Integrating Pulmonary and Systemic Transcriptomic Profiles to Characterize Lung Injury after Pediatric Hematopoietic Stem Cell Transplant

Abstract

Hematopoietic stem cell transplantation (HCT) is potentially curative for numerous malignant and non-malignant diseases but can lead to lung injury due to chemoradiation toxicity, infection, and immune dysregulation. Bronchoalveolar lavage (BAL) is the most commonly used procedure for diagnostic sampling of the lung but is invasive, cannot be performed in medically fragile patients, and is challenging to perform serially. We previously showed that BAL transcriptomes representing pulmonary inflammation and cellular injury can phenotype post-HCT lung injury and predict mortality outcomes. However, whether peripheral blood testing is a suitable minimally-invasive surrogate for pulmonary sampling in the HCT population remains unknown. To address this question, we compared 210 paired BAL and peripheral blood transcriptomes obtained from 166 pediatric HCT patients at 27 children's hospitals. BAL and blood mRNA abundance showed minimal overall correlation at the level of individual genes, gene set enrichment scores, imputed cell fractions, and T- and B-cell receptor clonotypes. Instead, we identified significant site-specific transcriptional programs. In BAL, expression of innate and adaptive immune pathways was tightly co-regulated with expression of epithelial mesenchymal transition and hypoxia pathways, and these signatures were associated with mortality. In contrast, in blood, expression of endothelial injury, DNA repair, and cellular metabolism pathways was associated with mortality. Integration of paired BAL and blood transcriptomes dichotomized patients into two groups, of which one group showed twice the rate of hypoxia and significantly worse outcomes within 7 days of enrollment. These findings reveal a compartmentalized injury response, where BAL and peripheral blood transcriptomes provide distinct but complementary insights into local and systemic mechanisms of post-HCT lung injury.

Figure 1.

(A) Geographic location of participating children’s hospitals. (B) Patients were followed from the time of conditioning chemotherapy for the development of pulmonary symptoms. If bronchoscopy with bronchoalveolar lavage was planned for clinical reasons, patients were enrolled and BAL with a paired blood sample was collected. Patients were then followed clinically through hospital discharge.

Figure 2.

(A) Genes differentially expressed in BAL vs peripheral blood are shown. (B) Expression levels of select genes specific to lung (SFTPC, MUC5AC) and blood (HBA2, HBB). (C) Gene set enrichment scores to the 50 MSigDB Hallmark Pathways were calculated and correlation between expression of each gene set was calculated within each body site and then contrasted to identify site-specific coregulation. Here we show unique co-expression of Hallmark Hypoxia and IFNg gene sets in BAL but not blood, and unique co-expression of Hallmark DNA Repair and E2F targets in blood but not BAL. (D) Correlation of MSigDB Hallmark pathways within blood (top right triangle) and within BAL (bottom left triangle) are shown. See Supplemental Figure __ for detailed labels. (E) BAL-blood correlation was calculated for expression levels of n=7,169 protein-coding genes and the distribution of correlation coefficients is plotted. Expression levels of CXCL8 in BAL and blood are shown as an example of minimal correlation. Gene set enrichment scores for the MSigDB Hallmark Inflammatory Response gene set measured in BAL and blood are also shown as an example of minimal BAL-blood correlation. BAL and peripheral blood cell type fractions were imputed using CIBERSORTx and reference atlases, and cell fractions across body sites were correlated using Spearman correlation coefficients, with neutrophils and CD8+ T-cells shown as examples. T- and B-cell receptor clonotypes were measured using Imrep and the number of unique clonotypes across body sites were correlated using Spearman correlation coefficients, with TRA shown as an example.

Figure 3.

(A) BAL gene expression differences in non-survivors. (B) Peripheral blood gene expression differences in non-survivors. (C) Overlap between BAL and blood genes associated with mortality. (D) Network of genes co-expressed in BAL of non-survivors (right) but not co-expressed in survivors (left). Examples of hubs genes (CEACAM6, CXCL17, NFAM1) are shown. Examples of differentially co-expressed genes linked to each hub gene are shown (e.g. CEACAM6-FN1 co-expression) to illustrate differential gene-expression. To the right, pathway enrichment for hub genes are shown. (E) Network of genes co-expressed in peripheral blood of non-survivors (right) but not co-expressed in survivors (left). Examples of hub genes (ZNF707, GARRE1, TMEM86B) are shown. Examples of differentially co-expressed genes linked to each hub gene are shown (e.g. ZNF707-BUB1B co-expression) to illustrate differential gene-expression. To the right, pathway enrichment for hub genes are shown.

Figure 4.

(A) Concept diagram for four post-HCT lung injury subtypes derived in the PTCTC SUP1601 cohort and validated in the University of Utrecht, the Netherlands cohort (Nature Med 2024). (B) Example BAL and blood genes differentially expressed in lung injury subtypes 2, 3, and 4 relative to subtype 1. (C) Differentially expressed BAL and blood genes underwent pathway analysis and pathways identified in BAL and blood gene are quantified to show greater overall differences detected in BAL as opposed to paired blood.

Figure 5.

(A) Paired BAL and blood transcriptomes underwent multi-omics factor analysis (MOFA) followed by dimensionality reduction (umap) and k-means clustering to show 2 groups of patients. (B) Post-HCT lung injury subtype was mapped onto the two integrated transcriptome clusters, showing that most patients from subtypes 2, 3, and 4 mapped to Cluster B. (C) Approximately twice as many patients in Cluster B required oxygen prior to BAL sampling, and twice as many patients died or required ongoing mechanical ventilation within 7 days of BAL sampling.