One-hit Models of Ventilator-induced Lung Injury: Benign Inflammation versus Inflammation as a By-product

IN acute respiratory distress syndrome (ARDS), ventilation strategies affect outcome.^1,2  These findings have raised the question of how mechanical ventilation injures the lungs, major possibilities being physical tissue injury and the biotrauma hypothesis. As a result, in recent years, ventilator-induced lung injury (VILI) has been the most frequently used model in experimental ARDS studies.³ 

How quickly mechanical ventilation can injure the lungs was already demonstrated 40 yr ago when Webb and Tierney⁴  showed that ventilation of rats without a positive end-expiratory pressure (PEEP) and with 45 cm H₂O inflation pressure destroys lungs within 20 min. Clinically, physical injury can be diagnosed as extraalveolar air, termed barotrauma.⁵  Retrospective studies in patients suggested a safety threshold of about 35 cm H₂O plateau pressure (p_plat), below which barotrauma is rarely observed.⁶ 

In the National Institutes of Health, National Heart, Lung, and Blood Institute (NIH-NHLBI) ARDS Network study, where barotrauma was similar in both groups, the best correlation with survival was observed for proinflammatory mediators.^2,7,8  This has lent credibility to the biotrauma hypothesis that may be posited in a weak and in a strong form: in its weak form, it states that ventilator-induced inflammation contributes to the mortality of ARDS patients in conjunction with other factors (hits); in its strong form, it states that the inflammation induced by mechanical ventilation may suffice to injure the lungs much stronger than explained by the mechanical forces alone. The strong form of the biotrauma hypothesis has stimulated the so-called one-hit models—with ventilation as the only hit—to examine its mechanisms.⁹  It appears that such studies may be separated into two groups: (1) One-hit models that lead to a catastrophic breakdown of the lung that is sudden, strong, and final^4,10–12  and that is best explained by self-enforcing mechanical failure. In such studies, PEEP is typically absent (or at least very low) and p_plat is above 25 cm H₂O at the start and above 40 cm H₂O at the end of the experiment. (2) One-hit models with p_plat usually well below 25 cm H₂O where inflammation is present, but where severe lung injury and hypoxemia are absent.^13–15 

Because clinically, ARDS is graded by the P/F ratio,¹⁶  in our opinion, a biotrauma model should show hypoxemia (Pao₂/Fio₂ ratio less than or equal to 300 mmHg) that is caused by ventilation-induced inflammation. Therefore, in the current study, it was our aim to develop a one-hit model of ARDS in mice that is (1) characterized by ARDS-like hypoxemia without a catastrophic breakdown and (2) dependent on inflammation.

To this end, we ventilated mice for up to 7 h with a PEEP of 2 cm H₂O and p_plat of 10, 24, 27, or 30 cm H₂O. When we found that all pressures above 24 cm H₂O caused a catastrophic lung failure with ventilation pressures rapidly precipitating after 3 to 6 h, we started a second series where we utilized a pressure valve to limit airway pressure to 34 cm H₂O—i.e., below the safety threshold described by Boussarsar et al.⁶  In order to address a possible contribution of inflammation in the progression of the disease, some animals were pretreated with the antiinflammatory steroid dexamethasone. Neither the pressure release strategy nor the steroid application prevented the catastrophic lung failure of animals ventilated with p_plat above 24 cm H₂O.

A complete description is available in Supplemental Digital Content 1 (https://links.lww.com/ALN/B392).

Female C57BL/6 N mice (aged 8 to 12 weeks, weighing 20 to 25 g) were kept under standard conditions. All experimental procedures were approved by the Landesamt für Natur, Umwelt und Verbraucherschutz Nordrhein-Westfalen (Recklinghausen, Germany; permission number: AZ 84-02.04.2013.A131).

Surgery was performed as described previously.¹⁴  In short, mice were initially anaesthetized by intraperitoneal bolus injection of pentobarbital sodium (75 mg/kg) plus fentanyl (40 μg/kg). Anesthesia was maintained with pentobarbital sodium (20 mg/kg) via an intraperitoneal catheter every 30 to 60 min and fentanyl (20 μg/kg) every 3 h. A catheter was inserted into the carotid artery to permanently monitor blood pressure (AP_mean) and to supply saline (300 μl/h) to prevent hypovolemia and thrombus formation. Heart rate (HR) was calculated online from a three-lead electrocardiogram, and pulse oximetry was performed with a thigh clip.

Mice were ventilated for 7 h via a tracheal catheter with a mechanical ventilator (MidiVent Type 849; Hugo Sachs Elektronik - Harvard Apparatus GmbH, Germany) equipped with a flow-pressure sensor (Type 382 Mouse; Hugo Sachs Elektronik - Harvard Apparatus GmbH). Dynamic pulmonary compliance (C_dyn) and resistance were calculated online by the Pulmodyn software v 1.1 (Hugo Sachs Elektronik - Harvard Apparatus GmbH). ΔC_dyn (C_dyn end of ventilation − C_dyn start of ventilation) was calculated at the end of the experiments. Control animals were ventilated with a plateau airway pressure (p_plat) of 10 cm H₂O at 180 breaths/min with recruitment maneuvers every 20 min to prevent atelectasis.¹⁴  Groups with p_plat of 24 cm H₂O or more were ventilated at a frequency of 90 breaths/min. End-tidal carbon dioxide (ETco₂) waveform was monitored with a microcapnograph (Type 340; Hugo Sachs Elektronik - Harvard Apparatus GmbH). A PEEP of 2 cm H₂O was applied, the fraction of inspired oxygen (Fio₂) was set to 0.3 and the inspiration/expiration ratio was 1:1 with a sinus-shaped waveform in all experiments.

At the end of experiment, mice were euthanized by exsanguinations via the carotid artery. Blood samples were analyzed for arterial oxygen saturation, arterial oxygen tension (Pao₂), Paco₂, and pH; the Horovitz ratio was calculated as Pao₂/Fio₂.

We performed and analyzed two different series of experiments separately. In the first series (series_1), the following p_plats were used in the ventilation protocol: (10 cm H₂O [p10], 24 cm H₂O [p24], 27 cm H₂O [p27], and 30 cm H₂O [p30]).

In the second series of experiments (series_2), a respiratory pressure monitor and limiter (I; IPML type 870/01 for mouse; Hugo Sachs Elektronik - Harvard Apparatus GmbH) with pressure release above 34 cm H₂O (valve open time: 150 ms) was used to limit the maximum pressure in all animals. This value was chosen because clinical data suggest that pressures above 35 cm H₂O favor pneumothorax formation.⁶  The following p_plats were used during ventilation with the pressure release valve: 24 cm H₂O (p24 I), 27 cm H₂O (p27 I), and 30 cm H₂O (p30 I).

The steroid dexamethasone was used to assess the inflammatory part in our model. Depending on group assignment, animals in the second series randomly received dexamethasone (1 mg/kg iv) directly after the start of mechanical ventilation. The following additional groups were examined: 24 cm H₂O (p24 I D), 27 cm H₂O (p27 I D), and 30 cm H₂O (p30 I D).

The right superior lobe was fixed in 4% formalin for histopathology, the right middle and inferior lobes were snap frozen in liquid nitrogen and stored at −80°C for real-time quantitative polymerase chain reaction (qPCR) measurements. The wet/dry ratio was obtained from the right postcaval lobe after weight constancy at 42°C. Bronchoalveolar lavage of the left lung was performed by instilling twice 200 μl ice cold saline via a tracheal catheter. From each lung, about 350 μl bronchoalveolar lavage fluid (BALF) was recovered, centrifuged, and the supernatant was stored at −80°C until quantification of cytokines. All samples were analyzed in a blinded fashion. Because of limited amounts of material in a few cases or as a consequence of predefined internal quality controls, we could not perform all analyses on every lung.

BALF cytokine levels of chemokine (C-X-C motif) ligand 1 (CXCL1, KC), chemokine (C-X-C motif) ligand 2 (CXCL2, MIP-2), chemokine (C-X-C motif) ligand 10 (CXCL10, IP-10), interleukin-6 (IL-6), and tumor necrosis factor (TNF) were quantified with enzyme-linked immunosorbent assays (R&D Systems GmbH, Germany) according to the manufacturer’s protocols.

Frozen lung tissues were grinded in liquid nitrogen. Total RNA was isolated from 20 mg lung powder with RNeasy^® Mini Kit (QIAGEN GmbH, Germany) automated on a QIAcube robot (QIAGEN GmbH) with additional DNase digestion. RNA was quantified in buffered 10 mM TRIS-HCl (pH 7.5) using a spectrophotometer (NanoDrop 1000; Thermo Fisher Scientific Inc., USA). The details of PCR reaction are described in the Supplemental Digital Content 1 (https://links.lww.com/ALN/B392). Primer pairs used for the detection of chemokine (C-X-C motif) ligand 1 (Cxcl1), (C-X-C motif) ligand 10 (Cxcl10), dipeptidylpeptidase 4 (Dpp4), Il6, myeloperoxidase (Mpo), neuropeptide Y (Npy), TATA box binding protein (Tbp), and Tnf mRNA levels are listed in the Supplemental Digital Content 2 (https://links.lww.com/ALN/B393).

Quantification of real-time qPCR was performed with crossing point values, acquired via the second-derivative maximum method. Advanced relative quantification was performed with the LightCycler 480 software v 1.5 (Roche-Diagnostics GmbH, Germany) and efficiency-corrected by inrun standard curves from all samples.¹⁷  Interrun calibrators from unventilated control animals were used to avoid interrun variations. Data were normalized to Tbp, which was validated as a very stably expressed reference gene out of 10 candidate genes, as determined by geNorm (average M ≤ 0.2).¹⁸  Gene expression is depicted as fold induction relative to unventilated controls. Real-time qPCR quality control was performed by inrun controls, melting curve profiles using the LightCycler 480 software (Roche-Diagnostics GmbH), and product separation in agarose gels.

The right superior lobe was fixed with 4% formalin and embedded in paraffin for sectioning. Hematoxylin and eosin staining was performed with 3-mm-thick sections. Histopathology was evaluated in a blinded manner as reported previously.¹⁴  In short, a scoring system based on four criteria was used: neutrophils in the alveolar or interstitial space, alveolar septal thickening, alveolar congestion, and formation of hyaline membranes. Each criterion scored one point, if present, resulting in a range of 0 to 4 points.

Based on the data from Protti et al.¹⁰  and on own preliminary data, the experiments were planned with a statistical power of 80% and an alpha error of 0.05 (corrected for multiple comparisons) in order to detect differences in the Pao₂/Fio₂ ratio of greater than 80 with an SD of 10 (JMP 10; SAS Institute Inc., USA). The data were analyzed using SAS® software v 9.4 (SAS Institute Inc.) using two-sided tests. Survival analysis was performed with log-rank tests (Proc Lifetest; SAS® software v 9.4), and P values were corrected by the Bonferroni–Holm procedure. Univariate tests were carried out using general linear mixed model analysis (Proc Glimmix; SAS® software v 9.4) assuming a log-normal distribution for all cytokine data and a normal distribution for all other parameters; residual plots were used as diagnostics. In case of heteroscedasticity (according to the covtest statement), the df were adjusted by the Kenward–Rogers method. P values were always adjusted by the simulated-Shaffer procedure.¹⁹ P < 0.05 was considered significant. Data were plotted with GraphPad Prism® software v 6 (GraphPad Software Inc., USA).

In the first series, mice were ventilated with 10, 24, 27, or 30 cm H₂O. All animals were ventilated for 7 h unless they died before. Death was defined as a drop in AP_mean < 80% of the initial value and is indicated (†). Since in the group p10 p_plat never exceeded 34 cm H₂O, these data are shown in the graphs of both experimental series (see below).

In series_1, animals ventilated with 10 cm H₂O or with 24 cm H₂O for 7 h showed stable p_plat (fig. 1A), HR, AP_mean, and petCO₂ (figs. S1–S3 in Supplemental Digital Content 3, https://links.lww.com/ALN/B394, Supplemental Digital Content 4, https://links.lww.com/ALN/B395, and Supplemental Digital Content 5, https://links.lww.com/ALN/B396). In contrast, most animals ventilated with 27 cm H₂O and all animals ventilated with 30 cm H₂O did not survive the 7-h period. Animals ventilated with 27 cm H₂O showed a mean survival time of 389 min and 80% mortality; during ventilation with 30 cm H₂O, the mean survival time was 282 min and the mortality was 100% (fig. 1B; table S2 in Supplemental Digital Content 6, https://links.lww.com/ALN/B397, showing survival proportions).

Remarkably, before animals died, they showed a precipitating increase in p_plat (fig. 1A) and petCO₂ (fig. S3A, Supplemental Digital Content 5, https://links.lww.com/ALN/B396), as well as a matching decrease in C_dyn (fig. 2, A and B). This behavior is best illustrated if the p_plat data are normalized to the time point of the maximum gain in p_plat because then all animals follow a common kinetic (fig. S4, A and B in Supplemental Digital Content 7, https://links.lww.com/ALN/B398) such that once p_plat had increased by more than 18%, death occurred within about 30 min. The slope of that curve (first derivative, fig. S4, C, D in Supplemental Digital Content 7, https://links.lww.com/ALN/B398) shows that airway pressure increased continuously, but moderately, until shortly before death, when a dramatic increase occurred. This observation matches the dynamic pulmonary compliance data: the C_dyn curve showed a decrease by more than 20% about 30 min before death (fig. S5, A and B in Supplemental Digital Content 8, https://links.lww.com/ALN/B399). The slope of that curve shows that the pulmonary compliance was decreasing slowly, until suddenly a strong decrease started, which peaked shortly before death (fig. S5, C and D in Supplemental Digital Content 8, https://links.lww.com/ALN/B399).

In the two groups ventilated with p_plat ≥ 27 cm H₂O, the wet/dry ratio was increased (above 7), the pulmonary compliance was decreased (ΔC_dyn< −9.7), and the Horovitz ratio was less than 200 mmHg. In contrast to these observations, all parameters remained normal in the two groups ventilated with p_plat ≤ 24 cm H₂O (fig. 2).

Animals ventilated with p_plat ≥ 27 cm H₂O showed increased expression of proinflammatory mediators. In comparison to the 24 cm H₂O group, increased transcript levels of Il6 (19-fold), Cxcl1 (17-fold), Tnf (5-fold), and Cxcl10 (4-fold) were detected for the p27 group. Expression ratios of those genes after ventilation with 30 cm H₂O were slightly less strongly increased, in comparison to those with 24 cm H₂O: Il6 (12-fold), Cxcl1 (11-fold), Tnf (4-fold), and Cxcl10 (3-fold) (fig. 3A–D). The cytokine levels in the BALF matched the mRNA patterns. After ventilation with 27 cm H₂O or 30 cm H₂O, BALF protein levels were enhanced in comparison to the p24 group: IL-6 (3- to 7-fold), CXCL1 (4- to 6-fold), TNF (6- to 8-fold), and CXCL10 (5- to 6-fold) (fig. 3E–H).

Histopathologic scoring demonstrated significant alterations in all lungs ventilated with p_plat ≥ 27 cm H₂O (fig. 4). In most lungs ventilated with high pressures, neutrophils in combination with alveolar septal thickening were observed, resulting in a score of 1.8 (p27 group) and 1.6 (p30 group). Ventilation with 10 cm H₂O or with 24 cm H₂O let to a largely normal histopathologic score (0.1 for both groups).

The steep increase in p_plat shortly before the death of the animals suggested a mechanical failure of the lung with a possible positive (catastrophic) feedback loop between p_plat and lung injury. To further examine this hypothesis, in series_2, lungs were ventilated with a pressure release valve (IPML, I) that did not permit p_plat > 34 cm H₂O (fig. 5A). In addition, to explore a possible pathogenic role of inflammation (i.e., biotrauma), some animals were pretreated with dexamethasone.

Using pressure release ventilation, the mean survival time was 372 min in the p27 I group with a mortality of 100% (fig. 5B). In the p30 I group, mean survival time was 366 min (as opposed to 282 min in the p30 group without IPML) and the mortality was 50% (table S3, in Supplemental Digital Content 9, https://links.lww.com/ALN/B400). In these groups, HR and AP_mean (figs. S1–S2B in Supplemental Digital Content 3, https://links.lww.com/ALN/B394 and Supplemental Digital Content 4, https://links.lww.com/ALN/B395) remained stable until lung failure occurred. Lung failure is reflected by an accelerating increase in airway pressure (fig. 5A) and petCO₂ (fig. S3B, Supplemental Digital Content 5, https://links.lww.com/ALN/B396) and by a precipitating decrease in compliance (fig. 6, A and B). In the p27 I and the p30 I groups, the pulmonary compliance was decreased (ΔC_dyn< −10.1; fig. 6B), the wet/dry ratio was increased (fig. 6C), and the oxygenation index was less than 200 mmHg (fig. 6D).

All physiologic parameters remained normal in the two groups with p_plat ≤ 24 cm H₂O: p_plat (fig. 5A), C_dyn, ΔC_dyn, wet/dry ratio, and Pao₂/Fio₂ (fig. 6A–D), as well as HR, AP_mean, petCO₂ (fig. S1–S3B, Supplemental Digital Content 3, https://links.lww.com/ALN/B394, Supplemental Digital Content 4, https://links.lww.com/ALN/B395, and Supplemental Digital Content 5, https://links.lww.com/ALN/B396).

In comparison to p24 I, pressure release ventilation with p_plat 27 or 30 cm H₂O led to strongly increased transcript levels of: Il6 (16- to 18-fold), Cxcl1 (13- to 19-fold), Tnf (6-fold), and Cxcl10 (4- to 8-fold; fig. 7). Transcript levels in mice ventilated with 24 cm H₂O were only a little higher than in those ventilated with 10 cm H₂O. Compared to the p24 I group, BALF levels were enhanced for IL-6 (4- to 10-fold), CXCL1 (2- to 5-fold), TNF (5- to 7-fold), and CXCL10 (2- to 4-fold) in the p27 I and p30 I groups (fig. 8). IL-6 and CXCL1 levels in mice ventilated with 24 cm H₂O were similar to those ventilated with 10 cm H₂O; TNF and CXLC10 levels were somewhat higher (fig. 8). As in series_1, histopathologic scoring revealed significant alterations in all lungs ventilated with p_plat ≥ 27 cm H₂O despite the pressure release valve (fig. 9). Neutrophils in combination with alveolar septal thickening were observed in most lungs, resulting in scores of 1.4 (p27 I group) and 1.5 (p30 I group). Ventilation with p_plat ≤ 24 cm H₂O did not affect the histopathologic score.

Dexamethasone pretreatment diminished mRNA transcription of Il6, Cxcl1, Tnf, and Cxcl10 (fig. 7) and decreased the BALF protein levels of IL-6 and CXCL1 (fig. 8) in all animals ventilated with p_plat ≥ 27 cm H₂O. TNF BALF content was slightly reduced by dexamethasone in the p27 I D group, whereas the TNF and CXCL10 levels in the p30 I D group were not affected compared to those in the respective groups without dexamethasone.

Although dexamethasone pretreatment reduced inflammatory parameters after ventilation with p_plat ≥ 27 cm H₂O (figs. 7 and 8), survival proportions were not affected (fig. 5B). In line with these data, dexamethasone pretreatment did improve neither lung wet/dry ratio (fig. 6B) nor ΔC_dyn (fig. 6C). Further, dexamethasone pretreatment did not improve gas exchange, in mice ventilated with neither 27 cm H₂O nor 30 cm H₂O (fig. 6D).

Histopathologic examination (fig. 9) revealed no effect of dexamethasone pretreatment on the ventilation-induced structural alterations. In most lungs, neutrophils were observed in addition to alveolar septal thickening.

We observed that—even with a PEEP as low as 2 cm H₂O—p_plat of 24 cm H₂O were well tolerated for at least 7 h. However, slightly above this pressure (greater than or equal to 27 cm H₂O), we observed a catastrophic type of lung failure that rapidly developed once a critical threshold value was reached. We believe that this behavior is best explained in terms of material fatigue of stress bearing elements. Our findings further suggest that under these conditions, inflammation is the result rather than the cause of lung injury and as such it may accelerate an inevitable organ failure.

We have recently described a setup for a mouse intensive care unit providing anesthesia, ventilation, volume support, and thermoregulation that yielded stable lung mechanics and hemodynamics for 6 h.¹⁴  Here, we extend those findings and show that mice tolerate ventilation with p_plat = 24 cm H₂O for at least 7 h. Within this period, lung functions, oxygenation, and cardiovascular functions all remained completely stable despite the presence of inflammation. Similar observations have been made in human patients ventilated for hours or even a week.^20,21  Notably, the intensity of inflammation was similar to that at p_plat = 10 cm H₂O, being higher than inflammation in naive mice,¹⁴  but considerably lower than that at p_plat ≥ 27 cm H₂O. These findings do further support our notion that pulmonary inflammation occurs readily but that this does not necessarily compromise lung functions.²²  We therefore believe that the current definition of experimental ARDS,³  which allows us to declare ARDS in perfectly functional lungs only because of inflammation, lacks discriminatory power.

Ventilation with p_plat ≥ 27cmH₂O resulted in a type of lung failure that always followed the same characteristic pattern, even though—depending on the individual animal—the actual time of onset differed. These results are comparable to those described by Protti et al.¹⁰  who observed in pigs that all animals ventilated with 22 ml/kg (initial mean p_plat = 16 cm H₂O) survived ventilation for 54 h, whereas most animals ventilated with 38 ml/kg (initial mean p_plat ≥ 26 cm H₂O) died between 8 and 50 h. Unfortunately, Protti et al.¹⁰  did not report the time course of lung injury development although they state that lung injury progressed quickly once it had started.

To visualize the pattern of lung failure, we shifted the time scales of each experiment, so that the time points of the maximum gain (slope) in ventilation pressure were superimposed on each other (Supplemental Digital Content 7, https://links.lww.com/ALN/B398, and Supplemental Digital Content 8, https://links.lww.com/ALN/B399, showing the time-shifted data for p_plat and C_dyn). This analysis revealed that once p_plat had increased by more than 18% or compliance had decreased by more than 20%, death inevitably occurred within 30 min.

This type of lung failure was characterized by two properties: (1) it occurred at random times and (2) once it occurred, it developed in a rapid, self-enforcing way. Such a behavior is typical for fatigue failure, in the sense used in structural material mechanics, such as the failure of brittle polymers.^23,24  Here, fatigue describes the weakening of a material that is subjected to cyclic loading and unloading with stresses below the material’s ultimate tensile stress limit. In many cases, material fatigue is a function of the number of cycles,²⁵  which may be relevant here, because not only controls were ventilated with twice the number of cycles (180 breaths/min) than the other groups (90 breaths/min), but also this may help to understand species differences (see paragraph after next). One brittle stress-bearing element in the lung is collagen, a material that can develop propagating cracks.^23,24  Unfortunately, the material fatigue of collagen fibers has not yet been explored, but it is well known that collagen fibers may rupture upon minimal strain (less than 5%).^23,26  Another potential stress-bearing element are the basement membranes, whose destruction would explain the severe hemorrhage in the current and several previous studies.²⁷  Clearly, the nature of the stress-bearing elements requires further study.

At p_plat ≥ 27 cm H₂O, murine lungs approach their total lung capacity,²⁸  which means that the stress-bearing elements experience cyclic load under these conditions. This interpretation is also supported by the results seen in series_2, where the pressure release valve limited p_plat to 34 cm H₂O, which is much lower than in many previous studies where lung failure occurred in the presence of p_plat > 40 cm H₂O.^4,10  The final consequences have been well described: rupture of the alveolar plasma membrane and of the thin pulmonary capillary wall with consequential hemorrhage.^{4,9,26,29–31} 

Certainly, p_plat alone is not sufficient to explain the threshold of mechanical lung failure, and the underlying forces need to be described by a more comprehensive approach such as the energy/power concept developed by Protti et al.³²  and Gattinoni et al.³³  According to their analysis, the strain (total lung capacity to functional residual capacity [FRC] ratio) is critical for the time to generate lung injury³⁴  and the relevant total lung capacity to FRC ratios of 2 to 3 are comparable across mammalian species. Differences between species are explained by lung dimensions and by differences in specific lung elastance (the bulk modulus calculated as FRC/compliance), which is about 7 to 12 cm H₂O in mice,^28,35,36  6 cm H₂O in pigs,¹⁰  and 12 cm H₂O in humans.³⁷  These values suggest that human lungs are more resistant to injurious ventilation than those of mice or pig. This analysis, however, cannot explain the survival differences between pigs¹⁰  and mice, which may be explained by the more fragile alveolar structures in mice or alternatively by the breathing frequency that was six times higher in mice: 90 versus 15 breaths/min; thus, after 7 h of ventilation, our mice had experienced the same number of cycles as the pigs had after 42 h.¹⁰  Notably, respiratory rate exponentially increases the mechanical power delivered to the lungs during mechanical ventilation.³³ 

In summary, we believe that material fatigue of stress-bearing elements in the lungs due to cyclic load is the most likely explanation for our findings. However, at present, we cannot rule out other explanations such as the repeated opening and closure of alveolar units that are also associated with high rates of stress.³⁸ 

Another alternative explanation for the massive lung damage would be biotrauma, i.e., the destruction of the lungs by inflammation. In fact, the current series of experiments was initiated in order to establish a model of biotrauma with ventilation as the only hit. The experiments with dexamethasone, a well-known antiinflammatory drug, were designed to examine this hypothesis. Despite being able to reduce most of the inflammatory responses, dexamethasone treatment had little effect on physiologic parameters (p_plat, C_dyn, petCO₂, wet/dry ratio, and Pao₂/Fio₂), histopathologic scoring, or on survival proportions. The fact that infiltrating neutrophil numbers were not affected by the pretreatment with dexamethasone can be explained by the presence of hemorrhage (fig. 9). The results of the dexamethasone experiments suggest that inflammation was a result and not the cause of the lung failure. However, application of dexamethasone is only one method to reduce inflammation, and we cannot exclude the possibility that other methods may be more effective although the random-type pattern of lung failure in the present study is rather different from the gradual and well-timed lung failure that is known to be induced by inflammation, e.g., in acid-induced lung injury.³⁹  We would also like to acknowledge that the mouse lung is too small for gravitational effects to come into play, while in large animal one-hit models, severe regional inflammation can be generated after barrier breakdown.

Our findings also demonstrate the necessity to continuously monitor cardiovascular and pulmonary functions. Without monitoring blood pressure, we might have ventilated dead animals, and without monitoring lung functions, the interpretation of the lung failure would have been difficult. In addition, the random occurrence of lung failure mandates sufficiently long ventilation times. A thought experiment might illustrate the problems that can arise from curtailed observation times. If we had stopped the experiments in series_2 only 34 min earlier (after 386 min instead of 420 min), the percentage of animals surviving would have increased from 50% (table S3, Supplemental Digital Content 9, https://links.lww.com/ALN/B400) to 100% (table S4, Supplemental Digital Content 10, https://links.lww.com/ALN/B401) and dexamethasone would have had a significant effect on survival (P = 0.04; fig. S6B, Supplemental Digital Content 11, https://links.lww.com/ALN/B402, showing the Kaplan-Meier plots with different censor times) and lung compliance (P = 0.03; fig. S7B, Supplemental Digital Content 12, https://links.lww.com/ALN/B403, showing C_dyn with different censor times). These results again indicate that inflammation may accelerate an impending mechanical lung failure. Notably, in 38 murine VILI studies reviewed by us,⁹  the mean duration was 3.6 ± 1.8 h and we wonder how many treatments had remained effective had the animals been ventilated longer.

Our findings demonstrate a relatively sharp distinction between ventilation with 24 cm H₂O that was well tolerated and ventilation with 27 cm H₂O—only 3 cm H₂O more—where nearly all animals died because of fatal lung failure. With the caveat that 7 h may have been too short to see lung injury in those animals ventilated with 24 cm H₂O, it seems possible that there may be a threshold for the occurrence of mechanical lung injury, a suggestion that was also made in pig studies that lasted for 3 days.¹⁰  Such a threshold would be reached when brittle stress-bearing elements are exposed to high degrees of cyclic load. As a result, the onset of lung failure occurs at random time points. Our findings also suggest—at least in mice—that in one-hit models of VILI, inflammation accelerates the inevitable lung failure and that the relationship between ventilation and inflammation in ARDS patients may better be studied in multiple hit models.

The authors gratefully acknowledge the invaluable technical support by Nadine Ruske and Birgit Feulner, Institute of Pharmacology and Toxicology, Rheinisch-Westfälische Technische Hochschule (RWTH) Aachen University, Aachen, Germany.

Supported by the Interdisciplinary Center for Clinical Research Aachen (IZKF Aachen), Faculty of Medicine, Rheinisch-Westfälische Technische Hochschule (RWTH) Aachen University, Aachen, Germany.

The authors declare no competing interests.