Hematopoietic cell transplantation (HCT) is a life-saving procedure for various hematologic malignancies and non-malignant conditions. However, it carries a significant risk of pulmonary complications, including lung injury and infections, which can contribute substantially to morbidity and mortality. The complex interplay between the lung microbiome, the host immune response, and the effects of the conditioning regimen (chemotherapy and/or radiation) remains poorly understood. This lack of understanding hinders the development of effective preventative and therapeutic strategies. Previous studies have indicated links between pulmonary microbial dysbiosis and poor outcomes after HCT, but a comprehensive understanding of the diverse pathobiological signatures of lung injury in this context is lacking. This study aims to characterize the complex lung microenvironment in pediatric HCT patients, using integrated microbiome and transcriptome profiling to identify specific microbial and host responses that are associated with fatal lung injury.

The existing literature highlights the high incidence and mortality associated with lung injury following HCT. Previous research has shown that the lungs are not sterile, and the resident microbiome plays a role in both health and disease. However, the composition and function of the pulmonary microbiome in pediatric HCT patients, and its interaction with the host immune system and the impact of therapeutic interventions such as antibiotics remain incompletely defined. Studies have shown associations between various aspects of the microbiome and lung function or other complications, but there is a need for a more detailed understanding of the pathobiological pathways leading to fatal lung injury. This study builds upon previous work by employing a more comprehensive approach to characterize the lung microenvironment and its relationship with clinical outcomes in a large, multi-center cohort.

This prospective, multicenter study enrolled 229 pediatric HCT recipients from 32 children's hospitals across the United States, Canada, and Australia. A total of 278 bronchoalveolar lavage (BAL) samples were collected between 2014 and 2022. BAL fluid underwent RNA sequencing (RNA-seq) to profile both the microbiome and human gene expression. Microbial alignments were transformed from counts to quantitative masses using a reference spike-in, and stringent contamination subtraction was performed. Human alignments were analyzed for gene expression, pathway analysis, cell-type deconvolution, and T and B cell receptor (TCR/BCR) alignments. Unsupervised analysis using multi-omics factor analysis (MOFA) and uniform manifold approximation and projection (UMAP) identified four distinct BAL clusters based on shared microbial-human metatranscriptomic compositions. Clinical data (demographics, medical history, HCT details, antimicrobial exposure, etc.) were correlated with cluster assignments. The identification of pathogens was assessed using both clinical microbiology data and metagenomics sequencing data, establishing a conservative threshold for pathogen identification. Antibiotic exposure scores (AES) were calculated to evaluate the impact of antibiotic therapy on the microbiome. Differential gene expression was analyzed across the clusters. Cell type fractions were imputed using CIBERSORTx and lymphocyte receptor repertoires were assessed using ImReP. Finally, a random forest classifier was trained on the derivation cohort and applied to an independent validation cohort from the Netherlands to confirm the findings and assess their generalizability.

The study identified four distinct BAL clusters characterized by unique combinations of microbial composition, host gene expression, and immune cell profiles. Cluster 1, the most common cluster, exhibited moderate microbial burden, low infection rates, predominantly alveolar macrophage-related signaling, and the lowest mortality. Cluster 2 demonstrated high bacterial burden, increased neutrophils, and moderate mortality. Clusters 3 and 4 showed microbiome depletion, enrichment of viruses and fungi, fibroproliferative gene expression, lymphocytic inflammation, and high mortality rates (Cluster 4 having the highest). Metagenomic sequencing significantly increased the detection of pathogens compared to clinical tests, revealing previously occult infections (viruses, bacteria, fungi, and parasites). Antibiotic exposure was strongly associated with reduced microbial richness and diversity, increased fungal growth (Ascomycota), increased respiratory RNA viruses and was linked to increased mortality. Greater AES was linked to depletion of commensal bacteria (Actinomyces, Fusobacterium, Gemella, Haemophilus, Neisseria, Rothia, Schaalia, and Streptococcus), which mediated the relationship between antibiotic exposure and mortality. Cluster 3 showed signs of fibroproliferation, while Cluster 4 showed markers of cellular injury and lymphocyte activation. The findings were successfully replicated in an independent validation cohort from the Netherlands, demonstrating the robustness and generalizability of the identified BAL clusters and their association with mortality.

This study provides compelling evidence for the crucial role of pulmonary microbial dysbiosis in the pathogenesis of fatal lung injury after pediatric HCT. The identification of four distinct BAL clusters, each with a unique combination of microbial and host factors, highlights the heterogeneity of lung injury in this population. The strong association between antibiotic exposure, commensal depletion, and increased viral and fungal burden underscores the need for careful antibiotic stewardship in this high-risk patient group. The findings suggest that future trials should consider targeted therapies based on the specific pathobiological signatures identified in these clusters, potentially including microbiome restoration strategies.

This large, multi-center study provides novel insights into the complex interplay between the pulmonary microbiome, host immune response, and antibiotic exposure in pediatric patients undergoing HCT. The identification of four distinct BAL clusters provides a framework for future research into personalized diagnostics and treatments tailored to specific lung injury phenotypes. Future studies should investigate the use of microbiome-targeted therapies, along with the development of rapid diagnostic assays to better guide antibiotic use and improve outcomes.

This study has several limitations. The clinical heterogeneity of the cohort requires careful interpretation of the findings. Clinical protocols were not fully standardized, and post-HCT care varied across centers. BAL collection was not standardized across centers. The study lacked controls from healthy children and detailed histopathology, which would have aided the understanding of the microorganisms' contribution to each patient's pulmonary disease. Clinical microbiological testing of BAL varied across hospitals. As an observational study, it cannot definitively establish causal relationships.