Pulmonary Shunt Is Independent of Decrease in Cardiac Output during Unsupported Spontaneous Breathing in the Pig

OXYGENATION improves when spontaneous breaths are allowed to alternate with or to take place on top of mechanical breaths during mechanical ventilation (MV).^1–3  Whether this improvement is exclusively attributable to spontaneous breathing (SB)–induced reopening of collapsed lung regions has been questioned. Recent studies suggest an SB-induced improvement of ventilation/perfusion matching independent of reopening collapsed areas.^4,5 

During MV, increasing cardiac output (CO) concomitantly augments pulmonary shunt, which is explained by a progressive inhibition of hypoxic pulmonary vasoconstriction (HPV). Raising CO increases mixed-venous oxygen tension (Pvo₂), and this results in higher oxygen tension at the HPV sensor site, attenuating the hypoxic vasoconstrictor response.^6–8  Moreover, a rising CO increases pulmonary artery pressures, which then distend/recruit pulmonary vessels perfusing nonaerated regions.⁹  However, the effect of CO on shunt has not been examined during SB. In a previous study, we did not find any major impact of CO on venous admixture (Q_va/Q_t) during unsupported SB in a porcine lung collapse model.¹⁰  Therefore, we speculated that, in addition to the above-mentioned mechanisms for CO-dependent shunt during MV, the increased airway pressure (P_aw) during MV could counteract HPV by augmenting impedance to blood flow toward ventilated lung regions. However, in that previous study, we did not control for changes in CO.¹⁰  Therefore, the primary aim of the present study was to examine whether pulmonary shunt is independent of CO during SB when CO is intentionally modulated. A secondary aim was to investigate the effect of airway pressures, when increased by continuous airway pressure (CPAP) or MV, on the CO dependency of the pulmonary shunt.

We hypothesized that CO had no major effect on Q_va/Q_t during SB, and that the higher airway pressures during MV and CPAP (spontaneous breathing at airway pressure corresponding to MV) were associated with CO dependence of the pulmonary shunt.

The Uppsala Animal Ethics Committee (Uppsala, Sweden) approved the protocol; the guidelines regarding animal care from the Institute for Laboratory Animal Research^†† were followed. Seven piglets weighing 24.3–27.3 kg were studied.

The piglets were premedicated with 2.2 mg/kg tiletamine plus 6 mg/kg zolazepam (Zoletil; Virbac, Carros, France). After IV induction with 8 mg/kg ketamine and 1 mg/kg morphine, anesthesia was maintained with IV infusion of 20 mg·kg⁻¹·h⁻¹ ketamine and 1 mg·kg⁻¹·h⁻¹ morphine. After a 10-ml/kg IV bolus of dextran 60 (Macrodex; Meda, Solna, Sweden), 10 ml·kg⁻¹·h⁻¹ of Ringer’s acetate was administered IV. In supine position, kept over the whole experiment, an endotracheal tube (Mallinckrodt, Athlone, Ireland) with 9.0-mm internal diameter, cut at 21-cm length, was inserted via a tracheotomy, and connected to the ventilator (Servo-i; Maquet, Solna, Sweden). A central venous line and a pulmonary artery catheter (CritiCath SP5107H; Becton Dickinson, Singapore) were inserted via the right internal jugular vein. For partial occlusion of the venous return, an additional pulmonary artery catheter was placed via a femoral vein cut-down, and floated with an inflated balloon cephalad (usually just beyond 50 cm, approximately at the level of the diaphragm) until blood pressure dropped, indicating decreased venous return. An arterial line was placed via a branch of the subclavian artery, and cystostomy was performed for urine drainage.

Hemodynamic signals were sampled at 2000 Hz and recorded on a computer for offline analysis (BioPac Systems, Santa Barbara, CA). CO was measured with the thermodilution technique (Solar 8000; Marquette, Milwaukee, WI) using three injections of 10 ml of ice-cold saline (injected at random over 4–5 s according to the manufacturer’s instructions^‖); the reported mean value is a result of three individual measurements (usually within ±0.2 l/min). Blood samples were taken for analyses of pH, Pvo₂, arterial oxygen tension, and arterial carbon dioxide tension (Paco₂) (ABL3; Radiometer, Copenhagen, Denmark); mixed-venous and arterial oxygen saturation was measured in a cooximeter calibrated for porcine hemoglobin (OSM3; Radiometer). Venous admixture, calculated with the Berggren equation,¹¹  was used as a proxy for pulmonary shunt; derived hemodynamic variables were calculated according to standard formulae.

All animals were initially breathing spontaneously with zero end-expiratory pressure and no pressure support (i.e., unsupported SB) with a triggering sensitivity of −1 cm H₂O and a fraction of inspired oxygen of 1.0 for 30 min. Tidal volume (V_T) was obtained by integrating the respiratory flow from a Fleisch pneumotachograph (Series 3700; Hans Rudolph, Inc., Shawnee, KS) placed between the endotracheal tube and the Y-piece of the ventilator tubing. At the end of this period, respiratory rate and V_T were averaged over 2 min (i.e., control period) and were then used for setting V_T and respiratory rate during the volume-controlled mechanical ventilation of the animal.

Three ventilatory modes were used to study whether P_aw impacts on the CO–Q_va/Q_t relationship: (1) unsupported SB with low P_aw, (2) MV with increased P_aw but respiratory rate and tidal volume corresponding to those during SB, and (3) spontaneous breathing during CPAP at mean P_aw corresponding to mean P_aw during MV. Eight cmH₂O of positive end-expiratory pressure, but no pressure support, was applied during CPAP settings.

CO was actively modulated by partial balloon occlusion of the venous return in the inferior vena cava.¹²  After 10 min on a particular ventilatory mode (i.e., MV, SB, or CPAP), baseline measurements were obtained and then the venous return was partially occluded by slowly inflating the balloon of the catheter placed in the inferior vena cava until mean arterial blood pressure fell by at least 40%; 5 min later, measurements were repeated. If CO was found to be less than 1 l/min, the occluding balloon was slightly deflated and the procedure was repeated after 5 min. After releasing the balloon, the next setting was delayed until hemodynamics were restored.

The effect of CO on Q_va/Q_t was investigated at two shunt levels. Pure oxygen breathing during the preparation with healthy lungs resulted in minor shunt⁵  (s_min); major shunt (s_maj) was induced by blocking the ventilation of the entire left lung with a 7-French spherical Arndt bronchial blocker¹³  (Cook Medical, Inc., Bloomington, IN), placed and inflated in the left main bronchus under fiberoptic control. Disconnecting the entire left lung from ventilation created a pulmonary shunt that was not recruitable by increased airway pressures. For setting MV with s_maj, V_T and respiratory rate were obtained during SB with s_maj, after the stabilization of arterial oxygen tension, after the occlusion of the left main bronchus. The s_min condition always preceded s_maj, whereas within s_min and s_maj all particular settings (SB, MV, and CPAP) were performed in random order, determined beforehand by Internet randomizer.^#

To suppress spontaneous breathing during MV, titrated doses of IV suxamethonium (0.5–1 mg/kg) were given; no SB/CPAP setting was started until the effect of muscle relaxant had worn off. No purposeful/spontaneous movements, other than respiratory, were observed during SB, justifying the use of muscle relaxation during MV. At the conclusion of the experiment, animals were killed by an overdose of IV potassium under deep anesthesia.

The CO–Q_va/Q_t relationship was the primary interest of the study; the sample size needed to detect a coefficient of determination (R²) ≥ 0.5, based on earlier observations,¹⁰  at α = 0.05 and β = 0.2 (80% power), was estimated to be six animals. As variables obtained by repeated measures are not independent observations, mixed-model ANOVA was used for group comparisons, whereas the association between variables of interest was analyzed by linear mixed-model regression¹⁴  on the raw data. Estimation of models and R² was likelihood-based;^** for hypothesis testing maximum likelihood, whereas for parameter estimates restricted maximum likelihood estimation was used.¹⁵  In mixed-model ANOVA, ventilatory mode (MV, CPAP, and SB), shunt level (s_min, s_maj), and partial occlusion of venous return were regarded as fixed effects, and random intercepts for individual animals were allowed. In mixed-model regressions, the null hypothesis was that the explanatory variable had no fixed effect on the dependent variable, which was tested individually for each combination of ventilator mode and shunt level (six comparisons); beside the fixed effect, random intercepts and slopes for the individual animals were allowed. Backward stepwise multiple regressions and P values for fixed effects were calculated with the maximum likelihood test¹⁶  (increase of −2 × log-likelihood on exclusion of the examined parameter). Details are provided in the Appendix. When multiple comparisons were made, P values were adjusted by the Holm procedure. Statistical significance was assumed for values of (adjusted) P ≤ 0.05. All statistical analyses were performed with R Environment for Statistical Computing (version 2.14; R Foundation for Statistical Computing, Vienna, Austria; with lme4 package version 0.999375–42). Values are presented as mean (95% CI).

Independent of venous return (VR), heart rate was slightly lower during MV than during CPAP and unsupported SB in s_min, but similar in all s_maj settings. At baseline, CO, mean arterial blood pressure, and mean pulmonary artery pressure were similar in all settings; with partial VR occlusion, CO, mean arterial blood pressure, and mean pulmonary artery pressure decreased (table 1).

Anesthesia, supine position, and pure oxygen breathing resulted in a minor pulmonary shunt. At both shunt levels, Q_va/Q_t was lower during unsupported SB and CPAP compared with MV. Q_va/Q_t decreased with partial occlusion of VR during MV and CPAP at both shunt levels, but not during unsupported SB (table 2).

Q_va/Q_t depended on CO during MV (R² = 0.75) and CPAP (R² = 0.73); by contrast, when animals were breathing spontaneously at zero end-expiratory pressure, no CO dependence of Q_va/Q_t could be seen (fig. 1). The Q_va/Q_t–Pvo₂ relationship was similar: Q_va/Q_t increased during MV (R² = 0.61) and CPAP (R² = 0.69) with Pvo₂, but not during unsupported SB (fig. 2).

Stepwise backward selection showed a significant effect of CO on Q_va/Q_t during MVs_min but not during CPAPs_min, when the effect of Pvo₂ had already been taken into account, whereas when the effect of CO was already included, Pvo₂ no longer had a significant effect on Q_va/Q_t.

The pattern with s_maj was the same as with s_min: during MV and CPAP, Q_va/Q_t depended on CO (MV, R² = 0.76; CPAP, R² = 0.79) (fig. 1) and Pvo₂ (MV, R² = 0.77; CPAP, R² = 0.86) (fig. 2); whereas during unsupported SB, Q_va/Q_t was not dependent on either CO or Pvo₂.

Stepwise backward selection showed a significant effect of CO on Q_va/Q_t during CPAPs_maj but not during MVs_maj, when the effect of Pvo₂ had already been taken into account. However, when the effect of CO was included, Pvo₂ no longer had a significant effect on Q_va/Q_t.

V_T, approximately 6 ml/kg, was similar in all settings. Respiratory rate, and subsequently minute ventilation, was lower during CPAP with s_min (P < 0.001) but not with s_maj. Mean airway pressures were comparable in MV and CPAP, both with s_min and s_maj, but lower during unsupported SB (P < 0.001 for both s_min and s_maj). Partial VR occlusion did not impact on mean airway pressures.

Paco₂ was similar in all settings and unchanged during partial VR occlusion in all groups except with CPAP, where Paco₂ decreased slightly (1 kPa [0.3–1.8]; P = 0.038) during partial VR occlusion. Arterial pH was at a similar level in all settings, as animals compensated for any Paco₂ increase (table 2).

Pvo₂ was similar in all settings at baseline and lower during partial VR occlusion (table 2). Pvo₂ paralleled CO in all groups (table 3). Oxygen consumption was independent of the ventilatory mode and not altered by partial VR occlusion; however, it was significantly, by approximately 35%, higher during s_maj (table 2).

We found, in accordance with a previous study from our group, that during unsupported SB, pulmonary shunt was not dependent on either CO or Pvo₂. In addition, we could confirm the well-known observation that pulmonary shunt does depend on cardiac output and Pvo₂ during MV.^7,8,17  Interestingly, CPAP was also accompanied by CO- and Pvo₂-dependent pulmonary shunt.

The CO dependency of pulmonary shunting during MV is a consistent phenomenon seen in different species,^17,18  including humans,¹⁹  as well healthy¹²  as focally²⁰  or diffusely¹⁷  diseased lungs, and irrespective of whether CO is modulated pharmacologically,^17,21  mechanically,¹⁷  or by bleeding/volume repletion.¹²  However, the responsible mechanisms have not been fully elucidated. Increased blood flow, increased Pvo₂,^22,23  increased mean pulmonary pressure, and decreased pulmonary vascular resistance¹⁷  are all associated with increased shunt. Sandoval et al. propose that Pvo₂ mediates the CO-dependent shunting,²²  and this raises the possibility that HPV would act as the link between CO and shunt, as HPV is attenuated not only by increased Pvo₂ but also by increased pulmonary artery pressure.⁹  In our data, Pvo₂ seemed not to be able to explain all the effects of CO on Q_va/Q_t; however, this study was not designed to answer such a question. In contrast, during SB, the pulmonary shunt was lower, and the pulmonary vascular bed accommodated an increased CO without a concurrent increase of the pulmonary shunt.

In all settings, and irrespective of the magnitude of the shunt, lower CO resulted in lower PvO₂ (as oxygen consumption was maintained), and, as expected,^22,23  lower Pvo₂ was associated with lower Q_va/Q_t during MV and CPAP. However, despite an equally decreased Pvo₂, lower Pvo₂ was not associated with altered Q_va/Q_t in unsupported SB settings, where Q_va/Q_t was already lower than during MV. This finding challenges the concept that during unsupported SB the main mechanism of pulmonary blood flow redistribution is HPV; as in that case, the HPV should have been less effective at higher Pvo₂.⁶  The Pvo₂-dependent HPV was either less important for the redistribution of pulmonary blood flow during SB (irrespective of the magnitude of the shunt) or a complementary, yet unidentified mechanism redistributing pulmonary blood flow independent of HPV was active during SB. The negative intrapleural pressure generated during SB might have a role in pulmonary blood flow redistribution, as negative-pressure ventilation has been recently shown to increase oxygenation and decrease pulmonary shunt compared with MV in surfactant-depleted rabbits²⁴  and in humans with acute respiratory distress syndrome.²⁵ 

The beneficial effect of SB on pulmonary blood flow redistribution was lost during CPAP, where Q_va/Q_t was found to be CO and Pvo₂ dependent. Increased mean P_aw was the common denominator of MV and CPAP compared with SB. Increased P_aw is transferred along the airways toward alveoli, augmenting the impedance of perialveolar vessels.²⁶  With the impedance of those perialveolar vessels of ventilated areas increased, HPV might not be able to effectively divert pulmonary blood flow away from atelectatic lung areas.

SB can promote alveolar recruitment,²  and affect ventilation distribution.²⁷  We cannot categorically exclude that regional ventilation, rather than regional perfusion, was modulated by the ventilatory mode. This is, however, rather unlikely, as (1) in our earlier studies with almost 40% of the lung tissue atelectatic we found no substantial recruitment of lung collapse within 30 min, during neither SB nor MV⁵; and (2) in supine pigs, during mechanical ventilation at a fraction of inspired oxygen of 1.0 and with CO comparable to that seen with unrestricted venous return in the present study, the multiple inert gas elimination technique shows complete absence of areas with a low ventilation/perfusion ratio (0.001–0.1) during one-lung ventilation,²⁸  whose improvement in regional ventilation could decrease pulmonary shunt. Furthermore, as each ventilatory mode served as its own control during CO modulation, lung recruitment could not explain our observations.

There are few experimental or clinical data on CO dependency of the pulmonary shunt during SB; in awake near-term women, no significant correlation between CO and Q_va/Q_t was seen.²⁹  Arterial oxygen tension decreases during strenuous exercise; opening of preformed anatomical shunts is observed in dogs³⁰  and humans.³¹  Vogiatzis et al., however, raise doubts³²  concerning the role of pulmonary shunting in this particular scenario: during strenuous exercise at a fraction of inspired oxygen of 1.0, CO increases up to 28 l/min but Q_va/Q_t is found to be only 0.5% in humans, making a CO dependency of pulmonary shunt under this particular condition rather unlikely.

First, calculated venous admixture was used as a proxy of pulmonary shunt, being less accurate than shunt determined by the multiple inert gas elimination technique. However, as a fraction of inspired oxygen of 1.0 was used throughout to avoid hypoxia, especially in s_maj settings, Q_va/Q_t should reasonably agree with “true” shunt values.^32,33  Second, our conclusions are valid only within the particular range of variables of the present study. Nothing can be concluded for the very high CO range,³²  whereas for the very low CO range we could not exclude that even for SB there is a CO threshold, at approximately 2.5 l/min, below which Q_va/Q_t might be CO dependent (fig. 1). Furthermore, we do not know whether a P_aw threshold exists above which Q_va/Q_t starts to depend on CO.

Third, both independent and dependent variables showed marked individual variations, inherent in living organisms, which makes the exploration of their interdependency difficult. This study was designed (powered) to evaluate the CO–Q_va/Q_t relationship. Correlations between other variables are solely aimed at generating hypotheses for further investigation; it may well be that those other correlations that are not statistically significant still reflect dependency. However, such could be judged only on the basis of a larger study. Fourth, each regression was calculated from seven sets of two data points (however, not on seven regressions of two data points). We interpreted each correlation estimated from a regression model to indicate that lower CO results in a lower shunt level. However, we did not interpret the actual magnitude of the estimated correlations to indicate a strong or weak correlation, as this is of limited meaning to our question. Fifth, piglets have a strong HPV compared with humans and many other species.¹⁸  That Q_va/Q_t did not change with Pvo₂ during unsupported SB does, however, hardly support HPV as a major mechanism of redistributing pulmonary blood flow in the unsupported SB settings of the present study. Sixth, a number of factors could have influenced HPV in this porcine model. Although morphine³⁴  and benzodiazepines³⁵  (zolazepam) do not affect HPV, ketamine³⁶  (possibly even tiletamine, also an N-methyl-d-aspartate receptor antagonist) and dextran³⁷  might have somewhat attenuated HPV. Respiratory acidosis, seen at both shunt levels, could theoretically also affect HPV; however this still seems to be an unsettled issue. In isolated lungs, hypercapnic acidosis improves HPV, but only after 3 h,³⁸  a considerably longer time than our experiments lasted. In intact animals, hypercapnic acidosis either increases pulmonary vascular resistance without affecting the hypoxic response,³⁹  or attenuates HPV through complex mechanisms.⁴⁰  In any case, as (1) the animals served as their own controls, (2) drug dosages were similar and unchanged over time, (3) CO was similar in different ventilatory modes, and (4) there were no significant differences in Paco₂ between ventilatory modes, it is unlikely that any of the above-listed factors introduced bias into the present study. Finally, implications for the clinical contexts would be premature, when considering differences in physiology of this porcine model and humans in the above-discussed aspects. Furthermore, one-lung ventilation, in the operating room, and major pulmonary shunt induced by acute lung injury or adult respiratory distress syndrome, in the intensive care unit, comprise many more physiologic and pathologic processes than just lung collapse and ventilation/perfusion mismatch.

In contrast to MV and CPAP (at similar mean airway pressure as with MV), pulmonary shunting did not substantially depend on CO during unsupported SB at zero end-expiratory pressure. The CO independence of shunt during unsupported SB seemed to be related to airway pressures, as shunt was CO dependent once airway pressures were raised with CPAP to a level comparable to MV. Higher Pvo₂ was not associated with increased pulmonary shunt during unsupported SB, not even with major shunt, challenging the role of HPV as the main mechanism of pulmonary blood flow redistribution in this setting.

The null hypothesis, that the explanatory variable has no fixed effect on the dependent variable, was individually tested for each combination of ventilator mode (MV, CPAP, and SB) and shunt level (s_min and s_maj), resulting in six comparisons. P values for fixed effects were calculated with the maximum likelihood test as described below.¹⁶  Variables are given in table 4. All statistical analyses were performed with R Environment for Statistical Computing (version 2.14; R Foundation for Statistical Computing, Vienna, Austria; with lme4 package version 0.999375–42).

• 1. Loaded model: dependent variable ~ fixed effect (explanatory variable) + random slope (explanatory variable) + random intercepts

• 2. No random slope model: dependent variable ~ fixed effect (explanatory variable) + random intercepts

Compare to loaded (1) model by maximum likelihood test.

• 3. If random slopes are significant: dependent variable ~ random slope (explanatory variable) + random intercepts

Compare to loaded (1) model by maximum likelihood test.

• 4. If random slopes are not significant: dependent variable ~ random intercepts

Compare to no random slope (2) model by maximum likelihood test.

Stepwise backward selections

• 1. Loaded model: Q_va/Q_t ~ CO_{fixed effect} + Pvo_{2,fixed effect} + CO_{random effect} + Pvo_{2,random effect} + random intercepts

• 2. No CO_{random effect} model: Q_va/Q_t ~ CO_{fixed effect} + Pvo_{2,fixed effect} + Pvo_{2,random effect} + random intercepts

Compare to loaded (1) model by maximum likelihood test: if P < 0.05 keep CO_{random effect} as significant random effect.

• 3. No Pvo_{2,random effect} model: Q_va/Q_t ~ CO_{fixed effect} + Pvo_{2,fixed effect} + CO_{random effect} + random intercepts

Compare to loaded (1) model by maximum likelihood test: if P < 0.05 keep Pvo_{2,random effect} as significant random effect.

• 4. Final loaded model: Q_va/Q_t ~ CO_{fixed effect} + Pvo_{2,fixed effect} + significant random effect(s) + random intercepts

• 5. No CO_{fixed effect}: Q_va/Q_t ~ Pvo_{2,fixed effect} + significant random effect(s) + random intercepts

Compare to final loaded model (4) by maximum likelihood test: if P < 0.05, CO_{fixed effect} is significant, when the fixed effect of Pvo₂ is already included.

• 6. No Pvo_{2,fixed effect}: Q_va/Q_t ~ CO_{fixed effect} + significant random effect(s) + random intercepts

Compare to final loaded model (4) by maximum likelihood test: if P < 0.05, Pvo_{2,fixed effect} is significant, when the fixed effect of CO is already included.