correlations between the activity of thalamic nuclei with both hypercapnic air hunger and with the sense of breathing effort induced by inspiratory resistive loading [7, 8]. Subregions specifically activated included the dorsomedial, ventrolateral and ventroposterior nuclei. The ventral posterolateral nucleus (VPL) forms part of the sensory thalamus lying posterior to the VIM and is thought to be a region which can amplify or suppress ascending pain signals [35]. The VPL has recently been shown to correlate with breathlessness anticipation and its intensity in athletes [36]. Ventroposterior groups have also become DBS targets for neuropathic pain relief, demonstrating their role in sensory processing [37]. The pulvinar nucleus, the most posterior group, has also been shown to become activated during induced hypercapnia [38]. The VIM itself has not previously been implicated in any brain imaging studies of breathlessness.

Modulation of air hunger via VIM DBS

The high frequencies (115–155 Hz) used to relieve tremor creates a “reversible lesion” within the field of stimulation preventing aberrant firing patterns within the VIM [39]. DBS disrupts inputs from outputs, determined from its similarities with effects of permanent lesion [40]. We raise the following possible mechanisms of air hunger relief by VIM DBS (noting that without further imaging analysis, these are highly speculative).

1) The VIM directly, or indirectly through its connections with nearby sensory nuclei, gates the air hunger signal to higher areas thus DBS may upregulate this gating control.

2) The field of stimulation extends to neighbouring sensory areas that transmit the air hunger signal. Spread of stimulation to the VPL would be expected to induce paraesthesia in contralateral limbs, as noted with VPL stimulation. The absence of this in the patients reported here undermines this proposed mechanism.

3) DBS of the VIM mediates network-wide effects in other areas involved in processing of air hunger perception, akin to motor cortical stimulation for relief of pain, that may act via conferring functional changes in subcortical areas including the thalamus [41] and areas involved in processing of sensory affect such as the anterior cingulate and insular cortex [42]. The motor cortex projects to the thalamus and zona incerta while receiving inputs from thalamic nuclei [43].

4) DBS of motor regions proximal to the VIM may have a top-down influence on respiratory control, as demonstrated in sheep [44]. However, no differences in resting breathing between OFF and ON DBS were observed.

5) DBS of the VIM has been shown to significantly improve depression [45]. Aggravation of dyspnoea is associated in those with depression and negative affect [46]. Although the causal relationship between the two remains unclear, depressive mood could generate a heightened perception of experimentally induced air hunger during OFF VIM DBS. Although pre- and post-operative scores of depression and anxiety were not measured in this study, none of the patients in this cohort had pre-existing depression. In an attempt to interrogate this, albeit in a crude manner, we compared heart rate variability as an indirect indicator of anxiety between ON and OFF DBS in 10 patients and found no differences in root mean square standard deviation of successive differences of R-R interval (supplementary figure S2).

Differences in extent of relief

The change in air hunger with DBS ON ranged from +7.5 to −52.5%VAS. This wide range may be accounted for by differences in the volume of tissue activated (VTA) (supplementary figure S6) due to appreciable individual variation in position of electrodes and active contacts but could also reflect natural individual variation in the response of the neural tissue to the stimulation.

We posit the following explanations for the variability of electrode positionings and VTA: 1) different electrode trajectories selected during surgery; 2) differences in active contact selections which were based on largest clinical relief and therefore varied between participants; 3) that atlas-based locations were used for DBS targeting, as the VIM is not visible on MRI, which is further confounded by individual variation in brain anatomy; 4) differences in amplitude of stimulation, the key determinant of VTA (table 1 and figure 3); and 5) accuracy by which Lead-DBS can localise electrodes.

Individual differences in strength of connection between the thalamus and regions of interest within the perceptual framework of dyspnoea may also contribute to the heterogeneity in air hunger relief. The insular cortex is considered to be a principal site for breathlessness perception and is universally and strongly activated in brain imaging studies of breathlessness. Significant connections between insular cortex and thalamus have been demonstrated using high angular resolution diffusion-weighted imaging [47].

Reports from functional resting state activity also show a connection between the thalamus and insula. Wiech et al. [48] reported the most significant region connecte