The research, led by Hussein Anani, Grégory Destras and colleagues working across the Hospices Civils de Lyon and Nantes Université in France, analysed 184 endotracheal aspirates collected from 94 intubated critically ill patients enrolled in a registered clinical trial. Endotracheal aspirates, samples drawn through the breathing tube that delivers mechanical ventilation, provide a direct window into the microbial community colonising the lower airways. Rather than simply cataloguing which microbes were present, the team combined viral metagenomics with metatranscriptomics, sequencing not only the genetic material of the lung microbiome but also measuring which genes were actively being transcribed at the time of sampling.

This functional approach matters because the presence of a microbe does not necessarily indicate that it is doing anything. In the lungs of ventilated patients, where colonisation and infection can coexist in complicated ways, transcriptional activity offers a more dynamic picture of the microbial ecosystem. By examining both the bacterial community and the viral community together, and by measuring their gene expression, the researchers were able to define what they call functional pneumotypes: reproducible configurations of the lung microbiome that carry distinct clinical consequences.

The high-risk pneumotype they identified showed a striking signature. It was enriched in virulent bacteriophages associated with Streptococcus and in temperate bacteriophages associated with Klebsiella. Bacteriophages, the viruses that infect bacteria, come in two broad lifestyles. Virulent phages replicate aggressively and lyse their bacterial hosts, while temperate phages can integrate into the bacterial genome and lie dormant, sometimes carrying genes that increase bacterial virulence. An overrepresentation of these phage groups, alongside reduced transcriptional activity of commensal taxa such as Streptococcus and Alloprevotella and increased activity of Klebsiella pneumoniae, marked the unfavourable state of the airway ecosystem.

Crucially, this unfavourable pneumotype was detectable before patients developed hospital-acquired pneumonia, one of the most common and dangerous complications of intensive care. Hospital-acquired pneumonia arises when the microbial balance of the lower airways tips toward pathogenic overgrowth, and it is a major driver of prolonged ventilation, sepsis and death. The observation that the phage-rich, inflammation-linked configuration precedes clinical deterioration raises the possibility that shifts in the bacterial-viral ecosystem are not merely a byproduct of infection but may participate in driving it.

To translate these microbial patterns into something clinically usable, the team built machine-learning models capable of predicting which patients carried the unfavourable pneumotype. A signature based on two viral factors achieved an area under the receiver operating characteristic curve, or AUC, of 0.8, while a signature based on four bacterial factors reached an AUC of 0.7. These values indicate discriminative performance that is meaningful but not yet definitive, typical of biomarkers in early development. The two-viral-factor signature in particular suggests that a small number of phage measurements could eventually be incorporated into risk stratification for ventilated patients.

Importantly, the associations were not confined to a single cohort. The researchers validated their findings in an independent group of 117 intensive care patients from a separate clinical trial, analysing 239 additional endotracheal aspirates. In this validation cohort, the bacterial and viral signatures retained their robust association with the risk of all-cause mortality, strengthening the case that the pneumotypes reflect a genuine biological phenomenon rather than an artefact of one particular patient population or sampling campaign.

The study went beyond correlation by applying causal inference methods to the combined microbial and host transcriptomic data. This analysis identified virulent bacteriophages associated with Streptococcus, together with host genes including IL18R1, ADM and PTX3, as key regulators of severe lung dysbiosis linked to mortality. Each of these host genes is biologically plausible in this context. IL18R1 encodes a receptor involved in interleukin-18 signalling, a pathway central to innate immune responses and inflammation. ADM, which encodes adrenomedullin, is a peptide with roles in vascular regulation and inflammatory responses that is often upregulated in critical illness. PTX3, pentraxin 3, is a well-established marker of local inflammation produced at sites of infection. Their co-identification with virulent phages as candidate regulators paints a picture in which phage activity, bacterial community composition and the host inflammatory response form an interconnected network shaping patient outcomes.

The technical achievement underlying these results should not be understated. Lung secretions from ventilated patients are notoriously difficult samples, dominated by host DNA and RNA and complicated by antibiotic exposure, mechanical forces and repeated colonisation events. Recovering viral metagenomic information from endotracheal aspirates requires careful depletion of host material and sensitive detection of phage genomes, while metatranscriptomics demands preservation and measurement of microbial RNA in a noisy clinical environment. Applying both approaches across hundreds of samples from two independent trials, and linking the results to hard clinical endpoints such as mortality, represents a substantial methodological step for the field of respiratory microbiome research.

The implications for intensive care medicine are considerable. Current approaches to hospital-acquired pneumonia rely largely on clinical suspicion, radiographic findings and cultures, all of which lag behind the underlying microbial events. A pneumotype-based classifier, particularly one built from a small number of viral and bacterial markers, could in principle allow clinicians to identify patients drifting toward dysbiosis before pneumonia develops, opening a window for preventive interventions. The study also raises intriguing therapeutic questions: if virulent phages contribute to severe dysbiosis, then modulating the phage community, rather than only the bacteria, might one day become a strategy for protecting the lungs of critically ill patients. The authors note that the work was funded through the European Union’s Horizon 2020 programme and the MSD Avenir foundation, and the researchers have filed a patent on respiratory microbiome composition in intensive care patients, signalling their intent to move these signatures toward clinical application. For now, the study stands as a demonstration that the lung virome, long an overlooked dimension of the respiratory microbiome, carries information that could help predict survival in the intensive care unit.

Subject of Research: Bacterial-viral lung microbiome signatures associated with inflammation and mortality in critically ill patients

Article Title: Identification of functional bacterial-viral pneumotypes associated with airway inflammation and all-cause mortality in critically ill patients

Article References: Anani, H., Destras, G., Semanas, Q., Martin, F. P., Petrier, M., Oddoux, E., Burfin, G., Bulteau, S., Poulain, C., Sinha, D., Motos, A., Gourain, V., Cremet, L., Montassier, E., Bressollette-Bodin, C., Roquilly, A., & Josset, L. (2026). Identification of functional bacterial-viral pneumotypes associated with airway inflammation and all-cause mortality in critically ill patients. Nature Communications. https://doi.org/10.1038/s41467-026-78146-z

Image Credits: AI Generated

DOI: 10.1038/s41467-026-78146-z

Keywords: lung microbiome, bacteriophages, pneumotypes, hospital-acquired pneumonia, intensive care, metatranscriptomics, viral metagenomics, Klebsiella pneumoniae, Streptococcus, airway inflammation, mortality prediction, machine learning