Methodological issues in airway microbiome studies

Abstract

Background Studies on the lung microbiome face unique methodological challenges tied to the low bacterial load of acquired samples and the increased susceptibility to bacterial DNA contamination. Contamination may be introduced from i) the upper airways during sampling and ii) reagents, kits and the general laboratory environment during laboratory processing steps. Few publications exist on validity and reliability of applied methods of sampling, laboratory processing and bioinformatics analysis. Objectives The objective of the thesis was to address some of the methodological issues that remain unresolved in the field of lung microbiome research. In the first paper, we sought to determine whether protected (via a sterile catheter) bronchoscopic sampling techniques would reduce the influence of bronchoscopic carryover from the upper airways. In paper II, we examine the impact of laboratory contamination on airway samples and explore the expected inverse relationship between sample bacterial load and influence of contamination. We also compare different bioinformatic strategies to dealing with contamination. In paper III, we sought to determine whether processing samples through longer laboratory workflows would increase susceptibility to contamination, and to explore impact of choice of 16S rRNA gene variable region (V3 V4 or V4) on the presentation of the airways microbiome. Methods Study samples were collected from participants enrolled in the Bergen COPD Microbiome study (short name «MicroCOPD»). Samples included oral washes (OW), bronchoscopically acquired protected specimen brushes (PSB), protected bronchoalveolar lavages (PBAL), small-volume lavages (SVL) and negative control samples (NCS) consisting of PBS used for collection of all samples. Bacterial DNA was extracted using a combination of enzymatic and mechanical lysis methods and processing through the FastDNA Spin Kit (MP Biomedicals). Bacterial community composition was determined by high-throughput sequencing of the bacterial 16S rRNA gene using the Illumina MiSeq sequencing platform. Three library preparation setups were included in the thesis, varying in number of PCR steps (1- or 2-steps) and target marker gene region (16S rRNA gene V3 V4 or V4): Setup 1 (2-step PCR; V3 V4 region); Setup 2 (2-step PCR; V4 region); Setup 3 (1-step PCR; V4 region). Papers I and II were based on setup 1. Paper III included all three setups. Bacterial load was determined by quantitative PCR targeting the 16S rRNA gene region V1 V2 (paper II). Bioinformatics processing steps were performed using the Quantitative Insights Into Microbial Ecology (QIIME) bioinformatic package, versions 1 (papers I and II) and 2 (paper III). Strategies for decontamination varied across papers and included i) keeping samples intact (i.e. do nothing), ii) removing all sequences observed in NCS and iii) the removal of sequences identified as contaminants using the Decontam R package tools. In paper I, sequences observed in NCS were removed. In paper II, all three strategies were applied and compared. In paper III, Decontam was used. Results Analyses for paper I were based on the underlying assumption that the more similar the bronchoscopically acquired specimens (PSB, PBAL and SVL) were the OW sample, the greater the influence of upper airway contamination. Between sample comparisons were made based on three parameters: i. taxonomy, ii. alpha diversity and iii. beta diversity. Across all three parameters, similarity to the OW sample decreased in order SVL>PBAL>PSB. In paper II, an estimated 10-50% of the bacterial community profiles for the lower airway samples (PSB, PBAL) were derived from laboratory contamination. This was determined based on comparison to a dilution series of known bacterial composition and load. The DNA extraction kit was identified as the main contamination source. On comparison of the three decontamination strategies, we found that the Decontam R package provided a balance between keeping and removing sequences found in both NCS and study samples. In paper III, we found that the number of sequences and ASVs decreased in order setup1>setup2>setup3. This appeared to be associated with increased taxonomic resolution when targeting the V3 V4 region (setup 1) and an increased number of small ASVs in setups 1 and 2. For setups 1 and 2, we interpreted this as a result of contamination in the 2-step PCR protocol and sequencing across multiple runs (setup 1). Analyses of taxonomic composition revealed that genera Streptococcus, Prevotella, Veillonella and Rothia dominated all setups, but that relative abundances differed. Analyses of beta diversity revealed that while OW samples clustered together regardless of number of PCR steps, samples from the lower airways (PSB, PBAL) separated. Removal of contaminants identified in Decontam did not resolve differences across setups. Conclusions We show that protected bronchoscopic sampling techniques (PSB, PBAL) may provide protection from oropharyngeal carryover and should be the preferred sampling technique in future studies (paper I). We demonstrate that bacterial load will vary across airway sample types and that bacterial contamination from the laboratory will have an increased impact on samples of lower bacterial load (paper II). We recommend that estimates of contamination are reported in all studies. We also recommend the use of contaminant identification tools based on statistical models that limit subjectivity (e.g. Decontam). Finally, we demonstrate that differences in number of PCR steps (1- or 2-steps) will have an impact on final bacterial community descriptions, and more so for samples of low bacterial load (e.g. lower airway samples) (paper III). Our findings could not be explained by differences in contamination levels alone, and more research is needed to understand the underlying mechanisms contributing to the observed protocol bias.