View Abstract

A5967 - Understanding Heterogeneity in Severe Asthmatics Using High Dimensional Analysis of Cellular Phenotypes
Author Block: M. Camiolo1, X. Zhou2, T. B. Oriss3, K. C. Nadeau4, S. E. Wenzel5, A. Ray6; 1Pulmonary Medicine, UPMC, Pittsburgh, PA, United States, 2Department of Medicine, Stanford University, Stanford, CA, United States, 3Univ of Pittsburgh, Pittsburgh, PA, United States, 4Stanford Univ, Palo Alto, CA, United States, 5Professor of Med PACCM, Univ of Pittsburgh Med Ctr, Pittsburgh, PA, United States, 6Pulm Allergy Crit Care Med, Univ of Pittsburgh Sch of Med, Pittsburgh, PA, United States.
RATIONALE: Asthma is a common disease, affecting more than 300 million people worldwide. Though well controlled in most, a subset of patients experience disease that is refractory to treatment and accounts for nearly half the health care expenditure on asthma in the United States. Recent efforts focused on understanding severe asthma pathophysiology suggest ontological heterogeneity. Most molecular phenotyping studies to date have relied on transcriptional profiling. The emergence of time of flight mass cytometry (CyTOF) offers an opportunity to characterize the inflammatory milieu present in bronchoalveolar lavage (BAL) cells at the protein level in unprecedented detail.
METHODS: Healthy controls (HC) as well patients with mild to moderate (MMA) and severe asthma (SA) as defined by ERS/ATS guidelines underwent bronchoscopy. BAL cells were stained with heavy metal conjugated antibodies directed against a panel of intracellular and extracellular proteins. Acquisition was performed on the Helios CyTOF instrument. Intact singlets were gated using Flowjo and clustering analysis performed using Phenograph. Metadata was used to inform principal component analysis (PCA). K-means clustering was then based off high dimension cell count.
RESULTS: Study of 23 samples from 3 HC, 10 MMA and 10 SA patients revealed 26 clusters based on staining of 33 surface markers. Patient level analysis using PCA and k-means clustering revealed 3 subsets of asthmatics. While the majority of MMA cases were similar in cell composition to HCs, PCA revealed multiple divergent subgroups within the known umbrella of clinically severe disease. These subgroups of 4 and 6 patients captured 80% of severe asthmatics and illustrated possible heterogeneity in cytokine signaling. They could be broadly defined as T-cell rich and T-cell poor and showed wide disparity in staining for IFN-ɣ, IL17 and type 2 cytokines. Divergence in clinical parameters such as lung function and T2 biomarkers was also evident. Machine learning classification of BAL samples proved superior to clinical disease severity assessment in predicting covariance of cell count and cytokine intensity. Correlation analysis between cell clusters and cytokine intensity across all samples supported the relationships illustrated by patient level grouping analysis.
CONCLUSIONS: Using machine learning algorithms to evaluate high-dimensional data generated from CyTOF analysis of lavage samples, we identified and characterized subgroups within clinically severe asthma. Understanding the differences between these patients in a more granular way may help guide decisions regarding precision medicine. Ongoing studies are focused on integration of clinical characteristics of the severe asthmatics with their BAL cell immune profile.