Comparison of DNA extraction methods for microbial community profiling with an application to pediatric bronchoalveolar lavage samples

Abstract

Barcoded amplicon sequencing is rapidly becoming a standard method for profiling microbial communities, including the human respiratory microbiome. While this approach has less bias than standard cultivation, several steps can introduce variation including the type of DNA extraction method used. Here we assessed five different extraction methods on pediatric bronchoalveolar lavage (BAL) samples and a mock community comprised of nine bacterial genera to determine method reproducibility and detection limits for these typically low complexity communities. Additionally, using the mock community, we were able to evaluate contamination and select a relative abundance cut-off threshold based on the geometric distribution that optimizes the trade off between detecting bona fide operational taxonomic units and filtering out spurious ones. Using this threshold, the majority of genera in the mock community were predictably detected by all extraction methods including the hard-to-lyse Gram-positive genus Staphylococcus. Differences between extraction methods were significantly greater than between technical replicates for both the mock community and BAL samples emphasizing the importance of using a standardized methodology for microbiome studies. However, regardless of method used, individual patients retained unique diagnostic profiles. Furthermore, despite being stored as raw frozen samples for over five years, community profiles from BAL samples were consistent with historical culturing results. The culture-independent profiling of these samples also identified a number of anaerobic genera that are gaining acceptance as being part of the respiratory microbiome. This study should help guide researchers to formulate sampling, extraction and analysis strategies for respiratory and other human microbiome samples.

Figure 1. Microbial community profiles for the mock community. 16 S libraries were normalized to 900 sequences and 97% OTUs were consolidated at the genus level.

The nine genera comprising the mock community are marked in black italics, while the starred genera in grey italics correspond to contaminants.

Figure 2. Examination of contaminants in the mock community.

(A) Relationship between DNA yield and percent of contaminating genera in the mock community. The equation for a power law regression with coefficient of determination are presented in the inset. (B) Relative abundances of known mock community and spurious (contaminating) genera in mock community profiles. Asterisks indicate data points which represent more than one genus.

Figure 3. Average weighted Unifrac distances with standard error.

Distances for the mock community are presented in (A) and for BAL samples in (B). Significant differences were evaluated using non-parametric exact Mann-Whitney U tests.

Figure 4. Principal components analysis based on weighted Unifrac distances for BAL samples and mock community extracted using five different extraction methods.

CF samples processed with DTT (Sputasol) are not included.

Figure 5. Microbial community profiles for BAL samples.

16 S libraries were normalized to 400 sequences and 97% OTUs were consolidated at the genus level. Red boxes indicate genera previously cultured during routine microbiology. Samples processed with DTT (Sputasol) are labeled in blue. Community profiles including all sequences are presented in (A), and profiles excluding sequences at less than 0.6% relative abundance are presented in (B).