A presumably overly robust inflammatory response has been associated with poor clinical outcomes in patients with acute respiratory failure, including patients with acute respiratory distress syndrome (ARDS) and sepsis (1). Likewise, both abnormal gut and respiratory microbiota patterns (termed “dysbiosis”) are also predictive of increased mortality among critically ill patients (2). The ambitious aim of the study by Kitsios and colleagues (pp. 1666–1677) in this issue of the Journal is to better define the interplay between the host inflammatory response and the lung microbiome and the impact of this relationship on clinical outcomes in a heterogenous population of critically ill patients with acute respiratory failure (3). The results of this investigation represent an important step in the process of developing a microbiome-guided or microbiome-based treatment for critically ill patients with acute respiratory failure. The cohort characteristics in the study by Kitsios and colleagues were typical of an ICU population of patients with acute respiratory failure requiring mechanical ventilation: extrapulmonary sepsis (18%), ARDS (24%), and pneumonia (40%) were common diagnoses, and 32% of the patients received antibiotics before admission to the ICU. At the time of enrollment (, 72 h postintubation), posterior oropharyngeal swab and endotracheal aspirate (ETA) samples were collected and analyzed with 16s ribosomal RNA gene sequencing to characterize the microbiota of these respective environments. Simultaneously, the following plasma inflammation–related biomarkers were measured: receptor of advance glycation end-products, soluble tumor necrosis factor receptor 1, IL-10 fractalkine, and angiopoietin 2. This biomarker data was used in conjunction with clinical variables to dichotomize the patients into either hyper-(23%) or hypoinflammatory (77%) phenotypes. Similar to prior, albeit smaller, microbiome studies of the critically ill, the upper and lower respiratory microbiota demonstrated reduced a and b diversity when compared with samples from healthy control subjects (4). Nonetheless, there was substantial heterogeneity in bacterial composition across samples from study patients, which was addressed by using Dirichletmultinomial models and Laplace approximation of model fitting to identify distinct microbial clusters among the upper and lower respiratory samples. Common to both the upper and lower respiratory sampling was a particular cluster (“cluster 2”) that was notable for having a lower a diversity and a high abundance of respiratory pathogens. In particular, cluster 2 samples found in lower respiratory samples demonstrated a high abundance of