Observation of cardiogenic flow oscillations in healthy subjects with hyperpolarized ³He MRI

Abstract

Recently, dynamic MRI of hyperpolarized ³He during inhalation revealed an alternation of the image intensity between left and right lungs with a cardiac origin (Sun Y, Butler JP, Ferrigno M, Albert MS, Loring SH. Respir Physiol Neurobiol 185: 468–471, 2013). This effect is investigated further using dynamic and phase-contrast flow MRI with inhaled ³He during slow inhalations (flow rate ∼100 ml/s) to elucidate airflow dynamics in the main lobes in six healthy subjects. The ventilation MR signal and gas inflow in the left lower lobe (LLL) of the lungs were found to oscillate clearly at the cardiac frequency in all subjects, whereas the MR signals in the other parts of the lungs had a similar oscillatory behavior but were smaller in magnitude and in anti-phase to the signal in the left lower lung. The airflow in the main bronchi showed periodic oscillations at the frequency of the cardiac cycle. In four of the subjects, backflows were observed for a short period of time of the cardiac cycle, demonstrating a pendelluft effect at the carina bifurcation between the left and right lungs. Additional ¹H structural MR images of the lung volume and synchronized ECG recording revealed that maximum inspiratory flow rates in the LLL of the lungs occurred during systole when the corresponding left lung volume increased, whereas the opposite effect was observed during diastole, with gas flow redirected to the other parts of the lung. In conclusion, cardiogenic flow oscillations have a significant effect on regional gas flow and distribution within the lungs.

in the literature, the term cardiogenic oscillation has been used to refer to the modulation of the pulmonary gas pressure, flow, or concentration produced by the cardiac cycle. Cardiogenic oscillations have been intensively observed and recorded in the past in the context of pulmonary physiology measurements made at the mouth with pressure transducers, pneumotachographs (1, 20, 29), or gas analyzers (6, 15) but also directly inside of the intrathoracic airways during bronchoscopy (30). The cardiac cycle is thought to be an important component of gas mixing within the lung (12, 13, 15) and leads to oscillations in the concentration of oxygen and carbon dioxide in expired gas (5, 6). Therefore, the cardiac action has an uncontested influence on lung function, but it is unclear how the different observations of cardiogenic oscillations are related to each other. In the present study, only cardiogenic flow oscillations (CO_f), present in the conducting airways, are considered. CO_f can be detected in most subjects during the whole breathing cycle and in all regions of the lung (11). However, very little data have been published on the influence of CO_f on airflow pattern and gas distribution within the lung.

On the other hand, the field of study of pulmonary airflow has recently benefited from advancements in imaging and computation methodology. Computational fluid dynamics (CFD) simulations using realistic image-based airway models have vastly improved general understanding of the local characteristics of gas flow in the airways (25). These findings are of particular interest for inhaled therapy research (4) or the study of regional deposition of particles (17) but have yet to take into account realistic physiological features, such as CO_f, and have traditionally relied only on validation with in vitro airway tree models (21). In vivo, rates of gas ventilation in the main airways and the periphery can be assessed with dynamic hyperpolarized (HP) ³He ventilation imaging (16, 31). In addition, phase-contrast velocimetry (PCV) sequences can be used to map flow velocity profiles directly in the major airways (3, 10, 19). Sun et al. (24) recently performed dynamic ³He ventilation imaging on seven healthy subjects and revealed an alternation of the MRI intensity between the left and right lung (“ventilatory alternans”) with a periodicity approximating the heart rate (unmeasured), which was presumably the effect of CO_f. The aim of the present study is to explore this phenomenon further by performing dynamic HP ³He ventilation imaging during inspiration in healthy subjects to assess the extent to which CO_f can influence the gas distribution inside the lobes and additionally, by measuring gas velocities inside the main bronchi with PCV sequences to investigate the airflow pattern within the lungs.

MATERIALS AND METHODS

Subject Characteristics, ³He Production, and Administration

Six healthy volunteers were recruited for this study (demographics and pulmonary function test results shown in Table 1). Approval from the National Research Ethics Committee was obtained for all experiments. ³He was polarized on-site with a regulatory-approved spin exchange polarizer to ∼25% (GE Healthcare, Amersham, UK). A 1-liter gas mixture of N₂ and HP ³He was delivered for the subjects to inhale inside of the MRI scanner. The subjects performed a slow and constant-rate inspiration (inhalation time varied from 8 to 15 s between subjects) from a Tedlar bag (Jensen Inert Products, Coral Springs, FL) containing 20% of HP ³He for dynamic ventilation and 30% for PCV imaging. Data were acquired during inhalation.

Table 1.

Subject characteristics and sequences performed

Subject (Sex)	Age, yr	Weight, kg	FEV1, liter (% Predicted)*	FVC, liter (% Predicted)*	Dynamic Imaging	1D Velocity Profile	2D Flow	1H MRI
1 (M)	28	79	4.95 (107.8)	6.38 (108.7)	X	X	X	X†
2 (F)	32	62	3.46 (115.8)	4.54 (131.8)	X†	X†		X†
3 (M)	31	74	3.65 (81.5)	5.1 (94.4)	X†	X†		X†
4 (M)	27	79	4.87 (113.0)	5.88 (108.7)	X	X†		X†
5 (M)	31	89	3.65 (88.9)	5.07 (102.1)	X†	X†	X	X†
6 (M)	24	77	5.29 (107.8)	6.56 (111.3)	X†	X†	X†	X†

M, male; F, female; FEV₁, forced expiratory volume in 1 s; FVC, forced vital capacity. 1/2D, 1-/2-dimensional.

^*See Quanjer et al. (22);

^†synchronized ECG recording available.

Image Acquisition

Imaging experiments were performed on a GE HDx 1.5T MR scanner with a maximum gradient strength of 33 mT/m and slew rate of 120 mT·m⁻¹·ms⁻¹. A quadrature flexible transmit-receive (T-R) ³He radiofrequency coil (Clinical MR Solutions, Brookfield, WI) was used for dynamic ventilation and one-dimensional (1D) velocity profile measurements. For 2D flow imaging across the left main bronchus (LMB), a home-built dedicated, loop-butterfly quadrature T-R coil was used with a ratio of quality factor Q, when unloaded to loaded, of Q_unloaded/Q_loaded ∼230/25 (Fig. 1). The heartbeat of the subjects was monitored during all experiments with a finger probe or an ECG. When ECG was available (see Table 1), the R-wave occurrences were recorded during the imaging sequences. Sequence details were as follows.

Fig. 1.

Picture of the quadrature transmit-receive radiofrequency coil used for 2-dimensional (2D) flow measurement in the left main bronchus (LMB; see inset, slice location 2, solid line). Inset: example of a hyperpolarized ³He ventilation image with indication of slice locations used for 1D (axial 1D projections, across the anterior to posterior direction, below the carina, dashed line; 1) and 2D (solid line; 2) phase-contrast velocimetry (PCV) sequences.

Dynamic ventilation imaging.

Dynamic coronal images of the six healthy subjects were obtained with a spoiled gradient echo sequence [40 × 32 cm² field of view (FOV), 64 readout × 52 phase encode matrix, Cartesian sequential phase encoding, single slice, 25 cm thickness, echo/repetition time (T_E/T_R) of 0.8/2.7 ms, ±31.25 kHz bandwidth, 3° flip angle, and 150 frames with a time resolution/frame of 140 ms].

1D Velocity profile imaging.

Axial 1D projections, across the anterior to posterior direction, below the carina (Fig. 1, slice location 1), were acquired in the six subjects with a PCV sequence, providing dynamic 1D velocity profiles across the LMB and right main bronchi (RMB). A field of speed (FOS) of [−120, +120] cm/s was chosen along the superior-to-inferior direction, which is the principal direction of gas flow (20 cm FOV, 128 points, 1 cm slice thickness, 20° flip angle, and 768 frames with a 20-ms time resolution).

2D Flow imaging.

A 2D oblique slice through the LMB (Fig. 1, slice location 2) was imaged in subjects 1, 5, and 6 with a PCV sequence. The FOS was set to [−160, +160] cm/s in the direction of the axis of the LMB (5 × 3.75 cm² FOV, 32 × 18 matrix with a partial Fourier factor of 0.75 in the phase direction and sequential Cartesian phase encoding, 1.5 cm slice thickness, T_E/T_R of 3.3/6.4 ms, ±15.63 kHz bandwidth, 18° flip angle, and 232.4 ms time resolution/image frame).

Cardiac-gated proton lung imaging.

To observe the deformation of the lungs around the heart during the cardiac cycle, a standard cine cardiac-gated, balanced, steady-state-free precession ¹H pulse sequence was used to obtain an axial stack of images of the lungs in the six subjects during breath holding (30 cm FOV, 256 × 256 matrix, 1 cm slice thickness, 60° flip angle, T_E/T_R of 1.8/4.3 ms, and 20 heart phases).

Image Analysis

For the dynamic ventilation images, four regions of interest (ROIs) in the right upper lobe (RUL), left upper lobe (LUL), right lower lobe (RLL), and left lower lobe (LLL) parts of the lungs (Fig. 2A) were chosen, and the time evolution of the ³He MR signal was computed in each. The mean signals in each ROI were divided by the noise estimated from the first image of the experiment (acquired before inhaling ³He) to compute signal-to-noise ratio (SNR) values. A Fourier analysis was performed for each SNR time curve to detect the frequency of signal oscillations. The phase difference between signals at the fundamental frequency was also computed (the phase of the LLL signal was taken as a reference). Cardiac-gated proton images were analyzed with ScanIP (Simpleware, Exeter, UK) to segment the volume of the lung cavity for each of the 20 cardiac frames. Pulmonary veins and arteries were not excluded. The relative volume changes (volume divided by the average volume during the cardiac cycle) of the left and right lungs around the heart during the cardiac cycle were derived from the segmentation.

Fig. 2.

A: time evolution of the signal-to-noise ratio (SNR) in the 4 regions of interest (ROIs): right upper lobe (RUL), left upper lobe (LUL), right lower lobe (RLL), and left lower lobe (LLL) of the Cartesian dynamic images of subject 1 (example image and ROI positions shown in the inset). B: frequency spectrum of the signal corresponding to the LLL of the lung (curve LLL in A) after a Fourier transform (FT). The peaks at 1.08 and 2.2 Hz correspond to the fundamental and the 2nd harmonic of the heart rate of the subject [65 beats/min (BPM)]. C–G: time evolution of the signal in the 4 ROIs for subjects 2–6, respectively. For subjects 2, 3, 5, and 6, the ECG was recorded, and the dashed, vertical lines correspond to the occurrence of R-waves. Note: an R-wave occurrence is missing (not recorded) in G (subject 6) at ∼7 s.

For the PCV data, velocity maps and profiles were generated offline from the raw data using in-house software developed in Matlab (MathWorks, Natick, MA). The phase-difference reconstruction was performed using the two interleaves of each frame to extract the ³He gas velocity component in the encoded direction for each pixel. ROIs corresponding to the LMB and RMB were selected manually, and the time evolution of the average gas velocity was calculated for each 1D profile. For the 2D flow imaging experiments, the flow was derived by integrating the velocities over the area of the LMB. An error analysis of the 1D average gas velocity and 2D flow values was performed. The following formula gives the statistical uncertainty of the velocity value σ_v derived from the phase-difference reconstruction

$${\sigma }_v=\frac{FOS}{2\pi }\sqrt{{\left(\sigma /S_1\right)}^2+{\left(\sigma /S_2\right)}^2}$$ (1)

where S₁ and S₂ are the magnitude values of the two interleaves, and σ is the SD of the noise (the same in both interleaves). In practice, S₁ ∼ S₂, and the uncertainty in each velocity value is therefore inversely proportional to the signal of each corresponding pixel. Since the signal intensity is velocity dependent, the SD of the velocity between pixels can be fairly different. Indeed, the volume of gas in a pixel experiencing high flow rate is renewed with “fresh” polarized gas (that has not undergone radio frequency destruction) more rapidly, whereas pixels with low flow rate have a lower signal and therefore, a higher variance σ_v². The variance of the 2D flow measurement was calculated from the sum of each pixel variance multiplied by the pixel area, whereas the variance of the 1D average velocity measurement was derived according to the following formula

$${\sigma }_{av}^2=\left({\displaystyle {\sum }_{ij}^n}{\sigma }_{vij}^2\right)/n$$ (2)

where σ_vij is the SD of the pixel ij, and n is the number of pixels in the selected ROI.

RESULTS

Dynamic Ventilation

The SNR and hence, gas inflow in the LLL of the lung were clearly found to oscillate at the cardiac frequency in all subjects (see Figs. 2 and 4, Table 2, and Supplemental Material for videos of dynamic ventilation images of each subject). The signals in the other parts of the lungs (RLL, RUL, and LUL) were also found to oscillate at the same frequency in five of the six subjects but with a markedly different phase compared with the LLL signal (139° was the average phase difference for the LUL signal, 167° for RLL, and 145° for RUL). For subject 5 (Fig. 2F), no obvious pulsation was observed in the time evolution of RLL, RUL, and LUL SNRs. The simultaneous ECG recording in subjects 2, 3, 5, and 6 showed that the R-wave occurrences corresponded to a minimum of SNR in the LLL (maximum SNR in the other regions for subjects 2, 3, and 6) and were followed by a rapid rise of the LLL signal.

Fig. 4.

Correlation and linear regression line of the fundamental frequency of the signal oscillations and the heart rate measured with ECG. +, From the SNR oscillations of LLL during the dynamic ventilation experiments; ⋆, from the 1D velocity measurement (in LMB); solid line, linear regression line [f(SNR) = 6.24 + 0.92 × heart rate, correlation coefficient: 0.99]. §, heartbeat measured by finger probe shortly before the experiments and not by synchronized ECG recording; these 3 values are shown on the graph but were not included into the regression line nor were they used in the calculation of the correlation coefficient.

Table 2.

Fourier transform analysis of the data from the dynamic ventilation experiments

Subject	Heart Rate	Frequency of SNRLLL oscillations*	Δ∅LUL = /OLUL − /OLLL	Δ∅RLL = /ORLL − /OLLL	Δ∅RUL = /ORUL − /OLLL
1	65†	65	108°	−171°	166°
2	85‡	83	−140°	169°	152°
3	69‡	70	132°	178°	144°
4	62†	60	−141°	147°	146°
5	96‡	93	§	§	§
6	86‡	87	−172°	168°	119°

^*In beats/min;

^†from finger probe shortly before the experiment;

^‡from synchronized ECG recording;

^§no clear oscillations were observed for subject 5 in the left upper lobe (LUL), right lower lobe (RLL), and right upper lobe (RUL), explaining why no values are quoted. SNR_LLL, signal-to-noise ratio in the left lower lobe; Δ∅, phase difference.

1D Velocity Profile

Although the experiments were performed during constant inhalation, the average velocities in the LMB and RMB fluctuated dramatically with a periodic pattern whose frequency matched the heartbeat (Figs. 3 and 4). The recorded velocities ranged from −50 to 150 cm/s and varied antagonistically. More surprisingly, negative values (backflows) were measured during a small part of the cardiac cycle in subjects 1, 3, 4, and 6, demonstrating a pendelluft effect at the carina bifurcation between the left and right lungs. Although the periodic patterns were quite different among subjects, recordings of R-waves (start of systole) always preceded a strong gas inflow period in the left lung (low inflow or backflow in the right lung), whereas the opposite phenomenon happened at the beginning of diastole. The mean uncertainty in the measured average velocity value was found to be ±6 cm/s.

Fig. 3.

1D PCV. Time evolution of the average velocity of ³He gas in the LMB/right main bronchi (RMB) in subjects 1–6 (A–F, respectively) during a constant inhalation (see Fig. 1, inset, slice location 1). The vertical, dashed lines correspond to the occurrences of the R-waves that were recorded with ECG (not recorded for subject 1; A). For subject 3 (C), a case of arrhythmia can be observed between 6 and 7 s. Mean estimated errors for each average velocity curve were the following: LMB: ±5 cm/s, RMB: ±5 cm/s (A); LMB: ±8 cm/s, RMB: ±7 cm/s (B); LMB: ±6 cm/s, RMB: ±5 cm/s (C); LMB: ±8 cm/s, RMB: ±6 cm/s (D); LMB: ±7 cm/s, RMB: ±6 cm/s (E); and LMB: ±5 cm/s, RMB: ±5 cm/s (F).

2D Flow

Measured velocity maps from the LMB confirmed the periodicity and dramatic change of gas flow pattern during the cardiac cycle (Fig. 5) and the existence of backflow in subjects 1 and 6. Flow values ranged from −40 to 220 ml/s. For subject 6, the ECG was recorded simultaneously, and the maximum flow rate into the left lung was observed after the R-wave occurrence. σ_v was ∼5 cm/s, and the uncertainty on the flow values was below ±2 ml/s.

Fig. 5.

2D PCV. Time evolution of the ³He gas flow in the LMB of subject 7, measured during a constant inhalation (see Fig. 1, inset, slice location 2; time resolution of 232.4 ms). Example velocity maps (1–13), from which the values of the flow curve were calculated, are shown around the plot (color bar in centimeters per second). The average statistical uncertainty of the velocity value for the given velocity maps was ∼5 cm/s. The flow error bars (on the vertical axis) were calculated according to the method described in Image Analysis and range between ±1 and ±2 ml/s. The “x error bars” on the flow curve represent the acquisition window for each velocity map.

Proton Imaging

Manual segmentation of left and right lung cavities surrounding the heart exhibited a similar relative volume time evolution for all subjects (Fig. 6 and online Supplemental Material for a video of the lung segmentation of subject 1 through the cardiac cycle). During systole, the segmented lung volume increased, on average, by ∼53 ml, which agrees well with the volume displaced per heartbeat of 60 ml quoted by Cotes et al. (5). Interestingly, 83% of this increase was found in the left lung. The volume contraction of the heart ventricles corresponded to a left lower lung expansion confirmed by the displacement of the pulmonary vasculature in the left lung around the heart (see online Supplemental Video). During the first half of diastole, the opposite effect occurred with a similar rate of volume change as during systole (Fig. 6). The second part of diastole did not show major lung volume changes.

Fig. 6.

Relative left and right lung volume changes and the corresponding time derivatives [rate of volume change over time (dV/dt); inset] after segmentation of a cardiac-gated series of proton images of the lungs of subject 1, acquired during breath holding (20 cardiac phases).

DISCUSSION

In this study, a dramatic effect of CO_f on pulmonary airflow pattern was observed in the lungs of normal subjects with functional ³He MRI. Whereas Sun et al. (24) noticed a ventilatory alternans between the left and right lungs during inspiration in healthy subjects, our data suggest that this observation is the consequence of stronger flow oscillation in the LLL of the lung due to the proximity of the heart, with accompanying weaker oscillations of opposite phase in the rest of the lungs. We believe that Sun et al. (24) did not observe this localized origin, most probably because of their choice of ROI in the left lung. Figure 1A of that study clearly shows that the LUL of the lung was omitted in the signal intensity calculation. However, the dynamic images from the subjects in that study (available as videos in the online Supplemental Material) showed an alternans between the LUL and LLL of the left lung. The increase in time resolution in the dynamic ventilation imaging protocol between Sun et al. (24) and the present studies (from 2.5 to 7.1 frames/s) also better facilitates the visualization of the alternans. Our data are also in agreement with previous reports, suggesting that the phase of CO_f is different in different parts of the lungs and that it is more marked on the left side (30). Interestingly, the phase of the LUL signal seems to change slightly in one subject (Fig. 2A), which suggests that the phase difference between the LLL and LUL of the left lung could depend on the lung inflation level in some cases. The signal oscillations could be observed in all subjects in the LUL but only in five of six subjects in the rest of the lungs. Similarly, the ventilatory alternans and the CO_f were observed in five of seven subjects in the imaging study of Sun et al. (24) and in 75% of patients in a study by West and Hugh-Jones (30) with a gas flow meter, which confirms that the existence and/or severity of CO_f are subject dependent. However, it is worth noting that certain experimental and physiological conditions, such as low flow rates, hyperinflated lungs (13), higher cardiac output per beat, and lower heartbeat frequencies, are more favorable for the observation of cardiogenic oscillations. Subject 5 had a faster heartbeat rate than the other subjects, which could partly explain why no clear signal oscillations could be detected in the rest of that subject's lungs.

In addition to dynamic ventilation imaging, the PCV MRI technique was developed to observe the effect of cardiogenic oscillations on the gas flow in the LMB and RMB, which revealed the existence of a pendelluft effect. The ³He gas velocity and flow values in the LMB and RMB were found to oscillate antagonistically, and reversed backflows were recorded in four of six subjects. The shape of the flow patterns, although having common features, was different for all subjects. Among all subjects, the average velocity in both LMB and RMB during the cardiac cycle was ∼38.7 cm/s. Assuming a Weibel cross-section of the airway with an area of 1.17 cm² for each main bronchus and neglecting the fact that the slice is not fully perpendicular to the bronchi directions (Fig. 1, slice location 1), an average flow of 91 ml/s can be derived. This is roughly consistent with the expected average flow of 85 ml/s deduced from the average time of 11.8 s that was required by the subjects to inhale the 1-liter bag of gas. PCV measurements have some limitations and possible scope for improvements. It would be desirable to perform flow measurement at the entrance of the main lobes and not only through the main bronchi; however, the localization of smaller airways is highly challenging, given the quality and resolution of MRI. In addition, to deduce the exact flow pattern produced by CO_f, the measurement should be repeated during breath holding. However, because the ³He signal is nonrenewable, fresh gas needs to flow constantly through the ROI, which led us to a dynamic experiment during the inspiratory phase. Nevertheless, estimates of maximum CO_f values in the LMB can be obtained from the 2D PCV measurements by inspection of the amplitude of the flow oscillations: ∼45 ml/s for subjects 1 and 5 and 70 ml/s for subject 6. These results are reasonably consistent with previously reported volume flow rates of 42 ml/s (30). The addition of simultaneous ECG recording with MR measurements showed that some features in the CO_f pattern were common to all subjects. Maximum flow rates in the LLL of the lung occurred during systole, whereas the opposite effect happened during diastole, with gas being redirected to the other parts of the lungs.

The origin of CO_f has been a matter of debate for some time, stimulated by contradictory results from different groups (11). Whereas early studies suggested an aspirating effect of the heart during systole, simply explained by a pressure change induced by a volume change of the lung, more recent works claim that CO_f are caused by pulmonary artery pulsatility (23, 29). It is doubtful that pulmonary artery pulsatility could explain our observations of regional phase difference between the LLL and the rest of the lungs. Moreover, the fact that these latter studies are based on pressure and flow measurements at the mouth only and were performed during open-chest conditions raises the question as to whether the same, related, or indeed potentially different phenomena are being observed and compared. An alternative and more satisfactory mechanism is given by Engel (11): the oscillatory motion of the heart, in addition to volume changes, produces deflation and inflation in the surrounding parts of the lung, resulting in a dynamic redistribution of the gas. It is clear from the segmentation of the cardiac-gated proton images that the mechanical action of the heart produces a volume change mainly affecting the LLL of the lungs (Fig. 6 and online Supplemental Video); we believe that this results in pressure changes causing the observed redistribution of gas within the lung. During systole, the blood redistribution from the ventricles to the systemic and pulmonary circulation results in a stretch of the LLL of the lung. Simultaneously, more gas is observed to flow into this part of the lung. During diastole, the heart volume increases at the expense of the surrounding left lobe, where expiratory flows are measured.

It is interesting to compare the proposed mechanism with recent models of pendelluft in the bronchial tree (14). The pendelluft effect has been reported at different airway levels, mainly in pulmonary diseases or under abnormal conditions, and is expected to occur when regions of the lung have different dynamics of regional inflation and deflation (for example, regions with different compliance and/or flow resistance due to lung disease). In this sense, the asymmetric volume change created by the cardiac cycle could lead to regional pressure differences and explain the pendelluft observed at the carina bifurcation, therefore reflecting the modeling predictions by Greenblatt et al. (14). However, the main difference in the observed pendelluft in the present study is that it results from the natural cardiac motion in subjects with healthy lungs.

The resulting oscillatory gas flows constitute an additional mechanism for gas mixing in the lung (12, 13, 15), but quantitative measurements are required to conclude on their significance. A recent study supports that the heart-lung interactions are a vital source of gas mixing (26). In addition, aerosol transport, mixing, and deposition could be strongly affected by CO_f, which is particularly relevant for inhaled therapeutics. Ma and Darquenne (18) and Darquenne et al. (8) proposed the existence of cardiogenic mixing to explain differences between measurement and simulations of aerosol dispersion and for the heterogeneity of particle deposition in microgravity, respectively. In a following study to Darquenne et al. (8), they addressed the effect of CO_f on the deposition and dispersion of 1 μm particles during breath holdings (7). Although gravitational sedimentation is inferred to be the main mechanism, data have suggested that CO_f has a larger effect in the central airways than in the periphery of the lung. We believe that the effect of CO_f has been largely underestimated in the literature, especially in the field of CFD simulations of airflow in the main airways (9, 17, 25), where the influence of the heart has not, to our knowledge, been taken into consideration. Finally, it would be interesting to study how the observed mechanism could contribute to the chaotic mixing of fine particles proposed by Tsuda et al. (27). The classical theory assumes that acinar flow is kinematically reversible due to a low Reynolds number gas flow deep in the lung. However, Tsuda et al. (28) and Butler and Tsuda (2) observed kinematic irreversibility and complex convective stretch and fold patterns in excised rat lungs, supporting the theory that chaotic alveolar flow arising from flow trajectory asynchrony governs aerosol transport and mixing in the lung periphery.

Further work to simulate the influence of CO_f on airflow patterns is required to supplement our findings, but we hope that this study will increase awareness of the effect of cardiac motion on gas flow and distribution within the lungs among the pulmonary community.

GRANTS

Support for this work was funded by European Union Seventh Framework Programme Airway Disease Predicting Outcomes through Patient Specific Computational Modelling (EU FP7 AirPROM), Engineering and Physical Sciences Research Council (EPSRC; #EP/D070252/1), and the National Institute of Health Research (NIHR). This article presents independent research funded by NIHR.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors. The views expressed are those of the authors and not necessarily those of the National Health Service, NIHR, or the Department of Health.

AUTHOR CONTRIBUTIONS

Author contributions: G.J.C., H.M., M.R., and J.M.W. conception and design of research; G.J.C., H.M., N.J.S., and D.C. performed experiments; G.J.C. analyzed data; G.J.C., H.M., and J.M.W. interpreted results of experiments; G.J.C. prepared figures; G.J.C. drafted manuscript; G.J.C., H.M., and J.M.W. edited and revised manuscript; G.J.C., H.M., M.R., N.J.S., D.C., and J.M.W. approved final version of manuscript.