Systems biology approaches to investigate mechanisms of obstructive lung diseases

Abstract

Both asthma and chronic obstructive pulmonary disease (COPD) are obstructive lung diseases with a large impact on global health, causing 400 000 and 3 million deaths respectively each year. The numbers may be underestimated, since COPD often contributes to death without being registered as the cause of death. There is currently no therapeutic cure for either of these diseases, only symptomatic relief. One problem is that the available therapeutics do not always work, since the diseases present a range of clinical phenotypes with possibly different endotypes, so-called umbrella diseases.

The aim of this thesis was to study asthma, COPD caused by smoking, and obstructive lung disease related to preterm birth using systems biology approaches. This includes studying several analytical platforms in a range of compartments in order to subphenotype the patients into subgroups and elucidate the related mechanisms. Identifying these endotypes increases the possibility of finding effective therapeutics for all patient groups.

Obstructive lung diseases are often studied by collecting bronchoalveolar lavage (BAL) and epithelial cells from the lungs during bronchoscopy. This procedure is invasive, and costly which is why the cohorts studied are often small. Subgrouping results in even smaller sample sizes, which decreases the power of statistical analysis.

Using multivariate analysis is a means of increasing power by taking all variables into account. A workflow for performing the multivariate method orthogonal projections to latent structures discriminant analysis (OPLS-DA) to compare groups one by one in small sample sizes was developed using the programming language R, and was formatted into an R package entitled roplspvs. The roplspvs package performs OPLS modeling using the package ropls in R, including variable selection to extract the variables driving the separation the most. As OPLS models are prone to overfitting, the significance of the models was investigated thoroughly using permutations. Using roplspvs on small sample sizes, it was shown that permutations performed before variable selection (termed “sans v.s.”) and permutations including the variable selection step (termed “over v.s.”) are better suited to estimate the level of model statistics achieved by random than permutations post variable selection. An example of running the package was shown using a publicly available metabolomics dataset.

The roplspvs packages, along with the commercially available software SIMCA and univariate statistics, were then applied to investigate alterations between groups in a range of projects, including three clinical cohorts of asthma, COPD and BPD, as well as a project investigating the degradation of proteins in the processing of blood samples prior to biobanking the evaluate protein stability.

COPD in smokers and ex-smokers was studied by investigating the miRNA content of small extracellular vesicles (EVs) using OPLS modeling as well as univariate analysis, with the finding that COPD gave highly altered miRNA content of small EVs compared to healthy subjects. After stratifying the analysis by gender, potential alterations compared to smokers were identified in males with significant p[CV-ANOVA]=0.05 and p[permutations over variable selection]=0.12, but permutations sans variable selection were highly insignificant. The alterations were connected to potentially affected pathways through pathway analysis of genes regulated by the altered miRNA. Pathway affected by COPD and smoking were mainly connected to cell- growth and death with the p53-pathway mostly altered, while the less pronounced miRNA alterations related to COPD alone was connected to degradation through autophagy and proteolysis.

Premature birth has been connected to lung obstruction in adults who developed bronchopulmonary disease (BPD) during the neonatal period. To characterize T-cells in adults with a history of BPD, FACS analysis was performed on BAL cells. Univariate analysis showed increased levels of CD8+ T-cells, and decreased levels of CD4+ T-cells in subjects with BPD. Applying OPLS and stratifying the analysis by gender, it was indicated that the alterations were mostly driven by females.

Asthmatic subjects were subphenotyped into clusters using four platforms from blood and urine into phenotypic groups, which were studied using OPLS models to compare the groups. Clinical features were extracted that separated a large portion of the groups.

Finally, the metabolome of urine was used to separate asthmatics into severe and mild asthma, stratifying the analysis by oral corticosteroids (OCS). It was found that carnitines, which were the strongest drivers for separating the groups, were not affected by OCS use. Using roplspvs, it was shown that the levels of carnitines were strongly affected by gender, with higher levels in males than in females.

In conclusion, it was shown that OPLS models can be used to investigate cohorts consisting of small sample sizes, and that permutation procedures including variable selection efficiently test the significance of the models. Subgroups of COPD and asthmatic subjects were compared, showing alterations in miRNA levels, metabolome, and lymphocyte composition, as well as in clinical data connected to potential endotypes with separate disease mechanisms. Stratifications by gender supported earlier findings that gender has a strong effect on obstructive lung diseases. Together with further analysis on the cohorts in this study using other platforms, this is a step towards finding candidates for diagnostics and therapeutics.