Chemogenetic modulation of the parafacial respiratory group influences the recruitment of abdominal activity during REM sleep

Abstract

Current theories on respiratory control postulate that the respiratory rhythm is generated by oscillatory networks in the medulla: preBötzinger complex (preBötC) is the master oscillator responsible for generating inspiration, while parafacial respiratory group (pFRG) drives active expiration through recruitment of expiratory abdominal (ABD) muscle activity. Research addressing the role of pFRG in ventilation and rhythm generation across sleep states is limited. We recently reported the occurrence of ABD recruitment occurring despite the induction of muscle paralysis during REM sleep. This ABD recruitment was associated with increased tidal volume and regularization of the respiratory period in rats. As pFRG generates active expiration through the engagement of ABD muscles, we hypothesized that the expiratory oscillator is also responsible for the ABD recruitment observed during REM sleep. To test this hypothesis, we inhibited and activated pFRG using chemogenetics (i.e. designer receptors exclusively activated by designer drugs) while recording EEG and respiratory muscle EMG activities across sleep–wake cycles in male Sprague–Dawley rats. Our results suggest that inhibition of pFRG reduced the number of REM events expressing ABD recruitment, in addition to the intensity and prevalence of these events. Conversely, activation of pFRG resulted in an increase in the number of REM events in which ABD recruitment was observed, as well as the intensity and prevalence of ABD recruitment. Interestingly, modulation of pFRG activity did not affect ABD recruitment during NREM sleep or wakefulness. These results suggest that the occurrence of ABD recruitment during sleep is dependent on pFRG activity and is state dependent.

Statement of Significance.

The purpose of this study was to determine whether the parafacial respiratory group (pFRG)—the brain region lateral to the facial nucleus, responsible for the occurrence of active expiration during periods of an increased respiratory drive—may also control the recruitment of abdominal (ABD) muscles during sleep. Our results indicate that pFRG contributes to the generation of expiratory ABD activity specifically during REM sleep. Active expiration could be a natural mechanism of the body to cope with respiratory instability during sleep. Therefore, understanding the mechanism underlying the occurrence of ABD recruitment during sleep could provide useful insights into the development of alternative treatments for sleep apnea.

Introduction

Breathing requires the synchronization of a complex motor program to control gas exchange and adequately respond to various physiological and environmental conditions. The central pattern generator that orchestrates the respiratory rhythm lies in the medulla and is hypothesized to be composed by at least two coupled oscillators [1, 2]. The preBötzinger complex (preBötC) is essential for respiratory rhythmogenesis through contraction of inspiratory muscles [3–6]. The parafacial respiratory group (pFRG), silent in resting conditions, is recruited during periods of increased respiratory demand and generates active expiration through contraction of abdominal (ABD) muscles [2, 7–9]. Recently, the existence of a third generator, the post-inspiratory complex (PiCo), has been proposed to regulate respiratory braking during post-inspiration [1, 10].

Rhythmic activity of neurons in the parafacial area was initially described in brainstem–spinal cord preparations of newborn rats [11–13]. Parafacial neuronal oscillations commence during embryonic development before preBötC activity inception [14] and eventually become silent by adulthood [7, 9]. Mounting evidence obtained in anesthetized adult and juvenile rats highlights the dependence on pFRG late-expiratory activity for the recruitment of ABD muscles during active expiration [7–9, 15, 16]. The convergence of several modulatory mechanisms into the parafacial area (e.g. GABAergic, Glycinergic, Cholinergic, and Glutamatergic inputs) suggests that pFRG activation and the generation of active expiration throughout different physiological conditions may be more complex than previously thought, depending not only on central and peripheral chemosensitive inputs but also on developmental stage and vigilance state [7, 9, 17–20].

In recent years, attention has been drawn to the functional implications of expiratory activity observed in sleeping healthy rodents. While the occurrence of ABD recruitment during REM sleep seems to be consistent, occurring throughout postnatal development and into adulthood, the ABD recruitment during NREM is present at a higher rate during the first days of postnatal development and gradually decreases until it becomes almost nonexistent by adulthood [17, 18, 20, 21]. Interestingly, the occurrence of ABD muscle recruitment during both REM and NREM sleep across development, was preceded by periods displaying higher respiratory disturbances (i.e. apneas and hypopneas) and breathing variability [17, 20]. Moreover, the recruitment of ABD muscles during REM sleep was associated with increased tidal volume and reduced breathing variability, whereas ventilation was not affected during NREM sleep [17, 20]. The urethane-anesthetized preparation has been proposed as a model of natural sleep [22–25]. In this model, despite the absence of respiratory disturbances, ABD recruitment can be occasionally observed during activated (REM-like) epochs, but not during inactivated (NREM-like) states [24]. This evidence suggests that factors like respiratory variability and brain states may affect the occurrence of ABD recruitment during both natural sleep and urethane anesthesia. However, the role played by the pFRG in the generation of these events is still to be determined. To investigate the role of pFRG in the generation of ABD recruitment across sleep states we used a chemogenetic approach to manipulate pFRG activity and evaluated ABD occurrence during sleep. Our results suggest that pFRG modulates the occurrence, strength, and prevalence of expiratory ABD activity, specifically during REM sleep.

Methods

Ethical approval

Experimental procedures were approved by the Animal Care and Use Committee (ACUC) of the University of Alberta (AUP#461) according to the guidelines defined by the Canadian Council of Animal Care.

Viral injections into the pFRG

To test the hypothesis that pFRG activation underlies the occurrence of ABD recruitment during sleep, we transfected pFRG neurons with a DREADD (i.e. Designer Receptors Exclusively Activated by Designer Drugs) adeno-associated virus expressing either an inactivating G_i protein-coupled receptor driven by the synapsin promoter (AAV2/2, hSyn-HA-hM4D(G_i)-IRES-mCitrine; n = 7; UNC Vector Core, NC) or a stimulatory G_q protein-coupled receptor (AAV2/2, hSyn-HA-hM3D(G_q)-IRES-mCitrine; n = 8; UNC Vector Core) [26]. An additional cohort of rats (n = 6) was transfected with viruses expressing Cre-eYFP only (AAV2/2-hSyn-IRES-Cre-eYFP; UNC Vector Core) to assess the possibility of nonspecific effects associated with viral expression, and the use of the DREADD receptors ligand clozapine-N-oxide (CNO), or its metabolite clozapine [27–29].

Sprague–Dawley adult male rats (250–300 g) were anesthetized via intraperitoneal injection of ketamine (90 mg/kg) and xylazine (10 mg/kg) and positioned prone on a stereotaxic frame. The incision area was trimmed, aseptically cleaned, and local anesthetic was applied (Bupivacaine 0.25% solution). Access to the medullary region was obtained through the partial removal of the occipital bone. Bregma was positioned 5 mm below lambda and viruses were pressure injected bilaterally into the pFRG (1.8 mm rostral, 2.5 mm lateral, and 3.4 mm ventral to the obex) through a sharp glass micropipette (30 µm tip diameter) at a rate of 100 nL/min (volume injected was 200–300 nL per side). Following injections, the micropipette was left in place for 5 min to avoid backflow of the virus along the track. Incisions were sutured closed, rats were provided with oral analgesics (Metacam, 2 mg/kg), and monitored for recovery during 5 days with food and water ad libitum.

Chronic instrumentation

Three weeks after viral injections rats were instrumented with EEG and EMG electrodes according to previously published protocols [17, 24]. Rats were anesthetized with ketamine–xylazine, and bipolar multistranded PFA-insulated stainless-steel EMG electrodes (Cooner wire, CA) were inserted into the oblique ABD, diaphragm (DIA), and neck muscles. The wires were then tunneled under the skin and attached to an electrical socket (Ginder scientific, Ontario, Canada) placed between the shoulder blades. Subsequently, rats were positioned prone on a stereotaxic frame with bregma and lambda at the same level. Bipolar, multistranded PFA-insulated stainless-steel EEG wires (AM-System, WA) were implanted in the neocortex (nCTX) and hippocampus (HPC) according to the following coordinates (in mm) relative to bregma: for nCTX: rostrocaudal (RC), +2.5; mediolateral (ML), +1.2; dorsoventral (DV), −1.5 to −2.0; for HPC: RC, −3.3; ML, −2.4; DV, −2.5 to −3.0 mm. The distance between the two ends of the wires was kept at approximately 1 mm for optimal recordings. Additional wires to be used as surface EEG and ground were soldered to jeweler’s screws and positioned in the parietal–frontal bones and occipital bone, respectively. All electrodes were fixed to the skull with dental acrylic and wires were then tunneled under the skin and attached to a second electrical socket positioned on the neck. Rats were monitored postoperatively and provided with oral analgesics (Meloxicam, 2 mg/kg) for 5 days thereafter.

Recording procedures

One week after implantation rats were accommodated to a whole-body plethysmograph (Buxco Respiratory Products, DSI) for 5 h on three consecutive days with constant airflow delivered (1.5 L/min). On days 4 and 5, signals of implanted EEG and EMG were recorded during the rats’ natural sleep period from 12 to 5 pm after systemic (IP) injection of either vehicle (CTRL, 0.9% sterile saline + 0.02% d-methyl sulfate on day 4) or clozapine-N-oxide (CNO dosage of 3 mg/kg of a 0.3 mg/mL solution, day 5). Data from EEG and EMG activity was sampled at 1 kHz and acquired using PowerLab (AD Instruments, CO). HPC and nCTX signals were amplified at ×1000 gain and filtered between 0.1 and 500 Hz (model P511, Grass Technologies), whereas EMG activity was amplified at ×10,000 and filtered between 100 and 500 Hz (model 1700 A-M Systems).

Histology

At the end of the in vivo recordings, rats were anesthetized with an IP injection of urethane (1.7 mg/kg) and transcardially perfused with saline 0.9% followed by 4% paraformaldehyde (PFA) in phosphate buffer (PB). Brains were collected, post-fixed in 4% PFA overnight and sectioned (50 µm) in a vibratome (model VT1000S, Leica Mycrosystems) for immunodetection of the viral reporter proteins and specific neuronal markers. All immunostaining procedures were performed at room temperature. Free-floating sections were incubated for 1 h with 0.3% Triton X-100 and 10% normal donkey serum (NDS) in saline PB (PBS) to increase membrane permeability to antibodies and reduce nonspecific binding. Following blocking, sections were incubated overnight with primary antibodies diluted in a solution containing PBS, 1% NDS, and 0.3% Triton X-100. Primary antibodies used were as follows: anti-hemagglutinin (HA; rabbit; 1:800; Cell Signaling Technologies), anti-green fluorescent protein (GFP; chicken; 1:800; Aves Labs Inc.), anti-choline acetyltransferase (ChAT; goat; 1:800; EMD Millipore, ON, Canada) anti-neuronal nuclei (NeuN;mouse; 1:500; EMD Millipore). The following day, sections were washed three times in PBS and incubated with specific secondary antibodies (1:200; Cy2-donkey anti-rabbit; Cy3-donkey anti-goat; Cy5-donkey anti-mouse; Cy2-donkey anti-chicken; Jackson Immuno Research) diluted in PBS and 1% NDS for 2 h. Subsequently, sections were washed with PBS three times, mounted and coverslipped with Fluorsave mounting medium (EMD Millipore). Slides were observed under an AxioCam2 Zeiss fluorescent microscope connected with Metamorph acquisition software or Evos FL fluorescent microscope (Thermofisher). Images were acquired as TIFF files and used for cell counting and analysis of injection sites using ImageJ. Expression of the virus was investigated in the parafacial area. Serial sections (150 µm interval) spanning the caudal end of the facial nucleus (VII) (from 650 µm caudal to 850 µm rostral to the caudal tip of VII) were investigated. The core of pFRG was identified as the non-cholinergic cells in a square area (Figure 1, A; 2.5–3.2 mm ML and 9.8–10.6 mm DV relative to bregma [30]) ventrolateral to the caudal tip of VII (pF_L; Bregma −12.00 mm, [30]) and extending for 600 µm rostrocaudal (from Bregma −12.35 mm to Bregma −11.75 mm). Expression of neurons in the pFRG (pF_L) was also assessed in relation to the more medial parafacial area (pF_M) where chemosensitive neurons of the retrotrapezoid nucleus (RTN) are localized [31].

Figure 1.

Expression of viral reporter proteins in the parafacial area. (A) Schematic representation of the area analyzed at the caudal tip of the facial nucleus (VII; Bregma −12.00 mm). Neurons within the gray area were considered to be located in the medial parafacial area pF_M/RTN (blue square) and lateral parafacial area pF_L/pFRG (red square). This logic was applied to every slice analyzed from Bregma −12.6 mm to Bregma − 11.4 mm. (B-i and B-ii) Immunofluorescent images were taken just caudal to the facial nucleus (VII_c, Bregma 11.85 mm). Red square in (B-i) is represented in (B-ii). Immunostaining for cholinergic markers (red) revealed immunopositive processes caudal to the tip of the VII (circled area in B-i). Immunostaining for viral reporter proteins expressed by DREADD (green) revealed the transfection of cells located lateral to the caudal tip of VII and adjacent to the spinal trigeminal tract (Sp5). ChAT = choline acetyltransferase, NeuN = neuronal nuclei marker, DREADD = designer receptor exclusively activated by designer drug. (C-i–C-iii) Rostro-caudal (RC) distribution of total transfected cells (C-i), as well as cells transfected within pF_M/RTN (C-ii), and pF_L/pFRG areas (C-iii). Values are percentages of total cells transfected in an RC span of 1200 µm (analysis of 1 transverse section every 150 µm). The shaded region on graphs indicates the area corresponding to the core of pF_M/RTN and pF_L/pFRG, respectively (from Bregma −12.35 mm to Bregma −11.75 mm).

Data analysis and statistics

Sleep scoring was assessed manually using traces from HPC, nCTX, and neck EMG according to previously established sleep scoring criteria [17, 24]. EEG traces were bandpass filtered between 0.2 and 50 Hz to reduce noise associated with movement of the tethered cables or electrical current. Wakefulness was identified by the presence of strong neck EMG activity associated with activated EEG in the nCTX (low amplitude fast signal) and HPC (theta frequency; 4–10 Hz). Muscle tone during sleep was generally low, reaching even lower levels during the REM stage. The distinction between NREM and REM sleep was performed by analyzing the EEG and neck EMG signals. NREM stage was identified due to the presence of low-frequency (delta waves: 2–4 Hz) and high-amplitude signals in nCTX, whereas REM sleep was identified due to the presence of theta activity in the HPC (lasting 10 s or longer) and minimal neck_EMG activity. Possible effects of CNO, its metabolite clozapine, or modulation of pFRG on sleep architecture [27–29] were evaluated through the analysis of the number and duration of REM and NREM epochs per hour of recording, as well as the percentage of time spent in each sleep stage. Additionally, sleep fragmentation was further analyzed through the calculation of the number of transitions between sleep stages throughout the recording sessions. Sleep characteristics were analyzed using LabChart 8-Pro (AD-Instruments, Sydney, AU), SleepSign (Kissei Comtec, Japan), Excel, and Origin 8 software.

Respiratory variables were analyzed using signals recorded from the DIA and ABD EMG. The absolute value of EMG traces was integrated with a time constant decay of 0.08 s to perform peak analysis. Integrated DIA traces were used to calculate the respiratory rate (RR) and period. DIA_EMG peak amplitude values during ABD recruitment and in baseline conditions were used to estimate relative changes in DIA amplitude (DIA_amp) and minute respiration (MR) (i.e. the product of RR and DIA_amp; MR) and used as a proxy for rats’ minute ventilation similar to previous studies [32]. Respiratory variability was calculated and is reported as the percentage of coefficient of variation (CV) of the respiratory period. Central and post-sigh apneas (PSA) were identified as pauses of 2 s or longer in the DIA_EMG signals. Additionally, sigh rate and the ratio of sighs in which PSA were observed were also calculated.

REM, NREM, and quiet wakefulness events were classified as ABD⁺ or ABD− based on the presence or absence of ABD recruitment during each epoch. ABD recruitment was determined as three or more consecutive ABD bursts occurring during the latter half of inspiratory inter-burst intervals with an amplitude 50% larger than baseline non-respiratory related tonic ABD muscle activity. The incidence of ABD recruitment was evaluated by calculating the ratio of epochs in which ABD recruitment was observed (e.g. ABD⁺/REM_Tot or ABD⁺/NREM_Tot). The effect of CNO on the occurrence of ABD recruitment was further tested by assessing the prevalence and strength of these events in CTRL and CNO conditions. Prevalence of ABD recruitment was evaluated through the ratio of ABD to DIA burst frequency, whereas strength was assessed through the average ABD_EMG amplitude relative to baseline non-respiratory tonic ABD activity in CTRL and CNO conditions within each sleep epoch. Respiratory parameters (i.e. RR, respiratory variability, relative DIA_amp, and relative MR) were further compared within sleep epochs: respiratory parameters were analyzed throughout the duration of ABD recruitment events and compared to a similar number of breaths before the onset of ABD recruitment. Traces were analyzed using LabChart 8 Pro (AD-Instruments, Sydney, AU), Excel, and Origin 8 (OriginLab Corp.) software, and images of recorded traces were prepared with IgorPro (Wavemetrics).

Respiratory parameters and ratios are reported as mean ± SEM. Assumptions of homogeneity of variance and normal distribution were tested through Brown–Forsythe (α = 0.05) and Shapiro–Wilk (α = 0.05) tests, respectively. All within-individual (CTRL vs CNO) and in-between groups (EYFP vs DREADD-G_i vs DREADD-G_q) comparisons were performed with a two-factor repeated-measures analysis of variance (two-way RM ANOVA; one repeated factor; α = 0.05). Repeated measures in the two-way RM ANOVA were composed of only two levels (CTRL and CNO); therefore, the sphericity assumption was always met. If significant main effects were found, two-way RM ANOVAs were followed by all pairwise multiple comparison post-hoc procedures (Holm-Sidak method; α = 0.05). When applicable, comparisons of CNO/CTRL ratios across different groups (EYFP vs DREADD-G_i vs DREADD-G_q) were performed using one-factor analysis of variance (one-way ANOVA; α = 0.05). When significant main effects were resolved, one-way ANOVAs were followed by Tukey’s honestly significant difference (HSD) post hoc test (α = 0.05). If the normality test failed (p < 0.05), we performed a Kruskal–Wallis one-way analysis of variance on ranks (α = 0.05) as a non-parametric equivalent of one-way ANOVA. If the main effects were found, the Kruskal–Wallis test was followed by all pairwise multiple comparison post-hoc procedures (Dunn’s method; α = 0.05). All statistical analyses were performed in SigmaPlot 14.0 (Systat Software Inc.).

Results

A total of 21 rats underwent successful pFRG viral injections and electrode instrumentation for the analysis of respiratory variables during natural sleep in healthy conditions: 6 for control experiments with Cre-eYFP viruses, 7 for the inhibitory experiments with DREADD-G_i, and 8 for the excitatory experiments with DREADD-G_q.

Histological analysis

To verify that viral injections were effectively located in the area that was previously identified as pFRG[1, 7, 8, 33], we inspected an area comprising 1,500 µm rostrocaudal, from Bregma −11.25 mm to Bregma −12.75 mm [30]. We analyzed a rectangular area in the perifacial region, which included both chemosensitive RTN neurons in the region ventromedial to the facial nucleus (pF_M) and late expiratory pFRG neurons in the region ventrolateral to the facial nucleus (pF_L) (Figure 1, A). The analyzed area extended from the lateral border of the pyramidal tract to the medial edge of the spinal trigeminal tract, mediolaterally, and from the ventral surface of the medulla to the dorsal edge of the facial nucleus, dorso-ventrally. The average number of cells expressing the reporter protein was 152 ± 5.3 per animal (n = 14; 1 section every 150 µm) in an area that extended 600 µm caudal and rostral to the caudal tip of VII. On average, the majority of the transfected cells (~129 ± 6.8 per animal or 88.7 ± 4.2% of the total number of transfected cells counted per animal) were located in the perifacial area that includes the RTN/pFRG region from Bregma −12.35 mm to −11.75 mm (Figure 1, B-i and C-i) according to the current literature [1, 7, 8]. Only a few of those cells (36.9 ± 3.4 per animal; Figure 1, C-ii) were located more medially, in the ventral parafacial area (i.e. pF_M, RTN; 1.0–2.4 mm lateral to the midline [8, 30] whereas the majority of transfected neurons were located in the lateral parafacial area (i.e. pF_L or pFRG; >2.4 mm lateral to the midline; Figure 1, C-iii). We also evaluated the spread of the virus to cholinergic motoneurons in VII. On average, 34 ± 4.7 ChAT positive motoneurons per animal also expressed viral vectors along the RC area analyzed (10.1 ± 1.4% of motoneurons in that area). No cell bodies expressing viral reporter proteins were observed in adjacent areas of the medulla (i.e. spinal trigeminal tract, Bötzinger Complex, preBötC).

Sleep architecture and general respiratory characteristics

It has been recently demonstrated that CNO can be reversibly metabolized to clozapine and therefore induce clozapine-mediated effects [28]. Furthermore, clozapine has sedative properties and may influence sleep architecture [34–36]. In our study, the frequency, average duration, and percentage of time spent in each sleep phase (i.e. REM and NREM sleep) remained unaffected after systemic injection of CNO in rats transfected with either Cre-eYFP, DREADD-G_i, or DREADD-G_q (p > 0.05; Table 1), suggesting that, even though CNO is metabolized to clozapine, the amount of circulating clozapine was not sufficient to induce alterations in the sleep architecture. Representative examples of hypnograms in each condition are shown in Figure 2, A-i–A-iii. A chronological analysis of the time spent in each sleep stage throughout the recordings (Figure 2, B-i–B-iii) indicates that rats spent 39%–48% of the recording time awake, whereas 49%–56% and 3%–5% of the recording time was spent in NREM and REM sleep, respectively. No difference was observed between CTRL and CNO and between groups for the percentage of time spent in REM and NREM sleep (two-way RM ANOVA, p > 0.05, Table 1). In addition, the frequency of transitions between sleep stages was not significantly different after CNO administration across groups (two-way RM ANOVA, p > 0.05; Figure 2, C-i–C-iii). Most of the NREM epochs ended in wakefulness events (NRW/h = 6.14 ± 0.78 vs NRR/h = 3.19 ± 0.31, two-tail unpaired Student’s t-test, p = 0.01) whereas REM epochs indistinctively finalized in either wakefulness or transitions to NREM sleep (RW/h = 1.69 ± 0.27 vs RR/h = 1.50 ± 0.10, two-tail unpaired Student’s t-test, p = 0.519; Figure 2, C-i–C-iii).

Table 1.

Effects of chemogenetic modulation of pFRG on sleep architecture

	Average epoch duration (min) (mean ± SEM)			Epochs/h (mean ± SEM)			Percentage of time spent in sleep stage (mean ± SEM)
	CTRL	CNO	P	CTRL	CNO	P	CTRL	CNO	P
NREM
eYFP-CTRL	3.14 ± 0.61	3.73 ± 0.74	0.47¤	11.81 ± 1.60	10.54 ± 1.76	0.25¤	48.76 ± 9.80	56.46 ± 4.86	0.88¤
DREADD-Gi	3.53 ± 0.44	4.76 ± 0.74	0.08¥	7.36 ± 0.92	8.28 ± 1.16	0.18¥	47.56 ± 3.55	56.29 ± 4.85	0.10¥
DREADD-Gq	3.78 ± 0.50	5.03 ± 0.69	0.74§	16.57 ± 4.87	8.03 ± 1.17	0.18§	52.45 ± 1.91	56.36 ± 4.21	0.86§
REM
eYFP-CTRL	0.78 ± 0.11	1.05 ± 0.24	0.69¤	2.49 ± 0.77	2.60 ± 0.68	0.36¤	3.00 ± 1.31	4.55 ± 1.34	0.40¤
DREADD-Gi	1.10 ± 0.10	1.05 ± 0.21	0.12¥	2.92 ± 0.37	3.29 ± 0.52	0.70¥	5.43 ± 0.97	5.74 ± 1.50	0.36¥
DREADD-Gq	0.92 ± 0.12	1.14 ± 0.09	0.32§	4.31 ± 1.13	3.15 ± 0.49	0.49§	5.43 ± 0.71	5.44 ± 0.92	0.63§

Systemic administration of the DREADD ligand clozapine-N-oxide (CNO) had no effect on the sleep architecture of rats across experimental groups. Sleep was monitored throughout 5 h during the natural sleep period of the rats after either vehicle (CTRL) or CNO administrations. Two-way repeated-measures ANOVA within individual factors (CTRL and CNO); in-between group factors (CTRL-eYFP, DREADD-G_i, DREADD-G_q). No significant main effects were resolved for in-between group comparisons (¤), within individual comparisons (¥) and interaction between within and in-between factors (§).

Figure 2.

Sleep architecture is not affected by the chemogenetic modulation of pFRG. (A-i–A-iii) Hypnograms from three rats transfected with CTRL-eYFP (gray scale), DREADD-G_i (blue scale), or DREADD-G_q (green scale) viruses. Hypnograms show the transitions between wakefulness (W), non-rapid-eye-movement (NREM) and rapid-eye-movement (REM) sleep throughout 5 h of continuous recordings. The percentage of time spent in each sleep stage (B-i–B-iii) and the rate of transition between sleep stages (i.e. sleep fragmentation; C-i–C-iii) were not affected by the systemic administration of CNO (darker tone in each color scale) in either CTRL-eYFP, DREADD-G_i, or DREADD-G_q rats.

Group analysis of the RR in CTRL conditions versus CNO and in between groups suggests that RR was not affected by modulation of pFRG during either NREM sleep or REM sleep (p > 0.05; two-way RM ANOVA for NREM and for REM). On average, pooled data of RR during REM sleep was not different from NREM sleep (89.4 ± 1.3 bpm in REM vs 90.1 ± 1.7 bpm in NREM, unpaired Student’s t-test, p = 0.52). Similarly, respiratory variability (measured by the CV of the period) was not different between CTRL and CNO conditions and across groups in either REM or NREM sleep (p > 0.05; two-way RM ANOVA for REM and NREM). Overall, analysis of the pooled data suggests that breathing during REM epochs was more variable than during NREM sleep (24.9 ± 1.5% in REM vs 17.8 ± 1.4% in NREM, two-tail unpaired Student’s t-test, p = 0.006). We conclude that the commonly used dose of CNO applied in this study (3 mg/kg) did not mediate nonspecific effects in the sleep architecture or respiratory characteristics of adult male Sprague–Dawley rats. Additionally, targeted inhibition and/or excitation of the pFRG area did not affect the generation of the respiratory rhythm across sleep states.

Effect of pFRG modulation on the recruitment of ABD muscles across vigilance states

To understand the effects of pFRG modulation in the generation of ABD recruitment across sleep states, we analyzed the percentage of Wake, NREM, and REM sleep epochs in which ABD recruitment was observed, as well as the ABD/DIA ratios and the amplitude of ABD_EMG signals during the different sleep stages. Vigilance states were considered ABD+ when three or more consecutive late expiratory ABD bursts surpassed the threshold of 50% of baseline activity (Figure 3, A-i–A-ii) during the duration of the epoch. All respiratory parameters analyzed, were not significantly different across groups in CTRL (vehicle) conditions (p > 0.05; two-way RM ANOVA). Therefore, all CTRL data were pooled for subsequent interpretation of the results. During quiet wakefulness and NREM sleep, ABD recruitment was sparsely observed (Figure 3, B-i and C-iii). Specifically, only 0.5 ± 1.3% of the wake events displayed ABD recruitment in CTRL and this value did not change significantly after administration of CNO across groups (p > 0.05; two-way RM ANOVA; Figure 3, B-i–B-iii). Similarly, ABD recruitment was present in only 3.0 ± 0.58% of NREM epochs in CTRL conditions and this value did not significantly change after CNO administration across groups (p > 0.05; two-way RM ANOVA; Figure 3, C-i–C-iii). Interestingly, ABD recruitment events occurred in approximately 27.1 ± 3.3% of REM sleep epochs in CTRL conditions and this value was significantly affected by modulation of pFRG (p < 0.05, two-way RM ANOVA; Figure 3, D-i–D-iii). Inhibition of pFRG resulted in a decrease of the percentage of REM sleep epochs in which ABD recruitment was observed (9.4 ± 5.4% in CNO, p = 0.04; Holm-Sidak post hoc test; Figure 3, D-ii), whereas excitation significantly increased them (50.4 ± 8.2% in CNO, p = 0.001; Holm-Sidak post hoc test; Figure 3, D-iii).

Figure 3.

Modulation of pFRG influences the occurrence of ABD recruitment during REM sleep. (A-i and A-ii) Identification of occurrence of ABD recruitment across vigilance states in representative integrated traces of diaphragm (DIA) and abdominal (ABD) activity. Vigilance states were considered to be ABD⁺ when three or more consecutive ABD bursts exceeded a threshold of 50% of baseline ABD activity (gray shaded area (Ai) and occurred during the latter half of the inter-inspiratory period (red shaded area (A-ii). (B-i–D-iii) Changes in the ratio of vigilance state epochs in which ABD recruitment was observed in CTRL-eYFP (gray scale), DREADD-G_i (blue scale), and DREADD-G_q (green scale) rats following systemic administration of vehicle (CTRL; lighter tone in each color scale) or CNO (darker tone). Modulation of pFRG had no effect on the number of sparse ABD recruitment events during quiet wakefulness (B-i–B-iii) or NREM sleep (C-i–C-iii), but it significantly affected the occurrence of ABD recruitment during REM sleep (D-i–D-iii). Graphs include mean values and SEM (square), as well as individual paired data (circles) for each group.

An in-depth analysis of the inspiratory and expiratory outputs during REM sleep suggests that both the prevalence and amplitude of ABD recruitment is affected by modulation of pFRG (Figure 4, A). The prevalence of ABD bursts within each REM epoch was analyzed through the ratio of ABD/DIA bursts (a value closer to 1 indicates a stronger prevalence of ABD burst, whereas a value closer to 0 indicates a lower prevalence of ABD bursts within the sleep epoch analyzed; Figure 4, B-i and B-ii).

Figure 4.

Modulation of pFRG influences the prevalence and strength of abdominal (ABD) recruitment during REM sleep. (A) Integrated traces of diaphragm (DIA), ABD, and neck electromyogram, as well as hippocampus (HPC) and airflow recordings during a REM epoch after systemic administration of CNO in CTRL-EYFP (black), DREADD-G_i (blue), and DREADD-G_q (green) rats. (B-i and B-ii) Procedure for the calculation of ABD to DIA ratios in representative traces. Black dashed lines and asterisks indicate inspiratory peaks, whereas red dashed lines and asterisks indicate late expiratory ABD bursts that surpassed the threshold (gray-shaded area). Red dashed lines without asterisks indicate a respiratory event in which ABD bursts did not cross the threshold. (C-i–C-iii) Changes in the ratio of ABD to DIA bursts in CTRL-eYFP (gray scale), DREADD-G_i (blue scale), and DREADD-G_q (green scale) rats following systemic administration of vehicle (CTRL; lighter tone in each color scale) or CNO (darker tone). Graphs include mean values and SEM (square), as well as individual data (circles) for each treatment. (D-i and D-ii) Relative change in the amplitude (ABD_amp CNO/CTRL; E) and frequency of ABD bursts (ABD_freq CNO/CTRL; F) after chemogenetic modulation of pFRG in rats transfected with CTRL-eYFP (gray) DREADD-G_i (blue) and DREADD-G_q (green) viruses.

The prevalence of ABD bursts within REM events was approximately 0.30 ± 0.04 in CTRL conditions and inhibition of pFRG significantly reduced this value to 0.11 ± 0.04 (Holm-Sidak post hoc test, p = 0.03; Figure 4, C-ii), whereas excitation increased it to 0.49 ± 0.08 (Holm-Sidak post hoc test, p = 0.015; Figure 4, C-iii).

Next, we evaluated the strength of the ABD_EMG signals in response to modulation of pFRG, specifically during REM epochs, by analyzing the ratio of the ABD_EMG amplitude and ABD_EMG frequency in CNO and CTRL conditions across groups. The amplitude of ABD_EMG signals after inhibition of pFRG in DREADD-G_i rats was approximately 27% of that observed in baseline (ABD_amp CNO/baseline = 0.27 ± 0.17; Figure 4, D-i), whereas excitation in DREADD-G_q rats induced an increase of approximately 82% on the ABD_EMG signal amplitude compared to baseline values (ABD_amp CNO/baseline = 1.82 ± 0.68; Figure 4, D-i). Similarly, excitation of pFRG after CNO administration in DREADD-G_q rats, increased the frequency of ABD bursts within REM events compared to baseline conditions (ABD_freq CNO/baseline = 3.11 ± 1.18; Figure 4, D-ii), whereas pFRG inhibition reduced the ABD bursting rate to approximately 45% of the baseline values (ABD_freq CNO/baseline = 0.45 ± 0.20; Figure 4, D-ii). Overall these results suggest that chemogenetic modulation of pFRG affects the occurrence of ABD recruitment, as well as the ABD burst prevalence and ABD_EMG signal amplitude during REM sleep, but it has no apparent influence on the recruitment of ABD muscles or the strength of ABD_EMG recruitment during quiet wakefulness and NREM sleep in adult male Sprague–Dawley rats.

Effect of ABD recruitment and pFRG modulation on respiratory variables

To evaluate the effects of pFRG modulation during REM sleep, we proceeded to analyze various respiratory parameters (i.e. RR, CV of the respiratory period, diaphragm amplitude and MR. More specifically, we evaluated the changes in these parameters associated with the recruitment of ABD muscles by normalizing these values to their baseline immediately before ABD recruitment occurrence (a value higher than 1 indicates that ABD recruitment was associated with a value increase, whereas a value lower than 1 indicates that ABD recruitment was associated with a decrease in those values; Figure 5, A). Normalized values of RR, CV, DIA_amp, and MR were not different between CTRL and CNO conditions across experimental groups (p > 0.05; two-way RM ANOVA). The RR during ABD recruitment was no different from the baseline values in either CTRL conditions (1.05 ± 0.06, paired Student’s t-test, p = 0.575; Figure 5, B-i) or after modulation of pFRG through the administration of CNO in DREADD-G_i (0.98 ± 0.06, paired Student’s t-test, p = 0.297; Figure 5, B-i) and DREADD-G_q rats (1.08 ± 0.03, paired Student’s t-test, p = 0.451; Figure 5, B-i). Interestingly, the occurrence of ABD recruitment was associated with a decrease in respiratory variability in CTRL conditions (normCV = 0.67 ± 0.08, paired Student’s t-test, p = 0.007; Figure 5, B-ii) and after excitation of pFRG (normCV = 0.61 ± 0.20, paired Student’s t-test, p = 0.029; Figure 5, B-ii), but had no significant effect after inhibition of pFRG (normCV = 1.16 ± 0.48, Student’s t-test, p = 0.297; Figure 5, B-ii). Similarly, ABD recruitment was associated with an increase in diaphragm amplitude (DIA_amp) and MR in CTRL (paired Student’s t-tests, normDIA_amp = 1.10 ± 0.03, p = 0.033; normMR = 1.16 ± 0.07, p = 0.035; Figure 5, B-iii and B-iv) and excitatory conditions (paired Student’s t-tests, normDIA_amp = 1.12 ± 0.03, p = 0.029; normMR = 1.15 ± 0.05, p = 0.043; Figure 5, B-iii and B-iv), but these enhancements in breathing were no longer observed after inhibition of pFRG (paired Student’s t-test, normDIA_amp = 0.95 ± 0.08, p = 0.484; normMR = 0.91 ± 0.05, p = 0.186; Figure 5, B-iii and B-iv).

Figure 5.

Effects of ABD recruitment and pFRG modulation on respiratory parameters. (A) Process of normalization for each of the respiratory parameters analyzed on representative electrophysiological traces. Respiratory parameters during the occurrence of ABD activity (ABD⁺; red shading) were normalized to baseline values (i.e. before the occurrence of ABD recruitment; ABD−; gray shading). (B-i–B-iv) Comparison of respiratory parameters after vehicle (CTRL) and CNO administration across experimental groups (gray scale: CTRL-eYFP; blue: DREADD-G_i; green: DREADD-G_q). CV = coefficient of variation, DIA_amp = diaphragm amplitude, MR = minute respiration, RR = respiratory rate. Values above or below dotted lines imply that ABD recruitment was associated with respective increases or decreases in the corresponding respiratory parameters.

We previously observed that the occurrence of ABD recruitment during sleep was usually associated with REM epochs in which respiratory disturbances were more predominant [17]. Therefore, we further examined the effects of pFRG modulation on the frequency and duration of apneas in rats transfected with Cre-EYFP, DREADD-G_i, and DREADD-G_q viruses. Chemogenetic modulation of pFRG had no influence in the number or the duration of apneas and PSA observed in both DREADD-G_i and DREADD-G_q treated rats (p > 0.05, two-way RM ANOVA; Table 2). On average, respiratory disturbances oscillated around 2.23 ± 0.6 apneas/h, whereas PSA was present at a rate of 8.33 ± 1.9 events/h. The durations of both apneas and PSA were similar and oscillated around 3.41 ± 0.21 s.

Table 2.

Effects of chemogenetic modulation of pFRG on respiratory disturbances

	Apneas/h (mean ± SEM)			Avg. apnea duration (s) (mean ± SEM)			Sighs/h (mean ± SEM)
	CTRL	CNO	P	CTRL	CNO	P	CTRL	CNO	P
eYFP-CTRL	2.17 ± 0.64	2.45 ± 0.62	0.46¤	3.48 ± 0.27	3.26 ± 0.22	0.18¤	20.58 ± 2.06	17.83 ± 1.96	0.33¤
DREADD-Gi	3.29 ± 0.97	1.93 ± 0.32	0.36¥	3.37 ± 0.19	3.38 ± 0.26	0.52¥	22.49 ± 2.62	23.00 ± 1.76	0.84¥
DREADD-Gq	1.87 ± 0.32	1.66 ± 0.58	0.34§	3.00 ± 0.19	2.88 ± 0.20	0.86§	21.91 ± 1.17	23.29 ± 2.42	0.46§
	PS apneas/h (mean ± SEM)			Avg. PS apnea duration (s) (mean ± SEM)			PS apneas/sighs (mean ± SEM)
	CTRL	CNO	P	CTRL	CNO	P	CTRL	CNO	P
eYFP-CTRL	9.45 ± 1.34	7.69 ± 1.72	0.25¤	3.84 ± 0.11	3.54 ± 0.18	0.44¤	0.47 ± 0.06	0.44 ± 0.08	0.30¤
DREADD-Gi	9.73 ± 1.75	10.37 ± 2.06	0.18¥	3.64 ± 0.22	3.75 ± 0.23	0.41¥	0.43 ± 0.08	0.48 ± 0.10	0.19¥
DREADD-Gq	7.32 ± 2.94	5.43 ± 1.56	0.19§	3.43 ± 0.12	3.35 ± 0.29	0.33§	0.26 ± 0.07	0.35 ± 0.06	0.19§

Chemogenetic modulation of pFRG following the systemic administration of CNO in transfected rats had no effect on the occurrence of respiratory disturbances. The rate of apneas and post-sigh (PS) apneas, as well as their duration was not affected by the modulation of pFRG. All values are mean and standard error of the mean (SEM). Two-way repeated-measures ANOVA within individual factors (CTRL and CNO); in-between group factors (CTRL-eYFP, DREADD-G_i, DREADD-G_q). No significant main effects were resolved for in-between group comparisons (¤), within individual comparisons (¥) and interaction between within and in-between factors (§).

We conclude that pFRG activity, in addition to being responsible for the occurrence of ABD recruitment within REM epochs, is also responsible for the regularization of breathing and enhancement in diaphragm amplitude and MR associated with ABD activity. Because inhibition of pFRG eliminated the expected enhancement in breathing otherwise caused by the recruitment of ABD muscles, it is possible that those effects could be mediated through central modulation [2]. However, the improvement in breathing observed in CTRL and excitatory conditions was not due to a reduction in the number of respiratory disturbances, but to stabilization of breathing on a burst to burst basis instead.

Discussion

We used a chemogenetic approach to investigate whether pFRG could be responsible for the ABD recruitment previously observed during sleep [17, 18, 20, 21]. Our results suggest that pFRG modulation impacts the frequency, prevalence and amplitude of ABD recruitment during REM sleep but does not seem to influence ABD activity during quiet wakefulness and NREM sleep. This evidence suggests that the mechanism underlying the generation of ABD recruitment during sleep may depend on the vigilance state. Furthermore, our study also found that the effects of potentiation of ventilation observed at the onset of ABD recruitment in CTRL and excitatory conditions were blunted in inhibitory conditions. Finally, we found that the modulation of pFRG and the occurrence of ABD recruitment did not stabilize breathing by reducing the number of respiratory disturbances but had an effect on a breath to breath basis.

Technical considerations

In this study, we used a chemogenetic approach to specifically manipulate the activity of the pFRG brain region. Although the DREADD technology is being portrayed as a powerful tool to remotely manipulate cell activity [37, 38], a few reports have highlighted some limitations that could compromise the interpretation of findings unless rigorous control experiments are performed [27–29, 39, 40]. For example, systemic administration of the “inert” ligand CNO may produce locomotion-related and behavioral effects that are possibly associated with CNO’s reverse metabolism into clozapine and N-desmethylclozapine [28, 39, 40]. Additionally, it was recently proposed that DREADD-specific effects may be mediated by clozapine rather than CNO, since CNO sparingly crosses the blood-brain barrier and DREADD receptors display a higher affinity for clozapine [27, 39]. Additionally, due to the slow conversion of CNO into clozapine, systemic injection of CNO may delay and extend the duration of DREADD-mediated effects, which could be advantageous for extended experiments (>2-h duration), as in our present study [29]. Despite its limitations, the DREADD technology remains a powerful tool for the specific manipulation of brain activity [29, 41]. In this study, we controlled for potential off-target effects of the CNO metabolic by-products by administering identical commonly used doses of CNO to rats transfected with both DREADD and non-DREADD viruses (i.e. the Cre-eYFP experimental group). Our results suggest that CNO did not produce non-DREADD mediated effects in the sleep architecture, respiratory variables or the recruitment of ABD muscles during sleep, suggesting that the observed results are not associated with off-target effects caused by CNO and its metabolites.

Histological analysis and extent of pFRG chemogenetic modulation

Since specific anatomical or genetic markers for the identification of pFRG cells remain to be determined, we based our injection on stereotaxic coordinates that target pFRG according to previous work that assessed recruitment of active expiration and late expiratory neurons by pharmacological, chemogenetic, and optogenetic manipulations of the area ventrolateral to VII [7–9, 15, 16].

In our study, viral expression extended 1200 µm rostro-caudally around the caudal tip of the facial nucleus, with most transfected cells being circumscribed to a smaller area of 600 µm lateral to the facial nucleus. Although the expression of DREADD receptors was not dependent on a specific marker and therefore may affect other respiratory and non-respiratory structures in the surrounding area, the majority of the cells were located ventrolateral to the facial nucleus. A fraction of transfected cells was also present in the more medial, chemosensitive area of the RTN. This area has an important role in driving both inspiratory and expiratory activities [42], and it displays an exquisite state-dependent activity [43–49]. Our results suggest that RTN function may have minimally, if at all, affected our results, given that: (1) DREADD-Gi activation did not change inspiratory activity at any given vigilance state; (2) DREADD-Gq and DREADD Gi activation had no effect on either inspiratory or expiratory activity when RTN neurons are more excitable, that is, during wakefulness and nREM sleep; and (3) DREADD-Gq activation increased EMG amplitude and expiratory activity in REM sleep, when expiratory activity has been previously observed [17, 50] and when RTN neurons are in their least excitable state [47, 48]. We, therefore, conclude that DREADD modulation of neuronal activity most likely affected the more ventrolateral region of the parafacial area corresponding to the putative region of the pFRG.

The nonspecific transfection of cells in the parafacial area, combined with the localized viral injections may also be responsible for the incomplete silencing or activation of ABD recruitment during sleep in our study. Alternatively, the influence of competing inhibitory and excitatory state-dependent modulatory mechanisms on pFRG neurons or on premotor and motor neurons could also explain the incomplete effects on the recruitment of ABD muscles in DREADD-G_i and DREADD-G_q rats in comparison to previous work where pFRG activity was pharmacologically modulated under anesthesia [7, 8, 15, 16, 19].

The mechanism for ABD recruitment during sleep may depend on vigilance state

Previous research has demonstrated the occurrence of expiratory ABD recruitment during REM sleep and to a lesser extent during NREM sleep [17, 18, 20, 21]. Our results show that chemogenetic modulation of the expiratory oscillator specifically influences the frequency of occurrence and strength of ABD_EMG activity during REM sleep and not during NREM sleep. This is in clear contrast to what has been observed with RTN manipulations [47]. Vigilance state-dependent changes in respiratory frequency, minute ventilation, breathing regularity, and/or diaphragm amplitude have previously been described in dogs, cats, and rats [21, 49–53]. Furthermore, important regulatory mechanisms such as chemosensation have been found to be reduced during REM sleep in comparison to wakefulness and NREM sleep [47, 52, 53]. The mechanisms underlying such state-dependent variations remain to be fully investigated and may account for the distinct effects of our chemogenetic manipulations across vigilance states.

The activity of pFRG is influenced by a combination of inhibitory and excitatory mechanisms that are still poorly understood [2, 7, 15, 19] and seem to be dependent on vigilance state and physiological conditions [17, 20]. Specifically, the cholinergic modulation of pFRG may be of particular interest to better understand state-dependent expiratory ABD activity [15]. Cholinergic neurons of the laterodorsal and pedunculopontine tegmental nucleus (LDT/PPT) seem to be involved in the initiation of REM sleep, whereas cholinergic cells in the basal forebrain have been demonstrated to influence the duration of REM epochs and the homeostatic sleep response in sleep-deprived animals [54–56]. It is possible that the state-dependent activity of cholinergic neurons in the LDT/PPT and basal forebrain regions could modulate the recruitment of ABD muscles either through direct or indirect stimulation of pFRG rhythmicity or through modulation of ABD motor neurons specifically during REM sleep. This could explain the observation of the strongest effects of pFRG modulation during REM sleep in the current study. Other potential explanations for the observation of ABD recruitment specifically during REM sleep could be the withdrawal of inhibitory inputs or the activation of glutamatergic pathways impingent on the pFRG [7, 19]. Previous studies have demonstrated fluctuations in the release of the neurotransmitter across sleep–wake cycles in various brain regions [57–59]. Therefore, it is possible that these vigilance state-dependent fluctuations in neurotransmitters release could also influence pFRG activity. Further studies will be required to confirm phenotype-specific anatomical connections to neurons in the pFRG and their possible implication in the modulation of ABD recruitment during sleep.

Breathing enhancement following the recruitment of ABD muscles

Our results show that increased excitability of pFRG through chemogenetic modulation produces an augmented occurrence of REM events displaying ABD recruitment, whereas its inhibition reduces the number of these events. Furthermore, in CTRL and excitatory conditions ABD recruitment events are associated with a reduction in respiratory variability, as well as an increase in ventilation, similar to previous reports [17, 20]. However, following chemogenetic inhibition of pFRG, ABD recruitment did not elicit reductions in respiratory variability or increased ventilation. It was previously proposed that activation of pFRG may stabilize breathing through either mechanical facilitation of air outflow with ABD muscle recruitment, or through direct and indirect excitation of preBötC [2]. As the facilitatory component of air outflow produced by ABD muscle recruitment is still present in the inhibitory conditions of our study, we hypothesize that the enhancements in breathing occurring with ABD activity could be due to central mechanisms associated with pFRG activation. Although anatomical evidence to support the idea of a direct or indirect excitatory connectivity from pFRG to preBötC is non-existent to our knowledge, it has previously been proposed in embryonic and neonatal stages [14, 16] and has been inferred from physiological evidence in adult rats [7].

Abdominal muscle recruitment in REM sleep is associated with reduced respiratory variability during its transient occurrence [17, 20]; present study). Therefore, we hypothesized that impairment of ABD recruitment would increase the number of respiratory disturbances observed in healthy animals, whereas potentiation of these events could reduce the respiratory disturbances naturally occurring in healthy rats. However, the analysis of respiratory disturbance frequency revealed that chemogenetic modulation of the expiratory oscillator had no influence on apnea occurrence.

Breathing is a robust process of vital importance for survival [1]. Previous studies have demonstrated that ablation and/or inhibition of specific cell populations at the level of the inspiratory oscillator can compromise the integrity of breathing [4, 6, 16, 33, 60–62]. However, pFRG is silent during restful breathing [7, 8] and further hyperpolarization of pFRG neurons (present study) decreases the occurrence of ABD recruitment and the enhancement of breathing associated with ABD activity, but does not induce further disruptions of breathing (i.e. apneas). Similarly, potentiation of pFRG increases the occurrence of ABD recruitment during sleep but does not affect naturally occurring apneas. Respiratory disturbances in healthy rats are minimal and occur at an average rate of 2 apneas per hour [63]; present study). Potentiation of pFRG activity in conditions in which apneas are nearly non-existent, as in the present study, may not allow for an accurate evaluation of the benefits associated with an enhancement of ABD recruitment during sleep, although effects may be beneficial in pathological conditions such as sleep-disordered breathing.

Funding

This work was financed by Canadian Institutes of Health Research (CIHR), Natural Sciences and Engineering Research Council of Canada (NSERC) and the Canadian Lung Association operating grants (to S.P.); A.P. is supported by Breathing As One - Lung Association and Alexander Bell - Canada Graduate Scholarship - Natural Sciences and Engineering Research Council of Canada; N.K. was funded by Women and Children Health Research Institute (WCHRI), NSERC and Alberta Innovates - Health Solutions (AIHS) undergraduate summer studentships.

Conflict of interest statement. None declared.