Assessing ventilatory instability using the response to spontaneous sighs during sleep in preterm infants

Abstract

Study Objectives

Periodic breathing (PB) is common in newborns and is an obvious manifestation of ventilatory control instability. However, many infants without PB may still have important underlying ventilatory control instabilities that go unnoticed using standard clinical monitoring. Methods to detect infants with “subclinical” ventilatory control instability are therefore required. The current study aimed to assess the degree of ventilatory control instability using simple bedside recordings in preterm infants.

Methods

Respiratory inductance plethysmography (RIP) recordings were analyzed from ~20 minutes of quiet sleep in 20 preterm infants at 36 weeks post-menstrual age (median [range]: 36 [34–40]). The percentage time spent in PB was also calculated for each infant (%PB). Spontaneous sighs were identified and breath-by-breath measurements of (uncalibrated) ventilation were derived from RIP traces. Loop gain (LG, a measure of ventilatory control instability) was calculated by fitting a simple ventilatory control model (gain, time-constant, delay) to the post-sigh ventilatory pattern. For comparison, periodic inter-breath variability was also quantified using power spectral analysis (ventilatory oscillation magnitude index [VOMI]).

Results

%PB was strongly associated with LG (r² = 0.77, p < 0.001) and moderately with the VOMI (r² = 0.21, p = 0.047). LG (0.52 ± 0.05 vs. 0.30 ± 0.03; p = 0.0025) and the VOMI (−8.2 ± 1.1 dB vs. −11.8 ± 0.9 dB; p = 0.026) were both significantly higher in infants that displayed PB vs. those without.

Conclusions

LG and VOMI determined from the ventilatory responses to spontaneous sighs can provide a practical approach to assessing ventilatory control instability in preterm infants. Such simple techniques may help identify infants at particular risk for ventilatory instabilities with concomitant hypoxemia and its associated consequences.

Statement of Significance

Ventilatory control instabilities are extremely common in newborns, some of which are obvious (i.e. periodic breathing) but others which may go unnoticed using standard clinical monitoring. However, quantitative methods to assess the degree of instability are lacking. Our study demonstrates that simple bedside recordings of infants respiratory patterns can be utilized to quantitatively assess the degree of ventilatory control instability. This clinically implementable tool may help identify infants at particular risk for ventilatory instabilities with concomitant hypoxemia and its associated consequences.

Introduction

Periodic breathing (PB; typically defined as clusters of breaths separated by intervals of apnea) is commonly observed in both preterm and term human infants [1–4], and its frequency is inversely related to postnatal age [5–7]. Importantly, PB can often be associated with severe bradycardia and rapid and profound falls in both arterial and cerebral oxygen levels [8–12]. Evidence from animal studies suggests that the persistence of repetitive hypoxemia can cause brain injury [13] and impair normal development and cognitive function [14]. Furthermore, exposure to repetitive hypoxemia augments carotid body chemoreceptor responses [15, 16], which is expected to further promote ventilatory instability in a vicious cycle. However, the mechanisms driving these dysfunctional patterns of breathing are not well understood.

The resemblance of PB to oscillations in man-made feedback control systems has led to the idea that the engineering concept of “loop gain” should give insight into the genesis and predisposition towards PB [17–19]. In the case of the respiratory system, the loop gain represents the sensitivity of the feedback loop controlling ventilation and is defined as the size of a “corrective” ventilatory response relative to the size of the ventilatory disturbance that elicits the correction [20, 21]. If the loop gain ratio exceeds 1 (at the natural frequency of the system), a small disturbance in ventilation results in an excessively large corrective overshoot, which manifests as oscillations in breath amplitude which are clinically observed as periodic central apneas. If the loop gain is less than 1, the disturbance will be damped and breathing generally stabilizes. Thus, knowledge of an individual’s loop gain allows the degree of instability underlying a breathing pattern to be assessed.

Existing techniques that either directly or indirectly assess loop gain, such as calculating the amount (or duration) of PB in a recording [17, 18], or assessing the duty ratio of PB [18, 22], require PB (and its associated deleterious consequences) to already be present. This is problematic, as many infants that either (1) do not demonstrate PB during a finite examination period or (2) only experience “oscillatory breathing” patterns (i.e. cyclic central hypopneas), may still have important underlying ventilatory control instabilities that are not objectively assessed/quantified by visual inspection. Furthermore, whether the source of this ventilatory variability is driven by chemical processes (i.e. chemoreceptor-driven) or other mechanisms (i.e. neural or mechanical) [23, 24] is largely unknown in these infants. Therefore, methods that can assess infants with “subclinical” ventilatory control instability (i.e. those with clear oscillatory breathing patterns) are needed to help better identify infants at risk who would otherwise be missed by conventional clinical criteria. Notably, methods to estimate an individual’s loop gain from the spontaneous breathing patterns now exist, and have been validated in adults with obstructive sleep apnea (OSA) [25, 26], but typically require a validated measure of airflow which can be challenging to obtain in neonates. Accordingly, the aim of the current study was to evaluate the predisposition towards ventilatory control instability using simple bedside ventilatory recordings in a cohort of preterm infants as they approach discharge from intensive care.

Methods

Participants

In this pilot study, we undertook a retrospective analysis of a select group of infants (n = 20) that were enrolled in the Prematurity and Respiratory Outcomes Project (PROP, Clinical Trials.gov NCT01607216) [27, 28], in which infants born between 24–28 weeks estimated gestational age had approximately 1 hour recordings of respiratory inductance plethysmography (RIP) and pulse oximetry made at 36 weeks post-menstrual age. Data from 20 spontaneously breathing infants were selected by one author (J.S.K.) because they demonstrated minimal (defined as zero time spent in PB, n = 10), moderate (spent between 0% to 10% of the recording in PB, n = 6), or a relatively large (spent greater than 10% of the recording in PB, n = 4) amounts of PB, compared to other PROP subjects. These data were selected prior to any assessment of ventilatory control stability/variability and irrespective of the infant’s estimated gestational age, or whether they were receiving nasal cannula flow or supplemental oxygen. The PROP protocol was approved by local institutional review boards at each of the clinical research sites and the PROP Observational Study Monitoring Board. Parents of the PROP enrollees provided written informed consent for all study procedures, including tidal breathing analyses.

Study design and experimental recordings

All infants were studied supine and all recordings were analyzed from epochs of behaviorally-determined quiet sleep. During each recording the total amount of sleep was quantified via observation [29, 30] as well as the percentage of the time the infant had spent in PB, with PB defined as more than three consecutive cycles of breathing and apnea (with apnea defined as a complete cessation of airflow for greater than 3 seconds) [31]. In order to assess the infants’ ventilatory control stability and variability, we assessed the ventilatory response to spontaneous sighs that occurred during these recordings with two separate techniques (detailed below).

Identification of spontaneous sighs

In order to identify the spontaneous sighs for analyses, all RIP and pulse oximetry signals were displayed using the Vivosense software package (Vivonoetics, Newport Coast, CA). Spontaneous sighs were then detected visually as an augmented breath that was at least twice the volume of preceding breaths [32] accompanied by elevated chest and abdominal movements during periods of quiet sleep. Once a sigh was identified, measurements of flow and tidal volume (both derived from the RIP signals [33]) as well as oxygen saturation (SpO₂%) were then exported for the 30 seconds prior and 90 seconds following each sigh (start sigh = time zero) into Matlab (Mathworks, Natick, MA) for analysis. In order to be included in our analysis we required a 30-second period of stable breathing prior to the sigh, but did not exclude the data if multiple sighs occurred after the initial sigh. Data were excluded if any movement-related artifact impaired the signal quality within the 120-second analysis window.

Determining ventilatory control stability (i.e. loop gain)

Loop gain is the input–output function of the feedback loop controlling ventilation, which determines the magnitude and time course of the ventilatory “response” that follows a ventilatory “disturbance” (i.e. a sigh). Breath-by-breath measurements of ventilation (normalized by the mean ventilation during the analysis window) were derived from RIP traces (30 seconds prior and 90 seconds following each sigh). A standard model of ventilatory control (gain, time-constant, delay) that transforms the normalized ventilatory fluctuations seen in response to a sigh into a ventilatory-drive signal that best matches observed ventilation (Figure 1a) was then used to calculate loop gain at the natural cycling frequency (i.e. the frequency of PB if breathing was unstable) [25]. Sighs were incorporated as an independent source of ventilatory drive and modeled using an additional parameter. A loop gain, time-constant, and delay were then determined for each window analyzed, which was then averaged to produce a single value for each infant. For inclusion in analysis we required a minimum of three windows for each infant.

Figure 1.

Measuring loop gain and VOMI from spontaneous sighs. (a) In order to calculate loop gain, sighs were incorporated as an independent source of ventilatory drive and were defined as a ventilation >2 times the resting mean. Briefly, a normalized ventilation trace (blue waveform, bottom panel) was calculated from the derived flow trace. This signal was then input into a mathematical model which determined the best set of parameters (gain, time-constant, delay, additional ventilation due to the sigh) that gave the closest match between ventilatory drive (chemical drive + response to sigh, green line) and ventilation. A loop gain value was determined for each data segment and then averaged to produce a mean value for each infant. (b) Illustration of the VOMI extraction procedure from the computed power spectral density of the interpolated signal. The oscillatory response to the sigh is characterized by the appearance of a peak on the power spectral density curve. The frequency at which this peak is centered would determine the cycle duration of these periodic oscillations. This particular example shows an oscillatory pattern whose cycle duration is approximately 12.5 seconds.

Determining the ventilatory oscillation magnitude index

Sources of ventilatory variability can be either attributed to mechanical, chemical, and/or neural mechanisms [23, 24]. Mechanical and chemical homeostatic processes would evoke variability that is deterministic in nature and can take the form of periodic oscillations such as those found in PB. Conversely, neural inputs associated with fluctuations in sleep/wake state will produce more random breath-by-breath variations that can sometimes give the appearance of periodic-like behavior. Total ventilatory variability can, therefore, be considered as the sum of deterministic and random variations [34, 35]. To quantify the degree of ventilatory variability associated with the oscillatory behavior that is present in the ventilatory responses to the sighs, we applied spectral analysis to the normalized ventilation time-series. First, all sigh breaths were eliminated from the RIP-derived ventilation time-series by using linear interpolation between the breaths that preceded and followed the sighs. Second, the interpolated time-series was resampled at 4 Hz to generate an equally spaced measurement and then linearly de-trended. Subsequently, the power spectral density of the preprocessed time-series was computed based on the fast Fourier transform periodogram method. Lastly, the area under the power spectral density curve over the low-frequency band (0.03 to 0.125 Hz) was calculated to yield the ventilatory oscillation magnitude index (VOMI; Figure 1b). This frequency band represents the range of periodicities (8 to 30 seconds) exhibited by PB in preterm infants [1, 36, 37]. The VOMI was finally log-transformed and multiplied by a factor of 10 to yield a measurement in decibels (dB), which is a unit used in science and engineering to express power. A purely sinusoidal oscillation in ventilation that just reaches apnea (rise and fall by 100% of eupnea) would have a VOMI of −1.5 dB and an oscillation that is half the size (rise and fall by 50% eupnea) would have a VOMI of −4.5 dB (i.e. 3 dB lower).

Statistical analyses

The correlation between our measures of ventilatory control stability (i.e. loop gain and VOMI) and the percentage of the time of the recording that infants spent in PB was calculated. Unpaired t-tests were used to compare the loop gain and VOMI values between infants that displayed zero PB during the recording against infants that displayed any PB. Receiver operating characteristic (ROC) curves were then calculated to determine the best threshold values (based on sensitivity and specificity) that differentiated infants based on whether or not PB was present. A p-value < 0.05 was considered statistically significant.

Results

Table 1 displays the anthropomorphic characteristics as well as the average amount of quiet sleep obtained and number of windows analyzed in the cohort. Within the recordings, the percent of sleep time with PB ranged from 0% (10 infants) to 87.3% (mean ± SD; 9.8% ± 19.9%).

Table 1.

Preterm demographics and sleep information

Variables	All infants	Infants without PB	Infants with PB
Gestational age (weeks) at enrollment	27 (25–28)	28 (25–28)	27 (25–28)*
Gender (M:F)	9:11	6:4	3:7
Birth weight (grams)	975.3 ± 233.7	1109.5 ± 226.2	841.0 ± 155.2*
APGAR scores at 1 minute	5 (1–8)	5 (1–8)	5 (1–8)
APGAR scores at 5 minutes	6 (1–9)	6 (3–9)	7 (1–9)
Post-menstrual age (weeks) at recording	36 (34–40)	36 (34–40)	37 (34–39)
Quiet sleep duration (minutes)	19.3 ± 7.0	19.2 ± 5.8	19.3 ± 8.4
Total recording duration (minutes)	53.9 ± 13.6	50.7 ± 15.1	57.1 ± 11.8
Sigh frequency (sighs/minute of total recording)	0.7 ± 0.2	0.7 ± 0.3	0.7 ± 0.2
Number of windows analyzed per infant	5.6 ± 2.2	5.7 ± 1.8	5.4 ± 2.7

Values are present as mean ± SD or median (range) as appropriate.

*p ≤ 0.05 when comparing infants with and without PB.

Relationship between %PB and ventilatory stability indices

%PB showed a moderate positive association with both loop gain (r² = 0.30, p = 0.013) and the VOMI (r² = 0.26, p = 0.023). One infant had a %PB of 87.3% that was well outside 2 SDs of the cohort. Removal of this outlier resulted in a stronger association between the %PB and loop gain (r² = 0.77, p < 0.001; Figure 2a), yet minimally affected the strength of the association with the VOMI (r² = 0.21, p = 0.047; Figure 2b). When comparing the relationship between loop gain and VOMI (using the entire dataset), we observed a moderate association between the two measures of ventilatory stability (r² = 0.25, p = 0.026; Figure 2c). Forward step-wise regression suggested that after accounting for loop gain, the inclusion of the VOMI did not significantly add to the ability to predict the %PB.

Figure 2.

(a) Loop gain was moderately associated with the percent PB (black dotted line) when using the entire dataset. After removal of the outlier, the association between loop gain with the percent PB (solid blue line) became stronger. (b) Ventilatory oscillation magnitude index (VOMI) was moderately associated with the percent PB regardless of whether the outlier was included (dashed black line) or not (solid blue line). (c) Loop gain was moderately associated with the VOMI.

Potential mediators driving the appearance of PB

The frequency of sighs in the recording showed no association with %PB, nor was it associated with loop gain, VOMI or average SaO₂. While the average SaO₂ during the windows analyzed did not correlate with the %PB (r² = 0.11, p = 0.14), loop gain was strongly associated with both the %PB and the average SaO₂ (r² = 0.39, p = 0.003). When comparing the breathing stability indices between infants with vs. without PB (Figure 3), we found that infants without PB had a lower loop gain (0.30 ± 0.03 vs. 0.52 ± 0.05; p = 0.0025), shorter chemoreflex delay (2.9 ± 0.2 seconds vs. 4.0 ± 0.1 seconds; p = 0.0005), faster cycling period (11.3 ± 0.8 seconds vs. 14.9 ± 0.4 seconds; p = 0.001) and lower VOMI (−11.8 ± 0.9 dB vs. −8.2 ± 1.1 dB; p = 0.026). Furthermore, the average SaO₂ during the windows analyzed was lower (i.e. more hypoxic) in the infants that displayed PB compared to those without PB (97.3% ± 0.7% vs. 92.4% ± 1.2%; p < 0.001). While there was no difference in frequency of sighs between infants with and without PB when studied, infants that displayed PB were typically born earlier and weighed less than those that displayed PB (Table 1).

Figure 3.

Differences in the ventilatory stability parameters between infants that displayed no PB vs. those in which PB was observed. In infants that displayed PB, (a) loop gain, (c) chemoreflex delay (d) cycling frequency, and (e) VOMI were all elevated whereas the (b) time constant remained unchanged.

ROC curve analysis using loop gain as the testing variable to predict the presence/absence of PB showed a significant area under the curve (AUC) of 0.90 ± 0.07 (p = 0.0025). Specifically, a loop gain of >0.319 provided a sensitivity of 100% (confidence interval [CI]: 69%–100%) and a specificity of 80% (CI: 44.4%–97.5%) to detect the presence of PB. Comparatively, using VOMI as the testing variable resulted in a significant AUC of 0.79 ± 0.121 (p = 0.03); a VOMI of >−10.12 provided a sensitivity of 80% (CI: 44.4%–97.4%) and a specificity of 90% (CI: 55.5%–99.75%) to detect the presence of PB.

Discussion

The major finding from this proof-of-principle study is that a simplified ventilatory control model appropriately fits the ventilatory responses to spontaneous sighs and captures the predisposition to PB in preterm infants (Figure 1). To our knowledge, this is the first report that quantitatively assesses loop gain in a preterm population. In addition, linear associations confirmed the longstanding hypothesis that the mechanism of PB in infants is elevated loop gain, but also demonstrated that the elevated loop gain is driven partly by increased chemoreflex delay. The increased delay may be driven by increases in circulatory time, but could also reflect chemical/neural sensory processing latency. Thus, our approach to assessing loop gain in infants may provide insight into the causes of unstable breathing. We also show that a considerable %PB is observed without loop gain exceeding 1, suggesting that a substantial amount of PB may be being sustained by ongoing perturbations in a stable (but near-unstable) system. Taken together, the techniques presented here may help identify infants at particular risk for ventilatory instabilities with concomitant hypoxemia and its associated consequences.

Comparison to previous literature

The concept of assessing the ventilatory response to sighs to determine the stability of the ventilatory control system has been used by others in the past. For example, using an animal model, Khoo et al. [38] has fitted the ventilatory response to sighs with a simple chemoreflex model which enabled an assessment of the peripheral chemosensitivity (i.e. controller gain)—a component of the overall loop gain. Furthermore, in a landmark study, Fleming et al. [32] conducted the first study to quantify the stability of the ventilatory control system in healthy term infants and how it changed over the first year of life. Ventilatory responses that occurred following sighs were fitted with second-order equations representing the critically or underdamped response. While the authors did not quantify the loop gain per se in their study, we have since been able to derive an estimate of the infants’ loop gain and reported how loop gain (on average) is altered by development [21]. Notably, while the methods used to assess loop gain in the current work differed from that used in our previous retrospective analysis [21] of the data by Fleming et al. [32], the range of loop gain values derived are quite similar.

Mechanisms of elevated loop gain

The preterm predisposition to PB is complex and is likely attributable to a variety of physiological factors [21]. In brief, the underlying mechanism believed to be responsible for PB at this age is an increased chemosensitivity that develops with arterial hypoxemia, due to deficient ventilation-perfusion matching in the immature lung [1, 39]. Indeed, in the present study, the preterm infants that displayed PB were more hypoxemic (as evidenced by the lower average SpO₂ during the analyzed windows), and notably had higher loop gains (Figure 3). While infants with PB in had significantly lower average SpO₂ values compared to those without PB, the degree of hypoxemia experienced did not correlate with the %PB. Conversely, loop gain was strongly associated with both the %PB and the level of hypoxemia present, supporting the argument that the presence of hypoxemia likely elevates loop gain (e.g. via effects on chemosensitivity) which in turn is responsible for the development of PB. Hypoxemia could also be a marker of a reduced effective lung gas volume in its role in increasing loop gain. Importantly, the finding that the increased loop gain is in-part explained by chemoreflex delay is novel; although the origins of the increased delay (i.e. increases in circulatory delay or chemical/neural sensory processing latency) require further investigation.

Clinical utility and future applications

To date, several methods have been employed to characterize the stability and variability of the ventilatory control from spontaneous breathing patterns. However, they often rely on long recordings over many hours, can only assess loop gain once PB with clear apneas are captured [18, 22], or require a validated measure of airflow (i.e. nasal pressure, pneumotach-derived flow). However, these methods are not capable of quantifying an individual’s loop gain during oscillatory breathing (without apnea) or during periods of relatively stable breathing. The method for assessing loop gain employed in the current work, adapted from the techniques to measure loop gain in adults [25] with OSA, does not require capturing epochs of PB and can provide an assessment an infant’s loop gain across the range of stable to unstable breathing patterns. Specifically, it uses the natural fluctuations in ventilation (i.e. ventilatory response) that accompany spontaneous sighs (i.e. ventilatory disturbance). As such, our method avoids the need for classic physiologic tests to measure chemoreflex sensitivity (e.g. exposures to hypercapnic, hypoxic or hyperoxic gases), which are technically challenging and not part of standard care for preterm infants. An additional advantage of our method is that it minimizes the recording time required to detect potential problems with ventilatory control because sighs are extremely frequent in preterm infants. Lastly and perhaps most importantly, this report is the first time loop gain has been assessed in a preterm population and was achieved using only RIP bands, making this measurement easy to perform in the neonatal setting.

A priori knowledge of an infant’s underlying loop gain may lead to the identification of infants that are at particular risk for developing unstable breathing during periods of acute pathology such as upper respiratory infections. Understanding how loop gain changes with development should provide an insight into whether an infant, at a particular post-menstrual age, can respond robustly to respiratory challenges (i.e. apnea). Such a tool could also be utilized to assess a number of other clinical questions: For instance, how does loop gain change with development? Is this trajectory/pattern different in preterm vs. term infants? Does an elevated loop gain early in life predict the development of other disorders (e.g. bronchopulmonary dysplasia or apnea of prematurity) or help explain the increased prevalence of OSA in ex-preterm infants [40]? Furthermore, this technique may also help predict the response to common treatments, such as caffeine and supplemental oxygen, for prolonged apnea in preterm infants (“apnea of prematurity”). Certainly, knowledge of the underlying loop gain has been helpful in predicting which adult OSA patients are likely to respond to a variety common non-CPAP therapies [41–45].

Ventilatory variability vs. stability/instability

It is important to note that inherently stable ventilatory control systems (i.e. systems with a low loop gain) can also exhibit breathing patterns with some periodicity, for example, periodic hypopnea [35, 46]. Such periodic-like behavior may be driven by either extrinsic (e.g. fluctuations in sleep–wake state and other physiological variables, behavioral inputs, neural variability external to chemoreflex feedback) or intrinsic (e.g. neural variability at the level of respiratory pattern generator or within chemoreceptor circuits in the medulla) factors (often termed biological noise). Notably, when the loop gain is low (as it typically is in a “stable” system), the magnitude of biological noise plays a key role in the pathogenesis of oscillatory breathing. In the current study, we aimed to quantify the degree of ventilatory variability in the system, periodic plus non-periodic, using the VOMI (i.e. magnitude of ventilatory oscillations with periodicities between 8 and 30 seconds), for the purpose of quantifying the size of both sub-clinical ventilatory oscillations as well as visually-detectable PB. Similar to our findings with loop gain, the VOMI moderately correlated with the %PB, and was lower in the infants that displayed no overt PB, indicating that they also had more regular breathing. Interestingly, when loop gain and VOMI were entered into a forward stepwise regression model to predict the %PB, only one of these measures is required to account for the observed variance (i.e. when loop gain is added first VOMI is excluded and vice versa). We interpret this to mean that both of these measures are tapping into the same construct, and although the VOMI is sensitive to a range of respiratory disturbances, it appears that in this population VOMI is primarily being driven by fluctuations in the chemoreflex feedback system.

Methodological considerations

There are a few limitations of the present work that need to be taken into consideration when interpreting our findings. First, this study was conducted as a pilot study and thus the sample size was small. Future studies in larger populations of infants will be required to confirm whether our findings are robust, and whether they are useful in predicting both short and long-term clinical outcomes. Second, our estimate of loop gain assumes that spontaneous sighs occurred as an external (i.e. independent) disturbance to the ventilatory control system. While the physiological mechanisms responsible for inducing sighs remain unclear, established indexes of respiratory drive do not appear to be predictive of an impending sigh [47], which would support our assumption that a sigh is an independent spontaneous disturbance. Lastly, the methods used to assess loop gain/ventilatory variability, are not able to discern all of the precise reasons that loop gain may be elevated (i.e. whether a high loop gain is driven by an enhanced chemoreflex sensitivity or via increases in the plant gain). While it is challenging to obtain high quality, breath-to-breath recordings in preterm infants, the inclusion of measurements of end-tidal oxygen and carbon-dioxide [48] in future investigations may help uncover the mechanisms of the elevated loop gain.

Summary and conclusions

The current pilot study highlights that measures of ventilatory control stability and variability (i.e. loop gain and VOMI) determined from the ventilatory responses to spontaneous sighs provide a practical approach to identifying otherwise unrecognized ventilatory control instability in “clinically stable” preterm infants. Such simple techniques may help identify infants at particular risk for ventilatory instabilities with concomitant hypoxemia and its associated consequences.

Funding

This work was supported by National Institutes of Health (NIH), National Heart, Lung, and Blood Institute (NHLBI) and National Institute of Child Health and Human Development (NICHD) through U01 HL101800 to Cincinnati Children’s Hospital Medical Center, AH Jobe and CA Chougnet; U01 HL101798 to University of California San Francisco, PL Ballard and RL Keller; U01 HL101456 to Vanderbilt University, JL Aschner; U01 HL101465 to Washington University, A Hamvas and T Ferkol. B.A.E. was supported by the National Health and Medical Research Council (CJ Martin Overseas Biomedical Fellowship, 1035115) and is now supported by a Future Leader Fellowship (101167) provided by the National Heart Foundation of Australia. M.C.K. and L.N-G. were supported by a National Institute of Biomedical Imaging and Bioengineering grant (EB001978). S.A.S. was supported by the American Heart Association (15SDG25890059) and the NHLBI (R01HL102321, R01HL090897, R01HL128658, R35HL135818, and P01HL094307). P.I.T. was supported by a project grant (1064163). S.A.L. is supported by NeuroSleep, an NHMRC Centre of Research Excellence (1060992) and the Monash University Faculty of Medicine Nursing and Health Sciences bridging postdoctoral fellowship. S.A.S. has received personal fees from Cambridge Sound Management, Nox Medical, and Merck unrelated to the current work.

Conflict of interest statement. None declared.