Abstract
In this methodology-oriented study, we show that methylation profiling can be easily integrated into sputum analysis workflows and exhibits a strong potential to contribute to the profiling and understanding of pulmonary inflammation. Wherever RNA degradation is of concern, in silico correction can be effective in improving both sensitivity and specificity of downstream analyses. We suggest that deconvolution methods should be integrated in sputum omics analysis workflows whenever possible in order to facilitate the unbiased discovery and interpretation of molecular patterns of inflammation.
Introduction
Induced sputum is a vital "liquid biopsy" for studying lung inflammation, but it presents two major hurdles for molecular profiling. First, cellular heterogeneity creates a complex mixture of macrophages, neutrophils, eosinophils, and epithelial cells. Because cell proportions vary drastically between patients, bulk omics analysis often reflects shifts in cell counts rather than changes in cellular behavior. Second, RNA degradation in biobanked samples introduces significant bias, leading to false results in transcriptomic studies. This paper introduces "Sputum Omics 2.0," combining whole-genome methylation analysis with mathematical deconvolution to overcome these barriers.
Methodology
The researchers analyzed sputum from 29 subjects across two German prospective cohorts: ALLIANCE (9 asthmatic, 10 healthy controls) and COSYCONET (10 stable COPD samples). Molecular profiling utilized Agilent microarrays for the transcriptome and Illumina 450k BeadChips for the methylome.
To address the technical challenges of sputum analysis, the study employed two primary computational techniques:
- In Silico RNA Correction: A linear regression model based on RNA Integrity Numbers (RIN) was used to correct for degradation bias, allowing the retention of samples that would typically be discarded.
- Quadratic Programming (QP) Deconvolution: Using manual differential cell counts as a "ground truth," the authors applied a QP-based regression to mathematically estimate the molecular profiles of specific cell types (e.g., macrophages) from bulk data without requiring physical cell sorting.
Key Results
Methylation as a Robust Classifier
Unsupervised clustering of methylation data separated asthmatics from controls more cleanly than transcriptomic data. This suggests that the methylome may provide a more stable and distinct signature of chronic respiratory disease than the transcriptome.
Impact of RNA Correction and Deconvolution
Correcting for RNA degradation improved analytical sensitivity, increasing the number of detectable differentially expressed genes (DEGs) in asthma from 64 to 86. The impact of deconvolution was even more significant; while bulk analysis found only 86 DEGs in asthma, deconvolution uncovered 155 DEGs in macrophages and 552 DEGs in neutrophils.
Unmasking Cell-Specific Signatures
Deconvolution revealed that the SMAD3 gene (a known asthma risk factor) and various HLA (Human Leukocyte Antigen) regions were specifically differentially methylated in the macrophages of asthma and COPD patients. These insights were previously masked in bulk samples by the high proportion of eosinophils.
Discussion and Conclusion
While limited by a small sample size (n=29), this study demonstrates the power of integrating deconvolution and methylome profiling into respiratory research. By mathematically "untangling" cellular heterogeneity, researchers can identify cell-type-specific molecular signatures that are otherwise lost. The authors suggest that these methods should be integrated into all future sputum omics workflows and could be applied to other complex samples, such as bronchoalveolar lavage (BAL) fluid, to better define disease endotypes.
Summary updated