Diagnosing lung transplant rejection currently depends on histologic assessment of transbronchial biopsies (TBB) with limited reproducibility and considerable risk of complications. Mucosal biopsies are safer but not histologically interpretable. Microarray-based diagnostic systems for TBBs and other transplants suggest such systems could assess mucosal biopsies as well. We studied 243 mucosal biopsies from the third bronchial bifurcation (3BMBs) collected from seven centers and classified them using unsupervised machine learning algorithms. Using the expression of a set of rejection-associated transcripts annotated in kidneys and validated in hearts and lung transplant TBBs, the algorithms identified and scored major rejection and injury-related phenotypes in 3BMBs without need for labeled training data. No rejection or injury, rejection, late inflammation, and recent injury phenotypes were thus scored in new 3BMBs. The rejection phenotype correlated with IFNG-inducible transcripts, the hallmarks of rejection. Progressive atrophy-related changes reflected by the late inflammation phenotype in 3BMBs suggest widespread time-dependent airway deterioration, which was especially pronounced after two years posttransplant. Thus molecular assessment of 3BMBs can detect rejection in a previously unusable biopsy format with potential utility in patients with severe lung dysfunction where TBB is not possible and provide unique insights into airway deterioration. ClinicalTrials.gov NCT02812290.
- basic (laboratory) research/science
- lung transplantation/pulmonology
- microarray/gene array