The effect of M‐current activation on controller gain and obstructive sleep apnoea severity: a randomised controlled trial using flupirtine

Abstract

Abstract

Ventilatory control instability, or high loop gain (LG), contributes towards upper airway collapse in approximately one‐third of people with obstructive sleep apnoea (OSA). A high LG can be the product of elevated chemosensitivity (controller gain) and/or an excessive ventilatory output (plant gain). Therapies such as carbonic anhydrase inhibitors (targeting plant gain) have been shown to reduce OSA severity; however, there is a lack of viable pharmacological options targeting controller gain. This study investigated the effect of flupirtine (400 mg), a KCNQ potassium channel opener, on LG and OSA severity in fifteen moderate‐to‐severe OSA patients through a randomised, double‐blind, placebo‐controlled trial. Despite the hypothesised potential of flupirtine to reduce LG by attenuating chemosensory activity, our findings revealed no significant effect on LG and OSA severity. The lack of overall efficacy of flupirtine is most likely due to multifactorial nature of OSA and the challenges of its management. Our findings suggest a need for a nuanced understanding of OSA pathogenesis and caution against the use of flupirtine in managing OSA. While, pharmacological modulation of ionic channels within the ventilatory control system presents a promising strategy, given the plethora of robust targets available, it remains to be determined whether an effective treatment can capitalise on a single predominant ionic current ubiquitous throughout the ventilatory system, or if a more successful approach necessitates the simultaneous modulation of multiple targets. This research enhances our understanding of the ventilatory control system's contribution to OSA and the complexity of finding a one‐size‐fits‐all treatment.

Key points

• Around one‐third of obstructive sleep apnoea (OSA) cases involve an unstable control of breathing, leading to airway collapse.

• This research examined whether the drug flupirtine could stabilise breathing control and reduce OSA severity in 15 patients.

• Flupirtine, which was expected to improve breathing control by reducing chemosensitivity, showed no significant benefit for OSA.

• While targeting ionic channels in the breathing system is promising, the search for an effective OSA treatment may require addressing multiple targets simultaneously.

Abstract figure legend Investigating the effect of flupirtine, a KCNQ channel opener, on ventilatory control instability (i.e. loop gain) and obstructive sleep apnoea (OSA) severity. Left panel: high loop gain has been shown to contribute to airway collapse in patients with OSA. Reducing chemoreflex sensitivity (controller gain, a key component of overall LG) is associated with improvements in OSA severity. Middle panel: flupirtine was hypothesised to reduce controller gain by inhibiting carotid body signalling via KCNQ channel activation, thereby lowering LG. Right panel: the study assessed whether KCNQ activation could decrease LG and improve upper airway collapsibility and OSA severity. Key findings: flupirtine treatment did not result in significant reductions in LG or OSA severity compared to placebo. However, we observed a varied response: one‐third of patients experienced improvements in OSA severity, while another third showed worsened outcomes, particularly with increased airway collapsibility.

Introduction

Destabilisation of the ventilatory control system has been linked to both initiating and exacerbating upper airway collapse during sleep leading to obstructive sleep apnoea (OSA; Deacon & Catcheside, ²⁰¹⁵). The purpose of the ventilatory control system is to maintain sufficient ventilation to enable adequate gas exchange. When disturbances in ventilation occur, commensurate changes in ventilatory drive and effort are produced in response to achieve homeostasis. However, it is also well known that oscillations in ventilatory drive lead to corresponding fluctuations in hyper/hypotonicity of the upper airway musculature. In an individual with a collapsible airway, these periods of low‐drive hypotonicity can trigger repeating cycles of airway collapse. In order to quantify the stability of the ventilatory control system experimentally, investigators often measure the ‘loop gain’ (Khoo, 2001) of the system. Simplistically loop gain (LG) is defined as the ratio of the response (i.e. the magnitude of the overshoot in ventilation) to a given disturbance (i.e. the reduction in ventilation caused by an apnoea) within the ventilatory feedback system. Importantly, LG is the product of feedback gains interacting between two system components: a controller (central respiratory centre) that directs a plant (lungs, blood tissue) to produce an output (CO₂ removal) according to a set‐point (eupnoea).

To date, several interventions targeting the ventilatory control system have been trialled in patients with OSA. These interventions have been shown to lower LG, reducing the accompanying ventilatory oscillations that contribute to airway collapse. This can be achieved by either carbonic anhydrase inhibitors (Edwards et al., 2012; Hoff et al., ²⁰²⁴; Schmickl et al., ²⁰²⁰) or supplemental oxygen (Edwards, Eckert et al., 2014; Sands, Edwards, Terrill, Butler et al., ²⁰¹⁸; Wellman et al., ²⁰⁰⁸) which predominantly reduce the plant and controller sensitivity/gain, respectively. Both interventions have been shown to reduce LG by ∼40% and ∼50%, respectively, leading to an approximate 50% reduction in OSA severity as measured by the apnoea‐hypopnoea index (AHI). Despite the encouraging outcomes of such interventions, there are currently no approved pharmaceutical therapies targeting a reduction in controller gain for the treatment of OSA.

The ventilatory chemosensory system (i.e. controller gain) has exquisite complexity and robustness that involves a multitude of sensors and transductive pathways; including but not limited to ion channels, mitochondrial mechanisms, neurotransmitters and gaseous messengers (Iturriaga et al., 2021; Ortega‐Sáenz et al., ²⁰²⁰). In the interest of avoiding polypharmacy options, an alternative strategy is to lower controller gain by identifying a common mechanistic target shared by both central and peripheral chemosensors (which determine the controller gain). Potassium (K⁺) ion channel closure is a crucial and necessary step in chemosensory transduction. Therefore, maintaining an activated outward K⁺ current may provide a mechanism to reduce chemosensory activity (Wang & Kim, 2018). The Kv7 refers to a family of voltage‐gated K⁺ conductance channels; the Kv7.2 and 7.3 subunits generate a subthreshold ionic current (M‐current) that can strongly dampen neuronal excitability, and KCNQ channel openers can enhance this action. Furthermore, these channels are expressed in the central (Mulkey et al., 2015) and peripheral (Buniel et al., 2008) chemoreceptors.

The aminopyridine drug flupirtine is a KCNQ opener and a novel therapeutic for diseases associated with cellular hyperexcitability (Jones et al., 2021). Channel activation by flupirtine leads to hyperpolarisation of the membrane potential which attenuates action potentials (Lawson, 2019). Flupirtine is also one of the few available KCNQ opening drugs which exhibits efficacy for the relevant Kv7.2 and 7.3 subunits (Surur et al., 2019). As a proof‐of‐concept to demonstrate the utility of pharmacologically reducing controller gain to reduce OSA severity, we sought to investigate the effect of flupirtine in moderate‐to‐severe OSA patients. We hypothesised that flupirtine would reduce LG (via reductions in controller gain) and this in turn would reduce OSA severity.

Methods

Ethical approval

All participants were informed of the study purpose and procedures and provided their written informed consent. Ethical approval was obtained through the Monash Health Human Research Ethics Council (no. 51271) and the trial was registered at anzctr.org.au (ACTRN 12 618001579280). The study conformed to the standards outlined by the Declaration of Helsinki (including clause 35 of the Declaration).

Participants

Fifteen moderate‐to‐severe OSA patients (4 women and 12 men; aged 40–70 years: see Table 1) were screened and recruited from the community via social media advertisement before enrolment in a phase II, single centre, randomised, double‐blind, placebo‐controlled, clinical trial. Participants were confirmed to have had a previous OSA diagnosis (AHI > 15 events/h, evidenced from a Type 1 or 2 Sleep Study Report) or were currently being treated with continuous positive airway pressure (CPAP)/oral appliance for their OSA. Participants were excluded based on (1) medical and concurrent medication exclusions in accordance with the most recently published summary of product characteristics for flupirtine, (2) concurrent or recent (within the past month) use of any medication known to influence sleep, arousal, circadian rhythm, breathing or muscle function, (3) if pregnant or breast feeding, (4) had a history of shift work or rotating shifts in the month prior, (5) were involved as a driver in a motor vehicle accident during the past 2 years where the cause of the accident was attributed to driver sleepiness and (6) inability to sleep supine. Participants were required to abstain from any current OSA treatment (i.e. CPAP or oral appliance) for a minimum of 5 nights prior to each assessment in order to ensure OSA severity has returned back to pre‐treatment values (Kohler et al., 2011).

Table 1.

Patient characteristics

Age (yrs)	54 (9)
Sex (female: male)	3:12
BMI (kg/m2)	30.7 (5.4)
Neck circumference (cm)	41 (3.9)
Waist circumference (cm)	105 (14.5)
Epworth Sleepiness Scale	6 (5)
Race, n(%)
Caucasian	13 (86.6%)
Asian	2 (33.3%)
Currently treated*, n(%)	9 (60%)
CPAP	9
Oral appliance	1
Use >6 h/night, n(%)	5 (75%)
Comorbidities, n(%)
Hypertension	3 (20%)
Diabetes	nil
Depression/anxiety	3 (20%)
Hypercholesterolaemia	1 (6.7%)
Arthritis	1 (6.7%)
Chronic pain	1 (6.7%)
GORD	2 (33.3)
Medications, n(%)
ACE‐I/ARB	3 (20%)
Non‐steroidal anti‐inflammatory	1 (6.7%)
Amino‐salicylate	1 (6.7%)
Anti‐rheumatic	1 (6.7%)
Antidepressant	3 (20%)
	2 (33.3%)

Abbreviations: ARB, angiotensin receptor blocker; ACE‐I, angiotensin‐converting enzyme inhibitor; AHI, apnoea‐hypopnea index; BMI, body mass index; GORD, gastro‐oesophageal reflux disease.

^*Note that 1 patient alternated between CPAP and oral appliance use. Data presented as the mean (SD) or n (%) as appropriate; n = 15.

Trial design

The trial used a two‐way crossover design: 2‐condition (active vs. placebo), 2‐period (1‐week washout), 2‐sequence (AB|BA). Participants were preassigned by a third party to a computer randomised treatment sequence based on order of enrolment to determine which period they would receive – either the active condition (400 mg of modified release flupirtine) or placebo (lactose), presented as a visually identical oral capsule. Each period consisted of an evening chemoreflex testing session followed by an overnight polysomnography (PSG) sleep study at the Monash University Sleep and Circadian Medicine Laboratory (Melbourne, AUS). The co‐primary outcomes of the trial were the AHI and LG determined from the overnight PSG. Secondary outcomes included subjective sleepiness (Stanford Sleepiness Scale), subjective sleep quality (worst:0, best:10), blood pressure (diastolic and systolic) and heart rate measured with an automatic sphygmomanometer, as well as sleep efficiency, hypoxic exposure, arousal index, and chemoreceptor sensitivity.

Protocol

Arrival for each test period was scheduled for 06.30 h. Average sleep quality in the last week was assessed with a 10‐point Likert scale (1 = worst, 10 = best). Anthropometric measurements (weight, height, neck and waist circumference) were also taken during the first testing period. The condition‐blinded researcher then administered the investigational drug product (either active or placebo) to the participant and waited 90 min prior to chemoreflex testing (described below). Following chemoreflex testing, participants were instrumented with PSG sensors (described below). Supine blood pressure was measured in triplicate 10 min pre‐sleep opportunity, and again 10 min after awakening in the morning. Lights were turned off at the person's habitual bedtime indicating the start of the sleep opportunity. Body position was visually confirmed by infrared video recording. Upon awakening, participants were asked to compare their subjective experiences of the sleep period in reference to their normal ‘everyday’ experience using 10‐point Likert scale in two domains: sleep quality (worst‐best) and awakening (unrefreshed‐refreshed). Alertness levels were also assessed using the Stanford Sleepiness Scale. Participants were asked about any noticeable side effects and/or effects that were outside their normal experience. Further assessments were conducted at the second week visit (1 week post visit 1) or by phone 1 week after visit 2. Any indication of adverse effects prompted administration of an adverse events questionnaire.

Polysomnography

Standard American Academy of Sleep Medicine (AASM) clinical PSG channels were recorded to identify sleep states and cortical arousals: EEG (frontal, central and occipital regions of both left and right hemisphere referenced to mastoid), left and right electrooculogram, and submental electromyogram (EMG); electrocardiogram; thoracic and abdominal respiratory effort (respiratory inductance plethysmography (RIP); blood oxygen saturation (finger pulse oximetry); snoring (suprasternal notch placed tracheal microphone); limb movements (bilateral anterior tibialis EMG) and central body position (sensor attached to thoracic RIP band). Calibrated airflow and pressure measurements were recorded using participants’ natural breathing through a pitot tube connected to a sealed oronasal mask (Resmed Airfit F10) modified to facilitate measurement of mask pressure (by differential transducer; Hans Rudolph Inc., Kansas City, KA, USA), and inspired/end‐tidal values of $P_{\mathrm{C}{\mathrm{O}}_2}$ and $P_{{\mathrm{O}}_2}$ (VacuMed, Ventura, CA, USA).

Chemoreflex testing

Each participant was first instrumented with electroencephalography (EEG; to monitor alertness levels during chemoreflex testing) and then fitted with a sealed oronasal mask. In addition, a 2‐way non‐rebreathing valve (Hans Rudolph Inc.) was connected (distal to the pitot tube) with an exhalation port expelling into the room. The inhalation valve was connected to a 3‐way stopcock valve via a 1.8 m CPAP tube (19 mm ID). This valve allowed switching between either room air from an adjacent control room or a 100 l balloon filled with a hypoxic/hypercapnic air mixture (14% O₂, 6% CO₂, N balance). The mask seal and circuit connections were tested for leaks using forced occlusions. Participants were instructed to relax and breathe normally in the supine position while watching a pre‐approved selection of emotionally neutral video material. Baseline resting ventilation (when breathing room air) was recorded for 5 min before being transiently exposed to hypoxic/hypercapnic air mixture administered for 30 s to create a simulated respiratory disturbance. A following period of 150 s was allowed before a further disturbance (30 s) was performed. The test continued until a minimum of 10 satisfactory disturbance‐response (30 + 150 s) cycles were achieved, after which participants rated the level of exertion they experienced using the Borg scale. Ventilatory signals were recorded at a sampling rate of 256 Hz using systems from both Cambridge Electronic Design Limited (Spike 2 and Power 1401; Cambridge, UK), and Compumedics (Profusion and Grael; Melbourne, Australia) which also recorded the EEG and PSG channels.

Data analysis

Sleep and respiratory events from recorded PSG data was scored using Profusion PSG software (Compumedics, Melbourne, Australia) by an experienced sleep scientist (S.H.) blinded to condition. Scoring criteria used the AASM 2012 recommended hypopnoea scoring criteria (Berry et al., 2012). The EEG arousals and respiratory event scoring was modified by a second blinded sleep scientist (L.T.) for the sole purpose of adjusting event onset/offset timings, which are necessary parameters required for accurate non‐invasive endotype analysis (described below).

Chemoreflex data was processed and analysed with customised software using MATLAB (The MathWorks Inc, Natick, MA, USA) as described previously (see Sands, Edwards, Terrill, Butler et al., 2018). Primary variables of interest were LG, measured at two resonance frequencies: natural (LGn) and 1‐cycle/minute disturbances (LG1). Measurements of controller and plant gain were also assessed.

Endotypes were measured using signal processing techniques applied to the overnight PSG data using validated techniques as previously described (Terrill et al., 2015; Sands, Edwards, Terrill, Taranto‐Montemurro et al., ²⁰¹⁸; Sands, Terrill, Edwards et al., ²⁰¹⁸). Briefly, the PSG data was divided into 7‐min windows to estimate breath‐by‐breath measurements of ventilatory drive (V _DRIVE). The OSA endotypes were then calculated based on the following principles (see Fig. 1):

• Loop gain (Terrill et al., 2015): Recorded flow signal was fitted to a standard ventilatory control model consisting of a gain, time constant and delay component (Chemical Drive, see Fig. 1A ), which produces the V _DRIVE signal (ventilatory drive, see Fig. 1A ). The gain term in this model represents the LG (herein reported as response to a 1‐cycle/min disturbance), where a higher LG reflects a more unstable ventilatory control system.

• Arousal threshold (Sands, Terrill, Edwards et al., 2018): The median value of V _DRIVE immediately preceding scored respiratory related arousals (dashed line, per Fig. 1A ).

• Collapsibility and muscle effectiveness (Sands, Edwards, Terrill, Taranto‐Montemurro et al., 2018): Three measurements of airway collapsibility inferred at different levels of ventilatory drive during scored obstructive events were taken (V _MIN, V _PASSIVE and V _ACTIVE). Measurements are expressed as percentage of median eupnoeic ventilation (V _EUPNOEA; when ventilation and V _DRIVE are matched), where lower values indicate greater airway collapsibility (see Fig. 1B ).

  • ○The V _MIN (ventilation at nadir ventilatory drive), and V_PASSIVE (median ventilation at eupnoeic V _DRIVE during a scored obstructed) reflect measures of passive airway collapsibility.
  • ○ V _ACTIVE is the ventilation observed just prior to the termination of a respiratory event when V _DRIVE and therefore respiratory muscle activation is at peak. It reflects how effective the upper airway muscles at reopening the airway and recovering ventilation. A lower value denotes a more collapsible airway, and thus reflects less effective muscle compensation.

Figure 1. Quantifying the obstructive sleep apnoea (OSA) endotype traits from polysomnographic (PSG) data.

A, top panel: electroencephalogram (EEG) power (black) showing scored arousals (green) and calibrated flow signal (l/min), with obstructive events (apnoeas or hypopnoeas) highlighted in purple. Bottom panel: breath‐by‐breath observed ventilation (purple line), along with retrofitted models of chemical drive (black line) and ventilatory drive (combining chemical and arousal components; green line), are shown. Loop gain is derived from the proportional increase in chemical drive (ventilatory response) following ventilation reductions (ventilatory disturbance). The arousal threshold is defined as the ventilatory drive preceding arousal‐associated events (e.g. dashed black line). The ventilatory response to arousal (VRA) is quantified as the difference between chemical drive (black) and overall ventilatory drive (green) during arousal‐related events. B, the endogram summarises how ventilatory drive (modelled) and ventilation (observed) vary across the whole sleep period (continuous black line shows median values and red shaded area represents interquartile range). Once this relationship is known the other remaining traits are derived. V _MIN represents the ventilation at the lowest decile of drive (minimal drive), V _PASSIVE corresponds to ventilation under passive neuromuscular conditions (when eupnoeic drive = 100%), and V _ACTIVE represents ventilation achieved during maximal neuromuscular stimulation (i.e. at the arousal threshold). The arousal threshold is quantified as the median level of drive preceding the termination of any scored obstructive event. [Colour figure can be viewed at wileyonlinelibrary.com]

Lastly, as an exploratory analysis, we utilised a recently published clinical score developed to identify OSA patients with high carotid body chemosensitivity (hCBC; Li et al., ²⁰²¹) based on three criteria obtained from PSG: (1) total AHI, (2) fraction of AHI events in REM sleep, and (3) longest obstructive apnoea duration. A hCBC score of ≥ 3 identifies OSA patients with greater carotid body chemosensitivity (i.e. controller gain) with a reported sensitivity of 79% and specificity of 88%.

Statistical analysis

All statistical analyses were performed using GraphPad Prism 8.4.3 (GraphPad Software Inc., San Diego, CA, USA). Main analyses used data recorded in all sleep states (Total = REM+NREM) not accounting for body position. V _PASSIVE and arousal threshold were square root transformed to improve normality, as done previously (Sands, Terrill, Edwards et al., 2018; O'Driscoll et al., ²⁰¹⁹). Effect of treatment condition on outcomes and dependent variables were assessed using paired t tests (placebo vs. flupirtine). Linear regressions were used to assess relationships between percentage change in endotypes and percentage change in AHI. To assess predictors of response, baseline endotypes were regressed against the change in AHI. Normal distributions were assessed using Shapiro‐Wilk test. Statistical significance was accepted at P < 0.05. Data are reported as the difference from placebo (Δ) using the mean (SD) or median [Q1–Q3], as appropriate.

Results

Participants

Fifteen OSA patients (AHI range = 10–109 events/h) were studied (see Fig. 2) – of these patients, nine were actively being treated with CPAP or an oral appliance (treatment duration ranged between 1 month and 7 years). All enrolled participants completed both study testing periods. Patient characteristics are detailed in Table 1. In brief, the patient sample comprised middle‐aged, overweight and predominantly male individuals.

Figure 2. Consort diagram.

    [Colour figure can be viewed at wileyonlinelibrary.com]

Effect on AHI (co‐primary outcome) and sleep parameters

Compared to placebo, flupirtine did not change OSA severity when assessed overall or during NREM and REM sleep (see Fig. 3 and Table 2). Flupirtine significantly increased the total sleep time (ΔPlacebo = +42 (±52) min, P = 0.007), which was consistent with overall better sleep efficiency and lower wake after sleep onset (WASO; Table 3); however, no differences were observed between sleep stage durations (as minutes or percentage) or on measures of sleep latency. There was no difference in oxygen desaturation or hypoxic burden between conditions (Table 2).

Figure 3. The overall apnoea‐hypopnoea index (primary outcome) was not different between placebo and flupirtine.

Black squares indicate the mean (SD) for each condition. Individual responses (open circles) showed wide variation (n = 15). Data were compared using paired t test (P = 0.780).

Table 2.

Effects of flupirtine on markers of sleep apnoea severity compared with placebo

	Placebo	Flupirtine	P value
AHI (events/h)
Total	48 (29)	49 (30)	0.780
REM, n = 13	42 (21)	52 (21)	0.097
NREM	47 (29)	48 (31)	0.890
NREM supine	63 (31)	62 (33)	0.745
Hypopnoea (%)	65 [47–98]	71 [57–84]	0.762
Apnoea duration (s)	25.0 (4.6)	24.0 (3.5)	0.798
Hypopnoea duration (s)	24.0 (5.0)	24.0 (4.0)	0.967
Hypoxic markers
Nadir $S_{\mathrm{p}{\mathrm{O}}_2}$ desaturation (%)	88 [85–91]	86 [84–90]	0.272
Average $S_{\mathrm{p}{\mathrm{O}}_2}$ desaturation (%)	5 [4–6]	5 [4–5]	0.656
Total ODI 3%	16 [10–53]	22 [4–59]	0.670
Total ODI 4%	10 [4–45]	16 [2–45]	0.453
Hypoxic burden ([%min]/h)	50 [31–141]	76 [21–117]	0.454

Data presented as means (SD) or median [interquartile range] and compared using paired t test (parametric or non‐parametric as appropriate). n = 15 for all values except REM AHI (2 patients did not achieve REM sleep on placebo).

Abbreviations: AHI, apnoea‐hypopnoea index; F _Hypopnoea, the amount of all respiratory events that were hypopnoeas as a fraction of total respiratory events; NREM, non‐rapid eye movement; ODI, oxygen desaturation index; $S_{\mathrm{p}{\mathrm{O}}_2}$, peripheral capillary oxygen saturation.

Table 3.

Effects of flupirtine on sleep and cardiovascular parameters compared with placebo

	Placebo	Flupirtine	P value
Total sleep time (min)	334 (83)	376 (50)	0.007
Supine sleep (%)	52 [40–82]	66 [53–76]	0.252
Sleep efficiency (%)	65 (17)	75 (9)	0.007
Sleep latency (min)	26 [18–35]	23 [18–43]	0.989
REM latency (min), n = 13	174 [93–271]	100 [79–188]	0.068
WASO (min)	127 (77)	77 (48)	0.002
NREM (%TST)	90 (8)	88 (7)	0.200
N1 (%TST)	13 [10–33]	21 [8–30]	0.524
N2 (%TST)	59 (12)	57 (10)	0.375
N3 (%TST)	6 [3–17]	8 [2–18]	0.188
REM (%TST)	10 (8)	12 (7)	0.200
Arousal index (arousals/h)
Total	45 (20)	46 (19)	0.699
Respiratory	31 [24–37]	34 [25–44]	0.252
Waking refreshed	5 [2–6]	5 [4–6]	0.637
Subjective sleep quality	4.7 (2.3)	4.9 (2.2)	0.787
Subjective sleepiness (SSS)	3 [2–3]	3 [2–4]	0.249
Evening heart rate (beats/min) *, n = 13	65 (11)	64 (16)	0.972
Morning heart rate (beats/min) *, n = 13	66 (10)	64 (10)	0.747
Blood pressure (supine)
Evening systolic (mmHg) *, n = 13	131 (16)	129 (14)	0.783
Morning systolic (mmHg) *, n = 13	136 (21)	128 (18)	0.026
Evening diastolic (mmHg) *, n = 13	78 (6)	76 (8)	0.613
Morning diastolic (mmHg) *, n = 13	82 (12)	79 (10)	0.395

Sleep efficiency measures time asleep/total sleep period; sleep latency measures time taken to sleep onset. Subjective sleepiness was measured with the Stanford Sleepiness Scale (SSS), a score above 3 was considered sleepy; subjective sleep quality and waking refreshed were measured using a visual analogue scale, a higher score indicates higher sleep satisfaction. Data presented as means (SD) or median [interquartile range] and compared using paired t test (parametric or non‐parametric as appropriate).

Abbreviations: N1, stage 1 sleep; N2, stage 2 sleep; N3, stage 3 sleep; NREM, non‐rapid eye movement; REM, rapid eye movement; WASO, wake after sleep onset.

^*Indicates a two factor RM‐ANOVA was used to compare the effect of group (placebo vs. flupirtine) and time (evening vs. morning) on heart rate, systolic and diastolic blood pressure (P values reported here represent post hoc comparisons (Tukey's)). n = 15 unless otherwise indicated.

Effect on chemoreflex sensitivity during wake and the endotypic traits during sleep

While flupirtine increased tidal volume (ΔPlacebo = +71 (±113) ml/min, P = 0.035), it had no effect on any other ventilatory parameters during wake (see Table 4). There was no impact of flupirtine on LG during wake or its components controller and plant gain (see Table 4), whether it was assessed using the natural frequency (LGn) or at a cycle of 1/min (LG1). Likewise, there was no difference in the clinical chemosensitivity (hCBC) score between conditions (P = 0.175). In addition, no relationship was found between the hCBC score (calculated under both placebo and flupirtine conditions) and our estimates of LG, whether they were measured during wake (placebo, active P values = 0.482, 0.782) or during sleep (co‐primary outcome: P values = 0.828, 0.762). The hCBC score was not associated with measures of controller gain (P values = 0.919, 0.976). Lastly, there was no evidence that flupirtine modified any of the key OSA endotypes overall (see Table 5) or via specific sleep states (see Appendix 1).

Table 4.

Effects of flupirtine on markers of awake baseline respiratory measures and chemosensitivity (i.e. loop gain) compared with placebo

	Placebo	Flupirtine	P value
Baseline respiratory variables
Minute ventilation (l/min)	8.9 (2.5)	9.9 (2.1)	0.170
Tidal volume (ml)	645 (204)	716 (171)	0.035
Total respiratory cycle time (s)	4.0 [3.6–5.3]	4.0 [3.6–4.6]	0.819
Inspiratory time (s)	1.7 (0.5)	1.7 (0.4)	0.965
End‐tidal carbon dioxide (mmHg)	38 (2)	37 (4)	0.238
Clinical chemosensitivity score (hCBC)	3 [2–3]	3 [3–6]	0.175
Dynamic loop gain
Loop gain (LGn)	0.16 [0.10–0.19]	0.17 [0.12–0.30]	0.060
Loop gain (LG1)	0.33 (0.14)	0.41 (0.24)	0.112
Controller gain	0.39 (0.24)	0.44 (0.24)	0.360
Plant gain	0.99 (0.37)	0.98 (0.35)	0.907
Delay (s)	5.2 [3.8–6.3]	5.4 [4.2–9.6]	0.268

Baseline respiratory variables and chemosensitivity data as measured during awake chemoreflex testing. Baseline respiratory variables were taken in the 5 min of baseline breathing before commencement of the chemoreflex testing. hCBC = high carotid body chemosensitivity score (higher score indicates greater chemosensitivity). Data presented as means (SD) or median [interquartile range] and compared using paired t test (parametric or non‐parametric as appropriate). n = 14 (one participant was not able to provide usable data in a chemoreflex test session).

Table 5.

Effects of flupirtine on measures of sleep apnoea endotypes compared with placebo

	Placebo	Flupirtine	P value
Loop gain	0.68 (0.19)	0.69 (0.18)	0.859
ArTH (%V EUPNOEA)	161 (34)	163 (33)	0.550
VRA (%V EUPNOEA)	31 [24–42]	31 [24–47]	0.359
V PASSIVE (%V EUPNOEA)	68 [58–72]	64 [51–72]	0.188
V MIN (%V EUPNOEA)	57 (15)	50 (21)	0.062
V ACTIVE (%V EUPNOEA), n = 14	83 (28)	73 (33)	0.093

Data presented as means (SD) or median [interquartile range] and compared using paired t test (parametric or non‐parametric as appropriate). n = 15 unless indicated otherwise.

Abbreviations: ArTH, arousal threshold; VRA, ventilatory response to arousal; V _eupnoea, eupnoeic ventilation; V _passive, ventilation when upper airway dilator muscles are hypotonic/passive; V _active, ventilation when upper airway dilator muscles are maximally activated; V _min, ventilation at lowest ventilatory drive.

Predictors of treatment response

There were no key OSA endotypes measured at placebo that predicted the response to treatment. However, the baseline (placebo) high hCBC score was a significant predictor of treatment outcome measured as change in AHI (r²  = 0.31, P = 0.029).

Additional outcomes

Despite the sedative properties of flupirtine, there was no difference between condition and post‐sleep subjective ratings of alertness (P = 0.637), sleepiness (P = 0.249), or perceived sleep quality (P = 0.787); see Table 3. Flupirtine also led to a decrease of 8.2 ± 9.5 mmHg in morning systolic blood pressure (P = 0.026). Overall, a single night dose of flupirtine was well tolerated with no serious adverse events reported. Regarding side effects associated with flupirtine, there was one report of a dry mouth upon awakening. In contrast, there were two reports of experiencing a dry mouth upon awakening associated with placebo in addition to one report made for each of nocturia, headache and excessive morning tiredness.

Post hoc responder analyses

As a sensitivity analysis, participants were ranked in order of percentage change in total AHI (ΔAHI = [placebo − flupirtine/placebo] × 100) and binned into tertiles. This approach resulted in the upper tertile being classed as ‘good’ response group (ΔAHI = −33%; n = 5) and the lower tertile as a ‘poor’ response group (ΔAHI = +58%; n = 5). We then conducted exploratory analyses using two‐way repeated measure ANOVAs (condition [placebo × flupirtine], responder [good × poor]) to determine any main effects and/or interactions that may help explain the variability in treatment response. Key findings are detailed below.

As shown in Fig. 4A , the AHI during the placebo condition was not different between the two groups (F [1, 8] = 3.1, P = 0.881); the good responder group was characterised by a decrease of 12.2 events/h in the AHI (95% CI [−24, −0.16], P = 0.047) while the poor responder group showed a significant increase in AHI by 17 events/h (95% CI [5, 29], P = 0.009). There were no other characteristics that separated the poor vs. good responder groups regarding age, sex or BMI. There were, however, several endotype differences between the two response groups. There were significant interaction effects (but no main effects; F values [1,8] P values = 0.130–0.953) that emerged on two key OSA endotypes: The arousal threshold (ArTH; F [1,8] = 7.6, P = 0.025) and collapsibility as measured by both V _PASSIVE (F [1,8] = 12, P = 0.009) and V _MIN (F [1,8] = 7.3, P = 0.027). For good responders, the arousal threshold remained stable between conditions but increased within poor responders after taking flupirtine (ΔArTH = +12 [−25 to −0.07] %Eupnoea; P = 0.049; Fig. 4B ). Similarly, the airway collapsibility, as measured by V _MIN, was unchanged within good responders (P = 0.596), but significantly decreased within poor responders (ΔV _MIN = −13 [0.5–26] %Eupnoea; P = 0.042; Fig. 4C ). When measured as V _PASSIVE, there was a significant difference in response to flupirtine between response groups (Mean Diff = 21 [1.3–42] %Eupnoea; P = 0.036; Fig. 4D ).

Figure 4. Exploratory responder analysis of two groups: good and poor responders.

Good responders, blue circles (n = 5); poor responders, red squares; n = 5. Main graphs show between‐condition effects as means and 95% confidence intervals and inset graphs show the individual data (black bars indicate means in inset). A, total AHI; B, arousal threshold; C, collapsibility measured as V _MIN; D, collapsibility measured as V _PASSIVE. Note all endotypes values were measured from total sleep time as %eupnoeic ventilation. Two‐way repeated measure ANOVAs (condition [placebo × flupirtine], responder [good × poor]) were utilised to determine any main effects and/or interactions that may help explain the variability in treatment response. The main graphs A–D show P values for post hoc tests comparing responder groups (good (blue circles) vs. poor (red squares)) across both drug conditions. Inset graphs show individual data separately for each responder group. Post hoc test P values reveal within group changes between drug conditions: placebo (P) and flupirtine (F). [Colour figure can be viewed at wileyonlinelibrary.com]

Discussion

The current pilot study demonstrated that 400 mg of slow release flupirtine did not show any appreciable effect on the AHI, LG, nor any of the other OSA endotypes. Despite the lack of overall group differences, there was significant individual variability in treatment response as determined by the AHI (range = −33% to +55%). A common feature of OSA treatment research is often the wide individual variance in treatment response; an expected outcome of attempting to resolve a multifactorial disorder with distinct individual differences in pathogenesis. An exploratory sensitivity analysis revealed two distinct response groups based on the percentage reduction in AHI. While these two groups were comparable during the placebo condition, their response to treatment differed regarding the drugs effects on collapsibility (both V _PASSIVE and V _MIN) and the arousal threshold.

Potential reasons for flupirtine's lack of effect on LG

Contrary to our expectations, flupirtine did not influence LG, a finding that was unexpected given our prediction that it would reduce the sensitivity of both central and peripheral chemoreceptors. The retrotrapezoid nucleus (RTN), recognised as the primary central chemosensory mechanism, is notably susceptible to modulation by drugs that alter M‐currents (Hawryluk et al., 2012; Mulkey et al., ²⁰¹⁵). KCNQ openers have been documented to decrease RTN firing rates by as much as 50% (Hawryluk et al., 2012), without affecting chemostimulation. It may be that flupirtine was effective in dampening neuromodulatory stimulation of the RTN, but it did not reduce LG due to the preserved phasic activation of the RTN's inherently CO₂‐responsive cells (Guyenet et al., 2019).

We also hypothesised that KCNQ channels would reduce peripheral chemoreceptor (i.e. carotid body) sensitivity. This hypothesis is supported by the evidence that KCNQ channels are expressed in the carotid body (Buniel et al., 2008), although their role in signal transduction has yet to be fully established. In contrast, there is abundant literature demonstrating high expression of BK_Ca channels (Iturriaga et al., 2021; Mondéjar‐Parreño et al., ²⁰²⁰; Wang & Kim, ²⁰¹⁸), which perform a role similar to that of KCNQ channels and are often co‐expressed (Ma et al., 2020). While the impact of modifying KCNQ channels in the carotid body remains unexplored, blocking BK_Ca channels has been demonstrated to significantly increase ventilation. The novel respiratory control modulator GAL‐021 inhibits the activity of both BK_Ca and KCNQ channels, stimulating ventilation by sensitising type 1 cells in the carotid bodies (Golder et al., 2015; Lu et al., ²⁰²⁰). Given the synergistic operation of KCNQ and BK_Ca channels, simultaneous activation of both may be necessary for effective desensitisation of the carotid body. For instance, Zavaritskaya and colleagues (2020) found that vasodilatation in a murine model was achieved only through combined activation of Kv 7.4/7.5 and BK_Ca channels. This effect was not observed when either channel was activated alone. Therefore, targeting a single potassium channel may not be an effective strategy for reducing carotid body hypersensitivity. Instead, successful modulation probably requires the co‐activation of multiple, strategically selected ion channels.

Differences in good versus poor response group

Post hoc analyses showed that response group differences were due to significant changes within the ‘poor’ response group, as we observed stable endotype physiology between placebo and treatment condition in the ‘good’ responders. In other words, it was not a beneficial effect of flupirtine on the OSA endotypes that led to an improvement within the good responders, but rather an unfavourable effect of flupirtine increasing collapsibility within poor responders (a finding that remained even after controlling for body position – data not shown). Furthermore, greater chemosensitivity (inferred from the hCBC score) and airway collapsibility (lower V _MIN) were predictors of a poor response. It is possible that the increased collapsibility was due to the myorelaxant properties of flupirtine (Harish et al., 2012). Kv7 channel activation can reduce skeletal muscle force and rigidity by suppressing synaptic reflexes mediated by glutaminergic NMDA receptors (Zagorchev et al., 2016). Kv7 channels are highly expressed on the XII‐MN and increasing their activation raises the XII‐MN membrane threshold (i.e. reduces excitability), leading to attenuated bursts of phasic activation. This would translate to less motor pool recruitment of the airway dilator muscles (Ghezzi et al., 2017, 2018). However, as both response groups had comparable collapsibility on placebo, it remains unclear why upper airway muscle tone was solely preserved within the good responder group (Fig. 4).

Predictors of responder group

Although the post hoc analysis comparing response groups (good/poor) found differences in collapsibility and arousal threshold, these endotypes (measured on placebo night) were not associated with changes in AHI. However, the hCBC score calculated during the placebo condition was useful as a response predictor. Caution should be noted though, as hCBC score did not correlate with any quantitative measures of LG or chemosensitivity measured within the current study, albeit in a small and potentially underpowered sample. Additionally, as the hCBC score developed by Li and colleagues (2021) was validated in a predominantly Chinese patient population, it is possible this tool may not be directly generalisable to the current, predominantly Caucasian, sample (n = 13/15). Compared to OSA patients of Chinese descent, Caucasians experience lower levels of airway collapsibility but more unstable ventilatory control (higher LG; O'Driscoll et al., ²⁰¹⁹). Nonetheless, having a clinical score to calculate markers of key OSA endotypes such as the arousal threshold have been previously developed (Edwards, Sands et al., 2014), and the hCBC score is another potentially useful and easily administered clinical tool to identify OSA patients with a high LG endotype. Future OSA endotype research should consider incorporating the hCBC score (and other relevant clinical tools) so further validation work can be done within larger and varied patient samples.

Methodological considerations

When interpreting our findings there are a number of limitations that must be considered. Firstly, given our hypothesis that flupirtine would reduce LG (via reductions in controller gain), it may seem odd that we did not screen and enrol OSA patients with high LG. Given that flupirtine had never been tested in OSA patients before and therefore we did not know which endotypes it might impact, we chose to recruit a heterogenous sample for its first trial in OSA patients. Such an approach is similar to other trials of novel pharmacotherapies (Malhotra et al., 2024; Sands et al., ²⁰²⁴; Taranto‐Montemurro et al., ²⁰¹⁹; Thomson et al., ²⁰²²; Walsh et al., ²⁰²⁵). Secondly, we only assessed the impact of flupirtine after one night of administration. Given that flupirtine was expected to bind to KCNQ2/3 potassium channels (particularly in the carotid bodies), its impact on reducing carotid body sensitivity (i.e. controller gain) should have been immediately evident as long as the dose was high enough (the dose given here was the standard clinical dose given to treat chronic pain). Nonetheless, chronic dosing studies within a larger and more targeted sample would be required to completely rule this drug out as a potential OSA therapeutic. Thirdly, our evening blood pressure measurements were performed approximately 1 h after the chemoreflex test and it is possible that blood pressure was elevated in response to the hypoxic/hypercapnic exposure. If true, then this may have contributed to why we did not see any change in blood pressure overnight. Lastly, given the exploratory nature of our subgroup analyses (good vs. poor responders) we did not correct for multiple comparisons; thus, caution should be employed when interpreting these results.

Conclusions

A single dose of flupirtine, a KCNQ potassium channel opener, did not reduce LG (or any other OSA endotype), and by extension did not successfully impact OSA severity within an unselected patient sample – although wide individual variance in treatment response was observed. There has been renewed research interest toward repurposing KCNQ activators as innovative treatments for conditions marked by neurocellular hyperexcitability or overactivity, such as schizophrenia, drug abuse, anxiety and neurodegenerative disorders (Costi et al., 2022; Hansen et al., ²⁰⁰⁷). Despite the speculative benefits of flupirtine, the findings of this study indicate that KCNQ activation may exacerbate OSA in a subgroup of OSA patients due to its myorelaxant effects and disruption of both central respiratory control and XII‐motoneuron function. Consequently, the use of flupirtine should be avoided in treatment strategies involving OSA patients, particularly those with higher airway collapsibility. This caution is particularly pertinent when considering KCNQ activators for managing drug‐resistant epilepsy, which frequently co‐occurs with OSA (Manni & Terzaghi, 2010; Villasana‐Salazar et al., ²⁰²⁰). Future strategies to lower controller gain by targeting the carotid body should be aware of these considerations and aim to mildly modify a multitude of key strategic targets involved chemoreception.

Competing interests

The authors declare no competing non‐financial interests.

Author contributions

This work was conducted within the Sleep and Circadian Research Laboratory at Monash University. Conception and design: I.C., P.J.B., B.E. Study coordination and data collection: L.T., J.C. Regulatory Management: L.T., B.E. Study physicians: S.J., G.H. Data analysis and Interpretation of results: All authors. Initial manuscript draft: L.T., B.E. All authors were responsible for revising it critically for important intellectual content. All authors have approved the final version of the manuscript and agreed to be accountable for all aspects of the work. All persons designated as authors qualify for authorship, and all those who qualify for authorship are listed.

Funding

This work was supported by funding from the Monash Research Impact Fund. L.T. was supported by scholarships from both Monash University and the Monash Lung and Sleep Institute. S.A.L. received project grant funds from the Monash Lung and Sleep Institute outside the current work. G.S.H. has received equipment to use in research projects from ResMed, Philips Respironics and Air Liquide Healthcare. G.S.H. is the current president of the Australasian Sleep Association. B.A.E. has received grant support from Apnimed, Myofunctional Research Company and the National Health and Medicine Research Council, as well as personal fees from Signifier Medical and EQT Partners outside the current work. All other authors report no disclosures. At the time of conducting the study, I.C. and P.J.B. were named inventors in patent applications covering the use of drugs that modulate Kv7 potassium channels to treat sleep‐disordered breathing and were entitled to share in any proceeds from the commercialisation of inventions claimed in these patent applications (at time of publication these patents have lapsed).

Supporting information

Peer Review History

Biography

Luke D. J. Thomson is a dedicated sleep scientist completing his PhD at Monash University in Melbourne, Australia. Luke began working in clinical sleep labs in 2012 while pursuing an Honours degree in Psychology at Griffith University. This was followed by a Graduate Diploma in Sleep Science at the University of Western Australia. In 2017, Luke joined the Monash University Sleep & Circadian Medicine Lab under the mentorship of A/Prof. Brad Edwards. The present study is part of Luke's dissertation work which focuses on pharmaceutical treatments for sleep apnoea, though he's still waiting for someone to invent a puppy CPAP for his snoring pugs.

Appendix 1.

Table A1

Table A1.

Effects of flupirtine on the sleep apnoea endotypes compared with placebo divided by sleep stage and body position

	Placebo	Flupirtine	P value
Total
Loop gain	0.68 ± 0.19	0.69 ± 0.18	0.859
ArTH (%V EUPNOEA)	161 ± 34	163 ± 33	0.550
VRA (%V EUPNOEA)	31 [24–42]	31 [24–47]	0.359
V PASSIVE (%V EUPNOEA)	68 [58–72]	64 [51–72]	0.188
V MIN (%V EUPNOEA)	57 ± 15	50 ± 21	0.062
V ACTIVE (%V EUPNOEA) n = 14	83 ± 28	73 ± 33	0.093
NREM
Loop gain	0.72 ± 0.19	0.76 ± 0.23	0.306
ArTH (%V EUPNOEA)	160 ± 35	162 ± 36	0.476
VRA (%V EUPNOEA)	28 [22–35]	31 [24–42]	0.639
V PASSIVE (%V EUPNOEA)	69 [59–73]	64 [52–70]	0.359
V MIN (%V EUPNOEA)	56 [48–71]	55 [29–69]	0.121
V ACTIVE (%V EUPNOEA) n = 14	93 [66–106]	76 [59–104]	0.091
NREM supine
Loop gain	0.79 ± 0.26	0.81 ± 0.24	0.712
ArTH (%V EUPNOEA)	168 ± 35	166 ± 35	0.617
VRA (%V EUPNOEA)	31 [20–35]	28 [24–44]	0.303
V PASSIVE (%V EUPNOEA)	53 ± 19	52 ± 25	0.855
V MIN (%V EUPNOEA)	52 ± 20	47 ± 22	0.240
V ACTIVE (%V EUPNOEA) n = 14	78 ± 44	76 ± 35	0.786
REM
Loop gain n = 10	0.38 [0.33–0.50]	0.46 [0.41–0.56]	0.418
ArTH (%V EUPNOEA) n = 10	154 ± 21	157 ± 26	0.598
VRA (%V EUPNOEA) n = 10	55 ± 29	56 ± 27	0.899
V PASSIVE (%V EUPNOEA) n = 10	64 ± 26	60 ± 25	0.576
V MIN (%V EUPNOEA) n = 10	68 [37–72]	57 [24–68]	0.266
V ACTIVE (%V EUPNOEA) n = 10	101 [75–109]	97 [34–102]	0.232

Data presented as means ± SD or median [interquartile range] and compared using paired t test (parametric or non‐parametric as appropriate). n = 15 unless otherwise indicated. One participant did not produce usable data to measure V _ACTIVE (in both conditions); 5 participants did not generate enough usable data in REM sleep to derive valid endotype measurements (all during placebo condition).

Abbreviations: ArTH, arousal threshold; V _ACTIVE, ventilation when upper airway dilator muscles are maximally activated; V _EUPNOEA, eupnoeic ventilation; V _MIN, ventilation at lowest ventilatory drive; V _PASSIVE, ventilation when upper airway dilator muscles are hypotonic/passive; VRA, ventilatory response to arousal.

Data availability statement

In addition to the individual data provided in the manuscript, data that support the findings of this study will be available upon reasonable request.