Inflammatory mechanisms of ventilator-induced lung injury: a time to stop and think?


  • M. R. Wilson,

    Lecturer in Physiology
    1. Section of Anaesthetics, Pain Medicine and Intensive Care, Faculty of Medicine, Imperial College London, Chelsea and Westminster Hospital, London, UK
    Search for more papers by this author
  • M. Takata

    Magill Professor in Anaesthetics, Corresponding author
    • Section of Anaesthetics, Pain Medicine and Intensive Care, Faculty of Medicine, Imperial College London, Chelsea and Westminster Hospital, London, UK
    Search for more papers by this author

  • You can respond to this article at

  • Editorial note: ‘Science Letters’ is an occasional feature in which we offer researchers working in basic science, or in a very specific area of biomedical science, an opportunity to explain their work to a wide clinical anaesthetist readership. We challenge them to outline the nature and clinical implications of their work, and its potential and actual relevance to anaesthesia, in a way that engages and enthuses readers to adapt their own, clinical researches so as to take better account of this science. Researchers willing to take up the challenge are invited to contact the Editor-in-Chief.

Correspondence to: M. Takata



Ventilator-induced lung injury (VILI) is the phenomenon by which mechanical ventilation exacerbates lung injury in critically ill patients. It is particularly relevant for those suffering from acute respiratory distress syndrome, in which the iatrogenic injury caused by VILI contributes to their high mortality. The innate immune system is widely accepted to play an important role during VILI. However, it is our belief that the identification of inflammatory mediators that are crucial during VILI, and thus may make useful therapeutic targets, has become obscured by the wide variety of pre-clinical animal models of VILI reported in the literature. We aim here to summarise some of our work addressing this issue over the last 10 years, and thus, we hope, make interpretation of a convoluted field a little clearer.

Mechanical ventilation is an indispensable tool for the support of critically ill patients, particularly those suffering from acute lung injury/acute respiratory distress syndrome (ALI/ARDS). However, mechanical ventilation can also damage injured lungs, a phenomenon known as ventilator-induced lung injury (VILI), with substantial negative impact on mortality and morbidity in ARDS. Although ‘protective ventilation’ – using a small tidal volume (VT) to decrease lung overstretching – may reduce the impact of VILI [1], it cannot be avoided entirely and mortality from ARDS remains 25–45% [2]. Thus, development of therapies to attenuate VILI is a matter of major clinical importance.

Over the last 10 years, numerous pre-clinical studies have addressed the pathophysiology of VILI, increasingly using mice, primarily due to the potential for genetic modification and similarities between the mouse and human immune systems. Many apparently vital biological mediators have been identified [3, 4] – an impressive achievement of basic science at first glance; however, a possible problem has gradually become apparent: are there now too many mediators vital to VILI? Conceptually, it is difficult even for ‘experts’ to believe that so many pathways and processes can be truly relevant, and deciding which mediators might make useful therapeutic targets, to be tested in clinical trials, is an impossible task. Here, we summarise our 10 years’ experience of laboratory-based VILI research, in an attempt to help a general readership make more informed judgements regarding data derived from rodent models of VILI.

What exactly are we modelling?

There has been much recent discussion surrounding animal models of ALI/ARDS, the consensus being that it has proven difficult to develop a model that fully replicates the human condition [5, 6]. Models should ideally incorporate four central features of the syndrome: histological evidence of injury; barrier disruption; inflammation; and physiological dysfunction [5]. It is worth highlighting that in many studies, animal models (of ARDS in general, as well as VILI) often did not fulfil such criteria, particularly clinically relevant features such as hypoxaemia and respiratory mechanics changes [7]. With regard to VILI, specifically there is an additional consideration: for models to have any translatable value, it is critical that the injury must definitively occur through the process of overstretching the lung. In our opinion, unfortunately this is not always the case.

One of the major factors to be considered with any model of VILI is VT. The findings of the ARDSNet trial made it clear that an inappropriately high VT, leading to overstretching of the lungs, increased patient mortality [1], with ~12−1 more injurious than ~6−1. One approach to modelling is therefore to mimic such clinical scenarios using a VT of ~12–20−1, assuming that this induces enough stretch within mouse lungs. The counter-argument is that within a non-human disease model, using a species with very different respiratory mechanics from humans (mouse lungs are much more compliant than human lungs) [8], imposing an artificial requirement to use ‘clinical’ ventilator settings is pointless; rather, experiments using a very high VT to ensure lung over-distension will be more meaningful.

Both types of model have been used experimentally [9-11]. However, it is important to appreciate that varying VT induces responses that are completely different qualitatively, not just quantitatively. We have recently shown that only very high VT strategies (30–40−1) induced substantial, non-recruitable changes in respiratory mechanics, gas exchange, oedema and inflammation in healthy mice [12]. In contrast, markers of injury produced by ‘clinically comparable’ (for humans) VT strategies (10–20−1) were due not to overstretching of the lungs, but predominantly other factors, e.g. atelectasis, and more disturbingly, various ‘experimental artefacts’ such as surgical stress, prolonged anaesthesia and unphysiological supine positioning for rodents. Although these factors, in particular atelectasis, may also impact on patients, we must reiterate that lung overstretching remains the only variable that has been shown conclusively to affect patient mortality in ARDS [13, 14]. We believe that identification of potential therapeutic targets for VILI must come from models in which stretch can be clearly shown to be the dominant pathophysiology. Such assurance can come from very high VT strategies and/or a careful study design that excludes ‘other’ factors (e.g. by proper detection of substantive differences between injurious high VT and non-injurious low VT).

TNF – new potential in an old target?

Having established whether the model used within a study is what it purports to be, our attention passes to the mediators/pathways identified. While completely novel mediators certainly look exciting, especially if reported in high-impact basic science journals, an important question for clinicians is ‘Is this pathway/mediator a plausible therapeutic target?’ This is not a straightforward question to answer, given the numerous processes reportedly involved in VILI, but we would suggest the following considerations to enhance a target's likely usefulness:

  1. The mediator should be as ‘upstream’ as possible (as VILI has a predictable onset, early mediators can be modulated prophylactically).
  2. The mediator should play a key regulatory role in the pathogenesis of VILI, rather than being part of the general cellular inflammatory/signalling machinery.
  3. The effect of modulating the target mediator/pathway should be impressive, objective and physiologically relevant, e.g. demonstrating substantial effects on hypoxaemia rather than subtle effects on semi-quantitative, subjective morphological findings.

With these in mind, we have spent the last decade exploring the role of the pro-inflammatory cytokine tumour necrosis factor-α (TNF) in VILI, using the aforementioned ‘very high VT’ mouse models. There are clear reasons to support TNF as a potential therapeutic target for VILI (and other aetiologies of ARDS): it is an ‘upstream’ target, rapidly produced in response to stretch [15]; and it regulates many pathophysiological processes linked to VILI, including barrier permeability, alveolar fluid clearance, adhesion molecule up-regulation, leucocyte recruitment/infiltration and apoptosis [16, 17]. On the other hand, a number of studies in (predominantly septic) patients receiving mechanical ventilation has shown anti-TNF treatment to be generally ineffective [18-23]. We suggest that such anti-TNF studies were carried out prematurely, before the biology of TNF was sufficiently understood to give clinical trials any chance of working.

Tumour necrosis factor signals through two cell-surface receptors, p55 (TNFRSF1a) and p75 (TNFRSF1b). Only recently, it has become apparent that these two receptors can stimulate different, in many cases opposing, intracellular pathways [24]. Using our VILI models, we have found that while normal ‘wildtype’ animals developed severe pulmonary oedema and deteriorations in respiratory mechanics and gas exchange over 2 h of ventilation, animals lacking the p55 receptor were completely protected [25]. In contrast, animals lacking both receptors were not protected, while those lacking p75 alone developed more injury. Thus, while TNF signalling through p55 promotes VILI, signalling through p75 appears beneficial. The exact mechanisms behind this are one of the prime focuses for our current work [26, 27], but this knowledge could explain why previous clinical studies using ‘total’ TNF blockers were unsuccessful.

Of course, protection in genetically modified animals is a long way from clinical translation, which requires efficacious, specific, pharmacological means of inhibiting p55. Conventional antibodies, with their two epitope-binding sites, are in general not very effective for receptor blockade, as they tend to stimulate rather than inhibit receptors, due to cross-linking. Recently, however, a ‘domain antibody’ (comprising the smallest functional antigen binding portion of the IgG molecule) has been developed that is specific for p55. Using this, we have shown that acute blockade of p55 signalling within the alveolar space dramatically attenuates physiological deterioration and pulmonary inflammation during VILI in mice [28], suggesting that further clinical studies of TNF-based therapies may be warranted.


Clinical trials of therapies for critically ill patients have proven generally fruitless. In our opinion, this arises partly because of a disconnection between the bench and the bedside, not least a difficulty in interpreting animal models. With regard to VILI, we believe that following some general recommendations may help to address this issue. First, clinicians should not rely on results from animal models that do not clearly show the key physiological features of VILI/ARDS. Second, evaluation of the model should not be based on whether the magnitude of the insult (e.g. VT) seems clinically relevant for humans at first glance, given that what causes stretch in one species may not do so in the other. Finally, the effect of treatment should be physiologically relevant rather than limited to subtle changes, especially in subjective parameters.

It is crucial that we make real efforts from both the bench and the bedside, to bridge the gap towards development of efficacious treatments in critical care. This seems particularly pertinent in the light of ever-increasing pressure on resources and the costs of clinical trials.

Competing interests

No competing interests declared. Our group's research has been funded predominantly by the Wellcome Trust (laboratory infrastructure, animal model work and studies using knockout mice) and partly by GlaxoSmithKline (p55 domain antibody work). GlaxoSmithKline has a financial interest in the use of domain antibodies, including those targeting p55, in the treatment of pulmonary and other diseases.